A conjunction conspiracy at the West Germanic left periphery | John R ...

March 16, 2018 | Author: Anonymous | Category: Documents
Share Embed

Short Description

Our conclusions differ from Benincà and Poletto (2004: 62), who suggest that RP .... director of.the scuola] è il mio ...


Advances in Comparative Germanic Syntax

Linguistik Aktuell/Linguistics Today (LA) Linguistik Aktuell/Linguistics Today (LA) provides a platform for original monograph studies into synchronic and diachronic linguistics. Studies in LA confront empirical and theoretical problems as these are currently discussed in syntax, semantics, morphology, phonology, and systematic pragmatics with the aim to establish robust empirical generalizations within a universalistic perspective.

General Editors Werner Abraham

University of Vienna / Rijksuniversiteit Groningen

Elly van Gelderen

Arizona State University

Advisory Editorial Board Cedric Boeckx

Christer Platzack

Guglielmo Cinque

Ian Roberts

Günther Grewendorf

Lisa deMena Travis

Liliane Haegeman

Sten Vikner

Hubert Haider

C. Jan-Wouter Zwart

Harvard University University of Venice

J.W. Goethe-University, Frankfurt University of Lille, France University of Salzburg

University of Lund

Cambridge University McGill University

University of Aarhus University of Groningen

Terje Lohndal

University of Maryland

Volume 141 Advances in Comparative Germanic Syntax Edited by Artemis Alexiadou, Jorge Hankamer, Thomas McFadden, Justin Nuger and Florian Schäfer

Advances in Comparative Germanic Syntax

Edited by

Artemis Alexiadou University of Stuttgart

Jorge Hankamer UCSC

Thomas McFadden University of Stuttgart

Justin Nuger UCSC

Florian Schäfer University of Stuttgart

John Benjamins Publishing Company Amsterdam / Philadelphia



The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data Advances in comparative Germanic syntax / edited by Artemis Alexiadou ... [et al.].        p. cm. (Linguistik Aktuell/Linguistics Today, issn 0166-0829 ; v. 141) Includes bibliographical references and index. 1.  Germanic languages--Grammar, Comparative. 2.  Germanic languages--Syntax.  I. Alexiadou, Artemis. PD99.A38    2009 430'.045--dc22 isbn 978 90 272 5524 2 (hb; alk. paper) isbn 978 90 272 8957 5 (eb)


© 2009 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

Table of contents

Advances in Comparative Germanic Syntax Artemis Alexiadou, Jorge Hankamer, Thomas McFadden, Justin Nuger & Florian Schäfer


part i.  Cartography and the left periphery On a (wh-)moved Topic in Italian, compared to Germanic Anna Cardinaletti


C-agreement or something close to it: Some thoughts on the ‘alls-construction’ Michael T. Putnam & Marjo van Koppen


Uncharted territory? Towards a non-cartographic account of Germanic syntax Jan-Wouter Zwart


Bootstrapping verb movement and the clausal architecture of German (and other languages) Gisbert Fanselow


A conjunction conspiracy at the West Germanic left periphery John R. te Velde


part ii.  Word order and movement Reconsidering odd coordination in German Hironobu Kasai The syntax and semantics of the temporal anaphor “then” in Old and Middle English Carola Trips & Eric Fuß



Jespersen’s Cycle and the issue of prosodic ‘weakness’ Patrizia Noel Aziz Hanna


Holmberg’s Generalization: Blocking and push up Hans Broekhuis


 Table of contents

part iii.  Thematic relations and NP realization The No Case Generalization Halldór Ármann Sigurðsson


The new impersonal as a true passive Jóhannes Gísli Jónsson


Anaphoric distribution in the prepositional phrase: Similarities between Norwegian and English Jenny Lederer


part iv.  Finiteness and modality Experiencers with (un)willingness: A raising analysis of German ‘wollen’ Remus Gergel & Jutta M. Hartmann


Finiteness: The haves and the have-nots Kristin M. Eide


Index of subjects & languages


Advances in Comparative Germanic Syntax Artemis Alexiadou, Jorge Hankamer, Thomas McFadden, Justin Nuger & Florian Schäfer

This volume contains a selection of the papers that were presented at the 21st Comparative Germanic Syntax Workshop, held from March 31st to April 2nd 2006 at the University of California, Santa Cruz, and at the 22nd Comparative Germanic Syntax Workshop, held on June 8th and 9th 2007 at the Universität Stuttgart. With the range of topics covered in the fourteen chapters we have put together a mixture that we think accurately reflects the feeling of the CGSW meetings. Alongside traditional themes of Germanic comparative syntax like verb second and Holmberg’s Generalization we find discussion of those driving current cutting edge work in comparative syntax and syntactic theory in general, for example the cartographic approach to clause structure and the nature of the syntax-morphology interface. We have organized the volume into four thematically defined parts. Not surprisingly, some of the chapters could be at home in more than one of the parts as they touch on several of the recurring themes of the volume. Within each part, the chapters are thus organized in such a way as to highlight the relationships between the various topics and to create a smooth transition from one part to the next. While each chapter is intended to stand on its own, this structure will hopefully facilitate a reading of the volume in its entirety.

Part I. Cartography and the left periphery In the first part of the volume we have collected five papers which deal with the inventory of functional projections at the left periphery of the clause, and in particular with the advantages and disadvantages of what has come to be known as the Cartographic approach (Rizzi 1997; Cinque 1999 and many others). In the first place, work in this tradition argues that the structure of the left periphery is far more articulated than had previously been assumed. Where the standard Minimalism of the mid-1990s (e.g., Chomsky 1995) had only C and T, eight or more functional heads are now commonly posited. In the second place, the functional heads in this structure are argued to play a crucial role in the the syntax and semantics related to discourse functions,

 Artemis Alexiadou, Jorge Hankamer, Thomas McFadden, Justin Nuger & Florian Schäfer

e.g., both driving the fronting of topic elements and encoding or supporting their topical semantics. A great deal of research over the past decade has focused on this cartographic approach, either supporting and developing it further (see e.g., the contributions in Cinque 2002 and Rizzi 2004) or arguing against it. In Part I of this volume we have collected papers of both types. The contribution by Anna Cardinaletti, “On a (wh-)moved Topic in Italian, compared to Germanic”, falls squarely into the pro-cartography camp. On the basis of data from left peripheral elements in Italian and English, Cardinaletti attempts to work out some of the details of the relationship between two of the functional heads that have been proposed in the cartographic approach to appear at the interface between the traditional I and C layers. The central proposal is that Resumptive Preposing (RP) in Italian (1a) should be treated as wh-movement to the highest topic position, like English Locative Inversion and Comparative Inversion but unlike Italian Clitic Left Dislocation (1b) and Focalization (1c), which target the low topic position. (1) a. b. c.

La stessa proposta fece poi il partito di La stessa proposta la fece poi il partito di La stessa proposta fece poi il partito di the same proposal (it) made then the party of

maggioranza maggioranza maggioranza majority

(nonˉunaˉanaloga) notˉoneˉsimilar

‘Theˉmajorityˉpartyˉthenˉmadeˉtheˉsameˉproposal (notˉaˉsimilarˉone).’

From this follow three consequences for RP. First, it is restricted to root contexts because only there is the highest TopicP projected. Second, preverbal subjects are impossible because the fronted XP moves through specFinP, satisfying the EPP and blocking subsequent subject raising. If wh-movement and focalization proceed via this same escape hatch, their incompatibility with RP is accounted for as well. Third, subjects must be heavy, because it is Heavy NP Shift that obviates what would otherwise be a Minimality violation – a light subject would intervene between the fronted element and the position where it originates. A second contribution which makes crucial use of cartographic assumptions to account for specific phenomena on the left periphery is that by Michael Putnam and Marjo van Koppen. In “C-agreement or something close to it: some thoughts on the ‘alls-construction’”, the authors examine a little-studied construction in Midwestern American English, exemplified in (2)

(2) Alls Jim and Carol want to do is stay at home.

Their proposal is to assimilate this ‘alls construction’ to the phenomenon of Complementizer agreement found in certain West Germanic languages, and they offer a

Advances in Comparative Germanic Syntax 

minimalist-inspired analysis based on this association. Specifically, the s ending on all signals agreement between the Force head and a non-2nd person definite subject. This agreement is subject to a strict form of locality, which becomes apparent when the overt complementizer that is present as in (3). (3) a. *All-s that b. All that


know is that Doris talks a lot. know is that Doris talks a lot.

When there is no that, the heads of the articulated C domain (Force, Top and Fin) are conflated, so that φ-features associated with Force can Agree with the subject, yielding the s ending. But if that is present, conflation is blocked, and Force is not sufficiently local to the subject for the requisite Agree relation. The next two chapters in Part I of the volume argue explicitly against the cartographic approach and propose alternative views of the left periphery and clause structure in general. In “Uncharted territory? Towards a non-cartographic account of Germanic syntax”, Jan-Wouter Zwart explores a strictly derivational approach to clausal structure, where syntactic relations are established dynamically rather than on the basis of a pre-determined hierarchy of functional elements. He argues that under a cartographic approach, where word orders are derived by a fixed hierarchy of positions, if A precedes B and B precedes C, then A should precede C even when B is not present. Cases where this prediction is not borne out, which he calls “transitivity failures”, are however attested in the left periphery, adverb ordering, argument-adjunct interactions and adjective ordering. Zwart thus proposes an implementation of a noncartographic alternative, and lays out how it can be made to work in four areas of Germanic syntax:  the structure of the left periphery, topicalization/focalization, subject placement and object placement. Gisbert Fanselow, in “Bootstrapping verb movement and the clausal architecture of German (and other languages)” similarly argues that the basic assumptions of the cartographic approach suffer from empirical and theoretical problems and proposes a decidedly non-cartographic approach to the derivation of German verb-second word orders. He makes two key claims. First, movement of the finite verb does not target COMP or any other pre-existing position in a fixed hierarchy of functional projections. Rather, it adjoins to TP via a reformulated bootstrapping version of head movement. Second, subsequent movement of an XP to the first position is triggered not by (pragmatic) features on the XP itself or by cartography-style Topic or Focus heads, but rather by an unrestricted edge feature on the finite verb. In the final chapter of Part I, “A conjunction conspiracy at the West Germanic left periphery”, John te Velde examines the role played by functional projections at the left edge of the clause in a special kind of coordinate construction. This Left-Edge Ellipsis, which is found in many of the Germanic verb second languages, involves a

Artemis Alexiadou, Jorge Hankamer, Thomas McFadden, Justin Nuger & Florian Schäfer

gap at the left periphery of the second (and subsequent) conjuncts, as in the German example in (4):  (4) Die Briefmarkeni zeigt Hans seinem Onkel ti und ei the stamps shows Hans his uncle and

verkauft sells

er seiner Tante ti. he his aunt ‘Hansˉshowsˉtheˉstampsˉtoˉhisˉuncleˉandˉsellsˉthemˉtoˉhisˉaunt.’

te Velde notes that previous accounts of LEE have been focused on examples where the gap is the subject and are not easily extended to those like (4) where it is an object, and that they are not consistent with Phase Theory and other recent developments in Minimalism. He thus proposes an alternative which avoids both shortcomings and is also consistent with a particular understanding of the Coordinate Structure Constraint (CSC, originally due to Ross, 1967). Specifically, the gap is taken to be the result of ellipsis rather than being a trace of across-the-board movement or a bound pronoun. The ellipsis itself is restricted syntactically, in that the conjunction must c-command the gap, and recovery is restricted semantically, by a requirement that the gap undergo Match at LF with its antecedent. The left edge requirement resulting from the c-command condition conspires with the V2 syntax of these languages such that any V2-fronted argument will be eligible for ellipsis if the conditions on Match at LF are also met. te Velde also presents evidence from asymmetries with subject, object and wh-fronting in favor of assuming three distinct landing sites in the left periphery, Spec-TP, Spec-TopP and Spec-CP respectively.

Part II. Word order and movement Discussion of the left periphery and cartography is usually concerned at least implicitly with questions of word order, as the overt ordering of elements is one of the main pieces of evidence for underlying structure. The chapters in Part I of this volume are no exception in this respect, and thus lead in to the four contributions in Part II, which are concerned more directly with word order issues. Not surprisingly for a volume on Germanic syntax, the verb-second phenomenon plays an important role in tying the parts together, featuring prominently in the contributions by Zwart, Fanselow, and te Velde in Part I and in those by Trips & Fuß, and Noel in Part II. The first chapter in part II picks up directly where te Velde’s chapter left off, with the discussion of empirical challenges to the CSC. In “Reconsidering odd coordination in German”, Hironobu Kasai proposes to handle German data which seem to violate the CSC – e.g., those in (5) and (6) – in terms of constraints on movement.

Advances in Comparative Germanic Syntax 

(5) Die Suppe wird der Hans essen und sich hinlegen. the soup will the Hans eat and self down-lie ‘Hansˉwillˉeatˉtheˉsoupˉandˉlieˉdown.’ (6) Äpfel ißt der Hans drei und zwei Bananen. apples eats the Hans three and two bananas ‘Hansˉeatsˉthreeˉapplesˉandˉtwoˉbananas.’

Kasai shows that a previous analysis of such examples proposed by Schwarz (1998) doesn’t actually work for all types of verbal ‘odd coordination’. He then argues that the CSC can be maintained in the face of these data if it is understood in terms of the Principle of Minimal Compliance (PMC, Richards 1998). Informally, this states that a well-formed dependency that satisfies a particular constraint can obviate a violation of that same constraint by some other dependency. For the CSC specifically, Kasai proposes that a functional head triggering movement out of a coordinate structure must establish an ATB dependency with that structure. Once it establishes one ATB dependency, however, the PMC implies that further movement dependencies can potentially be established with a single conjunct. This allows precisely those types of odd coordination observed in German. For example, in ‘verbal odd coordination’ like (5), ATB V-to-C movement satisfies the CSC, obviating a violation by subsequent movement out of one of the conjuncts to Spec-CP. In “The syntax and semantics of the temporal anaphor ‘then’ in Old and Middle English”, Carola Trips and Eric Fuß discuss the properties of the adverbs corresponding to then in the early history of English, þa and þonne. They identify two issues concerning these adverbs that have remained unexplained in the literature. First, they pattern together with wh-elements and negatives with respect to subject-verb inversion in OE and EME, in that they trigger inversion even with subject pronouns, whereas in other contexts such inversion generally does not occur. Yet, from a semantic point of view, they do not seem to form a natural class with such elements. Second, in the ME period, þa and þonne lose their ability to systematically trigger inversion. Trips and Fuß address these issues by proposing that, contrary to what is generally assumed in the literature, subject-verb inversion with the adverbs meaning ‘then’ is not derived in the same way as inversion with interrogatives and negatives. Rather, they occur in SpecTP in OE and EME, what is normally the surface position for pronominal subjects, forcing such subjects to stay in a lower, post-verbal position. Fronted wh-elements and negatives, on the other hand, are located in SpecCP. This explains why the former but not the latter could be preceded by fronted topics yielding V3 surface orders. The loss of inversion with þa and þonne in ME was caused by the development of an EPP feature requiring that SpecTP be filled by an actual subject, keeping the ‘then’ words out and also triggering the simultaneous rise of overt expletives in the period. The connection between the two changes is something that previous accounts, which treated þa and þonne the same as fronted operators, have not explained.

 Artemis Alexiadou, Jorge Hankamer, Thomas McFadden, Justin Nuger & Florian Schäfer

A second study focussing on historical forms of Germanic is Patrizia Noel’s contribution “Jespersen’s Cycle and the issue of prosodic ‘weakness’”. This chapter traces prosodic weakening of the negation particle from Proto-Indo-European through to Modern Standard German. Noel claims that prosodic weakening of the particle has actually occurred twice: first in the transition from Proto-Indo-European to Old High German, and then again in the transition from Old High German to Middle High German. Only in the latter case was Jespersen’s Cycle initiated. She argues that in Old High German mononegation was stable enough to tolerate the gradual reduction of the negative particle, whereas in Middle High German several factors – segmental, syntactic and prosodic – conspired to weaken its status to such a degree that strengthening by means of an additional element was necessitated. First, OHG avoided secondary stress on open syllables ending in a high vowel, which includes the negative particle ni. Second, the stabilization of V2 in declarative sentences prevented the negative particle from appearing in initial position where secondary stress would be expected. Third, the particle could not form a rhythmic foot on its own, nor could it easily resolve with the right-adjacent syllable to become the head of that foot, because it typically preceded a stressed verb-initial syllable. Word order in the lower domain of the clause is the focus of “Holmberg’s Generalization:  Blocking and push up”. This contribution, by Hans Broekhuis, takes Holmberg’s (1999) characterisation of the generalization as its point of departure, in terms of which HG is viewed as a condition on derivations, and Object Shift (OS) therefore necessarily emerges as a post-syntactic operation. Appealing to the fact that OS in Icelandic is semantically restricted, the author argues that this characterisation cannot be upheld, proposing instead that HG should be viewed as a filter on output representations. He also argues for a novel understanding of how HG actually manifests itself in specific sentences. In addition to the usual assumption that it simply blocks object movement whenever verb movement fails, he argues that it could result in object movement “pushing up” the verb to a higher position. Broekhuis concludes on the basis of data from Danish and Icelandic that in fact both strategies for satisfying HG are attested. He provides an Optimality Theoretic analysis which predicts which strategy will apply when and which can account for the observed cross-linguistic variation in the availability of OS and verb movement. To account for the full range of word-order patterns, Broekhuis proposes that there are in fact two different types of OS, one which is driven by φ-features on V, and another which is driven by Case features on v.

Part III. Thematic relations and NP realization This dovetails with part III of the volume, where we have collected three chapters concerned with the relationships between noun phrases and the rest of the clause,

Advances in Comparative Germanic Syntax 

including in particular thematic roles, binding and case. In the first contribution, “The No Case Generalization”, Halldór Ármann Sigurðsson argues specifically against the use of case to explain syntactic patterns such as that proposed by Broekhuis. Sigurðsson presents and defends the idea that, in fact, “Syntax has no case features”. Morphological case is instead a purely morphological phenomenon, determined and realized in the PF component. While this determination is based on and thus sensitive to the output of the Narrow Syntax, it cannot affect the pre-PF derivation, thus case can play no role in NP-licensing, movement or semantic interpretation. Sigurðsson starts by reviewing several types of evidence that have been presented in the literature supporting the No Case Generalization. First, NP-movement and the distribution of PRO are independent of morphological case. Second, the distribution of specific morphological cases has a complex and indirect relationship with the syntactic and semantic status of the NPs they mark. Third, patterns of case-assignment are subject to significant cross-linguistic variation, such that it would not make sense to posit (universal) features like [+nom], [+dat] or even [+oblique] in the syntax. He then moves on to outline an account of case assignment for Icelandic. Along the way he highlights relevant contrasts with other languages, especially German, and develops further the notion that case assignment is part of the non-isomorphic translation of abstract syntactic structure into morphological material. Sigurðsson notes several instances where this allows a more satisfying analysis than the assumption that case features are already present in the syntax. Also concerned with case is the contribution by Jóhannes Gísli Jónsson, “The new impersonal as a true passive”. This paper explores the nature of the Icelandic new impersonal construction, where a verb with passive morphology takes an impersonal subject and a (potentially definite) DP complement that bears the same case as it would in the active form, be it accusative, dative or genitive. Jónsson argues against the view that these are actually active sentences with null subjects, claiming that they are instead true passives, with an understood agent. At stake is the question of whether structural accusative case can be assigned in the absence of the nominative, and Jónsson argues that this is indeed possible. Also at stake is the status of the unexpressed thematic agent – whether an agent theta role is actually assigned to a phonologically null but syntactically present subject, or is simply implicitly understood, as is often thought to be the case with standard passives. The third contribution in Part III, “Anaphoric distribution in the prepositional phrase: Similarities between Norwegian and English” by Jenny Lederer, is also concerned with thematic issues, in particular with how they relate to the licensing of anaphora. This chapter studies the interaction between the distribution of “anaphoric” pronouns (i.e., reflexives and reciprocals) in PPs and the directional vs. locational sense of prepositions. As is well known, variation between anaphoric and nonanaphoric forms is found in this context, which is unexpected under the standard

 Artemis Alexiadou, Jorge Hankamer, Thomas McFadden, Justin Nuger & Florian Schäfer

Binding Theory. Lederer argues that an account of the pronominal distribution requires recognizing that the use of a reflexive pronoun encodes spatial semantics both in terms of directionality and in terms of location near the body. English and Norwegian, while using reflexives differently in terms of designating the moving thing and its destination, are united in favoring a reflexive when the moving thing and its destination end up in the same place or very close together. Lederer argues that a particular formal analysis by Tungseth (2003), which attempts to instantiate the notion “PATH” in the syntactic structure, is too simplistic. She proposes instead that the principles governing the choice between anaphoric and non-anaphoric pronouns must make reference to more detailed spatial semantics.

Part IV. Finiteness and modality The fourth and final part of this volume contains two papers that tackle issues of the tense-aspect-mood system. The chapter by Remus Gergel and Jutta Hartmann, “Experiencers with (un)willingness: A raising analysis of German ‘wollen’”, serves as a fitting transition from part III, as it is concerned with thematic aspects of certain modal constructions. The authors propose a raising analysis of the German modal wollen ‘want’ on its volitional reading and thereby support Wurmbrand’s (1999) Generalized Raising Hypothesis, as opposed to a treatment of modals that splits them up into Raising and Control verbs. Drawing on data from Spanish, Romanian and Dutch in addition to German, they present evidence that volitional wollen (and its congeners) induces coherence, and that this is necessary (though not sufficient) to allow an oblique argument to raise from the lower clause to become the subject of the modal, acquiring a second theta-role. The paper represents a contribution both to the study of the syntax of modal constructions and to the development of the theory of thematic roles. The final chapter of the volume, “Finiteness: The haves and the have-nots” by Kristin M.  Eide is also concerned with modals and more broadly with their relationship to finiteness. Eide identifies finiteness as the key player in the distribution of English phenomena that single out a particular class of verbs (the modals, the auxiliaries have and be, and the dummy do), and defines finiteness not in terms of tense or agreement but in its own right, as a primitive. The chapter shows how this perspective allows one to make sense of the difference between present-day English and Mainland Scandinavian. The latter is poorer than English in its agreement inflection, but richer in that it makes a four-way distinction in terms of both tense and finiteness for all verbs, whereas English makes such a distinction only for the auxiliaries have and be, not for lexical verbs. The generalization that emerges is that

Advances in Comparative Germanic Syntax 

only [+finite] verbs undergo the kinds of processes that set apart the lexical verbs (which are not marked for finiteness) from the modals, the auxiliaries have and be, and the dummy do (which are).

References Cinque, G. 1999. Adverbs and functional heads. Oxford: OUP. Cinque, G. (Ed.) 2002. Functional structure in DP and IP: The cartography of syntactic structures, Vol. 1.Oxford: OUP. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Holmberg, A. 1999. Remarks on Holmberg’s generalization. Studia Linguistica 53: 1–39. Richards, N. 1998. The Principle of Minimal Compliance. Linguistic Inquiry 29: 599–629. Rizzi, L. 1997. The fine structure of the left periphery. In Elements of grammar, L. Haegeman (Ed.), 281–337. Dordrecht: Kluwer. Rizzi, L. 2004. The Structure of CP and IP:  The Cartography of Syntactic Structures, Vol.  2. Oxford: OUP. Ross, J.R. 1967. Constraints on Variables in Syntax. Ph.D. dissertation, MIT. Schwarz, B. 1998. Odd coordination. Journal of Comparative Germanic Linguistics 2: 191–219. Tungseth, M. 2003. Two structural positions for locative and directional PPs in Norwegian motion constructions. Nordlyd 31(2): 473–487. Wurmbrand, S. 1999. Modal verbs must be raising verbs. WCCFL 18, 599–612.

part i

Cartography and the left periphery

On a (wh-)moved Topic in Italian, compared to Germanic* Anna Cardinaletti

Università Ca’ Foscari di Venezia

In this paper, I compare Romance and Germanic left periphery. I show that Italian Resumptive Preposing (RP) differs from both Clitic Left Dislocation (CLLD) and Focalization and shares many properties with fronting phenomena in English (Topicalization, Locative Inversion, Comparative Inversion). RP constituents are (wh-)moved to a high Topic position only available in root contexts and can co-occur with either preverbal pronominal subjects or post-verbal heavy subjects (while CLLD can target a Topic position lower than Focus – Rizzi 1997, Frascarelli and Hinterhölzl 2007 – and does not display any restriction on the subject). The analysis is based on Rizzi and Shlonsky’s (2006) account of English Locative Inversion and will lead us to the discussion of the interaction of Fin and Subj, i.e., the heads which lie at the interface of the I and C layers, and the comprehension of the different restrictions on the occurrence of preverbal subjects in Italian and English/German left-peripheral constructions, ultimately explaining the generalizations arrived at in Cardinaletti (2007).

1.  Introduction In this paper, the Italian Topic-initial construction called Resumptive Preposing (RP) in (1a) is compared with the other left-peripheral constructions of Italian,

*The audience of the Comparative Germanic Syntax Workshop, held in Stuttgart on June 8–9, 2007, is kindly thanked for comments and suggestions. Many thanks to Josef Bayer, Guglielmo Cinque, Silvio Cruschina, Mara Frascarelli, Carlo Geraci, Jacqueline Guéron, Liliane Haegeman, Terje Lohndal, Salvo Menza, and two anonymous reviewers for many comments on previous versions of the paper. I am very grateful to Artemis Alexiadou for making my participation at the CGSW very enjoyable.

Anna Cardinaletti

Clitic Left Dislocation (CLLD) in (1b) and Focalization in (1c) (sentence (1a) is taken from Benincà 2001a: 156):1 (1) a. b. c.

La La La the

stessa proposta fece stessa proposta la fece stessa proposta fece same proposal (it) made

(non una analoga). not one similar

poi poi poi then

il partito il partito il partito the party

di maggioranza. di maggioranza. di maggioranza of majority

‘The majority party then made the same proposal (not a similar one).’

RP shares properties with both CLLD and Focalization but ultimately differs from both of them. Differently from CLLD and similarly to Focalization, the clause-initial DP la stessa proposta in (1a) is not resumed by any clitic pronoun. Differently from Focalization and similarly to CLLD, the clause-initial DP is a Topic, not a (contrastive) Focus. Furthermore, as we will see, RP is restricted to root contexts, does not allow preverbal subject DPs, and requires that the postverbal subject be heavy. While the former two properties are shared by Focalization, the latter is not. Notice that the postverbal subject agrees with the verb, like all instances of postverbal subjects in Italian: (2) La stessa proposta fecero poi i ministri dell’opposizione. the same proposal made-pl then the ministers of-the opposition

RP has the following intonation: the RP constituent and the verb build a single prosodic unit with continuous intonation, while the postverbal heavy subject builds a prosodic unit by itself.2 To analyse RP, I compare it with partly similar constructions in English and German, namely Topicalization, Focalization, Left Dislocation, Locative Inversion, and Comparative Inversion. In addition, I discuss wh-movement. Following Rizzi (1997), (2001), (2004), I assume an articulated structure of the left periphery as in (3). Following previous work of mine (Cardinaletti 1997, 2004), I assume a multiple subject positions approach, according to which subject-related features are encoded in two heads below Fin:  phi-features are checked in specTP, where weak subjects occur (4a); DP subjects and strong pronouns further raise to specSubjP to check the Subject-of-Predication feature (4b):3

1.  The term Resumptive Preposing is taken from Cinque (1990: 86–89). Benincà (2001a) calls (1a) Anteposizione anaforica (Anaphoric anteposition). 2.  Things are slightly different when a parenthetical is present, as in (17). In this case, the verb builds a prosodic unit together with the postverbal heavy subject. 3.  TP in (4) corresponds to AgrSP in Cardinaletti (2004).

On a (wh-)moved Topic in Italian, compared to Germanic

(3) Force Topic Focus Modifier Topic Fin

(4) a. [SubjP [TP weak pronoun Vfin … [VP weak pronoun Vfin ]]] b. [SubjP DP [TP DP Vfin … [VP DP Vfin ]]]

Because of the interpretive import of the Subject-of-Predication feature and following Rizzi’s (1996) notion of criterion, Rizzi (2006) and Rizzi and Shlonsky (2006) take Subj to be a criterial head, on a par with the other heads of the left periphery. The Subject Criterion is usually satisfied by movement of the subject DP into specSubjP. Local c-command by a higher head can also be relevant, and as we will see, this is central in the analysis of RP. I claim that RP involves movement to the higher Topic projection in (3), which is only licensed in root clauses and can also be targeted by a CLLD constituent when it functions as an Aboutness-Shift topic. CLLD qua Familiar Topicalization instead targets the lower Topic projection. The RP constituent (wh-)moves to the high Topic position through specFinP and in so doing it satisfies the Subject Criterion along the lines of Rizzi and Shlonsky’s (2006) analysis of English Locative Inversion. A preverbal subject is thus impossible. If wh-movement and Focalization use the same escape hatch, specFinP, we also understand why RP and these constructions cannot co-occur. Finally, the restriction against postverbal light subjects can be understood as a Minimality effect: Heavy NP shift is a way to remove the subject from a position intervening between the fronted element and its merge position, similarly to Marginalization. The paper is organized as follows. In section 2, I discuss the discourse properties of RP. In section 3, the subject restriction is presented, and three potential analyses of this restriction (verb movement across the subject, movement of the RP constituent to the subject position, adjacency requirement between the RP constituent and the verb) are discussed and discarded. Section 4 shows that like English Topicalization, RP is a root phenomenon and displays an argument-adjunct asymmetry. Section 5 argues for the (wh-)movement derivation of this construction. In Section 6, I compare RP with English Locative Inversion. In section 7, I extend the analysis to wh-questions and Focalization, which display similar restrictions on the distribution of DP subjects. Section 8 discusses two further restrictions on the postverbal subject in RP: (i) as in English Locative Inversion and Comparative Inversion, the postverbal subject must be heavy; (ii) differently from Italian wh-questions and Focalization, the postverbal subject cannot be marginalized. Section 9 concludes the paper.

2.  The fronted element is (similar to) an Aboutness-shift topic RP is a Topic construction. Since it differs from the other Italian Topic construction, i.e., CLLD, it is necessary to establish more precisely what kind of Topic it contains.

Anna Cardinaletti

As Cinque (1990: 87) observes, “the pragmatic conditions under which” RP is “well formed are different and more restricted than those governing CLLD. The fronted phrase must either directly resume an identical phrase in the immediately preceding discourse or be inferentially linked to such a phrase”, as in the following examples taken from Cinque (1983: n.25): (5) a. Fece loro la proposta di assumere un nuovo usciere. [he] made to.them the proposal of taking.on a new porter. Tale proposta discusse poi con lo stesso direttore. This proposal [he] discussed then with the director himself b. Il presidente giurò di non avere avuto contatti con the Chairman swore to not having had contacts with

esponenti del governo straniero e la stessa cosa members of.the government foreign and the same thing

giurò anche il suo segretario. swore also the his secretary

As these examples show, RP constituents usually contain demonstratives like questo (this), tale (such), and adjectives like stesso, uguale (same), simile (similar), which explicitly mark their relation with the previous context (Benincà 2001a:  141f). The above-quoted description by Cinque however seems to be able to characterize CLLD as well, and as a matter of fact, CLLD can occur in the same contexts as RP. As shown in (6), RP in (6a) and CLLD in (6b) are both felicitous.4 In contexts such as (7), however, CLLD is possible, but RP is not: (6) Il ministro propose di votare il disegno di legge. the minister proposed to vote the bill a.

La stessa proposta




partito di maggioranza.

b. La stessa proposta  la fece poi il partito di maggioranza. the same proposal (it) made then the party of majority (7) Chi ha letto questo libro? who has read this book a. *Questo libro


letto il

mio amico Gianni.

b. Questo libro  l’ha letto il mio amico Gianni. this book (it) has read the my friend Gianni

4.  The difference seems to be a register one. While CLLD is typical of informal Italian, RP belongs to the formal/literary register and recalls Object Preposing in Old Italian, which did not require a resumptive clitic.

On a (wh-)moved Topic in Italian, compared to Germanic

To understand the difference between RP and CLLD, consider the Topic typology proposed by Frascarelli and Hinterhölzl (2007), where Aboutness-shift topics and Familiar topics are differentiated. Aboutness-shift topics are defined as follows: “what the sentence is about” (Reinhart 1981; Lambrecht 1994); in particular a constituent that is “newly introduced, newly changed or newly returned to” (Givón 1983:  8), a constituent which is proposed as “a matter of standing and current interest or concern” (Strawson 1964). Familiar topics are instead “textually given and d-linked with a previously established Aboutness-topic”. The RP constituent in (6a) looks most similar to an Aboutness-shift Topic, as is the CLLD constituent in (6b). The Topic of the first utterance is il ministro ‘the minister’; in the second utterance, the Topic changes to la stessa proposta ‘the same proposal’, which resumes the focus of the previous sentence. If this is correct, then an Aboutness-shift topic can appear syntactically both as a CLLD and a RP constituent. In (7), the Topic in the answer is familiar since it refers to a Familiar topic in the question (characterized by the Marginalization intonation, see Antinucci and Cinque 1977; Cardinaletti 2001, and section 8.3). In question-answer pairs, there is no informational shift. In answers, the focus is represented by the constituent that matches the wh-word in the question, while the topic is the same in both sentences (see Aboh 2007 for discussion). In this case, only CLLD as in (7b) is possible as an answer, while RP in (7a) is not. In other words, the RP constituent cannot be a Familiar topic.5 A further difference between RP and CLLD concerns the postverbal subject: in RP, it cannot be a focus by itself, as is instead the case of the other instances of postverbal subjects of transitive verbs in Italian, including postverbal subjects in CLLD, as in (7b) (Cardinaletti 2001; Belletti 2001, 2004). What characterizes RP is a Topic-Comment articulation (in Rizzi’s 1997: 286 sense) inside a new-information clause, connected to the previous sentence via discourse anaphora. In the context of (6), repeated in (8),

5.  Object scrambling (Cardinaletti 2001) is possible in both types of contexts. Consider (i), where the scrambled object is underlined: (i) a.

Il ministro propose di votare il disegno di legge. Fece poi the minister proposed to vote the bill. made then

la stessa proposta il partito di maggioranza. the same p  roposal the party of majority

b. A: Chi ha letto questo libro? B: Ha letto questo libro Gianni. who has read this book has read this book Gianni Either the scrambled object targets two different positions in (ia) and (ib), or the position targeted by object scrambling differs in feature content from both left-peripheral Topic positions.

Anna Cardinaletti

the RP construction is as felicitous as the reversed, unmarked sentence in (8b) with the subject first: (8) Il ministro propose di votare il disegno di legge. the minister proposed to vote the bill a. La stessa proposta fece poi il partito di maggioranza. the same proposal made then the party of majority b. Il partito di maggioranza fece poi la stessa proposta.

The fact that the RP constituent belongs to a new-information clause does not however mean that RP is a Focus construction.6 The RP constituent is neither a Contrastive Focus nor a New Information Focus by itself. First, although the RP constituent can be itself contrasted by means of a focalizing adverb (9), the construction does not have the phonological and semantic properties of (left-peripheral) Focalization. What follows the RP constituent is not destressed, and we do not find the Focus-Presupposition articulation (in Rizzi’s 1997: 287 sense): in (9), il partito di maggioranza is not presupposed, but new information. Second, RP is not possible as the answer to a question, (10B). As said above, RP cannot contain a topic subject:7 (9) Il ministro propose di votare il disegno di legge. Proprio la stessa the minister proposed to vote the bill exactly the same proposta fece poi il partito di maggioranza, non una analoga. proposal made then the party of majority, not a similar [one] (10) A: Cosa ha letto il professore di Maria? what [he] has read, the professor of Maria B: *Questo libro ha letto il professore di Maria. this book has read the professor of Maria

In conclusion, RP must be differentiated from both CLLD and Focalization: it is a newinformation clause with a Topic-Comment articulation, and the RP constituent is an Aboutness-shift topic.

6.  Our conclusions differ from Benincà and Poletto (2004: 62), who suggest that RP “involves a Focus-like anteposition” which activates “the Focus field”. 7.  RP also differs from Sicilian left-peripheral Focus, as discussed in Cruschina (2006) and exemplified in (iB), which can be a New Information Focus, as shown by the fact that it can answer a wh-question: (i) A: Chi scrivisti airi? B: N’articulu scrissi. what [you] wrote yesterday a paper [I] wrote

On a (wh-)moved Topic in Italian, compared to Germanic

3.  On the subject restriction In RP, the subject cannot occur between the RP constituent and the verb (11a). This is not the case in CLLD (11b): (11) a. *La stessa proposta il partito di maggioranza fece il giorno successivo. the same proposal the party of majority made the day after b. La stessa proposta il partito di maggioranza la fece il giorno successivo.

Some potential analyses of (11a) must be discarded. First, RP is not an instance of subject inversion due to verb movement across the subject. With compound tenses (12), the subject follows the past participle and cannot occur between the auxiliary and the lexical verb; the same is true of modal + infinitive sequences, as in (13): (12) a. La stessa proposta ha poi fatto [il partito di maggioranza]. the same proposal has then made the party of majority b. *La stessa proposta ha [il partito di maggioranza] fatto. (13) a. La stessa proposta potrà poi fare [il partito di maggioranza]. the same proposal will-be-able then make the party of majority b. *La stessa proposta potrà [il partito di maggioranza] fare.

Second, the ungrammaticality of (11a) is not due to a competition between the subject DP and the RP constituent in one and the same position, namely the preverbal subject position. RP constituents do not have the same distribution as subjects; they are for instance impossible in contexts where preverbal subjects can occur, such as complementizer-deletion sentences (14) and Aux-to-Comp constructions (15) (Cardinaletti 1997, 2004):8 (14) a. Gianni crede *(che) la stessa proposta fece / abbia fatto il Gianni believes  that the same proposal made / has made the

partito di maggioranza. party of majority

b. Gianni crede (che) Maria abbia fatto quella proposta. Gianni believes (that) Maria has made that proposal

8.  The following sentence is possibile without che (that):  Credo la stessa proposta fece il partito di maggioranza. Giorgi and Pianesi (2004) show that 1sg credo (I believe) is not a verb that projects a full clause, but an item inserted in a left-peripheral functional head. This sentence is therefore not a counterexample to the claim made in the text.

 Anna Cardinaletti

(15) a. *Avendo la stessa proposta fatto il partito di maggioranza, … having the same proposal made the party of majority, … b. Avendo Maria fatto quella proposta, … having Maria made that proposal, …

Complementizer-deletion and Aux-to-Comp sentences have a reduced structure without the left periphery: the former are TPs or SubjPs (Cardinaletti 2004: 131), the latter are FinPs (Rizzi 1997: 303). The ungrammaticality of (14a) without che (that) and of (15a) suggests that the RP constituent does not target these projections, but a higher projection of the left periphery. The CLLD sentences in (16) are also ungrammatical in the same contexts: (16) a. Gianni crede *(che) la stessa proposta la fece il partito Gianni believes  that the same proposal it made the party

di maggioranza. of majority

b. *Avendo(la) la stessa proposta fatto il partito di maggioranza, … having  it the same proposal made the party of majority, …

Third, the ungrammaticality of (11a) is not due to the lack of adjacency between the RP constituent and the verb. The two elements need not be adjacent, as shown by the fact that a parenthetical can intervene (17) (see (25a), (36) and (49b–d) for other examples of non-adjacency): (17) La stessa proposta, credo / secondo loro, fece il partito di the same proposal, [I] think / according.to them, made the party of maggioranza. majority

(17) shows that the RP constituent and the verb are not in one and the same projection (as is also the case of wh-questions: Chi, credi/secondo te, ha invitato, Gianni? ‘whom, [you] think/according to you, has invited, Gianni?’). Parentheticals are forbidden in instances of fronting which require the simultaneous filling of the specifier and the head of the left-peripheral projection, such as focalization in Hungarian and Gungbe (Cardinaletti 1997: 46). But this is not the case of RP. Finally, RP displays a different behaviour in root and embedded contexts, which subject DPs never display. This restriction is discussed in the next section.

4.  Resumptive Preposing is a root phenomenon Although RP can occur in embedded clauses that are complement to bridge verbs (18), it is impossible in structures with defective left periphery, namely central adverbial

On a (wh-)moved Topic in Italian, compared to Germanic

clauses, factive complements, and infinitival complements, (19a–c) (see Haegeman 2002, 2006 for the discussion of different types of sentential complements): (18) Gianni crede / ha detto [che la stessa proposta fece il partito Gianni believes / has said  that the same proposal made the party

di maggioranza]. of majority

(19) a. *[Se la stessa proposta fa anche l’altro candidato], if the same proposal makes also the other candidate non otterrai quel posto. [you] not will.get that position b. *Mi dispiace [che la stessa proposta fece anche il partito di [I] regret  that the same proposal made also the party of maggioranza]. majority c. *Gianni sostiene [la stessa proposta di fare subito anche lui]. Gianni claims  the same proposal to make soon also he

RP displays the same distributional restrictions as the pragmatically similar English Topicalization. A topic can occur in the complement to a verb like believe (20) (from Chomsky 1977: 93), but it cannot occur in clauses with defective left periphery (21) (from Haegeman 2006). Notice that this restriction does not apply to Italian CLLD. The CLLD counterparts of (19) and (21) are possible as shown in (22). As Cinque (1990: 58) points out, “the “left-dislocated” phrase of CLLD can occur at front of virtually any subordinate clause type” (see also Rizzi 1997: 309 for infinitival clauses):9 (20) I believe that this book you should read.

9.  The case of sentential subjects is less clear. For many speakers, RP in sentential subjects is only slightly degraded with respect to CLLD: (i) a. [Che la stessa proposta la fece il partito di maggioranza] è noto a tutti. b. ?[Che la stessa proposta fece poi il partito di maggioranza] è noto a tutti.   that the same proposal (it) made then the party of majority is well-known to everybody As for English Topicalization, the judgments found in the literature are not uniform. Authier (1992) considers (iia) ungrammatical, while Lasnik and Saito (1992: 77) provides the grammatical example in (iib). See Haegeman (2008) for discussion:

(ii) a. *[That this book, Mary read thoroughly] is true. b. [That this solution, I proposed last year] is widely known.


 Anna Cardinaletti

(21) a. *[If these exams you don’t pass] you won’t get the degree. b. *John regrets [that this book Mary read]. c. *My friends tend [the more liberal candidates to support]. (22) a. [Se la stessa proposta la fa anche l’altro candidato], non otterrai quel posto. b. Mi dispiace [che la stessa proposta la fece anche il partito di maggioranza]. c. Gianni sostiene [la stessa proposta di farla subito anche lui].

On the basis of the contrast between (21) and (22), Haegeman (2002: 161) and (2006) suggests that English Topicalization depends on the presence of Force and that the language does not have “any alternative way of relating a fronted topic to the associated clause”, where “Force anchors the clause to the speaker and to speech time”. The Topic is taken to occupy the higher Topic position in (3) licensed by Force, as shown in (23a). The CLLD constituent in (22) instead occupies the lower Topic position, which does not depend on the presence of Force, (23b). Since RP patterns with English Topicalization, Haegeman’s analysis can be extended to RP: the RP constituent targets the high Topic position, as shown in (23c):10 (23) a. [HighTopicP this booki … b. [LowTopicP la stessa propostai … c. [HighTopicP la stessa propostai …

[TP I don’t like ti]] [TP lai fece il partito … ti]] [TP fece il partito … ti]]

Haegeman’s proposal is coherent with the conclusion arrived at by Frascarelli (2007) and Frascarelli and Hinterhölzl (2007). The different types of topics they assume

Non-uniform judgments are also found in English as far as Topicalization in relative clauses is concerned (see Rizzi 1997: 306f). RP is however possible in relative clauses: (iii) a. La persona [a cui la stessa cosa confidò il preside della the person  to whom the same thing confided the director of.the

scuola] è il mio collega Giorgio Rossi. school is my colleague Giorgio Rossi

b. L’unica volta [che della stessa cosa si occupò anche il the only time  that of the same thing took care also the

segretario della scuola] i risultati furono disastrosi. secretary of.the school the results were disastrous

10.  Haegeman (2002: 159, 164) suggests that Sub, the position of complementizers, should be differentiated from Force and locates Force below FocusP, as in (i). The exact location of Force is not crucial here:


Sub Topic Focus Force Modifier (Topic) Fin IP

In more recent work, Haegeman (2008) assumes a movement analysis of adverbial and factive clauses and analyses (21) in terms of intervention effects. If this will turn out to be the correct analysis, it can be easily extended to RP.

On a (wh-)moved Topic in Italian, compared to Germanic 

occupy different positions: Aboutness-shift topics are highest in the structure, while Familiar topics occur immediately above the Fin head: (24) Force Shift Topic … Focus Familiar Topic Fin

If RP is an Aboutness-shift topic, as I have claimed in section 2, it is expected to occur in the highest Topic position, as proposed in (23c). The assumed structure predicts that a Shift topic linearly precedes a Familiar topic, and this is indeed what is found with the combination of RP and CLLD (on sentences like (25b) also see Benincà 2001a: 159): (25) a.

La stessa proposta a Gianni gli fecero gli studenti spaventati the same proposal to Gianni to-him made the students frightened

dall’esame. by the exam

b. *A Gianni la stessa proposta gli fecero gli studenti spaventati dall’esame.

Support for the High Topic analysis of RP comes from the following behaviour of RP embedded under bridge verbs. The RP constituent can only follow the complementizer (see (18), repeated here as (26a)); if it precedes che (that), as in (26b), we get an ungrammatical sentence. CLLD in (26c) is instead possible, albeit marginally for some speakers: (26) a. Gianni crede che la stessa proposta fece il partito di maggioranza. b. *Gianni crede la stessa proposta che fece il partito di maggioranza. c. ?Gianni crede la stessa proposta che la fece il partito di maggioranza.

This rather strange state of affairs can be analyzed as follows. The Topic position targeted by RP is licensed by Force, which is realized by che in (26a). In (26b), no head licenses the presence of the high Topic position, and the sentence is ungrammatical on a par with (14) above. In (26c), the presence of the resumptive clitic witnesses that we are dealing with an instance of CLLD: the left-dislocated item can occur in the low Topic position, while che occurs in Fin (for the marginal occurrence of che in Fin, see Cardinaletti 2004: 158,n.17; for a different interpretation of the data, see Benincà 2001b: 54f).11

11.  If Haegeman is correct in taking FocusP to also be licensed by Force, the sentence in (i), taken from Rizzi (2001: 289), patterns with (26b): (i) *Credo questo che avreste dovuto dirgli (non qualcos’altro). [I] believe this that [you] would-have had-to tell-him (not something else) ‘I believ that THIS you should have told him (not something else).’

 Anna Cardinaletti

Further support for this analysis comes from another restriction on RP:  as observed by Menza (2006: 107, 109), RP is impossible with imperatives (27a) and quasi-verbs like ecco ‘here (is/are), look here’ (27b): (27) a. *Il libro portami! the book bring-to.me b. *(Nessuno sospettava della signora in grigio.) La signora in grigio (nobody suspected of the lady in grey.) The lady in grey

ecco entrare all’improvviso nella stanza. here enter suddenly the room

Since imperatives and quasi-verbs cannot be embedded, it is reasonable to assume that they have a reduced structure, without the higher Topic position (see Jensen’s 2007 recent discussion of English and Scandinavian imperatives). It follows that in these two contexts, there is no space for a RP element. The sentences are possible with a clitic pronoun, as expected if CLLD can involve a lower Topic position, and as Focalization, as expected if the Focus projection is lower than the high Topic position (sentences (28a) and (29a) from Menza 2006: 109):12 (28) a. Il libro, portamelo! the book bring-to.me-it b. (Nessuno sospettava della signora in grigio.) La signora in grigio (nobody suspected of-the lady in grey.) The lady in grey eccola entrare all’improvviso nella stanza. here her enter suddenly into-the room (29) a. Il libro portami, non il quaderno! the book bring-to.me, not the notebook

Sentence (i) is however not fully ungrammatical for me. This difference in judgment recalls the non-unanimity among Italian speakers on Focalization in clauses with defective leftperiphery, as reported in Haegeman (2002: 152,n.24). Notice that the phenomenon illustrated in (26) is different from Bavarian Emphatic Topicalization discussed in Bayer (2001), where a XP can appear in front of the complementizer if the embedded clause is itself topicalized to the left periphery of the root clause. No amelioration is found in the following sentence with respect to (26b): *[La stessa proposta [che fece il partito di maggioranza]], Gianni non (lo) crede davvero ‘the same proposal that made the party of majority, Gianni not it believes at all’. 12.  As the enclitic pronoun in (28b) shows, ecco assigns accusative case. The selectional properties of ecco are similar to those of causative and perception verbs (Menza 2006: 27).

On a (wh-)moved Topic in Italian, compared to Germanic 

b. (Nessuno sospettava della signora in grigio.) La signora in grigio (nobody suspected of the lady in grey.) The lady in grey ecco entrare all’improvviso nella stanza, non sua sorella. here enter suddenly into.the room, not her sister

Once again, RP patterns with English Topicalization, which is also impossible in imperatives (Haegeman 2008): (30) *Your essay, leave in my pigeon hole this afternoon.

Haegeman (2008) observes that English Topicalization is also banned from yes-no questions (31a). The same is true of RP (31b): (31) a. *That book about shrimp, did you actually read? b. *La stessa proposta fece anche il direttore del museo? the same proposal made also the director of the museum?

Yes-no questions are another context where the higher Topic position is not available. CLLD and Focalization are instead possible in yes-no questions: (32) a. La stessa proposta la fece anche il direttore del museo? the same proposal it made also the director of.the museum? b. La stessa proposta fece il direttore del museo, o the same proposal made the director of.the museum, or

una diversa? a different [one]

4.1  PP fronting Cinque (1990: 86) analyses PP fronting without a resumptive clitic as in (33a) as an instance of RP. (33b,c) are two more examples of the same type: (33) a. Allo stesso modo, si comportò suo figlio. in.the same manner refl. behaved his son b. Alla stessa cosa pensò il Preside della scuola. to.the same thing thought the director of.the school c. Con la stessa sollecitudine (*ci) operò l’associazione with the same promptness with.it operated the association umanitaria in Afghanistan. humanitarian in Afghanistan

PP fronting in (33) displays the same properties as RP except one: Like Topicalization of adjuncts in English (34), PP fronting is also possible in clauses with defective left periphery, (35): (34) If on Monday we haven’t found him, we’ll call the RSPCA.

 Anna Cardinaletti

(35) a. Se [allo stesso modo si comporterà anche suo figlio maggiore], if  in.the same way will.behave also his son elder

il problema potrà essere risolto facilmente. the problem can be solved easily

b. Mi dispiace [che allo stesso modo si sia comportato anche suo [I] regret  that in.the same way has behaved also his

figlio maggiore]. son elder

c. ?Gianni sostiene [allo stesso modo di potersi comportare subito Gianni claims  in.the same way to be.able behave soon

anche lui]. also he

Haegeman (2002), (2006) suggests that in English, adjuncts can target a lower position than topicalized phrases, presumably the ModP of Rizzi (2004) (see (3)), which is also available in clauses with defective left periphery. The same can be suggested for the Italian sentences in (35). Given the structure in (3), the order “RP constituent – fronted PP” in (36a) is expected when RP and PP fronting to specModP co-occur. The same order is found with a temporal adverb in specModP, as shown in (36b): (36) a.

La stessa proposta con pari sollecitudine avanzò anche il the same proposal with equal promptness made also the

direttore del museo. director of the museum

b. La stessa proposta domani farà anche il partito the same proposal tomorrow will-do also the party

di maggioranza. of majority

5.  Resumptive Preposing involves (wh-)movement 5.1  The analysis of English Topicalization and Italian CLLD English Topicalization is analysed as in (37). This analysis goes back to Chomsky (1977) and is adopted by Rizzi (1997: 293) with one modification: the empty category in the merge position is not a variable but a null constant (nc), licensed by an anaphoric operator OP which connects it to the topicalized antecedent. This analysis accounts for the lack of Weak Crossover effects (WCO; see Lasnik & Stowell 1991). Compare (37) with wh-movement in (38):

On a (wh-)moved Topic in Italian, compared to Germanic 

(37) [TopicP Johni, [FinP OPi [SubjP hisi mother really likes nci]]] (38) *[FocusP Whoi [FinP does [SubjP hisi mother really like vbli]]]?

A similar analysis is proposed by Rizzi (1997: 293) for Italian CLLD, which also does not display WCO, modulo the presence of a resumptive clitic pronoun instead of the empty operator (39):13 (39) [TopicP Giannii, [FinP [SubjP suai madre loi ha sempre apprezzato nci]]] Gianni his mother him has always appreciated

Since WCO is displayed by RP (40a), the same analysis cannot be extended to this construction. The merge position must contain a real variable, as shown in (40b) (on the trace in specFinP, see section 6): (40) a. *[La stessa persona]i promossero anche i suoii dirigenti. the same person promoted also the his directors b. *[TopicP la stessa personai [FinP ti [SubjP [TP promossero i suoii dirigenti vbli]]]]

In addition to WCO effects, there are a number of properties which support the analysis in (40b), which treats RP as an instance of (wh-)movement.

5.2  On the (wh)-movement properties of Resumptive Preposing Cinque (1990:  88–89) discusses three properties of RP that make it resemble whmovement. They are listed in (41), and the examples, taken from Cinque (1990: 88), are provided in (42) (property (41a) has already been discussed, see (1a) vs. (1b)). No such properties are displayed by CLLD: (41) a. There is no resumptive clitic (42a). b. Parasitic gaps are possible (42b). c. The fronted element cannot co-occur with a wh-phrase (42c).14

13.  WCO effects are preserved under CLLD if the CLLD constituent contains a quantifier (Cinque 1990: 179,n.6): (i) a. *?Suai madre ama ognii bambino. his mother loves every boy b. *?Ognii bambino, suai madre loi ama. every boy, his mother him loves 14.  The two constituents cannot occur in the reversed order either:  (i) *… e a chi la stessa cosa disse suo figlio?  … and to whom the same thing said his son

 Anna Cardinaletti

(42) a. … e questo disse anche il Sottosegretario. … and this said also the Vice Minister b. La stessa cosa negò __ senza commentare __ il suo avvocato. the same thing denied without commenting the his lawyer c. *… e la stessa cosa a chi disse suo figlio? … and the same thing to whom said his son ‘To whom did his son say the same thing?’

Cinque (1990: 88) concludes that “these are all expected properties, if Wh-movement of a null operator is indeed involved” in RP. Other properties that RP shares with wh-movement are listed in (43):15 (43) a. b. c. d.

The preposed constituent and the verb need not be adjacent. The construction can be subject to successive cyclic movement. The construction exhibits reconstruction effects. Preverbal subject DPs are not possible, but preverbal weak subjects are.

Property (43a) has already been discussed in section 3 above (see (17)). The properties in (43b–d) will be discussed in the following sections.

5.2.1  On the movement possibilities of the RP constituent Like wh-movement and CLLD (Cinque 1990), RP is sensitive to strong islands (exemplified in (44a) by an adverbial clause), but possible out of weak islands (exemplified in (44b) by a wh-island and in (44c) by a factive island): (44) a. *La stessa cosa sono partito [prima che dicesse il nuovo the same thing [I] have left  before that said the new

presidente della società]. president of the company

b. La stessa proposta  ci chiediamo [a chi farà il the same proposal [we] refl ask  to whom will-make the

nuovo presidente della società]. new president of the company

c. La stessa cosa  ci dispiace [che abbia detto anche il direttore the same thing [we] regret  that has said also the director della scuola]. of the school

15.  CLLD differs from RP in that it cannot avail itself of successive cyclic derivations (Cinque 1990: 64) and allows for preverbal DP subjects, (11b).

On a (wh-)moved Topic in Italian, compared to Germanic 

The crucial test to establish the status of RP is to check adjunct extraction, which can only proceed successive cyclically. We have seen in section 4.1 that fronted PPs are also instances of RP. While adjuncts are impossible in long-distance CLLD (45a), they are possible in RP (45c), as in wh-questions (45b) (sentences (45a–b) are taken from Cinque 1990: 66):16 (45) a. *In modo definitivo  ha detto [che  l’aggiusterà]. in a definitive way [he] has said  that [he] it will.fix b. In che modo  ha detto [che   l’aggiusterà]? in which way [he] has said   that [he] it will.fix c. Allo stesso modo   ha detto [che si è comportato il figlio in the same manner [he] has said   that has behaved the son

di Maria]. of Maria

While CLLD only gives rise to long dependencies (Cinque 1990: 64–68), RP is like wh-movement in allowing both long and successive cyclic movement, (44b–c) and (45c) respectively. As we know on the basis of wh-movement, adjunct extraction is sensitive to weak islands (Cinque 1990: 29). This is also true of RP. Examples are provided for wh-islands in (46) and factive islands in (47) ((46a) is taken from Rizzi 1990: 104): (46) a. *In che modo non immagini [chi potrebbe essersi comportato]? in which way [you] not imagine   who could have behaved b. *Allo stesso modo non immagino [se si sia comportato anche in the same manner [I] not imagine   whether has behaved also il figlio di Maria]. the son of Maria (47) a. *In che modo   ti dispiace [che   si sia comportato]? in which way [you] regret   that [he] has behaved

16.  Like wh-movement, RP is trivially possible with successive cyclic movement of arguments: (i) a.

Che   cosa hanno detto che fece, il partito di maggioranza?  what [they] have said that did, the party of majority 


La stessa proposta   hanno detto [che fece il partito the same proposal [they] have said   that made the party

di maggioranza]. of majority

 Anna Cardinaletti

b. *Allo stesso modo mi dispiace [che si sia comportato anche il in the same manner [I] regret   that has behaved also the

figlio di Maria]. son of Maria.

In conclusion, RP shares the movement possibilities of wh-movement.

5.2.2  On reconstruction Like wh-movement (Cinque 1990:  131), RP exhibits reconstruction effects. In (48b), the quantifier due (two) can be understood in the scope of ogni partecipante all’incontro, as in (48a): (48) a. Ogni partecipante all’incontro fece poi due proposte simili. every participant at.the meeting made then two proposals similar b. Due proposte simili fece poi ogni partecipante all’incontro. two proposals similar made then every participant at.the meeting

5.2.3  On the restriction against strong preverbal subjects (11a), repeated here as (49a), shows that a full DP subject cannot occur between the RP constituent and the verb. RP however allows for preverbal weak subjects (pro, uno, tu in (49b–d)) or clitic subjects (si in (49c)), like the wh-questions in (50) and (51):17 (49) a. *La stessa proposta il partito di maggioranza fece poi. b. La stessa proposta pro fece senza consultarsi con nessuno. the same proposal [he] made without consulting with anybody c.

La stessa proposta ?uno/si può fare in tutta autonomia. the same proposal one can make in all autonomy

d. Credo [che la stessa proposta tu possa fare in I think  that the same proposal youweak can.subj make in tutta autonomia]. all autonomy (50) a. *Chi Gianni ha invitato? whom Gianni has invited? b. Chi pro ha invitato? whom he has invited?

17.  The examples in (51) are taken from Poletto and Pollock (2004: 293), Ordoñez and Olarrea (2005), and Chinellato (2003: 38–39), respectively. For arguments that (non specific) indefinite uno in (49c) and (50c) is a weak pronoun, see Cardinaletti (1997: 59,n.21) and (2004: 138). Indefinite weak pronouns are also found in English (one), French (on), and German (man).

On a (wh-)moved Topic in Italian, compared to Germanic 

c. ?Che cosa uno mangia d’estate, in questa città? what one eats in summer, in this town? (51) a. Où il va? a′. *Où Yves va? where he goes? where Yves goes? b. Qué tú quieres? b′. *Qué José quiere? what you want? what José wants? c. Cossa elo beve? c′. *Cossa what he drinks? what

Paolo beve? Paolo drinks?

In Italian (50), as in French (51a,a′) and Caribbean Spanish (51b,b′), a weak subject can occur between the wh-phrase and the verb, while a full DP is ungrammatical in the same context. The same contrast is found in the productions by aphasic patients who speak Northern Italian dialects, (51c,c′). These patients cannot produce the subject clitic pronouns which are typical of their dialects, and produce preverbal weak pronouns instead (51c). A DP subject between the wh-phrase and the verb is however never found (51c′). The contrasts in (49) are parallel to those in (50) and (51). Focusing on whquestions, Cardinaletti (2007) has expressed the paradigms in (50)–(51) in the generalization (52), which relies on the multiple subject positions approach developed in Cardinaletti (1997), (2004) (see (4)): (52) Generalization: Only subjects in specSubjP are excluded from occurring between the wh-phrase and the verb in wh-questions, whereas subjects in specTP (or lower subject positions) are permitted.

The generalization (52) can be extended to RP, as in (53): (53) Generalization (extended to RP): Only subjects in specSubjP are excluded from occurring between a wh-phrase or RP constituent and the verb, whereas subjects in specTP (or lower subject positions) are permitted.

5.3  Conclusion In conclusion, the properties listed in (41) and (43) support the analysis proposed in (40b), repeated here in (54) for the sentence in (1a): (54) [TopicP la stessa propostai [FinP ti [SubjP [TP fece poi il partito di maggioranza vbli]]]

This analysis implies that the construction contains an A′ relation involving genuine quantification, differently from English Topicalization and Italian CLLD, and similarly to wh-movement and Focalization. In what follows, I focus on one of the properties that RP shares with wh-movement and Focalization, namely the restriction against the occurrence of preverbal DP subjects, (43d).

 Anna Cardinaletti

6.  The comparison with English Locative Inversion We now address the above mentioned restriction that in RP, a DP subject must appear in postverbal position. English Topicalization does not display this property (see section 7.3), but another English construction does, namely Locative Inversion. In what follows, I am going to compare RP with English Locative Inversion and to extend Rizzi and Shlonsky’s (2006) analysis of this construction to RP. An example of Locative Inversion is (55), where the locative PP appears in front of the verb and the subject obligatorily occurs postverbally: (55) a. Into the room walked my brother Jack. b. In the room was sitting my old brother.

As we have done in (14) and (15) for Italian RP, it can be shown that the locative does not occur in the preverbal subject position:  e.g., it does not invert with the verb in questions, (56) (from Rizzi & Shlonsky 2006: 344):18 (56) *Is in the room sitting my old brother?

Like English Topicalization and Italian RP, English Locative Inversion does not occur in clauses with defective left periphery ((57) is taken from Stowell 1981): (57) a. *[That in the chair was sitting my old brother] is obvious. b. *I expect [in the room to be sitting my old brother].

18.  Differently from English Locative Inversion, Italian locative and dative PP fronting found with unaccusative verbs targets the preverbal subject position, specSubjP, which hosts full subject DPs: (i) a. [SubjP su Giannii [TP è [VP ti caduta una grande disgrazia]]] on Gianni has fallen a big misfortune b. [SubjP a Giannii [TP è [VP ti piaciuto il regalo ]]] to Gianni has pleased the present ‘John liked the present.’ Differently from the RP constituent in (15) and (19) above, a fronted PP appears in constructions such as Aux-to-Comp (see Cardinaletti 2004: 122–125) and is possible in sentences with defective left periphery (see Cardinaletti 2004: 156,n.8). In specSubjP, fronted PPs check the subject-of-predication feature (Cardinaletti 2004) and satisfy the Subject Criterion (Rizzi & Shlonsky 2006). To prevent PPs to occur in specSubjP in English, Rizzi and Shlonsky (2006: 345) suggest that T moves to Subj. An XP moving to specSubjP must also be able to check the Nominative Case feature of T. Since PPs cannot check Nominative Case, they cannot occur in specSubjP. In order to explain why sentences like (i) are possible, we can hypothesize that in Italian, T does not move to Subj, which implies that an XP moving to specSubjP does not need to check Nominative Case. Nominative Case is checked by the postverbal subject.

On a (wh-)moved Topic in Italian, compared to Germanic 

If in (55), the locative PP does not occur in specSubjP, but in a higher position, which element occupies the subject position? Since English does not possess null subjects, there seems to be no suitable element to occur in this position. How is the EPP or, in Rizzi and Shlonsky’s (2006) terms, the Subject criterion satisfied? Providing the structure in (58), Rizzi and Shlonsky (2006: 347) suggest that the Subject criterion is satisfied indirectly: (58) [FinP in the roomi Fin+Loc [SubjP Subj [TP was [VP ti sitting my old brother]]]]

Fin endowed with a nominal feature (Fin+Loc in (58)) satisfies the Subject criterion by being in a local head-head relation with Subj. No preverbal subject is thus necessary to satisfy the Subject Criterion. The derivation proceeds as follows: The uninterpretable locative feature on Fin attracts the locative PP into its specifier. Since Fin is not a criterial head, however, “it does not assign any special interpretive property to its Spec”. Therefore, the locative PP must further move to a criterial position, either specTopicP, as in (58), or specFocusP, as in (59) (from Rizzi and Shlonsky 2006: 344): (59) a. [FocusP in what roomi [FinP ti [SubjP is ti sitting my old brother]]]? b. [FocusP In the living roomi [FinP ti [SubjP is ti sitting my old brother]]] (, not in the bedroom).

Italian is a pro-drop language. A null subject can occur in subject position and satisfy the EPP, as in (49b), as can an overt weak pronoun, (49c,d). But why is the preverbal full subject in (49a) impossible? Adopting the multiple subject structure in (4), this amounts to asking why specSubjP cannot be occupied.19 Suppose that the analysis of Locative Inversion suggested by Rizzi and Shlonsky (2006) is adopted for Italian RP. The structure in (54) can be specified as in (60): (60) [TopicP la stessa propostai [FinP ti Fin+nom [SubjP Subj [TP fece poi il partito di maggioranza ti]]]]

Fin endowed with a nominal feature (Fin+nom in (60)) satisfies the Subject criterion by being in a local head-head relation with Subj. A full subject need not occur in specSubjP to satisfy the Subject criterion and can remain postverbally. The derivation proceeds as follows: the nominal feature on Fin attracts the Resumptive DP into its specifier. The DP further moves to specTopicP.

19.  The postverbal position of the subject is not surprising per se because this is independently possible in Italian. What is surprising is that, differently from other instances of postverbal subjects, subject inversion in RP is obligatory and does not depend on information structure (see section 2).

 Anna Cardinaletti

7.  On some consequences of the analysis 7.1  Wh-questions a nd Focalization The analysis of RP in (60) can be extended to wh-questions and Focalization, which in Italian display similar restrictions against preverbal subjects. As we have seen in (50a), wh-questions do not allow a preverbal subject DP, and for many speakers, the same is true of Focalization (61):20 (61) *A Maria Gianni parlò, non a Sandra. to Maria Gianni spoke, not to Sandra

I suggest that like RP, wh-movement and Focalization proceed through specFinP, as shown in (62):21 (62) a. [FocusP a chii [FinP ti Fin [SubjP Subj [TP parlò Gianni ti]]]]? to whom spoke Gianni b. [FocusP a Mariai [FinP ti Fin [SubjP Subj [TP parlò Gianni ti]]]] (non a Piero) to Maria spoke Gianni not to Piero

Fin endowed with the features that attract wh- and Focus phrases licenses Subj. The subject does not need to move to specSubjP to satisfy the Subject Criterion. This proposal thus accounts for the generalization in (52)–(53) above. Notice that subject DPs are possible in long wh-movement and Focalization, where the embedded Fin does not contain the features to attract wh- and Focus phrases and cannot satisfy the Subject Criterion. The embedded specSubjP can thus be targeted by the subject DP: (63) a. [FocusP a chii [FinP ti Fin [SubjP Subj [TP credi [ForceP che [FinP [SubjP Gianni to whom [you] think that Gianni abbia già parlato ti]]]]]]]? has already spoken b. [FocusP a Mariai [FinP ti Fin [SubjP Subj [TP credo[ForceP che [FinP [SubjP Gianni to Maria [I] think that Gianni abbia già parlato ti]]]]]]] (non a Piero) has already spoken not to Piero

20.  The judgment in (61) is not shared by all Italian speakers. I will account for this variability of judgments in section 7.3 below. 21.  Differently from Germanic languages, no T-to-Fin movement takes place in Italian wh-questions and Focalization. See Cardinaletti (2007) for discussion and section 7.4 below.

On a (wh-)moved Topic in Italian, compared to Germanic 

7.2  On specFinP as an escape hatch I have suggested that in Italian, RP, wh-movement, and Focalization proceed through specFinP. The same analysis has been proposed by Haegeman (1996b: 145) for whmovement in e.g., English and German, as shown in (64): (64) a. [FocusP whoi [FinP ti have [TP you seen ti]]]? b. [FocusP weni [FinP ti hast [TP du gesehen ti]]]?

SpecFinP can be regarded as the escape hatch for all movements into the left periphery. This is rather natural in view of the fact that Fin is the interface head between the IP and the CP layers. In a cartographic perspective, Fin is the most suitable C head to share its features with Infl (Chomsky 2007, 2008). Furthermore, it can be taken to encode the features of the head(s) activated in the left periphery: It attracts the XPs endowed with the features to be checked in the criterial heads in the left periphery. This analysis has many welcome consequences. It explains why in Italian RP and Focalization, only one constituent can be fronted. It also explains why a RP and a wh-constituent cannot co-occur in either order (see (42c) and (i) in n.14) and similarly, why a RP and a Focus constituent cannot co-occur in either order. See (65a) (from Benincà 2001a: 159), and (65b) with the opposite order: (65) a. *Ai nostri colleghi, le stesse cose ha detto il presidente. to-the our colleagues the same things has said the president b. *Le stesse cose ai nostri colleghi ha detto il presidente. the same things to.the our colleagues has said the president

If these movements all target specFinP as an escape hatch, it follows that only one element per clause can be fronted. Notice that, since RP and wh-movement/ Focalization target distinct specifiers, specTopicP and specFocusP respectively, the ungrammaticality of (42c), (i) in n.14, and (65) cannot be taken to arise from mutual exclusion from one and the same structural position (while question operators and focalized constituents compete for the same position, specFocusP, Rizzi 1997: 298).22

22.  As Rizzi (1997: 330, fn.18) and (2001: 291) show, things are more complex in embedded clauses, in which Focalization and wh-movement can co-occur (in this order and with dativeaccusative ordering):  (i)

Mi domando a Gianni che   cosa abbiano detto (non a Piero). I wonder to Gianni what [they] have said (not to Piero) 

The fact that in embedded clauses, specFinP is not involved in the same way as in main clauses has some consequences for preverbal DP subjects, which are more acceptable than in main clauses especially if subjunctive mood is chosen:  (ii)

Mi domando chi Gianni ?ha / abbia visto ieri al cinema.  I wonder whom Gianni has / has.subj seen yesterday at the movie

 Anna Cardinaletti

Differently from what happens in the other constructions targeting the left periphery, Italian CLLD allows more than one left-peripheral Topic and can co-occur with the other left-peripheral constructions (see (25a) for the co-occurrence of CLLD and RP). If CLLD does not make use of specFinP as an escape hatch (see (39)), this distributional freedom follows straightforwardly.

7.3  On operators in specFinP So far we have discussed the left-peripheral constructions in which a preverbal subject in specSubjP is not possible. Preverbal subjects however occur in other constructions, as shown in (11b) for Italian CLLD and (70) below for English. The distribution of subjects in left-peripheral constructions is summarized in (66): (66)

– subject in specSubjP

+ subject in specSubjP

Italian wh-questions Italian Focalization English Locative Inversion Italian Resumptive Preposing

(English wh-questions, see sect. 7.4) English Focalization English Topicalization Italian CLLD

Notice that the occurrence of a preverbal subject DP does not correlate with the type of left-peripheral position targeted, in other words with the type of features involved. Compare Italian and English Focalization, which involve the same left-peripheral position but behave differently with respect to the occurrence of subjects, or English Locative Inversion and Focalization, which can target the same position (see (59b)) but behave differently in this respect. In Rizzi and Shlonsky’s (2006) proposal, Fin works as an expletive-like element for the licensing of Subj when it is enriched with the feature of the left-peripheral constituent it attracts. Otherwise, Fin is not able to satisfy the Subject Criterion. This situation is for instance found in English Topicalization, where, as we have seen in (37), an empty operator is merged in specFinP, Fin cannot license Subj, and a DP subject occurs preverbally. (67) seems to hold: (67) In the presence of an operator in specFinP, Fin cannot satisfy the Subject Criterion and a preverbal subject is possible.23

But what about English Focalization, where a subject can occur in specSubjP? A look at Topicalization and Focalization in German is useful to answer this question. As (68a) shows, a resumptive d-pronoun is present in German Left Dislocation. The existence

23.  A full subject need not be present. A weak subject in specTP also gives rise to a grammatical output. In this case, SubjP is presumably not projected.

On a (wh-)moved Topic in Italian, compared to Germanic 

of this construction has been taken as evidence to adopt an empty-operator analysis of German Topicalization as in (68b), similarly to English (37) (Koster 1978; Cardinaletti 1984, 1986). The updated structures in (68) are taken from Haegeman (1996b: 146). Since a d-pronoun can also co-occur with a focalized constituent, (69a), the empty counterpart of the d-pronoun can be taken to also be present in German Focalization, as shown in (69b): (68) a. [TopicP die Kinderi [FinP diei hat [SubjP Hans [TP gesehen ti]]]] b. [TopicP die Kinderi [FinP OPi hat [SubjP Hans [TP gesehen ti]]]] (69) a. [FocusP die Kinderi [FinP diei hat [SubjP Hans [TP gesehen ti]]]], the children them has Hans seen nicht die Eltern. not the parents b. [FocusP die Kinderi [FinP OPi hat [SubjP Hans [TP gesehen ti]]]], nicht die Eltern.

Suppose that this is a general property of Germanic languages, and the same can be assumed for English. A structure containing a null operator is present not only in Topicalization, as in (37) and (70a), but also in Focalization, (70b) (but see n.25 for a refinement). In Focalization, a quantificational operator binds a variable in the argumental position, to account for the WCO effects found in English Focalization in contrast with Topicalization (Culicover 1992): (70) a. [TopicP this booki [FinP OPi [SubjP John should give nci to Paul]]] b. [FocusP this booki [FinP OPi [SubjP John should give vbli to Paul]]](not mine)

Since in (70), Fin does not contain the features of the left-peripheral constituents, it cannot license Subj. Preverbal subjects in specSubjP can co-occur with both topicalized and focalized left-peripheral constituents, (70a) and (70b), respectively. Things are however different in Italian, where a resumptive clitic pronoun is present in CLLD but not in Focalization, (71a,b) vs. (71c,d):24 (71) a. Gianni, Maria l’ha visto. c. *Gianni l’ha visto. Gianni Maria him has seen Gianni [she] him has seen b. *Gianni, Maria ha visto.


Gianni ha visto.

24.  That in Italian, clitic pronouns are not optional is also shown by the behaviour of constructions activating the right periphery (Cardinaletti 2001, 2002). When an anticipatory clitic pronoun is present, as in (i), we are dealing with Right Dislocation; sentence (ii) without a clitic pronoun is an instance of a different syntactic construction, namely Marginalization: (i)

Maria l’ha visto, Gianni. vs. (ii) Maria him has seen, Gianni

Maria ha visto, Gianni. Maria has seen, Gianni

 Anna Cardinaletti

As we have shown in (39), in (71a) no resumptive element occurs in specFinP, and specFinP is not involved. A preverbal subject in specSubjP is thus possible. The distribution of the resumptive clitic does not allow us to conclude anything about the presence of an empty resumptive operator in (71d). Suppose that in Italian, both analyses made available by Universal Grammar are possible:  either movement of the focussed element to SpecFocusP through specFinP, (72a), as I have assumed in (62b), or presence of an empty operator in specFinP, (72b), as in German (69b) and English (70b): (72) a. [FocusP Giannii [FinP ti Fin [SubjP [TP pro ho visto vbli]]]] (non Paolo) b. [FocusP Giannii [FinP OPi Fin [SubjP [TP pro ho visto vbli]]]] (non Paolo) Gianni [I] have seen not Paolo

This dual analysis can offer an account for the non-uniform judgments of the co-occurrence of preverbal subjects and focalized constituents mentioned in n.20. Italian speakers can either assume an analysis without an empty operator, as in (72a), which forbids a preverbal subject as shown in (61), or a structure like (72b), which is compatible with the presence of a preverbal subject (as said in n.20, some speakers find (61) grammatical).25 Judgments of wh-questions like (50a) with preverbal subjects are instead shared by all speakers of Italian:  a full subject cannot occur between the wh-constituent and the verb because an analysis like (72b) is not available for wh-questions. If the quantificational operator in specFinP binds the variable in object position, the whphrase, which is an operator, would have no variable to bind, and a violation of Full Interpretation arises (73). For the same reason, the analysis in (74) is ungrammatical for English and German wh-questions: (73) *Chii, OPi whom

hai visto vbli? [you] have seen

(74) a. *Whoi, OPi have you seen vbli? b. *Weni, OPi hast du gesehen vbli?

25.  Since both structures are quantificational, no different semantics is expected. A dual analysis could be hypothesized for English as well. The analysis with the quantificational operator assumed in (70b) accounts for the possibility of preverbal subjects. The analysis with movement into specFocusP, as in (i),


[FocusP this booki [FinP ti [SubjP I should give vbli to Paul]]] (not mine)

accounts for the differences between Focalization and Topicalization as far as subjacency and anti-adjacency effects are concerned (see Rizzi 1997: 329,n.11, 332,n.26 for discussion). On the basis of these differences, Rizzi suggests that English Focalization structurally contrasts with English Topicalization in not containing a null Operator.

On a (wh-)moved Topic in Italian, compared to Germanic 

In conclusion, we have seen that all left-peripheral constructions except Italian CLLD involve specFinP: this position is either occupied by the trace of the fronted element or by an empty operator identified by the fronted constituent. This analysis not only explains the distribution of subjects in left-peripheral constructions. It also explains why English and German have unique Topicalization and Focalization and display no co-occurrence of Topicalization, Focalization, and wh-movement.26

7.4  The interaction with T-to-C movement Consider now the distribution of subjects in constructions in which T-to-C movement occurs. In a split C perspective, T-to-C movement is T-to-Subj-to-Fin. Following Rizzi and Shlonsky (2006: 349), the finite verb moves because it is attracted by “verbal” Fin. This happens in V/2 languages like German (75a) and in English residual V/2 contexts, namely wh-questions (75b) and Negative Inversion (75c) (Rizzi 1996; Haegeman 1996a). The exact nature of the cases where verbal Fin occurs remains to be established. What is clear, however, is that in these cases, a DP subject can freely appear in SpecSubjP:27 (75) a. [FocusP den Johanni [FinP OPi hat [SubjP Hans gesehen ti]]] the-acc Johann has Hans seen b. [FocusP whoi [FinP ti did [SubjP John see ti]]]? c. [FocusP with no jobi [FinP ti would [SubjP John be happy ti]]]

Verbal Fin cannot license the Subj head because this has been attracted to Fin itself. A preverbal subject is thus needed to satisfy the Subject Criterion. In verbal Fin contexts, the subject position can always be occupied by a subject DP.28 The proposed

26.  In German, a left-dislocated constituent can precede a wh-phrase if the d-pronoun is not moved to specFinP but occurs in the Middle Field. See (ia), taken from Cardinaletti (1984: 112), whose structure is provided in (ib): (i) a.

Den Karl, wer hat den gesehen?  the Karl, who has him seen

[TopicP den Karl [FocusP weri [FinP ti hat [SubjP ti den gesehen]]]]


27.  Compare (75c) with (i), where the negative phrase is topicalized and no T-to-Fin movement applies (see Haegeman 1996a):


[TopicP with no jobi [FinP OPi [SubjP John would be happy ti]]]

Given the operator status of the negative phrase in (75c), the following analysis with an empty operator in specFinP cannot be correct (see (74)):


*[FocusP with no jobi [FinP OPi would [SubjP John be happy vbli]]]

28.  A full subject need not be present. A weak subject in specTP also gives rise to a grammatical output. In this case, SubjP is presumably not projected. This situation can be found in V/2 languages like German and Icelandic which allow expletive pro and middle-field subjects.

 Anna Cardinaletti

analysis thus captures the other generalization arrived at in Cardinaletti (2007) for wh-questions: (76) Subjects in specSubjP in wh-questions are possible in languages with T-to-C movement.

As we have seen in section 7.3, “non-verbal” Fin is instead compatible with both scenarios: SpecSubjP can be filled or not by a subject DP.

7.5  Summary of the interaction of Fin and Subj The following table summarises the interaction of the two clausal heads Fin and Subj, which lie at the interface between the IP and CP layers: (77)

+ “Fin satisfies the Subject Criterion” (– subject in specSubjP)

– “Fin satisfies the Subject Criterion” (+ subject in specSubjP)

+ T-to-Fin movement (verbal Fin) – T-to-Fin movement (non-verbal Fin)

not found

V/2 languages English wh-questions English Negative Inversion Italian Focalization (some sp.) English Focalization English Topicalization Italian CLLD

Italian wh-questions Italian Focalization English Locative Inversion Italian Resumptive Preposing

7.6  On German and Italian Topics The analysis in (68b) adopts the traditional account according to which German Topicalization involves an empty operator. Given the approach in Rizzi (1997: 292f), this analysis is forced by the hypothesis that null constants cannot be directly bound by Topic phrases. But if this restriction does not hold, what I said is also compatible with Topic movement through specFinP. In what follows, I will not enter the debate on whether German Topicalization (and Left Dislocation) is derived by movement to or merge in Topic position (see for instance the papers in Anagnostopoulou et al., eds., 1997 and Grewendorf 2002). I will only discuss the contrast in (78) on the basis of which Frascarelli and Hinterhölzl (2007) conclude that German topics must be derived through Movement, while Italian CLLD constituents are inserted in specTopicP: (78) a. La mia foto con Leoi luii non l’ha ancora mostrata. the my picture with Leo he not it has yet shown ‘My picture with Leoi, hei hasn’t shown it yet.’ b. *Mein Bild von Leoi hat eri wahrscheinlich Maria nicht gezeigt. my picture of Leo has he probably Maria not shown ‘My picture of Leoi, hei has probably not shown to Maria.’

On a (wh-)moved Topic in Italian, compared to Germanic 

The topic in (78a) cannot be derived via movement from the IP-internal position since reconstruction would yield a Principle C effect as in (79a). If Italian CLLD involves merger in specTopicP, Principle C is not violated because the DP Leo is not c-commanded by the subject pronoun lui (cf. Cinque 1990; Frascarelli 2004 for further arguments). Unlike Italian, the German topic in (78b) shows Principle C effects, which can be explained in terms of reconstruction into the object position in (79b): (79) a. *Luii non ha ancora mostrato la mia foto con Leoi. b. *Eri hat Maria mein Bild von Leoi nicht gezeigt.

Notice that the piece of evidence in (78) is not conclusive. There are two important, independent differences between (78a) and (78b). First, the PP inside the left-dislocated DP is headed by con (with) in Italian and von (of) in German. The Italian counterpart of von also produces an ungrammatical sentence: (80) *La mia foto di Leoi luii non l’ha ancora mostrata. the my picture of Leo he not it has yet shown

The difference in (78) is related to the different behaviour of PPs headed by functional vs. lexical prepositions with respect to binding and reconstruction. Since this would take us too far afield, it will not be discussed here. The second difference is that in Italian (78a), the strong pronoun lui is used, while in German (78b) the weak pronoun er is found. If the weak null subject is used instead of lui, as in (81a), an ungrammatical sentence is obtained as in German (81b) (where the lexical preposition mit ‘with’ is used instead of von to control for the above-mentioned property). Conversely, if a strong pronoun is used in German, as in (81c), a grammatical sentence is produced, like Italian (78a): (81) a. *La mia foto con Leoi, proi non l’ha ancora mostrata. b. *Mein Bild mit Leoi hat eri wahrscheinlich Maria nicht gezeigt. c. Mein Bild mit Leoi hat eri/deri wahrscheinlich Maria nicht gezeigt.

The grammaticality of (78a) crucially depends on the presence of the strong pronoun lui. Whatever the analysis of the usage of the null subject pro vs. the strong pronoun lui in Italian (see Frascarelli 2007 for a recent discussion), it must be extended to the usage of weak er vs. strong er in German (see Cardinaletti & Starke 1996). The important observation is that pro in (81a) behaves like weak er in (81b), and strong er/der in (81c) behave like strong lui in (78a). In conclusion, the contrast in (78) does not seem to be due to the way the left-peripheral constituent is merged in the two languages. For the reasons mentioned before, I adopt an operator/null constant analysis of German Topicalization as in (68b).

 Anna Cardinaletti

8.  Two further restrictions on the subject in RP In this section, two further restrictions on the subject in RP are discussed: the postverbal subject must be heavy and cannot be marginalized. The first property is shared by English Locative Inversion discussed in section 8.1 and Comparative Inversion briefly introduced in section 8.2.

8.1  Heavy NP shift English Locative Inversion displays a further restriction on the subject: Not only must the subject occur in postverbal position; with unergatives, it must also be heavy. See (82), taken from Culicover and Levine (2001), Rizzi and Shlonsky (2006: 351).29 This property is shared by Italian RP, which always involves transitive and intransitive verbs. All the RP examples discussed so far contain heavy subjects. A light postverbal subject as in (83a) is impossible; (83b) shows that CLLD does not display the same restriction: (82) a. Into the room walked Robin carefully. b. *In the room slept Robin fitfully. c. In the room slept __ fitfully [the students in the class who has heard about the social psych experiment that we were about to perpetrate]. (83) a. *La stessa proposta fece poi Gianni / lui. b. La stessa proposta la fece Gianni / lui. the same proposal (it) made then Gianni / he

The heaviness of the subject correlates in Italian with the peculiar intonation normally associated to RP, as we have seen in section 2. Rizzi and Shlonsky (2006: 352f) account for (82b) as follows: Assuming the structure in (84a), where Loc is analysed as belonging to the class of phi-features, attraction of the locative to specFinP is blocked by Relativized Minimality: the thematic subject Robin intervenes. The grammatical (82c) undergoes the derivation in (84b), where Heavy NP Shift removes the subject from the intervening position. The subject moves to specTP to check phi-features, and then, it is right-adjoined to FinP: (84) a. Fin+Loc Subj T+Phi [Robin v [sleep in the room fitfully]] b. Fin+Loc Subj t′DP T+Phi [tDP v [sleep in the room fitfully]]

29.  Jacqueline Guéron (p.c.) does not detect any difference between (82a) and (82b). She requires a heavy subject in (82a) as well. I have nothing to say about the English-internal variation indicated by the different judgments of (82a).

On a (wh-)moved Topic in Italian, compared to Germanic 

Rizzi and Shlonsky adopt a theory of locality in which whole chains and not just positions are relevant for the calculation of locality. The traces of the heavy DP in (84b) are only part of the chain of the shifted phrase. Suppose that something similar happens in Italian RP. Heavy NP Shift removes the subject from the intervening position, and Fin can attract the RP constituent, as we have assumed in (60). To account for (49b–d), in which a preverbal weak pronoun occurs in Italian RP, consider the possibility that weak pronouns are in chain with empty DPs in the left periphery of the clause (as proposed for null subjects by Frascarelli 2007 and for clitic pronouns by Belletti 2008). Then, only part of the chain of the weak pronoun raised to specTP intervenes between specFin and the RP constituent, and no violation of locality is produced, as expected. The analysis can be easily extended to wh-questions and Focalization, where Heavy NP Shift of the subject is also possible: (85) a.

Con chi parlò [l’amico inglese di Gianni che vive in Italia with whom spoke   the friend English of Gianni who lives in Italy

da tanti anni]? since many years? b. Con Maria parlò [l’amico inglese di Gianni che vive with whom spoke   the friend English of Gianni who lives

in Italia …] in Italy …

The proposal developed so far can account for one further property of RP. While subject extraction is possible across the Topic of CLLD (86a) (see Rizzi 1997: 306), it is impossible across the topic of RP (86b), as is the case in English Locative Inversion, (86c) (from Stowell 1981): (86) a. Chi   pro credi [che questa proposta, l’abbia fatta all’ultimo who [you] think   that this proposal it has made at the last congresso]? meeting b. *Chi   pro credi [che la stessa proposta abbia fatto all’ultimo who [you] think   that the same proposal has made at the last congresso]? meeting c. *What does John say [that near his house lies t]?

Since in Italian, the postverbal subject position can be the extraction site of whmovement (Rizzi 1982: Ch.4, 1990: 62ff), the ungrammaticality of (86b) confirms that in RP, the postverbal subject does not occur in the usual postverbal subject position. The ungrammaticality of (86b) must derive from a restriction on movement

 Anna Cardinaletti

out of the position targeted by Heavy NP shift, presumably linked to the fact that the wh-phrase is not heavy.30 Notice finally that both English Locative Inversion with heavy subjects (87a, from Rizzi & Shlonsky 2006: 356) and Italian RP (87b) are possible in raising structures: (87) a.

Into the room appeared to be walking a very large caterpillar.

b. La stessa proposta sembrò aver fatto il partito di maggioranza. the same proposal seemed [to] have made the party of majority

In (87), the thematic subject is moved by Heavy NP shift, while the locative PP and the RP constituent, respectively, are attracted by matrix Fin.

8.2  Another piece of comparison: English Comparative Inversion English displays another inversion construction that shares some properties with Locative Inversion and RP, namely a special instance of Comparative Inversion. The constituent in first position is a topic which establishes a similarity with some element mentioned previously, as does RP with words like stesso (same), simile (similar), etc. Leech and Svartvik (1975: 179), from which the example is taken, comment on (88) as follows:  “Occasionally subject-verb inversion occurs with a complement as topic when the complement expresses a comparison”. What they call a “complement” is the predicative AP equally strange: (88) For a long time, he refused to talk to his wife, and kept her in ignorance of his troubles. Equally strange was his behavior to his son.

Leech and Svartvik (1975: 178) use the term subject-verb inversion also to refer to what we have called Locative Inversion and observe that the two constructions belong to the same literary style. They claim that in subject-verb inversion, “the fronted topic is […]

30.  A similar restriction on wh-movement is found with locative and dative PP fronting in Italian (see n.18) and inverse copular sentences: (i) a. *Cosa su Gianni è caduto ieri? what on Gianni has fallen yesterday? b. *Chi la causa della rivolta è stato? who the cause of the riot has been? Notice that inverse copular sentences have certain similarities with RP in that they involve raising of (a part of) a predicate to the status of topic and are root structures in English (Guéron 1994). In Italian, however, they are non-root and pattern with locative and dative PP fronting (Cardinaletti 2004: 125).

On a (wh-)moved Topic in Italian, compared to Germanic 

useful in giving end-weight to a long subject”.31 The parallelism we have established between Italian RP and subject-verb inversion constructions in English is confirmed by their common syntactic (postverbal heavy subjects) and pragmatic (non-subject topics) peculiarities.

8.3  Marginalization Differently from what we have seen in (86b), subject extraction is possible out of Focalization sentences (89a), as is from wh-islands (89b, from Rizzi 1990: 62): (89) a. Chi   credi [che questa proposta abbia fatto who [you] think   that this proposal has.subj made all’ultimo congresso] (non quella)? at the last meeting, (not that [one]) b. L’uomo che non so [che cosa abbia detto] … the man that [I] not know   what has.subj said

In Focalization and wh-questions, a postverbal subject can be heavy (see (85)), but differently from RP, it does not need to. Postverbal DP subjects can be light, and in this case they are marginalized, i.e., they are topics (see section 2) that appear destressed at the end of the clause (Antinucci & Cinque 1977; Cardinaletti 2001, 2002): (90) a. Questo libro ha scritto, Gianni / lui, non quello. this book has written, Gianni / he, not that [one] b. Che cosa ha scritto, Gianni / lui? what has written, Gianni / he?

(89) and (90) show that marginalized subjects occupy a position from which extraction is possible, either their base position (Cardinaletti 2001: 122) or a Topic position above VP (Belletti 2004). In wh-questions and Focalization, Marginalization can be taken to have the same function as Heavy NP Shift with RP. If the subject is not marginalized, it would be interpreted as focused and would intervene between the fronted focus element and its first merge position. Notice that while a heavy subject can occur in all types of sentences that we are discussing (section 8.1), a marginalized subject is possible in Focalization and wh-phrases, but not in RP. In a RP sentence like (91), the destressed light subject in

31.  This construction is different from so-inversion, as in (i), which belongs to ordinary English and does not display a restriction against light subjects. The postverbal subject can be a (stressed) pronoun and gets end-focus (Leech & Svartvik 1975: 179, Merchant 2003: 63): (i)

a. b.

I’ve seen the play. So have I. I enjoyed the play and so did my friends.

 Anna Cardinaletti

postverbal position is right-dislocated, not marginalized (and pro occurs in preverbal position, see (49b)). This analysis of (91) is confirmed by the fact that a negative quantifier, which can occur in wh-questions (92a), is not possible in RP, (92b): (91) La stessa proposta   pro ripeté, Gianni, non senza qualche esitazione. the same proposal [he] repeated, Gianni, not without some hesitation (92) a. Dove non è andato nessuno? where not has gone anybody? b. *La stessa proposta   pro non ripeté, nessuno, neanche the same proposal [he] not repeated, nobody, not.even per far piacere a Maria. to please Maria

A marginalized negative quantifier, which occurs in an IP-internal position, is c-commanded by negation non, as required, (92a). This is not possible in (92b). If (92b) is an instance of Right Dislocation, as we have hypothesized, the clause-final subject lies outside of the clause, hence outside of the scope of non (Cardinaletti 2001, 2007). A negative quantifier is therefore ungrammatical. Why are RP and Marginalization incompatible? As argued in section 2, RP makes a new-information subject belong to the comment of the clause, while Marginalization does the opposite, leave a topic subject to the right. The two constructions thus imply contradictory features on the postverbal subject.

9.  Conclusions In this paper, Italian RP has been analysed in detail. This construction is distinct from both Italian CLLD and Focalization and shares a number of restrictions with English Topicalization, Locative Inversion, and a special instance of Comparative Inversion. To account for the restriction to root contexts, I have proposed that RP involves the highest Topic position only licensed in root clauses. Italian CLLD qua Familiar Topicalization instead targets the lower Topic position. The restriction against preverbal DP subjects can be accounted for if we assume that the fronted XP moves through specFinP and in so doing it satisfies the Subject Criterion along the lines of Rizzi and Shlonsky’s (2006) analysis of English Locative inversion. This movement also takes place in wh-questions and Focalization, but not in CLLD. The restriction against postverbal light subjects can be understood as a Minimality effect. While in English only Heavy NP Shift has the function of removing the subject from the intervening position, Italian can resort to both Heavy NP Shift and a different mechanism in the case of light subjects, namely Marginalization. A marginalized subject can occur in wh-questions and Focalization, but not in RP.

On a (wh-)moved Topic in Italian, compared to Germanic 

As we have seen, the grammatical “ingredients” made available by UG appear in different combinations in the fronting constructions found in Italian and English/German. While the availability of CLLD crucially depends on the existence of a specific class of lexical items, namely clitic pronouns, the other constructions are potentially available in all languages. Which ones are indeed instantiated in each particular language, and how the different combinations of grammatical options are chosen remain to be established on the basis of a detailed comparative study which I hope to address in future work.

References Aboh, E.O. 2007. Information Structuring begins with Numeration. Ms, Universiteit van Amsterdam. Anagnostopoulou, E., van Riemsdijk, H. & Zwarts F. (Eds) 1997. Materials on Left Dislocation. Amsterdam: John Benjamins. Antinucci, F. & Cinque, G. 1977. Sull’ordine delle parole in italiano: l’emarginazione. Studi di grammatica italiana 6: 121–146. Authier, J.-M. 1992. Iterated CPs and embedded topicalisation. Linguistic Inquiry 23: 329–336. Bayer, J. 2001. Asymmetry in emphatic topicalization. In Audiatur Vox Sapientiae, C. Féry & W. Sternefeld (Eds), 15–47. Berlin: Akademie Verlag. Belletti, A. 2001. Inversion as Focalization. In Subject Inversion in Romance and the Theory of Universal Grammar, A. Hulk & J.-Y. Pollock (Eds), 60–90. Oxford: OUP. Belletti, A. 2004. Aspects of the low IP area. In Rizzi (Ed.), 16–51. Belletti, A. 2008. Properties of the edge of the (small) clause. Paper presented at the Incontro di grammatica generativa, Padua, February 2008. Benincà, P. 2001a. L’ ordine degli elementi della frase e le costruzioni marcate. In Grande grammatica italiana di consultazione, L. Renzi, G. Salvi & A. Cardinaletti (Eds), Nuova ed., Vol. 1, 129–208. Bologna: il Mulino. Benincà, P. 2001b. The position of topic and focus in the left periphery. In Cinque & Salvi (Eds), 39–64. Benincà, P. & Poletto C. 2004. Topic, focus, and V2:  Defining the CP sublayers. In Rizzi (Ed.), 52–75. Cardinaletti, A. 1984. Die Syntax der linken Peripherie des deutschen Satzes. Tesi di Laurea, University of Venice. Cardinaletti, A. 1986. Topicalization in German:  Movement to Comp or base-generation in TOP? Groninger Arbeiten zur germanistischen Linguistik 28: 202–231. Cardinaletti, A. 1997. Subjects and clause structure. In The New Comparative Syntax, L. Haegeman (Ed.), 33–63. London: Longman. Cardinaletti, A. 2001. A second thought on emarginazione: Destressing vs. ‘Right Dislocation’. In Cinque & Salvi (Eds), 117–135. Cardinaletti, A. 2002. Against optional and zero clitics. Right dislocation vs. marginalization. Studia Linguistica 56: 29–57. Cardinaletti, A. 2004. Toward a cartography of subject positions. In Rizzi (Ed.) 115–165.

 Anna Cardinaletti Cardinaletti, A. 2007. Subjects and wh-questions. Some new generalizations. In Romance Linguistics 2006:  Selected papers from the 36th Linguistic Symposium on Romance Languages (LSRL), Rutgers University, New Brunswick, N.J., March 31-April 2, 2006, M.J. Cabrera, J.  Camacho, V.  Déprez, N.  Flores Ferran & L.  Sanchez (Eds), 57–78. Amsterdam: John Benjamins. Cardinaletti, A. & Starke, M. 1996. Deficient pronouns:  A view from Germanic. A study in the unified description of Germanic and Romance. In Studies in Comparative Germanic Syntax, Vol. II, H. Thráinsson, S.D. Epstein & S. Peter (Eds), 21–65. Dordrecht: Kluwer. Chinellato, P. 2003. The recovery of subject clitics in mild agrammatism: A generative approach to treatment. Rivista di grammatica generativa 28: 31–44. Chomsky, N. 1977. On wh-movement. In Formal Syntax, P.W. Culicover, T. Wasow, & A. Akmajian (Eds), 71–132. New York NY: Academic Press. Chomsky, N. 2007. Approaching UG from below. In Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax-Semantics, H.M. Gärtner, & U. Sauerland (Eds), 1–10. Berlin: Mouton de Gruyter. Chomsky, N. 2008. On phases. In Foundational Issues in Linguistic Theory. Essays in Honor of Jean-Roger Vergnaud, R.  Freidin, C.P. Otero & M.L. Zubizarreta (Eds), 133–166. Cambridge MA: The MIT Press. Cinque, G. 1983. ‘Topic’ constructions in some European languages and ‘connectedness’. In Connectedness in Sentence, Discourse and Text, K.  Ehlich & H.  van Riemsdijk (Eds), Tilburg: KUB. Reprinted in Anagnostopoulou et al. (Eds), 93–118. Cinque, G. 1990. Types of A’ Dependencies. Cambridge MA: The MIT Press. Cinque, G. & Salvi, G. (eds) 2001. Current Studies in Italian Syntax. Essays Offered to Lorenzo Renzi. Amsterdam: Elsevier. Cruschina, S. 2006. Informational Focus in Sicilian and the left periphery. In Frascarelli (Ed.), 363–385. Culicover P.W. 1992. Topicalisation, inversion and complementizers in English. OTS Working Papers, Going Romance and Beyond, D. Delfitto et al. (Eds), Utrecht: University of Utrecht. Culicover P.W. & Levine, R.D. 2001. Stylistic inversion in English: A reconsideration. Natural Language and Linguistic Theory 19(2): 283–310. Frascarelli, M. 2004. Dislocation, clitic resumption and minimality:  A comparative analysis of left and right Topic constructions in Italian. In Romance Languages and Linguistic Theory, R.  Bok-Bennema, B.  Hollebrandse, B.  Kampers-Manhe & P.  Sleeman (Eds), 98–118. Amsterdam: John Benjamins. Frascarelli M. (Ed.) 2006. Phases of interpretation. Berlin: Mouton. Frascarelli M. 2007. Subjects, topics and the interpretation of referential pro. An interface approach to the linking of (null) pronouns. Natural Language and Linguistic Theory 25: 691–734. Frascarelli M. & Hinterhölzl, R. 2007. Types of topics in German and Italian. In On Information Structure, Meaning and Form, S. Winkler & K. Schwabe (Eds), 87–116. Amsterdam: John Benjamins. Giorgi A. & Pianesi, F. 2004. Credo (I BELIEVE): Epistemicity and the syntactic representation of the speaker. Ms., University of Venice and ITC, IRST, Trento. Givón, T. 1983. Topic Continuity in Discourse: An Introduction. In Topic Continuity in Discourse: A Quantitative Crosslanguage Study, T. Givón (Ed.), 5–41. Amsterdam: John Benjamins. Grewendorf, G. 2002. Left dislocation as movement. Georgetown University Working Papers in Theoretical Linguistics 2: 31–81.

On a (wh-)moved Topic in Italian, compared to Germanic 

Guéron, J. 1994. Beyond predication: The inverse copula construction in English. In Paths Towards Universal Grammar. Studies in Honor of Richard S. Kayne, G. Cinque, J. Koster, J.-Y-  Pollock, L.  Rizzi & R.  Zanuttini (Eds), 173–187. Washington DC:  Georgetown University Press. Haegeman, L. 1996a. Negative inversion, the NEG-criterion, and the structure of CP. GenGenP 4.2: 93–119. Haegeman, L. 1996b. Null subjects and root infinitives in the child grammar of Dutch. GenGenP 4.2: 133–175. Haegeman, L. 2002. Anchoring to speaker, adverbial clauses and the structure of CP. Georgetown University Working Papers in Theoretical Linguistics 2: 117–180. Haegeman, L. 2006. Argument fronting in English, Romance CLLD, and the left periphery. In Negation, Tense, and Clausal Architecture: Cross-Linguistic Investigations, R. Zanuttini, H.  Campos, E.  Herburger & P.  Portner (Eds), 27–52. Washington DC:  Georgetown University Press. Haegeman, L. 2008. The internal syntax of adverbial clauses. To appear in Lingua. Koster, J. 1978. Locality Principles in Syntax. Dordrecht: Foris. Jensen, B. 2007. In favour of a truncated imperative clause structure: Evidence from adverbs. Working Papers in Scandinavian Syntax 80: 163–185. Lambrecht, K. 1994. Information Structure and Sentence Form: Topic, Focus and the Mental Representations of Discourse Referents. Cambridge: CUP. Lasnik, H. & Saito, M. 1992. Move α. Cambridge MA: The MIT Press. Lasnik, H. & Stowell, T. 1991. Weakest cross-over. Linguistic Inquiry 19: 335–391. Leech, G. & Svartvik, J. 1975. A Communicative Grammar of English. London: Longman. Menza, S. 2006. Il paraverbo. L’interiezione come sottoclasse del verbo. Alessandria: Edizioni dell’Orso. Merchant, J. 2003. Subject-auxiliary inversion in comparatives and PF output constraints. In The Interfaces:  Deriving and Interpreting Omitted Structures, K. Schwabe & S.  Winker (Eds), 55–77. Amsterdam: John Benjamins. Ordóñez F. & Olarrea, A. 2006. Microvariation in Caribbean/non Caribbean Spanish interrogatives. Probus 18: 59–96. Poletto C. & Pollock, J.-Y. 2004. On the left periphery of some romance Wh-questions. In Rizzi (Ed.), 251–296. Reinhart, T. 1981. Pragmatics and linguistics:  An analysis of sentence topics. Philosophica 27: 53–94. Rizzi, L. 1982. Issues in Italian Syntax. Dordrecht: Foris. Rizzi, L. 1990. Relativized Minimality. Cambridge MA: The MIT Press. Rizzi, L. 1996. Residual verb-second and the Wh-criterion. In Parameters and Functional Heads. Essays in Comparative Syntax, A. Belletti & L. Rizzi (Eds), 63–90. Oxford: OUP. Rizzi, L. 1997. The fine structure of the left periphery. In Elements of Grammar, L. Haegeman (Ed.), 281–337. Dordrecht: Kluwer. Rizzi, L. 2001. On the position ‘Int(errogative)’ in the left periphery of the clause. In Cinque & Salvi (Eds), 287–296. Rizzi, L. 2004. Locality and left periphery. In Structures and Beyond, A. Belletti (Ed.), 223–251. Oxford: OUP. Rizzi L. 2006. On the form of chains:  Criterial positions and ECP effects. In Wh-Movement: Moving On, L. Cheng & N. Corver (Eds), 97–133. Cambridge MA: The MIT Press.

 Anna Cardinaletti Rizzi, L. (Ed.) 2004. The Structure of CP and IP. The Cartography of Syntactic Structures, Vol. 2. Oxford: OUP. Rizzi L. & Shlonsky, U. 2006. Satisfying the Subject Criterion by a non subject: English locative inversion and heavy NP shift. In Frascarelli (Ed.), 341–361. Stowell, T. 1981. Origins of Phrase Structure. Ph.D. Dissertation, MIT. Strawson, P. 1964. Identifying reference and truth values. Theoria 30: 96–118.

C-agreement or something close to it Some thoughts on the ‘alls-construction’* Michael T. Putnam & Marjo van Koppen Carson-Newman College/University of Utrecht

In this paper we sketch out an account for an until now undiscussed phenomenon in generative syntax, namely the so-called “alls-construction” in Midwestern American English. In this construction, an s-ending is added to all under certain circumstances. We compare and contrast this construction with complementizer agreement in West Germanic. The alls-construction is similar to complementizer agreement in that the s-ending on all, just like the inflection on the complementizer in West Germanic, is sensitive to the agreement features on the embedded subject. Contrasted with complementizer agreement, however, the alls-construction does not allow inflectional morphology to appear on any other constituent than all. Furthermore, inflection on all is only possible when all is introducing an all-pseudo-cleft. We will mainly focus on the construction internal restrictions of the inflection on all. Based on Van Craenenbroeck & Van Koppen (2002), we assert that the alls-construction in Midwestern American English is in structure quite similar to complementizer agreement in West Germanic. We come back to the external restriction on the alls-construction in the final section, where we briefly discuss some issues concerning the pseudo-cleft status of the alls-construction.

1.  Introduction Complementizer agreement (henceforth C-agreement) is widely attested in West Germanic languages and dialects (cf. a.o. Haegeman 1992; Zwart 1993; Carstens 2002). An example from a dialect of Dutch is provided in (1).

*We would like to extend a special thanks to Jim Hartman for his insightful suggestions and critiques of this work. We would also like to thank the audiences at Baylor University, the University of Michigan Syntax Support Group and CGSW 21 for their comments, especially Christopher Becker, Vicki Carstens, Catherine Fortin, Hamid Ouali, Ian Roberts, Peter Svenonius, John te Velde and Jan-Wouter Zwart. We would also like to thank Jeroen van Craenenbroeck, Doris English, Richard Gray, Kleanthes Grohmann, William Keel, Sara Rosen and Thomas Stroik for their assistance and suggestions on previous drafts of this work. Lastly, we are grateful for the comments from an anonymous reviewer whose remarks undoubtedly strengthened our arguments. All remaining shortcomings are our own.

 Michael T. Putnam & Marjo van Koppen

(1) Datt-e we toffe jongens zijn, dat wille me wete. that-infl we great boys are that want we know ‘That we are great boys, that is what we want to know.’

Rotterdam Dutch

In the example (1) the complementizer dat ‘that’ bears inflectional morphology that agrees with the φ-feature set of the subject jongens ‘boys’.1 This form of C-agreement is assumed to be restricted to SOV-languages (cf. Zwart 1993). However, in Midwest vernacular variants of American English (an SVO-language) a construction resembling C-agreement is attested which we label: the alls-construction.

(2) All-s Jim and Carol want to do is stay at home.

(3) All-s I know about Mike and Dena is that they are old.

Midwest AE

As illustrated in (2) and (3) above, all exhibits the inflectional marker -s. Historically, it appears that the all-construction hails from Early Modern English “all as”:2

(4) [all as] → [alls]

The s-ending on all can therefore be argued to be a reduced complementizer on historical grounds. Evidence shows, however, that the s-ending on all is much more than just a phonological reduction. Much like C-agreement phenomena in other

1.  However, see Fuß (2005, 2007) for the argument that the Probe-Goal relation activating C-agreement morphology agrees with the entire φ-set of T rather than just the subject alone. 2.  Carter (1992) remarks that the “all as” contraction to “alls” appears “…to be similar to ‘there’s some as …’ for ‘there are some who …’” which is a substandard form found in his home dialect of East Kent English and “presumably influenced by Scandinavian where som can appear and function as a relative pronoun and the equivalent of ‘as’”. Teaman’s (1992) remark that the alls-construction is a “frozen idiom without any uncontracted variant” is also worthy of mention. For example, it is impossible to create sentences with the all as construction in place of alls:

a. All-s I know about cars I learned from my grandpa. b. *All as I know about cars I learned from my grandpa.

Following up the observations of Carter and Teaman, Liberman (2004) draws attention to the fact that alls could potentially be a colloquial form influenced by the German word alles ‘everything’ and that there exists certain dialects of English where as can take the place of the complementizer that (although based on data obtained throughout this study, that does not appear to be the case for the dialect of Midwestern American English that we analysed in this paper). As demonstrated, an in-depth historical analysis of the alls-construction would deviate greatly from the synchronic analysis we offer here; however, we duly note that this topic is indeed worthy of future research for gaining a better understanding of the historical processes involved in creating the modern alls-construction.

C-agreement or something close to it 

West Germanic languages and dialects, the alls-construction appears to be sensitive to the feature specification of the embedded subject, as illustrated in example (5)–(6).

(5) *All-s a woman wants is to make a lot of noise.

(6) All a woman wants is make a lot of noise.

If the alls-construction was merely a phonological reduction, the ungrammaticality of (5) above, but also of (7) below would be unexpected.

(7) *All-s that I know is that Doris talks a lot.

(8) All that I know is that Doris talks a lot.

In example (7) we see that the alls-construction is subject to some sort of adjacency constraint with the subject of the sentence; intervening elements such as that appear to bleed the s-ending on all. With the appearance of intervening items such as that, the s-ending on all cannot appear (cf. (8)). Naturally, the fact that the s-inflection present on all in the alls-construction is sensitive to the feature identity of the subject and respects adjacency conditions proves that structural considerations and syntactic operations may be at play. On the surface, the similarities between C-agreement phenomena as in (1) and the allsconstruction are striking, however in this paper we will investigate to what degree both constructions are similar and, at the same time, distinct from one another. The purpose of this paper is to see to what extent the alls-construction resembles other C-agreement phenomena in West Germanic languages and dialects. Furthermore, we will also discuss the formal features of subjects that license the alls-construction in Midwest AE dialects. By employing Chomsky’s (2000 and later subsequent work) theoretical construct known as Agree (i.e., Probe-Goal relation), we show that traditional C-agreement and the alls-construction are similar to the extent that both structures rely upon the feature-valuation and subsequent PF-reflex of Agree between both C0 and the lower subject. Where C-agreement and the alls-construction critically differ with one another, however, is the closeness of the CP-related phi-features in the respective languages that license these two constructions: In West Germanic SOVlanguages that license C-agreement, the C0 Probe is often very close to the subject. In contrast, the C0 bearing φ-features in alls-construction dialects in many cases appear much higher (i.e., ForceP).

2.  General properties of C-agreement As mentioned in the introduction, the inflection present on C0 in languages exhibiting C-agreement agrees with the φ-features of the embedded subject. For example,

 Michael T. Putnam & Marjo van Koppen

the paradigm of C-agreement in West Flemish (WF) distinguishes all person and number combinations (cf. Law 1991; Haegeman 1992). (9) a. K weten dan-k (ik) goan wegoan. I know that-I (I) go leave ‘I know that I am going to leave.’

C + 1sg (WF)

b. c. d. e. f. g. h. i.

C + 2sg C + 3fsg C + 3msg C + 1pl C + 2pl C + 3pl C + NPsg (Jan) C + NPpl (Jan en Pol)

K weten da-j (gie) goat weggoan. K weten da-se (zie) goat weggoan. K weten da-tje (jij) goat weggoan. K weten da-me (wunder) goan weggoan. K weten da-j (gunder) goat weggoan. K weten dan-ze (zunder) goan weggoan. K weten da Jan goat weggoan. K weten dan Jan en Pol goan weggoan.

The rich paradigm of C-agreement inflections that exist for WF are the exception rather than the norm for this language family; most C-agreement paradigms in West Germanic are defective (cf. Zwart 1997:159). Languages and dialects such as East Netherlandic and Frisian, for instance, only have an agreeing C0 in the presence of 1st person plural and 2nd person singular subjects respectively. Other dialects such as South Hollandic and Munich Bavarian license two licit inflectional categories (i.e., for South Hollandic 1st person and 3rd person plural and for Munich Bavarian 2nd person singular and plural). Another general property of C-agreement in West Germanic involves the inflectional morphology present on finite verbs. The finite verb in inversion contexts displays the same agreement ending as the complementizer (cf. Hoekstra & Smits 1999; Zwart 1993). (10) a.

da-de gullie that-2pl you2pl

kom-t come-2pl

Brabant Dutch (Zwart 1993:177)

b. Wanneer kom-de/*-t gullie? When come-2pl youpl ‘When do you come?’ c. Gullie kom-t/*-de youpl come-2pl ‘You are coming.’

The complementizer in Brabant Dutch exhibits a de-ending in (10a), as does the inverted verb in (10b). The finite verb in subject-initial main clauses does not (necessarily) show the same inflection (10c). However, the majority of the dialects do not make a distinction between the inflection on inverted and non-inverted finite verbs. In those cases the inflection on the finite verb always equals the inflection on the complementizer.

C-agreement or something close to it 

Lastly, C-agreement appears on the lowest C-related item (cf. Zwart 1993:44): (11) a.

Ik weet niet wat of datt-e I know not what if that-infl

de jonges gedaan hebbe. the boys done have

b. Ik weet niet watt-e de jonges gedaan hebbe. I know not what-infl the boys done have ‘I don’t know what the boys have done.’ South Hollandic

In (11a) the complementizer dat ‘that’ displays C-agreement morphology in coordination with the subject de jonges ‘the boys’. Since the example in (11b) appears to lack a complementizer, the Agree relation (i.e., Probe-Goal) is represented on the wh-item wat ‘what’.3

3.  Similarities between C-agreement and the alls-construction The very fact that the alls-construction in vernacular Midwestern American English is sensitive to the φ-feature identity of the embedded subject is perhaps the clearest indication that it may be a closely related phenomenon to C-agreement. (12) a. b. c. d. e. f.

All-s I want to do, is drink coffee. *All-s yousg want to do, is drink coffee. All-s he wants to do, is drink coffee. All-s we want to do, is drink coffee. *All-s youpl want to do, is drink coffee. All-s they want to do, is drink coffee.

Although the inflectional paradigm appears to be fairly robust, much like other West Germanic languages and dialects Midwestern AE exhibits a weakened, defective paradigm. The dialects speakers interviewed for this project, all of whom hail from Southern Ohio and Kentucky, do not allow the alls-construction in combination with 2nd person singular and plural subjects; however, there are other varieties of nonstandard American English that do not exhibit this same restriction of 2nd person morphology and its ability to license the alls-construction.4 As pointed out to us by an anonymous reviewer, the fact that the 2nd person subjects display different behavior than the other subjects might not be a surprise since 2nd person pronouns have two

3.  Zwart (1997) argues that there is a reduced complementizer in these examples. This means that the inflection according to his analysis does not sit on the wh-word, rather it is a reduced complementizer. 4.  For example, the restriction on 2nd person singular and plural subjects occurring with the alls-construction is acceptable to most speakers of Ozark-English dialects.

 Michael T. Putnam & Marjo van Koppen

distinct traits that separate them from 1st and 3rd person forms:  (1) only the pronoun YOU does not distinguish singular from plural forms (e.g., I/we, he,she/they), and (2) Only YOU does not distinguish nominative from non-nominative forms (e.g., I/us, he/she/them). Later in our discussion of the affinities that the alls-constructions shares with pseudo-clefts in Section 7, we will see how these properties help to explain why exactly YOU (i.e., 2nd person morphology) is excluded from appearing in this pseudo-cleft alls-construction. The restriction on the alls-construction as an ungrammatical construction with 2nd person subjects regardless of number specification is not the only constraint involving this construction. (13) a. ?All-s the man wants to do, is drink coffee. b. ??All-s a man wants to do, is drink coffee.

As shown in the examples in (13), there appears to be a restriction on the ability of a subject to license the presence of s-inflection on all and its own inherent definiteness. Pronouns and full DPs (13a) – being quite high on any definiteness scale – are almost always permitted barring other restrictions on the φ-features of the embedded subject. Indefinites (13b) occurring with the alls-construction are highly marked structures. There is also a structural adjacency requirement similar to the one in effect for C-agreement in West Germanic (cf. the West Flemish example in (14) from Ackema & Neeleman 2005; cf. also Carstens 2002). This example shows that C-agreement is impossible when there is an adverb intervening between C0 and the subject. The same restriction is found with the alls-construction: Any impeding constituents disrupting a direct path between the C0 and the embedded subject will result in an ungrammatical structure (15b).5 (14) … dat/*datt-e op den wärmsten dag van ’t joar wiej tegen oonze that/that-infl on the hottest day of the year we against our wil ewärkt hebt. will worked have

“…that we have worked against our will at the hottest day of the year.”

(15) a. All-s I know about Joe and Kelly is that they smell bad. b. ??All-s that I know about Joe and Kelly is that they smell bad.

If the alls-construction is the result of a local syntactic operation Probe-Goal  – a hypothesis that we both support and will provide evidence for in this paper – it only

5.  Unfortunately, it is impossible to put an adverb in between all and the embedded subject in Midwest AE. Therefore, it cannot be tested if adverbs lead to the same intervention effects as in West Germanic.

C-agreement or something close to it 

makes sense that intervention effects would nullify this Agree relation between the C0 (e.g., all) and the embedded subject.6

4.  Differences between C-agreement and the alls-construction One of the most interesting and revealing pieces of data about the true nature of the allsconstruction and its relationship to other C-agreement phenomena in West Germanic returns us once again to the topic of intervention effects. Although it has been shown that neither West Germanic languages and dialects exhibiting C-agreement (cf. (14)) show C-agreement when there is an intervener between C° and the embedded subject, it is important to note that C-agreement in West Germanic has the potential to appear on other items than the embedded complementizer. For the sake of the reader, we repeat the data from South Hollandic (cf. (11)) below as (16) (cf. Zwart 1993:44). (16) a. Ik weet niet wat of datt-e de jonges gedaan hebbe. South Hollandic I know not what of that-infl the boys done have b. Ik weet niet watt-e de jonges gedaan hebbe. I know not what-infl the boys done have ‘I don’t know what the boys have done.’

In (16a) and (16b) C-agreement respectively appears on the true complementizer dat ‘that’ or the wh-item wat ‘what’. The examples in (16) from South Hollandic demonstrate that C-agreement in West Germanic languages occurs on other C-related items when the C0-position is empty. American English dialects that license the allsconstruction, however, do not permit the s-inflection to appear on any C-related element other than all. (17) *All that-s I know about Joe and Kelly is that they smell bad. (18) All-s I know about Joe and Kelly is that they smell bad.

When considered in tandem with the intervention effects previously mentioned in this paper for the alls-construction, we observe that the complementizer that cannot bear the s-inflection (cf. (16)). However, when the lower complementizer that is removed from the sentence as in (17), the alls-construction is once again licit due to the now open path that the higher Probe (unvalued φ-set on C0) has to its now accessible Goal (interpretable φ-features of the subject). The removal of the intervening that eliminates any unwanted intervention effects that would cause the derivation to crash.

6.  We refer the reader to Van Koppen (2005) for more subtle data concerning adverb intervention.

 Michael T. Putnam & Marjo van Koppen

These data reveal more than the validity of intervention effects in the licensing of the alls-construction; they also provide evidence that all occupies a higher position in the CP-layer (cf. Rizzi 1997) in comparison to that which rests in a structurally lower position. It appears that the Probe-Goal relation is blocked by the presence of a complementizer in the American English dialects, whereas the complementizer does not have an effect on the absence or presence of C-agreement in the West-Germanic dialects. Another characteristic of the alls-construction distinguishing it from C-agreement in West Germanic is the lack of any sort of s-inflection in inversion contexts. Recall that in West Germanic languages and dialects that license C-agreement the inverted finite verb displays the same agreement inflection as the complementizer. The examples from Brabant Dutch (cf. (10)) are repeated here in (19) for the sake of the reader. (19) a.

da-de gullie kom-t that-2pl you2pl come-2pl

Brabant Dutch (Zwart 1993:177)

b. Wanneer kom-de/*-t gullie? When come-2pl youpl ‘When do you come?’

Although inflection on inverted finite verbs in Brabant Dutch are valued by the φ-features of C0, any sort of s-infection affiliated with the alls-construction is simply not possible in American English dialects that license the alls-construction as illustrated in (20). (20) *Does I ever behave like an idiot?Midwest AE

This disconnect might be due to the historical development of the alls-construction juxtaposed to C-agreement:  the s-inflection on alls is the result of a phonological reduction of the larger unit [all as] that has been re-analyzed to interact with the φ-features of the subject. Nonetheless, it is a point of distinction contrasting the alls-construction and ‘traditional’ C-agreement in West Germanic.

5.  Interim summary At this point we have observed that the alls-construction resembles C-agreement in West Germanic in two key areas: First, both the alls-construction and C-agreement are sensitive to the feature specification of the embedded subject. This leads us to believe that the alls-construction is principally syntactic in nature, licensed by some sort of formal feature in the derivation. Second, like C-agreement, the alls-construction can only occur when no intervening elements intercede between the C0-Probe and the embedded subject. On the other hand, we have shown three ways in which the allsconstruction starkly contrasts with C-agreement:  For example, unlike C-agreement

C-agreement or something close to it 

languages which allow complementizer-inflection to appear on any lexical item (e.g., complementizers and wh-items) that resides in C0, the s-inflection can only appear on all and no other item, even lower intervening complementizers such as that (cf. (16)). This supports the hypothesis that the complementizer all is in a structurally higher position in the CP-layer than that in American English. Connected with the previous point, a second difference between the two constructions has to deal with the historical development of the alls-construction from the larger unit [all as]. Third, the inflection borne on finite verbs in inversion contexts are different in West Germanic languages and dialects that license C-agreement as compared to American English dialects that allow the alls-construction. In the former, agreement morphology takes the form of the inverted finite verb, whereas in dialects where the alls-construction is grammatical this feature is not possible. In the second half of this paper we will put forward a working analysis of the alls-construction within the framework of minimalist syntax.

6.  A working analysis of the alls-construction At this point we are forced to return to the principal question addressed in this paper, namely, are the s-inflection in the alls-construction and inflectional morphology in West Germanic C-agreement structurally related to one another, or are they genetically dissimilar and unrelated, independent operations? As discussed thus far, the allsconstruction shares certain characteristics with C-agreement in SOV West Germanic languages, and in some cases significantly deviates from them. Our proposal is that the alls-construction and West Germanic C-agreement are in this regard quite similar to one another, i.e., the s-ending involved in the alls-construction is inflection valued through a Probe-Goal relation allowing the unvalued φ-features of C0 to interact with the interpretable φ-set of the embedded subject.

6.1  Prerequisite one: A brief analysis of C-agreement Before discussing the alls-construction in any detail, we must first provide a brief theoretical analysis of C-agreement. The minimal assumption, of course, is that inflection on the complementizer signals the presence of φ-features on C0 (cf. also Bennis & Haegeman 1984; Carstens 2002; Van Craenenbroeck & Van Koppen 2002; Van Koppen 2005). As suggested throughout this paper, the φ-features on C0 are believed to be unvalued, and can hence be classified as Probes in the sense of Chomsky (2002). The mechanism labeled Agree must then seek a suitable local Goal, in this case, the interpretable φ-feature set of the subject. This Agree relation is schematized in (21).

 Michael T. Putnam & Marjo van Koppen

(21) Agree between C0 and the subject CP | C′ C0 [uphi]

TP subject [iphi]


T′ T0

VP …

As we will demonstrate in the section 6.3, the minimalist analysis for C-agreement by means of Probe-Goal (Agree) relations is essentially the same for the alls-construction.

6.2  Prerequisite two: Properties of the alls-construction Before we can provide our analysis of the alls-construction in Midwest AE, we first have to discuss some characteristics of this construction in more detail.

6.2.1  The phi-features playing a role in the alls-construction At this point it is necessary to determine exactly which features on embedded subjects actively participate in the alls-construction. We would like to suggest two potential factors that serve as triggers for the alls-construction. As indicated in the contrastive data in (12) above, re-represented as (22) below, alls appears when the subject is either 1st or 3rd person. A subject in 2nd person results in a highly marked structure. (22) a. All-s I want to do is drink coffee. b. *All-s yousg want to do is drink coffee. c. All-s he wants to do is drink coffee d. All-s we want to do is drink coffee. e. *All-s youpl want to do is drink coffee. f. All-s they want to do is drink coffee.

To account for the fact that 2nd person pronouns cannot license the alls-construction, we propose that the feature triggering s-agreement in the dialect of English we are investigating in this paper, is [+number].7 As we already noted above, second person

7.  The s-ending on all in these dialects has a different distribution than the s-ending on the finite verb in English. In the latter language the s-ending also does not appear with first person singular subjects and not with plural subjects. Kayne (1994) argues that this is the case because the feature specification of this affix in standard English is [+singular]. Furthermore,

C-agreement or something close to it 

pronouns are different from first and third person pronouns in that they do not encode number. If second person pronouns are unspecified for number in the American English dialects under discussion and the s-ending is sensitive for the presence of a number specification, it is immediately clear why the s-ending does not appear with second person pronouns: they do not encode number.8 Second, as displayed in examples (23) – (24) below, the alls-construction can only be licensed by a definite DP (23), but not by an indefinite DP (24).9 (23) ?Alls/All the bar down the road does is make money on the weekends. (24) ??Alls/All a woman wants to do is eat chocolate without gaining weight.10

As evidenced by the data present in (23) – (24), the alls-construction is therefore licensed on a definiteness scale; highly definite XPs (e.g., pronouns and full definite DPs) serving as subjects are able to license the alls-construction whereas indefinites lead to a less grammatical sentence. This hierarchy, where a construction is sensitive to the definiteness of the argument, is at work in more constructions than the alls-construction. For instance, Van Craenenbroeck & Van Koppen (2002) show that subject doubling in Dutch dialects is sensitive to the definiteness of the subject.

6.2.2  The status of all Another factor that needs to be sorted out is the status of all. Why can the inflection appear on this quantifier? What position does it occupy? First of all, it has to be noted that the inflection on all is only possible if all is part of a so-called all-pseudo-cleft. The

Bernstein et al. (2008) argue that the s-ending on the finite verb in Appalachian English is sensitive to person features of the subject. 8.  We should not be surprised to find dialects in which second person pronouns actually do trigger the s-ending on all since there are English dialects which make a distinction between singular and plural. Actually a brief survey among a subset of our informants suggests that the s-inflection is much better when the subject is not simply you, but you guys as in (i) below. Alls you guys want to do …


We will leave this subject as a topic for further research.

9.  The alls-construction and the more regular all-pseudo-cleft are also ungrammatical with a bare indefinite. (i)

a. b.

*All restaurants do these days is/are increase the prices of beer. *Alls restaurants do these days is/are increase the prices of beer.

10.  The appearance of indefinite subjects in the alls-construction, but also in “regular” all-pseudo-clefts is restricted to indefinite subjects which can have a generic interpretation. We have no analysis for this observation at this point.

 Michael T. Putnam & Marjo van Koppen

inflection cannot appear on all when it is heading a simple relative clause acting as an object to the verb. See the examples in (25). (25) a. I want to give him *alls/all he wants to have.Midwest AE b. They want to give me *alls/all I want to have. c. Kelly wants to give him *alls/all he wants to have.

Also all cannot carry the s-ending when it is heading a subject relative clause. See the examples in (26a–b). (26) a. *Alls/all John does gives me hope b. *Alls/all John does scares me.

(subject clause) (subject clause)

Apparently, it is important that the all is part of a pseudo-cleft in order for it to show inflection in the dialect under consideration. We do not have a full-fledged analysis for this aspect of the alls-construction, but we briefly come back to it below (cf. also Van Koppen & Putnam 2008). We assume, following among others Collins 1991, the “relative clause” analysis of pseudo-clefts.11 This means that we assume that all in the all-pseudo-cleft has the same status as the head of so-called th-pseudo-clefts the one/the thing/the reason, namely the normal position heads of headed relative clauses occupy. This means that all is not the Probe for phi-features itself, as Probes are usually heads, but that the features of a head in the CP-domain are spelled out on all. This is comparable to what happens with C-agreement in West Germanic SOV-languages, cf. example (11) above.

6.3  The analysis of the alls-construction The fact that the alls-construction (i) does not appear in every environment in which the quantifier all can appear in (and hence that the s-ending seems to be dependent on the syntactic environment of all) and (ii) is sensitive to the feature identity of the embedded subject (cf. (12) and (13)) is perhaps the most critical evidence in favor of labeling this phenomenon as a syntactic operation similar to C-agreement. In order to develop our proposal that the same Probe-Goal mechanism involved in licensing C-agreement in West Germanic is by design the same relationship active in generating the alls-construction, we put forward the minimal hypothesis that the s-ending on all indicates the presence of phi-features on a head in the CP-domain. The analysis of an example with the alls-construction now goes as follows. Consider the example in (27) and its derivation in (28).

11.  See Den Dikken 2005 for an extensive discussion on the status of pseudo-clefts as relative clauses or wh-questions. We come back to the status of the all-pseudo-cleft in more detail in Van Koppen & Putnam (2008).

C-agreement or something close to it 

(27) All-s I want to do is smoke a cigarette. (28) C-agreement in American-English dialects NP N′ N0 All

CP C′ C0 [uphi] -s


TP I [iphi]

In this example C0 contains unvalued phi-features. It searches its c-command domain and finds the subject as its suitable Goal. C0 and the subject enter into an agreement relation and this relation gets realized as an s-affix on the complementizer. Now consider the example in (15b), repeated here as (29). (29)

*All-s that I know about Joe and Kelly is that they smell bad.

The question arises why the complementizer that blocks agreement on all. Note that in this respect the American-English dialects under discussion differ significantly from the West-Germanic examples in (11), repeated here as (30). (30) a. Ik weet niet wat of datt-e de jonges gedaan hebbe. South Hollandic I know not what of that-infl the boys done have b. Ik weet niet watt-e de jonges gedaan hebbe. I know not what-infl the boys done have ‘I don’t know what the boys have done.’

In these South Hollandic examples the agreement affix either appears on the complementizer or, if the complementizer is absent, on the question word introducing the embedded clause. A potential way to handle this difference between the West-Germanic dialects with C-agreement and the American-English ones is based on an analysis of C-agreement by Van Craenenbroeck & Van Koppen (2002). They argue that you only get C-agreement if the subject is sufficiently local to the phi-features in Cº. More in particular, they show on the basis of complementizer agreement in Dutch dialects, that-trace and that-omission data from English and tense agreement in Irish that an Agree relation is spelled out if the Goal (or a copy of the Goal) is in the direct syntactic vicinity of the Probe. They define locality as follows: the local domain of a Probe “should include

 Michael T. Putnam & Marjo van Koppen

its specifier (cf. Chomsky’s 2001b12 internal merge), its complement (cf. Bobaljik 1995; Bobaljik & Thráinsson 1998), the specifier of that complement (Bobaljik 2001; Polinsky & Potsdam 2001), and the head of that complement (Bobaljik 2001)”. This definition of locality is summarized in the following tree structure. (31) Local domain of a Probe (Van Craenenbroeck & Van Koppen 2002) XP YP

X′ X




Following Rizzi (1997, 1999) they suggest that the CP-domain should be split up in several projections. The phi-features leading to C-agreement are associated with Rizzi’s higher CP-projection ForceP. They argue that if there is nothing special going on in the embedded clause, i.e., when there is not topicalization, the CP-domain is conflated. The features of Force0 are part of this conflated CP and hence local to the subject (cf. 32a). However, if there is an adverb intervening between the complementizer and the subject, cf. (32b), then the CP-domain can no longer be conflated, since the adverb targets the Top-projection inside the left periphery. In this scenario, the features of Force0 are not “local” to the subject and C-agreement disappears. (32) a. [CP [C0datte] [TP wiej [T′ tegen oonze wil ewärkt hebt]]]. that we against our will worked have “…that we have worked against our will.” b. [ForceP [Force dat] [TopP op den wärmsten dag van ’t joar [FinP [TP wiej that on the hottest day of the year we

tegen oonze wil ewärkt hebt.]]]] against our will worked have

“…that we have worked against our will at the hottest day of the year.”

Van Craenenbroeck & Van Koppen (2002) argue that the complementizer in English is present when the CP-domain is not conflated. When the complementizer is absent (in that-trace and that-omission contexts) the CP-domain is conflated. More precisely, they argue that the alternation between the zero-complementizer and that in English has the same properties as C-agreement in the West-Germanic dialects: that is absent when the subject is local to C0 and it is present when the subject is not local to C0.


This work is referred to as Chomsky (2002) in our paper.

C-agreement or something close to it 

The Dutch complementizer dat ‘that’ is not sensitive to the locality of the subject. Hence, the presence or absence of the Dutch complementizer does not correlate with the conflation of the CP-domain. If we combine this observation with the data in (29) from the American-English dialects under discussion, it becomes clear why the presence of the complementizer blocks the agreement on all. The s-ending which results from checking the features of C0 locally with the subject are spelled out on all which occupies the highest specifier of CP. When the complementizer is not present, the CP-domain is conflated and the phi-features of Force0 are local to the subject. This is illustrated in (28) for the American-English dialects and in (32a) for the West-Germanic dialects. However, when the complementizer is present in American-English, then the CP-domain is not conflated, see the structure in (33) (33) Checking C-agreement at a distance in American-English dialects ForceP All


Force0 uφ

TopP Top′ Top0

FinP Fin′ Fin0 that


checking at a distance

In English the presence of the complementizer signals that the CP-domain is not conflated. This means that the phi-features of Force0 are not local to the subject and no local checking can take place. As a result C-agreement is not phonologically realized and hence all appears without inflection. In the West-Germanic dialects the agreement ending can appear on the complementizer, since the complementizer does not indicate that the CP-domain is extended.

7.  Remaining questions There are two questions that have not been addressed before, but which represent crucial properties of the alls-construction. The first one is why the inflection is

 Michael T. Putnam & Marjo van Koppen

restricted to pseudo-cleft contexts. Secondly, the question arises why the inflection in the alls-construction only appear on the quantifier all and not on other items introduction pseudo-clefts, like what, cf. the ungrammatical example in (34). (34) What/*What-s I want to do is have fun.

Although we do not have full-fledged answers to these questions, we would like to sketch routes of investigation here. First of all, it should be noted that the s-inflection on all-pseudo-clefts is only possible when the pseudo-cleft is preceding the main finite verb. Consider the examples in (34). (34) a. All-s Vicki brought to the party was bread. b. Bread was all (*all-s) Vicki brought to the party.

This seems to suggest that the appearance of the s-inflection is not only sensitive to features internal to the all-pseudo-cleft, but also to external properties. We refer the reader to Van Koppen & Putnam (2008) for a detailed analysis of this aspect of the alls-construction. Finally, we would like to discuss briefly why the s-ending is restricted to allpseudo-clefts and why it cannot appear on what introducing more regular pseudo-clefts. One difference between all-pseudo-clefts and what-pseudo-clefts is that in the former construction the introducing element, i.e., all, is outside the CP, whereas what irrespective of the exact analysis of the pseudo-cleft (as free relative (cf. Collins 1991) or as a wh-question (cf. among other Den Dikken et al. 2000) is assumed to be inside the CP. Wh-phrases, according to Rizzi (1997, 1999), target a specific position in the left periphery, namely FocP. This means that the presence of a wh-item like what also requires the CP-domain to be split up, in order to be able to target this specific left peripheral projection. However, when the CP-domain is not conflated, then the phi-features in ForceP are not local enough to the subject and hence the s-ending cannot appear.

8.  Conclusion In this paper, we have shown that the alls-construction, which hails from the ‘all as’construction, displays certain similarities with C-agreement: (i) just like C-agreement it is sensitive to the feature specification of the embedded subject and (ii) the allsconstruction can only appear when there is no intervening C-related item between all and the embedded subject. However, the alls-construction also differs from C-agreement. Firstly, the inflectional ending of the alls-construction cannot appear on another, “lower”, C-related element, like the complementizer that. West Germanic

C-agreement or something close to it 

C-agreement can appear on another, “higher”, C-related element, like a wh-word. Secondly, in contrast to C-agreement, the inflection on all in the alls-construction does not equal the inflection on the finite verb in inversion contexts. We have argued that the alls-construction Midwest AE is comparable to C-agreement in West Germanic, in the sense that both constructions are the result of phi-feature checking between a head in the CP-domain and the embedded subject. However, the differences between the alls-construction and C-agreement can be accounted for if we assume that the presence of a complementizer in the West-Germanic dialects with C-agreement does not indicate the non-conflated CP-structure, whereas the presence of the complementizer in English does.

References Ackema, P. & Neeleman, A. 2005. Beyond Morphology, Interface Conditions on Word Formations. Oxford: OUP. Bennis, H. & Haegeman, L. 1984. On the status of agreement and relative clauses in West Flemish. In Sentential Complementation. Proceedings of the International Conference Held at UFSAL, W. de Geest & Y. Putseys (Eds), Dordrecht: Foris. Bernstein et al. 2008. One form for different features:  Micro-syntactic variation in English. Handout. Linguistic Society of America. Bobaljik, J.D. 1995. Morphosyntax: The Syntax of Verbal Inflection. Ph.D. dissertation, MIT. Bobaljik, J.D. 2001. Agreement domains. Handout. Console X, Leiden, 7–9 December, 2001. Bobaljik, J.D. & Thráinsson, H. 1998. Two heads aren’t always better than one. In Syntax 1(1): 37–71. Carter, M. 1992. All’s I know. LinguistList 3.131. http://linguistlist.org/issues/3/3-131.html (accessed August 26, 2008). Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Honour of Howard Lasnik, R. Martin, et al. (Eds), Cambridge MA: The MIT press. Chomsky, N. 2002. Beyond explanatory adequacy. MIT Occasional Papers 22. Carstens, V. 2002. Rethinking complementizer agreement:  Agree with a Case-checked goal. Linguistic Inquiry 34(3): 393–412. Collins, P. 1991. Pseudo-cleft and cleft constructions: A thematic and informational interpretation. Linguistics 29: 481–519 Dikken, M. den, Meinunger, A. & Wilder, C. 2000. Pseudoclefts and ellipsis. Studia Linguistica 54: 41–89. Dikken, M.  den. 2005. Specificational copular sentences and pseudo-clefts. In The Blackwell Companion to Syntax, Vol. IV, Ch. 61. M. Everaert & H. van Riemsdijk (Eds), Malden MA: Blackwell. Fuss, E. 2005. The Rise of Agreement. A Formal Approach to the Syntax and Grammaticalization of Verbal Inflection. Amsterdam: John Benjamins. Fuss, E. 2007. Cyclic Spell-Out and the domain of post-syntactic operations: Evidence from complementizer agreement. In Dynamic Interphases -  Linguistic Analysis, Kleanthes K. Grohmann (Ed.), 33(33–4): 267–302.

 Michael T. Putnam & Marjo van Koppen Haegeman, L. 1992. Theory and Description in Generative Syntax:  A Case Study in West Flemish. Cambridge: CUP. Hoekstra, E. & Smits, C. 1999. Everything you always wanted to know about complementizer agreement. Proceedings of WECOL 19. Hofherr, P.C. 2003. Inflected complementizers and the licensing of non-referential pro-drop. In The Role of Agreement in Natural Language, TLS 5 Proceedings, W.E. Griffin (Ed.), 47–58. Kayne, R. 1994. The Antisymmetry of Syntax. Cambridge MA: The MIT Press. Law, P. 1991. Verb-movement, expletive replacement, and head government. The Linguistic Review 8: 253–285. Liberman, M. 2004. Language Log:  All’s I know. http://itre.cis.upenn.edu/~myl/languagelog/ archives/001253.html (accessed August 26, 2008). Polinsky, M. & Potsdam, E. 2001. Long distance agreement and topic in Tsez. Natural Language and Linguistic Theory 19(3): 583–646 Richards, M. 2005. Object Shift and Scrambling in North and West Germanic: A Case Study in Symmetrical Syntax. Ph.D. dissertation, University of Cambridge. Rizzi, L. 1997. The fine structure of the left periphery. In Elements of Grammar, L. Haegeman (Ed.), 281–337. Dordrecht: Kluwer. Rizzi, L. 1999. On the Position ‘Int(errogative)’ in the Left Periphery of the Clause. Ms, University of Sienna. Teaman, B. 1992. All’s I know. LinguistList Re:  3.131. http://linguistlist.org/issues/3/3–155. html#3 (accessed August 26, 2008). Van Craenenbroeck, J. & van Koppen, M. 2002. The locality of agreement and the CP-domain. Handout. GLOW 2002, Amsterdam. Van Koppen, M. 2005. One Probe-Two Goals: Aspects of agreement in Dutch dialects. Ph.D. dissertation, University of Leiden. Van Koppen, M. & Putnam, M. 2008. Alls there is to say about the alls-construction. Ms, University of Utrecht-Carson-Newman College. Zwart, C.J-W. 1993. Dutch Syntax:  A Minimalist Approach. Ph.D. dissertation, University of Groningen. Zwart, C.J-W. 1997. A Minimalist Approach to the Syntax of Dutch. Kluwer: Dordrecht.

Uncharted territory? Towards a non-cartographic account of Germanic syntax Jan-Wouter Zwart

University of Groningen This article discusses the consequences of a strictly derivational approach—where syntactic relations are construed dynamically as the derivation proceeds—to the analysis of key areas of Germanic syntax. It discusses the nature of syntactic positions from a non-cartographic point of view. Evidence supporting a non-cartographic approach is found in word order transitivity failures in various domains (the left periphery, the order of adverbs, the adjective-noun construction). The implications of a non-cartographic approach are discussed in four key areas of Germanic syntax (the fine structure of the left periphery, topicalization/focalization, subject placement and object placement).

1.  Introduction* In a common implementation of the Minimalist Program (Chomsky 1995), syntactic structure is built from the bottom up by a recursive operation Merge, which combines two elements into a constituent. On this view, the computational system of human language is inherently derivational, in the sense that a syntactic object is defined in terms of a sequence of applications of the operation Merge. As argued by Epstein et al. (1998), syntactic relations can likewise be defined in terms of the operation Merge; on the strictest implementation, such relations are restricted to pairs of sisters created by Merge.

*This article was written in March 2006 and presented on April 2 of that year as one of the keynote lectures at the 21st Comparative Germanic Syntax Workshop, University of California, Santa Cruz. I would like to thank the organizers for including it in the program. Thanks are also due to Vera Lee-Schoenfeld, Jim McCloskey, Sandra Chung, Jeroen Van Craenenbroeck, Jan Koster, and the editors and reviewers for this volume. An earlier version of this article, without the benefit of reviewer’s comments, appeared in Groninger Arbeiten zur germanistischen Linguistik 45 (2007), 55–75.

 Jan-Wouter Zwart

This ‘derivational turn’ of the theory of grammar raises the question whether derivations are guided by (innate) global considerations of syntactic architecture, or whether they proceed on a strictly local basis, caring only about the syntactic and semantic relations between members of sister pairs. Much current work still seems to assume that derivations work towards a fixed goal, a universal syntactic structure characterized by strict hierarchies among functional elements and by the rules of phrase structure (e.g., Cinque 2002 and following publications). However, from the derivational point of view, the question is justified whether such a universal structure guiding the derivation needs to be assumed. As long as the operation Merge and the syntactic objects it yields are clearly defined, a more economical implementation of the Minimalist Program might dispense with global considerations and adopt a more flexible, dynamic approach to syntactic structure. In this article I would like to discuss the question of fixed vs. flexible structure in the context of Germanic syntax. The standard approach to Germanic generative syntax is firmly rooted within a strict cartographic tradition. For instance, the analysis of the verb second pattern of Continental West-Germanic, North-Germanic, and earlier stages of English, based on Den Besten (1977), assumes that all main clauses have the same expansion (CP, in the standard terminology since Chomsky 1986), and describes the verb second pattern as resulting from movements of heads and phrases to fixed positions within the clausal architecture. As we will see, one of the consequences of the non-cartographic approach is that reference to fixed positions becomes meaningless: positions are defined in terms of sisterhood relations (essentially, in terms of Merge), not in terms of a preinstalled map of the clause. What I set out to do in this article is chart the consequences of a non-cartographic approach to Germanic syntax (in particular, Continental West-Germanic syntax) in a limited set of key domains. The article has the following contents. In section 2, I discuss a central concept to the issue at hand, namely ‘syntactic position’ (cf. Nilsen 2003), opposing the rigid, cartographic approach and the flexible, non-cartographic approach. In section 3, I discuss evidence, some from the literature, some new, suggesting that even a weak cartographic approach is unable to account for certain word order patterns, namely those which cannot be derived from a hierarchical scale needed to account for other word order patterns (transitivity failure). Then in section 4, I briefly sketch an implementation of the non-cartographic approach in major areas of Germanic syntax.1

1.  I would like to acknowledge here the tradition of a flexible approach to syntax pursued in recent years by various researchers from Utrecht University (see Neeleman & Weerman 1999; Koeneman 2000; Nilsen 2003). An important recent contribution, Neeleman and Van de Koot (2008), particularly relevant to key aspects of Continental West-Germanic syntax,

Uncharted territory? 

2.  Syntactic positions I would like to begin by opposing two ways of defining syntactic positions, and then introduce some terminology needed in the discussion to follow.

2.1  The cartographic vs. the dynamic approach to syntactic structure First, we might define a syntactic position in terms of a fixed map of the clause: the cartographic approach. The map itself is the outcome of the application of the rules of phrase structure (the X-bar theory, e.g., Chomsky 1986) to empirical observations. A strong version of the cartographic approach holds that observations regarding a construction X in language A allow us to draw conclusions regarding the structure of another construction Y in language A, or about the structure of X and Y in languages other than A. Likewise, we may on this approach use observations regarding constructions X and Y—even if they do not cooccur in any language—to piece together a general (universal) structure of the clause in which both X and Y find expression. Second, we might define a syntactic position in terms of its local environment: the dynamic approach. On this approach, positions are emerging properties of derivations, created by the structure building procedure. The approach assumes that syntactic operations (essentially the single operation Merge which combines two elements into a constituent) are triggered by some local requirement, and take place without consideration of overall syntactic architecture. It is possible that the two approaches represent two sides of the same coin. On their strongest formulations, this is certainly not the case. The strong cartographic approach, for instance, entails that underlying a simple sentence like John left is an entire structure containing the full array of functional projections identified in work by Rizzi (1997) and Cinque (1999). The map describing that structure is universal and possibly an integral part of the faculty of language, and all clauses have the same expansion. This is incompatible with a dynamic approach, which on its strongest formulation denies the existence of universal phrase structure rules; what is universal is the way elements are merged, and what the operation yields; other than that, what you see is what you get, and there is no requirement stating that every clause needs to be expanded up to the CP-level, for instance.

regretfully came to my attention only after the present article was completed. Neeleman and Van de Koot argue that word order phenomena involving topic-comment/focus-ground need to be analysed as a function of a (flexible) mapping between information structure and syntactic structure, arguing explicitly against a range of implementations of the cartographic approach to this domain of facts.

 Jan-Wouter Zwart

Weaker versions of the cartographic and dynamic approaches might seem to be compatible. A weak cartographic approach might accept that not all projections need to be realized in full; just that when they are realized, they are bound to appear in a certain hierarchical order. On a weak dynamic approach, we might say that the map of the clause is an abstraction of the structure of different clauses or clause types. However, there is reason to believe that such an abstraction is impossible, because of the existence of word order transitivity failures (see section 3 and references cited there). Such transitivity failures suggest an inherent flexibility to the way structure is created, which is impossible to capture even on a weak version of the cartographic approach. The next section discusses a number of transitivity failures relevant to the question at hand. Before we turn to those, I would like to sketch very briefly the outlines of a non-cartographic approach to phrase structure.

2.2  A non-cartographic approach to phrase structure We start from the assumption that syntactic structure is the product of a single operation, Merge. We therefore aim to define syntactic positions in terms of the operation Merge. We take Merge in its simplest form to be an assignment operation which takes an element from a certain resource (the Numeration of Chomsky 1995) and assigns it to a workspace (the current derivation under construction). As pointed out by Jaspers (1998), such an operation is inherently asymmetric, in the sense that its product contains a previously existing part (the current stage of the derivation) and a newly added element (the element merged). We will therefore say that an element α is merged to a workspace δ (instead of α and δ merging together). We can now define the position in which α is merged to a workspace δ as the occurrence of δ, to use a term coined for this purpose by Chomsky (2000:115). Let P be a derivation (i.e., a syntactic object derived by Merge). Then before merger of α to δ, P = δ, and after merger of α to δ, P is the ordered pair 〈α,δ〉. At that point, we define the occurrence (occ) of δ in P as P minus δ. Hence,

(1) given a workspace δ of a derivation P, and an element α merged to δ, the position of α = occ(δ) in P.

It follows that only elements merged to the workspace have a syntactic position. (Thus, P and δ have no syntactic positions. The operation Merge creates them, but does not position them. It remains to be seen whether this is a desirable consequence.) We now posit that positions are created (i.e., elements are merged) because the workspace needs them. This ‘need’ is standardly described in terms of (uninterpretable) features which must be eliminated to prevent a derivation from crashing

Uncharted territory? 

(cf. Frampton & Gutmann 2002). I tentatively propose a slightly different take, where what the workspace needs is the resolution of an inner conflict (see also Platzack 1996; Koeneman 2000; Nilsen 1997, and Van Craenenbroeck 2006 for earlier similar approaches to movement). I believe the EPP of Chomsky (2001) to be essentially a requirement of this type. Examples of inner conflicts that might arise in this context are subjects contained within predicates, topic elements contained within a focus domain, operator elements contained within their scopal domain, etc.2 The idea would be that the ‘movement’ triggered by this inner conflict is externalization of the offending element. To be precise, we do not assume that material is extracted from the workspace (i.e., there is no such operation as ‘internal merge’, cf. Chomsky 2004 & Koster 2007). Every operation Merge assigns an element from the resource to the workspace. Note that we have not stipulated that elements merged are eliminated from the resource, nor will we (in fact, we may maintain that each stage of the workspace is properly included in the resource, if it is possible to ‘move’ a syntactic object created in the course of the derivation). Therefore, an element from the resource which causes an inner conflict within the workspace is still available in the resource to be assigned to the workspace a second time, and this element may then be stricken from the workspace in its original position, under a condition of identity (leaving a gap or trace).3 The technicalities of the operation are not crucial at this point. What is crucial is the hypothesis that merger is triggered by properties of a workspace. It follows from this hypothesis that positions are not absolute, but relative to a given workspace. For example, the position of subject of a clause is defined as the occurrence of any workspace which may function as the clausal predicate. This is different from the traditional definition in which the subject position is the specifier position of a functional head T (tense) (see section 4.3). If this approach to structure and ‘movement’ is on the right track, it implies that caution is advised when describing a syntactic process as targeting fixed positions, such as Spec,CP. It would have to be established in each particular case that the sister of Spec,CP is characterized by an inner conflict requiring externalization of an offending element. Beyond that, there are no particular requirements associated with either the moving category or the position it moves to which could force such an operation.

2.  As pointed out by a reviewer, this approach to the trigger for Merge does not extend to the first operation Merge in a derivation. 3.  The question of the fate of the lower copy is not particular to the approach to ‘movement’ contemplated here, so I will not discuss it any further at this point.

 Jan-Wouter Zwart

3.  Transitivity failures Empirical evidence against even the weak cartographic approach has accumulated in recent years, focusing on failures of word orders expected on the basis of reasoning by transitivity (Bobaljik 1999; Nilsen 2003; Van Craenenbroeck 2006). Given a structure (2), where A, B, C are absolute positions, we may infer the reasoning in (3).


A′ A

B′ B

C′ C




(> = ‘precedes’)


i.e., *C > A

In other words, since A precedes B, and B precedes C, we may infer that A precedes C and not vice versa. Sections 3.1–3.4 discuss a range of cases where such word order transitivity fails to hold. Useful testing grounds for questions of this type are provided by Rizzi’s (1997) fine structure of the left periphery, Cinque’s (1999) adverb hierarchy, Vendler’s (1968) adjective scale, etc. If positions are relative, the structure in (2) follows if Z needs C, C′ needs B, B′ needs A and A′ does not need B or C. But nothing a priori excludes a derivation like (2) in which A′ does need B or C. For example, the word order where C precedes A may occur when, after (2) has been derived, an inner conflict is caused by C within A′ (for example, when A creates a focus domain and C is a topic). As we will see, transitivity failures discussed in the literature are often of this type.

3.1  Van Craenenbroeck (2006) on the left periphery Based on studies of word order phenomena within the left periphery of Italian, Rizzi (1997) concludes that functional projections within the left periphery are hierarchically ordered as in (4):

(4) Force > (Topic) > Focus > (Topic) > Finite

In this analysis, wh-elements appear in the specifier position of Focus, and clitic left dislocated elements in the specifier position of Topic. As Van Craenenbroeck (2006)

Uncharted territory? 

observes in Venetian (data from Cecilia Poletto by p.c.), Topic precedes Focus (5a), and the complementizer che follows (wh-)Focus (5b), but precedes Topic (5c): (5) a. Topic > Focus > che Me domando el premio Nobel a chi che i ghe lo podarìa dar I wonder the Nobel prize to who that they should give it to him ‘I wonder who they should give the Nobel prize.’ b. Focus > che Me domando chi che Nane ga visto al marcà I wonder who that Nane saw at the market ‘I wonder who Nane saw at the market.’ c. che > Topic Me dispiase che a Marco i ghe abia ditto cussi I’m sorry that to Marco they told him so ‘I’m sorry that they told Marco so.’

It follows that there is no simple reasoning by transitivity which would derive the position of the complementizer in (5a). Van Craenenbroeck (2006) proposes to understand the word order in the Venetian left periphery as follows. The key factor explaining the distribution of Topic elements in Venetian is their inability to remain inside a focus-marked domain. In our terms, a topic element inside a focus-marked domain creates an inner conflict within the focus-marked constituent, and must be merged anew (leading to erasure of the offending element). Van Craenenbroeck assumes that in unmarked clauses, IP constitutes a focus-marked domain, forcing externalization of any topic element contained inside it. As the externalization takes place as soon as the problem arises (i.e., as soon as IP is created), the topic element will be merged prior to the introduction of the complementizer in the derivation, leading to the che-Topic order of (5c). The focus-marked wh-elements are merged in Spec,CP in Van Craenenbroeck’s analysis, yielding the order of (5b) on standard assumptions. However, introduction of a wh-element creates a new focus-marked domain, which forces the topic element to be merged again, this time later than the merger of the complementizer, yielding the order of (5a). Van Craenenbroeck notes that this analysis implies that elements in the left periphery do not have a fixed landing site, and hence that the cartographic approach cannot be maintained. (I refer to Van Craenenbroeck’s paper for discussion of alternatives within a cartographic approach adopting (4).)

3.2  Nilsen (2003) on adverb ordering In the extended adverb hierarchy of Cinque (1999), a modal adverb like possibly precedes an aspectual adverb like always. As Nilsen (2003:10f) notes, this ordering

 Jan-Wouter Zwart

is attested in Norwegian as well, where possibly precedes negation (6a), and always follows it (6b). However, the inverse order (always possibly) is also found (6c): (6) a. possibly > neg Ståle har muligens ikke spist hvetekakene sine Ståle has possibly not eaten his weaties b. neg > always Ståle hadde ikke alltid spist hvetekakene sine Ståle had not always eaten his weaties c. always > possibly ..hvor spillerne alltid muligens er et klikk fra å vinne $1000 where players always possibly are one click away from winning $1000

The crucial observation here is that the order of possibly and always is fixed only relative to negation, but not relative to each other. In Nilsen’s (2003) analysis, the order in (6a) is explained by the circumstance that possibly is a positive polarity item, which—to use our terminology—creates an inner conflict when possibly is contained within a negative-marked domain. The order in (6b) Nilsen derives from the general inability of universal quantifiers to outscope negation; as we would say, negation inside a universal quantifier-marked domain creates an inner conflict. But no inner conflict is created by merging possibly and always in either order, and hence both orders occur. Again, this is not predicted on any cartographic approach, while the more flexible derivation contemplated here makes it possible.

3.3  Bobaljik (1999) on the argument-adjunct interaction The adverb hierarchy of Cinque (1999), collapsed somewhat coarsely in (7a), contains no information regarding the distribution of the grammatical functions; yet these are also strictly ordered according to the scale in (7b): (7) a. adverb hierarchy speech act > evaluative > temporal > aspectual > manner b. grammatical function hierarchy subject > indirect object > direct object

As Bobaljik (1999) observes, there appears to be no fixed combination of the two hierarchies, across languages, or even within a single language. In (8a), a low grammatical function is seen to precede a high adverb, whereas the situation is reversed in (8b): (8) a. direct object > speech act adverb    ..dat Jan Marie het boek eerlijk gezegd niet gegeven heeft (Dutch)     that John Mary the book frankly not given has ‘..that frankly John didn’t give Mary the book.’

Uncharted territory? 

b. manner adverb > indirect object    ..dat Jan snel Marie het boek gegeven heeft    that John quickly Mary the book given has ‘..that John gave Mary the book quickly.’

This suggests that both hierarchies play in different dimensions, frustrating attempts to reach a unified cartographic representation of the clause.

3.4  Adjective order It is well-known that adjectives appear in certain orders, summarized in the scale in (9) from Vendler (1968):4

(9) quality > size > shape > color > origin

This hierarchy is observed in (10a), but not in (10b): (10) a. color > origin a red Hungarian car b. origin > color a Hungarian red wine

These examples illustrate that adjectives (from any semantic class) may be construed in two ways, which have been called direct vs. indirect modification (Bolinger 1967; Sproat & Shih 1988; Cinque 2003). There are various aspects to the direct-indirect opposition, but a quick grasp of it is provided by thinking of indirect modification as being predicational, and direct modification as being non-predicational. Thus, we may say that a red wine is actually a deep purple, but not so with a red car; hence red is indirectly modifying in (10a) and directly modifying in (10b). As observed by Sproat & Shih (1990), the adjective hierarchy in (9) is observed only with direct modification adjectives, indirect modification adjectives showing more syntactic freedom. In Mandarin Chinese, indirect modification adjectives are construed with the noun via a linker, whereas direct modification adjectives are bare; only the latter show the fixed ordering: (11) a. size > shape xiao de lü de huaping small link green link vase


b. shape > size lü de xiao de huaping green link small link vase

4.  Unlike the preceding sections, the argument presented in this section is not taken from the existing literature.

 Jan-Wouter Zwart

c. size > shape xiao lü huaping small green vase d. *shape > size lü xiao huaping green small vase

Moreover, when direct and indirect modification adjectives are combined, the former invariably appears nearer to the head noun. In languages with prenominal adjectives, this means that the scale in (12) applies, regardless the semantic class to which the adjectives belong. (12) indirectly modifying > directly modifying

This accounts for the order in (10b), where the directly modifying adjective red occupies a position unexplained by the adjective hierarchy in (9). Another example is provided in (13), where visible has the two readings in (14): (13) the visible visible stars (14) a. indirect modification: not blocked from sight b. direct modification: sufficiently luminous, not too distant, etc.

Direct vs. indirect modification here has to do with permanent (direct) vs. contingent (indirect) properties. The only sensible interpretation of (13) is one in which the left occurrence of visible has the indirectly modifying interpretation of (14a) and the right occurrence has the directly modifying interpretation of (14b). (See Cinque 2003 and Larson & Marušicˇ 2004 for further discussion of the direct/indirect modification contrast.) Various observations suggest that indirectly modifying adjectives are construed with the head noun in a different, more loose way than directly modifying adjectives. Some of these observations suggest that directly modifying adjectives are heads and indirectly modifying adjectives phrases. For instance, directly modifying adjectives resist premodification (15), while discontinuous construal of adjectives appears to be restricted to the indirectly modifying type (16): (15) a. a ridiculously red Hungarian car b. #a Hungarian ridiculously red wine (16) dan-da kunya-a walbu-wa nga-ku-l-da kurrka-n ! this-nom small-nom raft-nom 1-inc-pl-nom take-negimp jungarra kurrka-tha walbu big:nom take-imp raft:nom ‘Let’s not take this small raft! Take the big raft.’

(Kayardild; Evans 1995:249–250)

Uncharted territory? 

In (15b), red loses the interpretation of (10b) (‘type of wine’), and has the strict color reading. In (16), the discontinuous construction of the boldfaced adjective and head noun yields a predicative (indirectly modifying) interpretation.5 Other observations suggesting a different syntactic construal between directly and indirectly modifying adjectives are illustrated in (17)–(18):  directly modifying adjectives tend to display morphological reduction (cf. also the absence of a linker in Chinese languages, (11)) and certain adjective positions allow only a directly modifying interpretation: (17) a. een vlot-te spreker (Dutch) a wellpaced-agr speaker ‘a fluent speaker’ (manner reading = direct modification) ‘a speaker who moves with ease, is well-dressed, etc.’ (characteristic of the person = indirect modification) b. een vlot spreker a wellpaced speaker ‘a fluent speaker’ (direct modification) *‘a speaker who moves with ease, is well-dressed, etc.’ (indirect modification) (18) a. un homme grand a man great ‘a great (significance) man’ (direct modification) ‘a great (size) man’ (indirect modification)


b. un grand homme a great man ‘a great (significance) man’ (direct modification) *‘a great (size) man’ (indirect modification)

These observations suggest that adjectives may be construed with nouns in two syntactically different ways, perhaps as heads in direct modification constructions, and as phrases in indirect modification constructions. Consider how these observations bear on the issue at hand, the (non-)cartographic structure of the clause. Since adjectives can be construed in two different ways, transitivity failures abound: given (12), a low adjective on the Vendler scale (9) with an indirect modification reading will always precede any directly modifying adjective, regardless its position on the adjective scale. (10b) is just one example. It follows that placement of the adjectives is not explained by the layout of a clausal map, but by local

5.  I have not found any languages where a discontinous adjective has a direct modification reading, but further investigation is necessary.

 Jan-Wouter Zwart

requirements forcing merger of nouns and adjectives (adjective phrases) in particular ways, depending on the intended interpretation.

3.5  Conclusion The data described in this section all point to the conclusion that the process of syntactic construction allows significant flexibility, thwarting efforts to design a uniform clausal cartography guiding the placement of syntactic objects in particular positions and orders. In the remainder of this article, we briefly consider some consequences of a more flexible approach to sentence construction for Germanic syntax.

4.  Some consequences for Germanic syntax I propose to discuss four areas here: the CP-domain (section 4.1), the position of topics and wh-elements (section 4.2), the subject position (section 4.3), and the object position (4.4).

4.1  The CP-domain The fine structure of the left periphery in Germanic syntax has been charted in much work predating Rizzi (1997), including Hoekstra (1993), Müller and Sternefeld (1993), and Hoekstra and Zwart (1994). An early illustration of a typical cartographic approach is Zwart (2000) (from 1996), where CP is taken to contain the three layers in (19), where C1 is occupied in Dutch by the demonstrative complementizer dat ‘that’ (associated with topics), C2 by the interrogative complementizer of ‘if/whether’, and C3 by the conditional/comparative complementizer als ‘if/when’): (19) [CP3 spec als [CP2 spec cond

of wh

[CP1 spec dat top

[TP (…) ]]]]

The approach is typically cartographic in that the order of the CPs is based on the attested pairs of complementizers in (20a), and on the absence of the pairs in (20b). (20) a.

als-of als-dat of-dat

‘as if ’ b. ‘that’ ‘whether’

*of-als *dat-als *dat-of

This allows us by inference to draw a complete map, even if the triple *als-of-dat is not found in any order. As shown in Zwart (2000), the structure in (19) makes correct predictions regarding the order of relative pronouns and complementizers in (dialects of)

Uncharted territory? 

Dutch: a demonstrative relative pronoun precedes dat but follows of, an interrogative demonstrative pronoun precedes of and dat, etc. This is explained on the reasonable assumption that the relative pronouns occupy designated specifier positions in the structure in (19). At this point we may wonder whether these results are lost under a more flexible approach. Quite the contrary, I believe. On a non-cartographic approach, we may assume that the derivation at a certain point (say, when a full subject-predicate combination [‘TP’] has been created), merges a complementizer to the current stage of the derivation (the workspace). A priori, we do not know anything about the feature make-up of this complementizer, so let us assume that it has only the categorial features (C). Following the analysis of agreement via sisterhood of Zwart (2006), we may assume that the complementizer’s morphological realization is the function of a sisterhood relation between a newly merged element (in this case, the relative pronoun) and the workspace in which the complementizer is contained. Concretely, when a relative pronoun is merged to the workspace, it shares certain features with its sister, which may then be realized on the sister’s head, C. The structures in (21) now illustrate how this might work with particular relative pronouns (where wie is an interrogative relative pronoun, and die a demonstrative relative pronoun, and the arrow indicates feature sharing/agreement): (21) die

wie C > of


C > dat


Dialect variation regarding the morphology of the complementizer (e.g., some dialects have wie dat rather than wie of) may be ascribed to the particular feature-to-form conversion of each dialect (adopting a morphology after syntax approach, as is common within minimalism; cf. Halle & Marantz 1993). With the exception certain rare patterns reported in the literature (of die order in the Amsterdam dialect reported by Hoekstra 1994:316, of met wie for the Strijen dialect reported by Van Craenenbroeck 2004:34), this suggests that the cartographic and dynamic approaches are equally well equipped to deal with the range of variation attested in the left periphery of relative clauses in Dutch (dialects). There is however a not uncommon order type which is puzzling from the cartographic perspective, but finds a natural analysis in the flexible approach. This is the order where two relative pronouns precede a single complementizer, as in (22) from Maastrichts.6

6.  The order die wad in Maastrichts is reminiscent of Bavarian der wo.

 Jan-Wouter Zwart

(22) de vrouw die wad of iech gezeen had the woman rel rel if I seen had ‘the woman I saw’

(Maastricht Dutch)

Here interrogative wad appears to occupy the specifier position of the interrogative complementizer of, but the position of demonstrative die is unexpected. Based on (19), we would expect the order *wad of die. On a more flexible approach, we expect a derivation like (23) to be possible, where wad is merged first, triggering agreement on C (23a), and further merger of die has no observable effect (23b): (23) a.

b. die

wad C > of

wad C = of

On this approach, agreement is a function of Merge, i.e., of the operation itself, and not a matter of valuation of preinstalled uninterpretable features (as in Chomsky 2001). The example of relative clauses in Dutch dialects illustrates that the observations which earlier gave rise to a cartographic analysis can easily be captured in a non-cartographic approach. In addition, certain facts which are puzzling from a cartographic point of view receive a straightforward analysis in the more flexible approach contemplated here.

4.2  The topic/wh-position The cartographic structure in (19) specifies designated landing sites for topics (spec,CP1) and wh-phrases (Spec,CP2). In Dutch and other continental West-Germanic languages, topics and wh-phrases are indeed fronted, i.e., externalized from TP, as illustrated in (24), and in the partial structure (25): (24) a. topic Dat boek ken jij niet that book know:2sg.inv you not ‘You don’t know that book.’


b. wh-phrase Welk boek ken jij niet which book know:2sg.inv you not ‘Which book don’t you know?’ (25) [CP2

welk boek C2


dat boek


[TP jij niet (…) ]]]

Depending on the type of clause, the verb ken occupies the C1 or C2 position in (25), yielding the verb second effect typical of Continental West-Germanic main clauses.

Uncharted territory? 

On a more flexible approach, we would have to describe the fronting of topics and wh-phrases following Van Craenenbroeck’s (2006) lead. Hence, a topic wants to be externalized from a domain which we may mark as ‘comment’ (26a), and a wh-phrase (a focus-marked category) from a ‘ground’ domain (26b).7 (26)



[ jij [dat boek]




jij [welk boek]

niet kent


niet kent


On this approach, we may follow Zwart (2005) and describe verb second as the positional dependency marking of the comment/ground domain after merger of the topic/wh-phrase (i.e., the verb in the verb second position functions as a linker between the newly merged topic/wh-phrase and the dependent comment/ground domain). In connection with this, the terms ‘comment’ and ‘ground’ strictly speaking apply only after merger of the topic/wh-phrase, creating a dependency where the dependent category, often TP, is a proposition which comes to function as comment/ground to the newly merged topic/focus element (cf. also Neeleman & Van de Koot 2008:144). The ‘inner conflict’ underlying topicalization and wh-movement, then resides in the circumstance that unmarked propositions in the relevant languages do not tolerate internal topic or focus elements.8 Verb second also occurs in subject-initial main clauses (hence SIMC) in Continental West-Germanic languages, which suggested to Den Besten (1977) that the subject in SIMCs occupies a position in what was later defined as the CP-domain. It is clear that this is not a necessary conclusion, even within a cartographic approach (see Travis 1984; Zwart 1993). However, within the non-cartographic approach contemplated here, the issue does not arise, as a statement of the type ‘The verb always moves to C’ or ‘Some category always moves to Spec,CP’ crucially refers to positions in terms which the non-cartographic approach does not recognize.9

7.  Note that the arrows in (26) represent a more complicated process, where no movement takes place, but dat/welk boek is erased after another token of the same item is merged to the comment/ground domain. 8.  See Neeleman and Van de Koot (2008) for a thorough analysis of the relevant word order patterns as the effect of externalization of topics/foci from their comment/ground. As Neeleman and Van de Koot (2008:146) correctly observe, the externalization requirement does not apply to new-information (‘wide’) focus in these languages. 9.  If Zwart (2005) correctly indentifies verb second as a mechanism marking part of the clause as dependent of a newly merged category, the possibility cannot be excluded that verb second applies after merger of a subject in the structural subject position (see section 4.3).

 Jan-Wouter Zwart

Arguments in the literature addressing the question of whether SIMCs are CP or TP are typically couched within a rigid cartographic approach (e.g., Schrijnemakers 1999). Just one example may serve to illustrate this. Schrijnemakers (1999:47–48) observes that adverbs in Dutch may be adjoined to TP in embedded clauses (27a). She then argues that if SIMCs are TPs, (27b), where the adverb is adjoined to TP, should be grammatical. However, as (27c) shows, in such cases the adverb is moved to Spec,CP and verb second applies. (27) a. ˉ…ˉ dat gisteren Jan het boek gelezen heeft that yesterday John the book read:part have:3sg ‘… that John read the book yesterday.’


b. *Gisteren Jan heeft het boek gelezen yesterday John have:3sg the book read:part c. Gisteren heeft Jan het boek gelezen yesterday have:3sg John the book read:part ‘Yesterday John read the book.’

In a non-cartographic approach, however, it is not clear that gisteren ‘yesterday’ occupies different positions in (27a) and (27c). The argument presupposes that gisteren is a topic, so let us assume that. We then assume, as in (26a), that topics are removed from unmarked propositions (‘TP’) in Continental West-Germanic, i.e., merged anew, followed by erasure of the topic from its position inside the proposition. This, then, applies in both main and embedded clauses, so that on a local, derivational definition of ‘position’, yesterday occupies the same position in (27a) and (27c). The differences between main and embedded clauses are caused by the trivial fact that a complementizer is merged to the derivation in embedded clauses but not in main clauses, and by the circumstance that the verb is used to mark dependency by position in main clauses but not in embedded clauses (explaining (27b)). The complementizer is merged only after the topic has been externalized, as in the Venetian examples discussed by Van Craenenbroeck (2006), cf. (5c). Also as in Venetian, the complementizer defines a topic domain from which (focus) wh-elements need to be removed, yielding the order in (28): (28) … welk boek of / dat / of-dat Jan gelezen heeft which book if / that / if-that John read:part have:3sg ‘… which book John read.’


As is well-known, topics do not appear in the pre-complementizer position in Continental West-Germanic embedded clauses, a mystery under the cartographic analysis which puts topics in Spec,CP in main clauses (cf. (27c)). On the non-cartographic approach, topics are invariably merged outside the propositional domain (as in (26a)), which may or may not be followed by merger of a complementizer, depending on whether topicalization takes place in an embedded clause or in a main clause.

Uncharted territory? 

To conclude, the non-cartographic approach does not describe fronting as movement to a particular landing site (say, Spec,CP), but as externalization of particular elements out of a certain stage of the derivation (essentially TP). On this approach, there is no ‘strong feature’ residing in C which forces Spec,CP to be filled. Hence, nothing is gained by describing the syntax of Continental West-Germanic main clauses as involving movement of the subject to Spec,CP. In non-cartographic terms, such a movement would be forced only if some inner conflict in TP were to force the newly merged subject to be externalized from TP again.

4.3  The subject position In the tradition of generative grammar it is standardly assumed that clauses have a structural subject position, which in recent years has been identified as the specifier position of TP (Chomsky 1981, 2001). Movement of the subject to this position is triggered by a mysterious EPP-feature residing in T, which attracts elements with particular categorial features (Chomsky 2001). The EPP (extended projection principle) simply states that clauses must have a subject (which seems right; cf. Chomsky 1982:10). The operation of subject placement, therefore, comes close to the kind of mechanism we have been assuming is involved in the placement of topics and focus/wh-elements: a given stage of the derivation needs a certain element to be merged outside of it. If there is substance to the claim that Spec,TP is the structural subject position across languages, it would have to be the case that Tense (the head of TP) brings something to the derivation which is in need of a subject. At the same time, Tense must be adding something to the constituent it is merged to (say, VP or vP), or else it would not have been included in the derivation. I will tentatively assume the following: (29) a. VP/vP represents a lexical domain (a structure of a verb with its arguments) b. Tense adds tense/aspect features, turning the derivation into an event c. the Subject adds a center to the event

A lexical domain as intended in (29a) (cf. Travis 2000) lacks anchoring in time, and hence is insufficient for reference to a state of affairs. This is why VP/vP needs to be supplemented with Tense features, yielding an event. But an event is incomplete without expression of a subject: the element to which the event applies. I propose to call the subject the ‘center’ of the event, and a derivation to which Tense and subject have been merged a ‘centered event’. The EPP may now be formulated as in (30), and ‘proposition’ may be defined as in (31): (30) EPP An event must be centered

 Jan-Wouter Zwart

(31) Proposition A proposition is the expression of a centered event

If the proposals in (29) are on the right track, it may be possible to derive the fact that Spec,TP is the universal subject position. This has nothing to do with features residing in T which attract the subject. Instead, it is an inherent characteristic of Tense features that they add a property to the derivation which creates the need for a subject. The relation between Tense and the subject therefore is indirect (there is no direct Tense-subject relation), but nonetheless very real. Returning to the status of the subject-initial main clause (SIMC) in Continental West-Germanic now, there appears to be no reason to believe that SIMCs are more than just TPs (i.e., derivations to which Tense and Subject have been merged). In this sense there is a symmetry between SIMCs and embedded clauses, where (in cartographic terms) the subject invariably occupies the Spec,TP position. It is therefore somewhat confusing that the analysis of Travis (1984) and Zwart (1993), where the subject is in Spec,TP in both main and embedded clauses, has been termed ‘asymmetric’ in the critical literature (e.g., Schwartz & Vikner 1996). A further question that may be raised concerns the ‘Force’ of a clause. In cartographic approaches, Force (i.e., declarative, interrogative, imperative) is a feature associated with a functional head in the CP-domain (see (4)). This may be taken over in a non-cartographic approach, albeit that it is going to be difficult to argue that the Force elements are functional heads rather than operators. However, for declarative force it is unclear that such an analysis is required. A viable alternative would appear to be that [declarative] is the unmarked interpretation of an unmodified proposition as defined in (31).

4.4  The object position Objects in Continental West-Germanic languages occupy a position in the ‘middle field’, i.e., between verbal elements on either end of the clause (more precisely, between the verb second position and the verb-final position).10 Within the cartographic tradition, it has proved difficult to define the object position. Chomsky (1989) proposed designated functional projections for hosting objects (AgrOP), but withdrew the proposal in later work (Chomsky 1995), on the grounds that the features relevant to these projections were included in the derivation only to help the derivation along. Since then, the standardly accepted position appears to have been that objects are licensed by ‘little v’, the element of agentive/causative semantics associated with transitive verbs

10.  This section deals with objects in canonical constituent orders, i.e., objects appearing in A-positions as a result of Object Shift (see Vanden Wyngaerd 1989).

Uncharted territory? 

(Chomsky 1995:315). This requires that multiple specifiers are associated with ‘little v’, as its projection also hosts the external argument of the verb. On a non-cartographic approach, multiple specifiers are unobjectionable: Merge can be reiterated without considerations of overall syntactic architecture. However, the association of objects with ‘little v’ proposed by Chomsky is problematic in light of facts discussed in Zwart (2001), where objects appear in the functional domain associated with unaccusative and passive verbs (which lack a ‘little v’ of the type that could license an object): (32) ..dat ze hem niet schijn-t te ken-nen  that 3sg.fem:nom 3sg.masc:acc not seem-3sg to know-inf ‘..that she doesn’t seem to know him.’


In (32), hem ‘him’ is an argument of the embedded verb kennen ‘know’, but it has been shifted to the left to a position in the matrix clause (i.e., to the left of the matrix negation niet ‘not’), where it finds itself in the functional domain associated with the unaccusative verb schijnen ‘seem’. From these and similar observations it may be concluded that languages display a process of object placement similar to subject placement, and therefore to be accounted for in similar terms, i.e., through some ‘EPP for objects’ (Lasnik 2001). For that we would have to know what the nature of the object position is, or better put, what the properties are of the workspace to which objects must be added. Work conducted in this area from a cartographic point of view suggests that the object position varies with the object’s discourse status (De Hoop 1992; Diesing 1992). For example, the object de telefoon ‘the telephone’ is interpreted as given in (33a), where it appears to the left of the discourse particle even (lit. ‘a little while’), and as new in (33b), where it appears to the right of it: (33) a. Wil je de telefoon even pakken ? want:2sg.inv you the phone prt take:inf ‘Please get the phone.’ b. Wil je even de telefoon pakken ? want:2sg.inv you prt the phone take:inf ‘Please get the phone.’


Example (33a) is most felicitous when both speaker and hearer are aware of the telephone (because it is ringing, for instance), while (33b) may be uttered when the telephone is new to the hearer (for instance when she is assisting the speaker who is packing to move). We know from Krivonosov (1977) that discourse particles of the type of even mark the watershed between old and new information. In the terminology applied above, we may say that a discourse particle defines a focus domain. In the situation which makes (33a) felicitous, de telefoon represents old information, which would cause an inner

 Jan-Wouter Zwart

conflict when contained within the focus domain defined by even. Hence, object shift in this situation is an instance of the type of externalization seen above, where a topic element is merged to a comment constituent. We may conclude from this example that objects may be merged at various time points in a derivation, depending on the nature of the current state of the derivation, and the intended semantic contribution of the object.11 This suffices to account for the placement of indefinite noun phrases, discussed by Diesing (1992) in strictly cartographic terms. Diesing observes that indefinites receive different interpretations, depending on whether they precede or follow discourse particles. In (34a), feuerwehrleute ‘firemen’ is interpreted generic (firemen are always available), while in (34b) feuerwehrleute receives an existential interpretation (there are firemen available): (34) a. … weil feuerwehrleute ja doch verfügbar sind since firemen prt available be:3pl ‘… since firemen are available.’


b. ˉ…ˉ weil jaˉdoch feuerwehrleute verfügbar sind since prt firemen available be:3pl ‘… since there are firemen available.’

Diesing (1992) proposes that existentially interpreted indefinites are inside VP, while those indefinites which receive a non-existential (e.g., generic) interpretation are outside VP. This assumes that discourse particles like ja doch ‘as we know’ mark the VP-boundary. However, this assumption is questionable, given the fact that the discourse particles may be realized further to the left; in that case, the two interpretations both are still available, if the prosodic cues present in (34) remain the same (Krifka 1991; Zwart 1995): (35) a. … weil ja doch feuerwehrleute verfügbar sind since prt firemen available be:3pl   ‘… since firemen are available’ (generic) b. … weil ja doch feuerwehrleute verfügbar sind since prt firemen available be:3pl ‘since there are firemen available’ (existential) (= (34b))

In (35), small caps indicate the syllables carrying primary pitch accent. These and similar observations (ja doch may also follow the indefinite and the prosody may still trigger an existential interpretation) suggest that the assumption that discourse particles mark the VP-boundary is too strong. If generic indefinites must be outside VP,

11.  See also Neeleman and Van de Koot (2008, section 3).

Uncharted territory? 

ja doch must be higher than VP in (35a). Hence it is difficult to map a structure of the clause on examples like (34)–(35). Underlying Diesing’s (1992) analysis of noun phrase placement is the idea that certain portions of the clause map onto certain portions of semantic representations, so that noun phrases in one position will receive a different interpretation from noun phrases in another position. This idea (phrased in cartographic terms by Diesing) is fully compatible with the non-cartographic approach. The assumption from Diesing’s work that appears to be untenable is that the relevant portions of the clause are defined with fixed phrase structure labels like VP, TP (see also Ter Beek 2008). What seems to be the case is that various factors (prosody, positioning of particles) contribute to the definition of certain subdomains of a proposition, and that these subdomains are relevant to semantic interpretation. It is precisely their relevance to semantic interpretation which may force leftward shift of objects of certain types (essentially externalization from the relevant subdomain, via remerge and erasure, as discussed above). While much remains unclear about the distribution of objects, the logic of the idea of an EPP for objects dictates that objects, like subjects, do not remain in their VP-internal argument position but are remerged to a certain stage of the derivation, deriving their position in the middle field. In this connection, it is important to note that even indefinite objects need not be adjacent to the verb in Continental West-Germanic languages (Zwart 1994; Ter Beek 2008). For example, adjunct clauses containing parasitic gaps may appear between a shifted indefinite object and the verb: (36) … dat er iemand een boek [zonder uit te lezen] that there someone a book   without out to read:inf terug gebracht heeft back bring:part have:3sg


  ‘… that someone returned a book without finishing it.’

This suggests that objects of any kind can be seen to shift to the left, vacating their original argument position inside the VP. The observations discussed in this subsection, then, suggest that the point in the derivation where the object is merged is not fixed. Hence it is impossible to identify any designated object positions. It does, however, leave the possibility that OV-languages with object shift are underlyingly head-initial (cf. Kayne 1994; Zwart 1994) wide open.

5.  Conclusion This article has made the following points. In a strictly derivational approach, syntactic positions can be defined in terms of their local environment, i.e., as a function of the

 Jan-Wouter Zwart

sisterhood relation created by the operation Merge. It follows that word order generalizations can be (and, from the point of view of theoretical economy, should be) defined in terms of local environments, not by reference to absolute, cartographically defined positions. On this approach, syntactic structure is inherently dynamic: each time a new element is merged to the current derivation, new features are imported, potentially creating ‘inner conflicts’ necessitating externalization of offending elements (i.e., new operations Merge followed by erasure of the offending element in its original position). The order of operations, then, is not determined by global considerations of syntactic architecture, but locally, on the basis of emerging properties of the derivation. If so, there is no way of guaranteeing fixed word orders, creating a flexibility which I believe is needed to describe language internal and crosslinguistic variation. We have shown how this non-cartographic approach is supported by a range of phenomena where word orders cannot be derived via reasoning by transitivity based on a fixed hierarchy of syntactic heads and projections. Finally, we have discussed a number of consequences of the approach for the analysis of Germanic syntax. Briefly, it appears unnecessary to maintain the full fine structure of the left periphery of Rizzi (1997) and others. Fronting of topics and focus elements can be described as forms of externalization, forcing relevant elements to appear outside the core proposition (‘TP’). Subjects, on the other hand, are by definition internal to the core proposition, leading to an analysis of subject-initial main clauses as being less developed than inversion constructions (in line with Travis 1984 and Zwart 1993). I have proposed that the EPP be understood as an externalization requirement, where a syntactic object representing an event needs to be combined with a noun phrase providing the event’s ‘center’. Finally, I have suggested that a similar requirement should hold of objects at an earlier stage in the derivation, explaining object shift as the result of a similar externalization requirement applying to objects of all kinds, but differently depending on the object’s intended discourse function. Within the confines of this contribution, it was regrettably not possible to proceed very far beyond the programmatic stage. Hopefully, this chapter serves its modest aim to raise a number of issues which might be addressed in future applications of the Minimalist Program to Germanic syntax.

References Bobaljik, J.D. 1999. Adverbs: The hierarchy paradox. Glot International 4: 27–28. Bolinger, D. 1967. Adjectives in English: Attribution and predication. Lingua 18: 1–34. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. 1982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge MA: The MIT Press. Chomsky, N. 1986. Barriers. Cambridge MA: The MIT Press.

Uncharted territory? 

Chomsky, N. 1989. Some notes on economy of representation and derivation. MIT Working Papers in Linguistics 10: 43–74. Chomsky, N. 1995. Categories and transformations. In The Minimalist Program, 219–394. Cambridge MA: The MIT Press. Chomsky, N. 2000. Minimalist inquiries: the framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, R. Martin, D. Michaels & J. Uriagereka (Eds), 89–155. Cambridge MA: The MIT Press. Chomsky, N. 2001. Derivation by phase. In Ken Hale: A Life in Language, M. Kenstowicz (Ed.), 1–52. Cambridge MA: The MIT Press. Chomsky, N. 2004. Beyond explanatory adequacy. In Structures and Beyond, A. Belletti (Ed.), 104–131. Oxford: OUP. Cinque, G. 1999. Adverbs and Functional Heads. Oxford: OUP. Cinque, G. (Ed.). 2002. Functional Structure in DP and IP:  The Cartography of Syntactic Structures, Vol. 1. Oxford: OUP. Cinque, G. 2003. The dual source of adjectives and XP- vs. N-raising in the Romance DP. Paper presented at NELS 34, Stony Brook, November 9. De Hoop, H. 1992. Case Configuration and Noun Phrase Interpretation. Ph.D. dissertation, University of Groningen. Den Besten, H. 1977. On the interaction of root transformations and lexical deletive rules. Ms., MIT and University of Amsterdam. Published in Groninger Arbeiten zur Germanistischen Linguistik 20 (1981), 1–78. http://gagl.eldoc.ub.rug.nl/root/1981-20/ Diesing, M. 1992. Bare plural subjects and the derivation of logical relations. Linguistic Inquiry 23: 353–380. Epstein, S.D., Groat, E.M., Kawashima, R. & Kitahara, H. 1998. A Derivational Approach to Syntactic Relations. Oxford: OUP. Evans, N. 1995. A Grammar of Kayardild. Berlin: Mouton de Gruyter. Frampton, J. & Gutmann, S. 2002. Crash-proof syntax. In Derivation and Explanation in the Minimalist Program, S.D. Epstein & T.D. Seely (Eds), 90–105. Malden MA: Blackwell. Halle, M. & Marantz, A. 1993. Distributed morphology and the pieces of inflection. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, K. Hale & S.J. Keyser (Eds), 111–176. Cambridge MA: The MIT Press. Hoekstra, E. 1993. Dialectal variation inside CP as parametric variation. In Dialectsyntax [Linguistische Berichte Sonderheft 5], W. Abraham & J. Bayer (Eds), 161–179. Opladen: Westdeutscher Verlag. Hoekstra, E. 1994. Overtollige voegwoorden en de volgorde of + interrogativum/relativum. De Nieuwe Taalgids 87: 314–321. Hoekstra, E. & Zwart, C.J.W. 1994. De structuur van de CP: Functionele projecties voor topics and vraagwoorden in het Nederlands. Spektator 23: 191–212. Jaspers, D. 1998. Categories and recursion. Interface 12: 81–112. Kayne, R.S. 1994. The Antisymmetry of Syntax. Cambridge MA: The MIT Press. Koeneman, O. 2000. The Flexible Nature of Verb Movement. Ph.D. dissertation, Utrecht University. Koster, J. 2007. Structure preservingness, internal merge, and the strict locality of triads. In Phrasal and Clausal Architecture:  Syntactic Derivation and Interpretation, S.  Karimi, V. Samiian & W. Wilkins (Eds), 188–205. Amsterdam: John Benjamins. Krifka, M. 1991. Focus, semantic partition, and dynamic interpretation. Paper presented at the 8th Amsterdam Colloquium, December 17–20.

 Jan-Wouter Zwart Krivonosov, A. 1977. Deutsche Modalpartikeln im System der unflektierten Wortklassen. In Aspekte der Modalpartikeln, H. Weydt (Ed.), 176–216. Tübingen: Niemeyer. Larson, R. & Marušicˇ, F. 2004. Indefinite pronoun structures with APs. Linguistic Inquiry 35: 268–287. Lasnik, H. 2001. Subjects, objects, and the EPP. In Objects and Other Subjects:  Grammatical Functions, Functional Categories and Configurationality, W.D. Davies & S. Dubinsky (Eds), 103–121. Dordrecht: Kluwer. Müller, G. & Sternefeld, W. 1993. Improper movement and unambiguous binding. Linguistic Inquiry 24: 461–507. Neeleman, A. & van de Koot, H. 2008. Dutch scrambling and the nature of discourse templates. Journal of Comparative Germanic Syntax 11: 137–189. Neeleman, A. & Weerman, F. 1999. Flexible Syntax. Dordrecht: Kluwer. Nilsen, Ø. 1997. Adverbs and A-shift. Working Papers in Scandinavian Syntax 59: 1–31. Nilsen, Ø. 2003. Eliminating Positions: Syntax and Semantics of Sentential Modification. Ph.D. dissertation, Utrecht University. Platzack, C. 1996. Germanic verb-second languages - Attract vs. Repel: on optionality, A-bar movement and the symmetrical/asymmetrical verb second hypothesis. In Deutsch -  typologisch, E. Lang & G. Zifonun (Eds), 92–120. Berlin: Walter de Gruyter. Rizzi, L. 1997. The fine structure of the left periphery. In Elements of Grammar, L. Haegeman (Ed.), 281–337. Dordrecht: Kluwer. Schrijnemakers, P. 1999. To C or not to C, that is the question: Over de structuur van subjectinitiële zinnen. MA thesis, Utrecht University. Schwartz, B. & Vikner, S. 1996. The verb always leaves IP in V2 clauses. In Parameters and Functional Heads, A. Belletti & L. Rizzi (Eds), 11–62. Oxford: OUP. Sproat, R. & Shih, C. 1988. Prenominal adjective ordering in English and Mandarin. Proceedings of NELS 18: 465–489. Sproat, R. & Shih, C. 1990. The cross-linguistic distribution of adjective ordering restrictions. In Interdisciplinary Approaches to Languages:  Essays in Honor of S.-Y. Kuroda, C. Georgopoulos & R. Ishihara (Eds), 565–593. Dordrecht: Kluwer. Ter Beek, J. 2008. Dutch indefinites, word order and the Mapping Hypothesis. In Groninger Arbeiten zur Germanistischen Linguistik 46, 55–72. http://gagl.eldoc.ub.rug.nl/root/ Volume46/ Travis, L. 1984. Parameters and Effects of Word Order Variation. Ph.D. dissertation, MIT. Travis, L. 2000. Event structure in syntax. In Events as Grammatical Objects:  The Converging Perspectives of Lexical Semantics and Syntax, C.  Tenny & J.  Pustejovsky (Eds), 145–185. Stanford CA: CSLI. Van Craenenbroeck, J. 2004. Ellipsis in Dutch Dialects. Ph.D. dissertation, University of Leiden. Van Craenenbroeck, J. 2006. Transitivity failures in the left periphery and foot-driven movement operations. Linguistics in The Netherlands 2006: 52–64. Vanden Wyngaerd, G. 1989. Object shift as an A-movement rule. MIT Working Papers in Linguistics 11: 256–271. Vendler, Z. 1968. Adjectives and Nominalizations. The Hague: Mouton. Zwart, C.J.W. 1993. Dutch Syntax:  A Minimalist Approach. Ph.D. dissertation, University of Groningen. Zwart, C.J.W. 1994. Dutch is head-initial. The Linguistic Review 11: 377–406.

Uncharted territory? 

Zwart, C.J.W. 1995. Word order, intonation, and noun phrase interpretation in Dutch. Proceedings of WECOL 7: 279–289. Zwart, C.J.W. 2000. A head raising analysis for relative clauses in Dutch. In The Syntax of Relative Clauses, A.  Alexiadou, A.  Meinunger, P.  Law & C.  Wilder (Eds), 349–385. Amsterdam: John Benjamins. Zwart, C.J.W. 2001. Object shift with raising verbs. Linguistic Inquiry 32: 547–554. Zwart, C.J.W. 2005. Verb second as a function of Merge. In The Function of Function Words and Functional Categories, M.  den Dikken & C.M. Tortora (Eds), 11–40. Amsterdam: John Benjamins. Zwart, C.J.W. 2006. Local agreement. In Agreement Systems, C.  Boeckx (Ed.), 317–339. Amsterdam: John Benjamins.

Bootstrapping verb movement and the clausal architecture of German (and other languages)* Gisbert Fanselow

University of Potsdam In the mainstream analysis of verb second clauses, the finite verb moves to Comp or one of the various heads present in the cartographic approach to CP. We show that such analyses are not satisfactory empirically, and cannot even be formulated within minimalist syntax. Verb movement – and head movement in general – should rather be analysed in terms of a ‘bootstrapping’ movement, in which the displaced head reprojects in its landing site.

1.  Introduction Bierwisch (1963) and Thiersch (1978), the first two milestones in the modern analysis of German syntax, proposed an analysis of main clause verb movement in which the finite verb does not target a pre-existing structural slot. Rather, the verb adjoins to TP in verb second (V2) constructions (if we permit ourselves to use current terminology when we refer to work written more than three or four decades ago). Bierwisch and Thiersch differed in their treatment of the first position in V2 constructions. Bierwisch “squeezed” the verb in between the first and the second constituent of TP, while in the

*This paper relates an updated version of ideas first presented at the 2001 UCLA-Potsdam Head Movement Workshop, to the approach developed in Fanselow & Lenertová (accepted). The research reported here was partially supported by DFG grants FOR 375 (A3) and SFB 632 (A1) to the University of Potsdam, and by a grant of the Humboldt-Foundation to Anoop Mahajan and myself. For helpful comments on this and earlier versions of the paper, I am in particular grateful to Artemis Alexiadou, Joanna Błaszczak, Werner Frey, Denisa Lenertová, Gereon Müller, and Ralf Vogel. Many thanks also go to the anonymous and the non-anonymous (viz., Olaf Koeneman) reviewer of the present paper. Some of the ideas in this paper have been presented at the University of California at Los Angeles, the Humboldt University at Berlin, and the Universities of Leipzig, Poznan, Stuttgart, Thessaloniki, Utrecht, and Warsaw. Many thanks go to the audiences for discussion.

 Gisbert Fanselow

model proposed by Thiersch, the adjunction of the verb to the TP-root is followed by a further transformational process, viz. the fronting of some arbitrarily chosen constituent to the left periphery of the clause (targeting the node “COMP”). This paper argues that these early accounts of the V2 phenomenon are superior to the CP-model developed later. Neither does the finite verb move to Comp, nor is movement to the first position triggered by (non-trivial) features characterizing the XP moved to the left periphery. Section 2 discusses a number of problems for the idea that the verb moves to a pre-existing head position in V2 constructions. Section 3 reviews (and rejects) some recent proposals trying to overcome the theoretical difficulties of classical head movement in various ways. In section 4, we argue for a model in which the verb moves to the left by adjoining to the root. It moves there as a head rather than as the remnant of vP, just as proposed by Bierwisch and Thiersch. Verb movement is carried out in a bootstrapping (‘Münchhausen’) fashion:  the verb carries both the attracting and the attracted feature (Section 4). This can be shown to be the only way in which a head can move. Section 5 complements the picture by investigating the first position in V2 clauses.

2.  Empirical and theoretical problems of movement to Comp 2.1  Movement to Comp in a non-cartographic account The Bierwisch-Thiersch model of V2 constructions described above (henceforth BTM) was replaced by the CP-approach (henceforth CPA) illustrated in (1), in which the verb goes to Comp and the first constituent occupies Spec,CP, for mainly two reasons:  (a) the presence of an overt complementizer seems to exclude V2 movement while its absence appears to force it, and (b) the verb in second position and lexical complementizers seem to play similar roles as hosts of weak pronominals (the “Wackernagel position”), or as loci of verum focus accents. Observations of these kinds were first discussed in den Besten (1977/1983/1989). (1) [CP [das Buch]i [[C hatk][TP die Maria ti dem Jakob gegeben tk ]]]   the book has the Mary the.dat Jakob given “Mary gave the book to Jakob”

Den Besten’s arguments appeared quite compelling in the grammatical model in which they were first formulated, so that the CPA soon became the standard for German and the other Germanic V2 languages. However, later insights imply that the arguments for the CPA are not conclusive. Consider, first, the argument involving the complementarity of overt complementizers and V2 movement. If V2 movement is a substitution operation, the fact that it cannot apply when there is an overt complementizer can indeed be explained by

Bootstrapping verb movement 

assuming that the finite verb moves into Comp. This argument for the CPA has force when one confines one’s attention to Dutch and Standard German, two languages in which one never1 finds V2 movement in clauses introduced by an overt subordinating complementizer.2 This seemingly supports the view that the finite verb and overt complementizers compete for the same position. However, in many if not all other languages with the V2 property (Frisian, Yiddish, the Scandinavian languages, Kashmiri) V2 constructions do occur in clauses introduced by an overt complementizer (see Vikner 1995; Bhatt 1999) under appropriate contextual conditions (see Bentzen, Hrafnbjargarson, Hróarsdóttir & Wiklund 2007; Julien 2007). The discussion in Vikner (1995) shows how difficult it is to capture embedded V2 constructions if one presupposes the CPA. All such difficulties disappear when one ceases to ground one’s analysis of verb second on a parochial property of German and Dutch, viz. the absence of verb fronting in CPs with overt complementizers. Abandoning the CPA furthermore allows a uniform treatment for Germanic verb second constructions, similar second position effects in Slavic languages (which occur irrespective of the presence of a complementizer) and verb fronting in Bantu languages such as Kinande (Schneider-Zioga 2007). Note also that while the absence of an overt complementizer may be a necessary condition for verb fronting in German and Dutch, it is not a sufficient one. Reference to the status of Comp thus explains very little. Indirect constituent questions as in (2) are incompatible with verb fronting in German, while embedded clauses involving the ‘topicalization’ of some constituent tolerate (and require) verb movement, as in (3). V2 movement is thus not causally related to the phonetic emptiness of Comp. The identification of the trigger of verb fronting in the CPA is a non-trivial issue, and it seems fair to say that no really satisfactory solution has been proposed so far. (2) a. Es ist unklar wen sie eingeladen hat it is unclear who she invited has “it is unclear who she invited” b. *Es ist unklar wen hat sie eingeladen (3) [DP die Behauptung [CP den Mann habe sie entführen wollen]] the claim the.acc man has she kidnapped wanted “the claim that she had wanted to kidnap the man”

1.  Freywald (2007, in preparation) documents a number of cases of V2 movement in clauses introduced by wenn ‘if ’ and even dass ‘that’ which she found in various corpora. Not all of them can, apparently, be explained away in terms of construction breaks. This suggests that German allows for embedded V2 in a very mild form, too. 2.  Some non-subordinating conjunctions such as denn ‘for’ allow (and require) V2 complements. This suggests that they do not occupy the position of complementizers.

 Gisbert Fanselow

Overt complementizers and finite verbs in second position also do not behave alike with respect to the specifiers of CP they are compatible with. Overt complementizers co-occur with wh-phrases in Spec,CP in embedded questions as in (4) in several dialects of German (see, e.g., Bayer 2004) and it is not clear why a finite verb in Comp is never able to do so (2b). In contrast, overt complementizers coexist with overt topicalization into Spec,CP (as exemplified in (5)) under very restricted circumstances only (both in terms of pragmatics and in terms of dialectal distribution, see, e.g., Bayer 2001 for a discussion), while topicalization in embedded verb second clauses as in (3) is as unrestricted as it is in main clauses. In the grammar of embedded clauses, complementizers and verbs in second position have little in common, quite an unexpected property if the CPA were correct. (4) I woass ned wer dass des dõa hot(Bavarian German) I know not who that this done has “I do not know who has done that” (5) An Mantl dass da Xaver kafft hot hot neamad glaubt (Bavarian German) a coat that the Xaver bought has has nobody believed “As for a coat, nobody believed that Xaver bought one”

The major argument in favor of the CPA is thus problematic because it makes a uniform treatment of Germanic V2 constructions and second position phenomena in other languages very difficult. Let us now turn briefly to Den Besten’s second argument for the CPA. It involves the observation that finite verbs in second position and overt complementizers are targeted in the same way by clitic-like elements such as weak subject pronouns. These pronouns have to immediately follow the complementizer in embedded clauses, and the finite verb in V2 matrix clauses with non-subject topicalization. The empirical facts can also be captured if one assumes that these weak pronouns target the highest overtly realized head of a clause, or appear directly above TP (if Comp is the only head potentially appearing above TP, see Zwart 1993 for more discussion). They do not force the CPA. If we abandon the CPA, we must have an answer as to why V2 constructions are root phenomena in a certain sense. After all, even if there are embedded V2 construction, these are often confined to particular pragmatic conditions. According to Bentzen et al. (2007), embedded V2 is possible only in contexts in which the embedded clause could be the ‘main point of the utterance (MPU)’ (but the embedded V2 clauses need not actually constitute the MPU, and in embedded clauses that are the MPU the verb need not move to second position). Julien (2007) argues that embedded V2 constructions have an independent illocutionary force (IIF), viz., assertion, of their own (see also Meinunger 2006). Embedded clauses with an MPU/IIF potential behave like root clauses in other respects as well. E.g., the occurrence of topic markers such as Japanese wa in embedded clauses is also restricted by the nature of the embedding verb (Kuroda 2005) in a way reminiscent of what one observes for Germanic V2 constructions.

Bootstrapping verb movement 

Structurally, these insights can be represented in different ways. CPs that allow for V2 constructions may involve a head (such as Force, see below, as proposed by Julien 2007) that CPs lacking an IIF do not have and that is the target of verb movement. Alternatively, clauses without an MPU/IIF potential may possess a feature (value) that is selected3 by, e.g., factive verbs and that is incompatible with V2 movement or the presence of an independent topic-focus articulation. There is a tradition beginning with Kiparsky & Kiparsky (1970) that attributes noun phrase-like properties to finite CPs, see in particular Müller (1995). In the sense of this tradition, one can assume that the feature excluding MPU potential is nominal in character (N or D). This assumption is in line with the finding that noun phrases have an articulated information structure licensing focus sensitive particles only when they involve a nominalization, i.e., when they embed a verbal projection in which the focus sensitive particles are actually situated (see Kleemann-Krämer, accepted). This nominal feature can then be made responsible for the overall impossibility of V2 movement in clauses without an MPU potential. German dass-clauses have an MPU/IIF potential when embedded in appropriate contexts. In line with this fact, it has also been observed that the low position for topics argued for by Frey (2004), and the position for contrastive foci and topics (Haider & Rosengren 2003; Frey 2006) can be filled in embedded clauses introduced by the overt complementizer. The absence of embedded V2 movement in German (and Dutch) is thus not due to the absence of root-clause like properties in certain embedded contexts, rather it reflects a purely formal property coming with the complementizer. We will return to the issue below.

2.2  A cartographic approach to V2 movement? If the verb does not target the position of the complementizer, it might still move to some pre-existing head position. Indeed, the rich functional structure postulated for the left periphery of the clause by Rizzi (1997) sketched in (6) offers a set of potential landing sites for the finite verb that seems to be large enough for allowing a satisfactory analysis of verb second constructions. An overview of some pertinent proposals can be found in Poletto (2002)

(6) [α Force [β Topic [γ Focus [δ Topic [ε Fin TP]]]]]

The enriched structure is in principle flexible enough to capture both languages in which overt complementizers and V2 do not co-occur and languages in which they do, because overt complementizers can be merged in different positions in different

3.  As Julien (2007) shows, the type of CP that a matrix predicate accepts may change when the latter is negated. One therefore should not speak of ‘selection’ in the sense of lexical selection/subcategorization.

 Gisbert Fanselow

languages. In particular, if German dass ‘that’ is merged as a Fin (later moving up all the way to Force), the finite verb cannot even move to Fin, as observed in Pili (2000), taking up an idea proposed by Poletto (1995). It has to stay in TenseP when the complementizer is present. For there to be an option of V2 movement in clauses with an overt complementizer, one merely needs to assume that this complementizer can be merged in a position higher than Fin. Of course, the most urgent question within the cartographic approach is: which head does the verb move to? As we shall see, only Force and perhaps Fin (as originally proposed by Poletto 2002) can host the verb, and the Force node involved in V2 movement is different from Rizzi’s Force (see Julien 2007). V2 movement thus does not really fit into the cartographic approach of Rizzi (1997). In the cartographic approach, V2 movement is a process triggered by a feature attracting an inflected verb present on some or all of the functional heads in (6). For German, we need to assume that the verb always goes to Force, so that the clause-initial XP is in Spec,ForceP. Movement to lower positions would not work If the clauseinitial DPs in (7) would occupy Spec,FinP, Spec,FocP, and Spec,TopP, respectively (in line with their pragmatic functions), Fin, Foc, and Top would have to attract the finite verb. Furthermore, the finite verb must not raise too high: it can only be placed behind the highest specifier that is overtly realized. This cannot be guaranteed easily if the left edge XPs in (7) would occupy different positions. Either we would have to make the implausible4 assumption that the structure in (6) is never projected beyond the highest overtly filled specifier, or we would have to invoke a constraint implying that the heads in (6) can attract heads only if they also attract specifiers. Both solutions are not very attractive. (7) a. es hat in Falkensee geregnet it has in Falkensee rained “it rained in Falkensee” b. wen hat Katharina gesehen? who has Katharina seen “who has Katharina seen? c. (Any news about Helena?) Die Helena hat ein Hund gebissen The Helena has a dog bitten “Helena has been bitten by a dog”

4.  This solution is not tenable if one assumes that the property of being the (contrastive) focus is structurally represented by agreeing with/being the specifier of a Focus head. In a sentence such as (5a), FocusP would thus have to be present when in Falkensee is a (contrastive) focus, even though es cannot raise beyond beyond FinP.

Bootstrapping verb movement 

In her analysis of related Scandinavian facts, Julien (2007) opts for a third solution, in which V2 clauses are always ForcePs. The finite verb in (7) is placed into Force, which possesses an EPP-feature attracting the closest argument to its specifier. The highest of the specifiers of FinP, FocP, and TopP thus ends up in Spec,ForceP. This is an effective way of making sure that the verb is placed immediately behind the highest overt XP. Unfortunately, this solution comes with a cost: if V2 movement targets Force, embedded V2 constructions show that Force must not be equated with the location of the complementizer – rather, we need an additional functional layer above ForceP to host the complementizer, as Julien observes herself. Within the cartographic model, we expect that the verb can also stop in the other functional head positions below Force, in a language-particular fashion. In other words, there should be variation concerning the position of finite verbs. Is this prediction borne out? At first glance, it seems we can give a positive answer, but the variation is quite restricted, as we shall see. Let us consider a few languages. The finite auxiliary nearly always appears in second position in Warlpiri (Legate 2002), i.e., it immediately follows the focus γ when there is no higher topic β, while it appears between β and γ when the higher topic position is filled, but only if the topic is not base generated as a ‘hanging topic’. Warlpiri is thus more or less just like German. The situation is different in Breton (Legendre 2001), Hungarian (Bródy 1990), and Greek (Tsimpli 1995). Here, the verb follows a displaced focus constituent, but it never goes further to the left, i.e., it follows the specifier γ of a Focus-phrase in the cartographic approach. This could be interpreted as representing a linguistic system in which Focus is the highest head possessing a strong head-attracting feature. Closer inspection suggests, however, that the difference to German and Warlpiri is located in a different part of grammar. Topics preceding the focus are always base-generated in their high position in Breton, Hungarian and Greek, as evidenced, e.g., by the fact that topics are obligatorily resumed by pronouns in the clause. We propose that these base-generated topics are adjoined to the highest clausal projection, so that a unified account of German, Warlpiri, Breton, Greek, and Hungarian is possible: the finite verb always goes to the highest head position Force,5 with the base-generating topics being adjoined6 to its maximal projection (or a higher one). Bantu languages such as Kinande constitute a further subcase of verb fronting above TP. The verb is not placed immediately behind the focus constituent in Kinande (Focus is represented by an independent overt head in this language), but there is a

5.  Apparently, this could also capture the array of facts found on Rhaeto-Romance (Poletto 2002), but a discussion is beyond the scope of the present paper. 6.  Unlike TPs (as in yesterday, John came) CPs are not, however, targets for adjunction by movement.

 Gisbert Fanselow

‘verb second’ phenomenon in the constituent following the focus head: in addition to ‘unmarked’ SVO order, objects and locatives can be fronted across subjects if the verb inverts with the subject. One further difference to Germanic verb second constructions lies in the fact that the verb agrees with the fronted constituent rather than with the subject in Kinande OVS and loc_V_S sentences (see Schneider-Zioga 2007). In a cartographic system, one could analyse this kind of verb movement as a preposing to the second (lower) Topic head (as assumed by Schneider-Zioga) or to Fin, with the fronted (post-focal) constituent occupying the position of δ. Kinande might thus be interpreted as showing that the verb always climbs to the projection hosting the highest specifier – but not if the head of that projection is lexicalized independently (as in the case of Focus in Kinande), in which case the verb stops in the projection of the second highest specifier. Alternatively, we could say that the verb either goes to Force or to Fin (see also Poletto 2002). The sample of languages we have discussed is certainly much too small for being representative. Still, it is certainly revealing that there is no evidence for the finite verb stopping in Topic and Focus,7 and that the Force node targeted by V-movement (Poletto 2002; Julien 2007) is not the Force node argued for by Rizzi (1997). In other words, the nodes which the verb goes to (Force, Fin) neither find an independent lexicalization, nor do they correspond to overt inflectional morphemes.8 This makes V2 movement much different from V-to-Tense movement, where the target position of movement (Tense) is independently motivated. The theory of V2 movement must explain a very striking fact: the head the verb is claimed to adjoin to is always phonetically empty. The idea that the verb moves to one of the heads in the Comp domain may be attractive if the overall grammatical model would allow substitution operations, in which the verb could move “into” one of the Comp nodes. In such an analysis, it may be plausible that the presence of an overt head blocks a further substitution, or that substitution into a head eradicates any material previously dominated by that head.

7.  This may cast some doubt on the existence of these nodes. Note that a model postulating Topic and Focus faces further difficulties outside the domain of verb placement. Some languages such as Venetian (see van Craenenbroek 2006; Zwart 2006) employ ordering requirements such as topic > focus, focus > che, che > topic, which cannot be translated into a cartographic system because of the obvious transitivity problems that arise when such ordering statements are translated into hierarchical relations between heads. Samek-Lodovici (2006) argues convincingly that the left peripheral focus of Italian is in fact situated at the right edge of the clause, being followed by right-dislocated material. This seriously undermines the line of reasoning in Rizzi (1997) in favor of a Focus-phrase in Italian. See also Lahne (2008) for further difficulties of the cartographic model. 8. 

Recall that Fin comes in addition to the Tense node representing verbal inflection.

Bootstrapping verb movement 

But if the merger of two syntactic objects is the only available operation in syntax, as assumed in minimalism, then substitutions cannot arise. If the finite verb adjoins to some head in the context of V2 constructions, it is an urgent question why adjunction can take place in the V2 context only if the head adjoined to is phonetically empty. Why does the verb refrain from adjoining to an overt Focus head in Kinande? Why does the verb refrain from adjoining to a lexical complementizer? Why are only those heads targets for adjunction that are never realized overtly? These are serious questions that should find an answer in a theory of V2 movement.

2.3  Movement targets the root only It is both interesting and reassuring that the empirical and conceptual problems that head adjunction theories of V2 constructions and the CPA face correspond to serious theoretical difficulties of head movement in general. In the minimalist program (Chomsky 1995, 2000, 2005), adjunction to the root of a phrase marker is the only way how movement may be executed, and it is obvious that this excludes adjunction of α to a head H as in (8), because the movement of α targets a position immediately dominated by the root of the phrase marker, not the root itself.

(8) [HP [H H] [YP … α … ]]

→ [HP [H α H] [YP … tα … ]]

Head movement of α understood as an adjunction of α to some other head H is impossible in the minimalist program, and there are good reasons amply discussed in Mahajan (2003), Fanselow (2001, 2003), Matushansky (2006), Suranyi (2005) and Müller (2004) (among others) for taking this consequence to be a welcome one. E.g., α does not c-command its trace in (8), in contrast to what holds for phrasal movement in general. The empirical observations discussed in 2.1. and 2.2 suggest that the V2 construction does not involve an adjunction of the finite verb to Comp or any other head above TP, and minimalist syntax predicts that such a adjunction structure cannot be created. The CPA and the idea that V moves to Force cannot be formulated in minimalist syntax.

3.  V2 as movement to a specifier position 3.1  The remnant movement approach to V2 constructions If the movement of a head is a syntactic operation (see, e.g., Lechner 2005; Matushansky 2006 for arguments in favor of this position) it cannot but be executed as an adjunction to the root. In other words, the only way of forming a V2 construction within the limits of the minimalist program is by adopting the BTM! In current terminology, Bierwisch and Thiersch differ as to whether the preposed verb may be “tucked in” (in the sense of Richards 2001) or not (but see below).

 Gisbert Fanselow

There are several ways in which the idea that the verb adjoins to the root may be made precise. The displacement of the verb might involve head movement in a strict sense, or it might be executed as remnant movement of the verb phrase. In the latter case, the verb (phrase) is attracted to a specifier position. The former model can also be formulated in such a way that the verb goes to a specifier. That head movement of X can be reinterpreted as the remnant movement of an XP out of which all material but the head was extracted was first proposed by Sportiche (1998) and Mahajan (2003). The idea was applied to German V2 phenomena by Müller (2004), see also Nilsen (2003) for a related approach. (9) Den Peter fragt die Irina nie the.acc Peter asks the Irina never ___________ ____ _________________ α β Σ _________________ _________________ γ Σ “Irina never asks Peter”

(analysis A) (analysis B)

A ‘conservative way’ of executing the remnant movement idea lies in having fragt move as a remnant vP (or, perhaps, TP9) in (9), followed by the leftward movement of some further constituent α. This is represented as analysis A in (9). The movement of α and β could be triggered by empty head(s) X. In a bolder move, Müller (2004) proposes that the first constituent and the finite verb of a V2 clause actually form a constituent γ, the vP (analysis B).10

9.  Finite TPs cannot undergo any type of fronting in German, however (see e.g., Wurmbrand 2008), so that it would be quite a mystery that remnant TP movement is allowed. This constitutes a serious problem for the remnant movement theory of V2 if the finite verb goes to Tense in German. That it in fact does is made likely by the grammaticality of structures such as (i). The category sitting before the finite verb denke of the matrix clause is, arguably, a vP out of which the head kommst was moved inside the complement CP. Tense is the only plausible landing site for this movement. (i) [vP [ trocken] [durch den Regen] t] denke ich nicht dass du kommst dry   through the rain think I not that you come “I don’t think you’ll get through the rain dry” Likewise, the German future auxiliary werden behaves much like its English counterpart (it accepts no complements but vPs, it cannot be embedded under any other auxiliary, it cannot show up in infinitives), which suggests it is a Tense element, too. Of course, it shows up in second position in main clauses, and would thus force a remnant TP analysis of V2 constructions. 10. 

See also Biberauer & Roberts (2004) and Lechner (2007) for a critique of Müller (2004).

Bootstrapping verb movement 

Remnant movement analyses appear attractive because they can be formulated in quite ‘traditional’ terms. The fronting of the vP can be conceived of as being triggered by an attracting feature of a head, and remnant movement of verbal projections is a standard process in German or Dutch (see Thiersch 1985; den Besten & Webelhuth 1990; Müller 1998). The remnant movement theory of V2 constructions loses much of its appeal, however, when one realizes that the shape of the to-be-fronted constituent must be fully stipulated (in contrast to what holds for well-established instances of remnant movement). E.g., Müller’s (2004) “Edge Domain Pied Piping Condition (EPC)” explicitly demands that a moved vP may only contain the highest head of vP and its highest specifier. Thus, the shape of V2 clauses is, essentially, stipulated in a construction specific way by the EPC. Analyses of type A have to guarantee that the vP that moves to second position does not contain more material than fits into a single head, and they do so by employing one kind of ‘complexity’ filter or the other. Somewhat ironically, theories working with constraints restricting the complexity of vP-remnants are more plausible if their shape constraints do not have to obeyed rigidly. If the second position of V2 constructions is always only filled by a single word, this property follows from a head movement theory while it needs to be stipulated in a remnant movement approach, so the former model is preferable if it can be made fit the overall syntactic architecture. If, on the other hand, the second position of a V2 clause is sometimes more complex than a syntactic word, the head movement theory is at a loss, while remnant movement approaches capture such constellations quite easily. Nilsen (2003) claims that the category moving to second position is in fact sometimes larger than a single verb, an observation that would lend considerable support to the remnant movement analysis of V2. According to his view, (10) shows that weak pronouns are pied-piped when the verb is preposed in Norwegian (as in the other Mainland Scandinavian languages). Certain adverbs may intervene between the constituent in first position (derfor) and the preposed verb (10a). The weak pronouns always have to immediately follow the finite verb, independent of the verb’s position, as (10c) shows. If (10b) derives from (10a) by a movement of svarte, the obligatory pied-piping of the pronouns shows that verb preposing can transport more than just a single head, as expected in a remnant movement model. (10) a.

Derfor vanligvis bare svarte ’n ’a ikke. Therefore usually just answered he-her not

b. Derfor svarte ’n ’a vanligvis bare ikke. c. *Derfor svarte vanligvis bare ’n ’a ikke.

The argument is not really compelling, however. The different linear positions occupied by the verb in (10a–b) do not force an analysis in which the verb actually moves from one position to the other. Rather, it suffices to allow that certain adverbs also adjoin to the first projection of the head in which the verb sits when it is in second

 Gisbert Fanselow

position (in addition to adjunction to TP).11 This also accounts for data such as (11). The additional adjunction option is absent in German, which makes the V2 property there very strict (see (11)). (11) a. Jens nesten  gråt Jens nearly cried b. Jens gråt nesten (12) a. *Jens fast weinte Jens nearly cried b. Jens weinte fast

There are no compelling examples in the V2 domain that would show that remnant vP movement can transport more than just a single head (or more than a specifier and a single head, in the EPC model). The remnant movement analysis of V2 thus needs to invoke a miracle, viz. the miracle that all material but the finite verb leaves vP. Such miracles can be indeed happen if forced by constraints coding the structure of the construction one wants to describe, but a theory that does not depend on such principles is certainly preferable. Remnant movement accounts of verb fronting even need more machinery in order to work properly. There must be shape conservation constraints that guarantee that the material extracted from vP or TP before remnant movement is arranged in identical order before and after it is extracted from vP. Note also that head movement is more local than phrasal movement (as was established in detail by Travis 1984), and that the choice between head and phrasal movement is governed by a principle such as the Head Movement Generalization (13) of Pesetsky & Torrego (2000). (13) Head Movement Generalization Suppose a head H attracts a feature of XP as part of a movement operation. (i) If XP is the complement of H, copy the head of XP into the local domain of H. (ii) Otherwise, copy XP into the local domain of H.

The stricter locality of head movement and principles such as (13) are a mystery in remnant movement accounts:  why should the linear distance between the attractor and the attractee be correlated with complexity restrictions on the specifier position of the attractor? The model must also allow that elements that cannot undergo “normal” movement (such as the negative head nicht “not” in German (14a–c), or verbal particles (15a–c)) nevertheless participate in the evacuation operation preceding remnant movement (14d, 15d).


See, e.g., Collins & Thraínsson (1996) for adverb adjunction to intermediate projection.

Bootstrapping verb movement 

(14) a. dass es heute nicht regnet that it today not rains “that it does not rain today” b. *Nicht regnet es heute Not rains it today “It does not rain today”

(no movement to the left periphery)

c. *dass nicht es heute regnet

(no scrambling) (=14a)

d. es regnet heute nicht


(15) a. dass er das Lied an-fängt that he the song at-catches “that he began with the song” b. *dass eran das Lied fängt

(no scrambling)

c. *An fängt er das Lied At catches he the song “he begins with the song”

(no movement to the left periphery)12

d. Er fängt das Lied an

Analysis B in (9) is confronted with a further problem. Phrases that originate in a very high position of the clause such as sentential adverbs constitute excellent left edges of a V2 sentence. If sentential adverbs originate in TP or higher (as seems necessary), the grammaticality of (16) requires that TPs may undergo remnant movement, but normally, finite TPs cannot be moved at all (see endnote 9). (16) Wahrscheinlich liebt er Probably loves he “Probably, he loves Vera”

Vera Vera

Furthermore, the model needs to make sure that the movement of some XP out of vP/TP for the sake of preparing V2 movement does not have the typical consequences of movement such as the creation of scope ambiguities (mediated by the trace of movement) or freezing. There is thus a long list of ancillary assumptions one needs to make if the remnant movement theory of V2 constructions is to work. This is not to say that remnant movement models cannot be properly formulated, it just means that when we analyse verb movement as “remnant movement”, we are not really employing the kind of remnant movement that we know from other domains of grammar, rather, it is a second type of

12.  However, particles occur in the left peripheral position of German main clauses in the context of remnant vP movement, when all material but the particle has been removed from vP. What is at stake is not whether a particle may be part of the remnant undergoing movement, what is at stake is whether the particle may move itself and be evacuated from the remnant.

 Gisbert Fanselow

remnant movement with radically different properties13 (see Müller 2002 for a detailed discussion) that has no raison d’etre but allowing the reanalysis of head movement as phrasal movement, and that has very little in common with standard movement.

3.2  Head movement and M-merger V2 movement cannot be analysed as a substitution, since such an operation cannot be formulated in the minimalist program. We saw in section 2 that V2 movement should not be implemented as an adjunction to a pre-existing head either, and we found remnant phrasal movement accounts of head movement quite unattractive because of the additional machinery they presuppose, and because of the lack of independent motivation. We are thus left with no alternative but BTM: the verb itself adjoins to the highest clausal projection in an instance of head movement proper. The BTM can be made precise in more than a single way. After adjunction to the highest clausal projection, the verb can either figure as a specifier or as a head. Let us begin our discussion of BTM with the former option. Matushansky (2006) does not only provide us with a very lucid discussion of the differences and commonalities of head and phrasal movement, she also proposes that the movement of a head H is triggered by the c-selection features of some other head K. H thus becomes the specifier of K after movement. This yields a constellation for head movement such as (17). (17) [KP H [[K] HP]]

If, e.g., K is Tense and H is a verb, the two categories involved in head movement should end up forming a morphological unit, and Matushansky argues that such units are formed by an operation independent of the syntactic movement, viz. m-merger, that is part of the morphological component and transforms (17) into (18). Note that complex heads such as V + Tense act as syntactic units for further derivational steps. Consequently, Matushansky proposes that m-merger applies immediately after head movement. More generally, she assumes a cyclic concept of spellout applying after each step in the derivation. (18) [KP [[K H K] HP]]

One very appealing aspect of her analysis lies in the separation of head movement proper from the m-merger component. She shows that the two operations often apply independently of each other.

13.  Consequently, there are no independently motivated expectations concerning which elements can be extracted from vP before remnant movement of vP to second position, which makes it close to impossible to falsify the approach.

Bootstrapping verb movement 

Matushansky’s approach is able to overcome the theoretical problems arising with the traditional view that heads adjoin to heads only, i.e., it specifies a concept of head movement that is compatible with core assumptions of minimalist syntax. In a very obvious sense, however, when applied to verb second constructions, her proposal shares the negative aspects of the CPA – for the simple reason that movement is again effected by an additional attracting head. Why does this attracting head have to be phonetically empty? Why can the verb not move in front of a focus marker or a complementizer without m-merging with that category?

4.  The bootstrapping way of verb movement There is a tradition (Ackema, Neeleman & Weerman 1993; Koeneman 1995, 2000; Bury 2000; Fanselow 2001, 2003) that upheld the BTM for German and Dutch in its purest form. This tradition is grounded in a proposal Holmberg (1991) made for the analysis of VP-shells: substitutions into empty heads should be replaced by adjunctions of heads to maximal projections followed by a re-projection of the moved head. It is the model we adopt in this paper: in V2 constructions, the finite verb adjoins to its own projection and re-projects in its landing site. In section 4.1, we will develop a restricted theory of head movement. Together with standard assumptions on the location of strong features, the Chain Uniformity Condition will be shown to imply that head movement is strictly local and of a bootstrapping nature: the element that moves possesses the strong feature that its projection checks. Section 4.2 applies this model to V-to-Tense movement. The bootstrapping analysis of V2 constructions is the topic of section 4.3.

4.1  Head movement is bootstrapping in nature Let us begin with a sketch of what a theory of syntactic head movement should look like. Movement is an operation that can target the root of a phrase marker only. Therefore, heads adjoin to the root, too, when they move. A head movement constellation thus seems to always look like (19). Head X is adjoined to a non-minimal projection of some head Y attracting it, rather than to Y itself. (19) [YP X [[Y] [XP …. tX ….]]]

(19) is not yet the correct structural representation of head movement, but it comes close to it, and – as in Matushansky’s proposal- it solves most of the theoretical problems arising in the ‘traditional’ way of carrying out head movement: the trace of movement is c-commanded by its antecedent, and the head merges at the root of the phrase marker. However, (19) violates the “Chain Uniformity Condition” (Chomsky 1995:253), that requires that the phrasal status of a category does not change during the movement

 Gisbert Fanselow

process. Therefore, (19) cannot be correct. X is a maximal projection in (19), while its trace is a minimal one. Recall that Speas (1990) and Chomsky (1995) established a configurational definition of projection levels. Σ is a maximal projection unless its mother is a projection of Σ. Σ is a head if Σ does not dominate further (non-terminal) material. If a head H adjoins to another category α, its mother fails to be a projection of H if the structure [α H α] results. Therefore, in [YP [X [Y Y]] [XP …. X ….]]], the trace of X is not maximal, while the moved head acquires that status in its landing site. This violates the Chain Uniformity Condition, and excludes head movement (unless we exempt certain positions from the algorithm computing phrasal status, as proposed by Chomsky 1995). This problem for head movement disappears once we give up an assumption that characterized generative research for many years, viz. that moved items cannot project in their landing position. The idea that moved items can ‘reproject’ was first formulated in van Riemsdijk (1989), but it was not allowed into mainstream minimalist syntax before Chomsky (2005), citing Donati (2005) as a primary source of evidence for the need of granting the moved item projection potential. That both the category adjoined to and the adjoining category are, in principle, capable of projecting after movement is also conceptually very natural if movement is a special case (‘internal merge’) of the merge operation, that does not by itself impose any constraints on which element is able to project. The license to project after movement is also the cue to an understanding of verb movement, as argued for in the works cited at the beginning of this section. The constellation created by head movement thus looks more like (20). (20) [XP X [ … [Y] [XP …. tX ….]]]

Let us work with the default assumption (21). In conjunction with the Chain Uniformity Condition, (21) implies (22), as can be seen easily (cf. Fanselow 2001, 2003). Consider (23) in this respect. In (23a), the target Σ of movement has projected further after α was adjoined to it. In this case, α is maximal (because its mother is not projected from it), so that α must have been maximal in its root position, too. Otherwise, the Chain Uniformity Condition would be violated. When some α adjoins to a category that projects after movement, α has thus undergone phrasal movement. When α projects after movement (as in (23b)), it is a head by definition, so that α must have been a head in its original position, too. In a head movement constellation, it is always the moved head that projects after it is adjoined. Its sister inevitably becomes a complement. (21) After the attraction of α to the root of Σ, either α or Σ may project. (22) When a head moves, it re-projects after movement. (23) a. [Σ α Σ] b. [α α Σ]

If this consequence is interpreted in a proper theory of checking, the strict locality of head movement can be derived, which constitutes a major argument in favor of the

Bootstrapping verb movement 

approach proposed here. It will also follow that the only way by which heads can move is a bootstrapping one. This is also a welcome consequence. These conclusions hold if the assumptions in (24) are correct. (24) a. α can merge with head H as a specifier or complement only if α checks a feature of H b. If a strong feature f is checked in [H α] or [α H1], f is a feature of H.

Both assumptions are common in current versions of minimalism. (24a) follows Chomsky (2005) since he assumes that (external) merge and move (internal merge) do not come for free, they must be triggered by a syntactic feature. Given (24b), only heads possess the strong features triggering syntactic processes such as movement. Together with the Chain Uniformity Condition, the assumptions in (24) imply the strictly local and bootstrapping nature of head movement. In bootstrapping movement, the moving item possesses both the strong feature f and the matching feature f+ checking f. We will show that head movement cannot arise with a different distribution of the features involved in the checking operation. So suppose that H has a strong feature f, and suppose that Δ possesses the matching feature f+. Movement is not bootstrapping in nature when H ≠ Δ. If H ≠ Δ, there can be no head movement. Let us see why: The most economical way of checking f consists of (externally) merging Δ with H as a complement, or as a specifier (should the complement position already be filled). In this case, either [HΔ] or [ΔH1] are generated, and no movement (in particular, no head movement) is licensed in a constellation such as (24b). Suppose, then, that Δ bearing the feature f+ that matches a strong requirement of H is properly embedded in α in one the constellations in (24). Where the locality conditions on checking permit it, f may still be checked in the context of merging α (containing Δ) with H.  This would give rise to either [H α] or [α H1], without any movement being involved. If the locality conditions on checking disallow it that Δ checks f on H when α merges with (a projection of) H, then f must be checked by extracting Δ or some Γ properly containing Δ. In the latter case, the moved item is more complex than a head ex hypothesi, so that we can disregard this situation in our discussion of head movement. Hence, we must consider the constellation (25) only. (25)

[Ώ Δ [Θ H [Π … tΔ …]]]

(25) represents a proper case of phrasal movement, in which case H projects so that Ώ = HP and Θ = H1. But it cannot be a head movement constellation. In that case, Ώ would be a projection of Δ, and HP would be equal to Θ. f+ on Δ would check the strong feature f on H/HP, which is not immediately compatible with (24b): The category possessing the strong feature HP is a complement and not a head (as required by (24b) if (25) is headed by Δ.

 Gisbert Fanselow

The only situation not yet excluded by (24b) is one in which f+ on Δ is strong, too – or in which there is an additional strong g on Δ checked by Θ.14 (24b) would then be satisfied if we interpret it in such a way that at least one strong feature must sit on the head when strong features are checked in a derivational step. We are thus considering a case of head movement in a context of the mutual checking of strong features. Note, however, that (25) would be ruled out in the situation under consideration if Π is a projection of Δ, because the features of Δ would then have been accessible to H and be checked by this head when Π was merged with H. Mutual checking of features leading to head movement of Δ would therefore presuppose the more complex constellation in (26), in which ΔP is not a complement of H, but rather embedded in at least one further intervening maximal projection headed by some R. (26)

[Ώ Δ [Θ H [Π ..R [ΔP … tΔ …]]]]

(26), however, is not compatible with the principle of the strict cycle. Chomsky (1995) proposes to implement this principle in terms of the ‘timing’ of the checking of strong features. Strong features must be checked immediately. In particular, no YP must contain any unchecked strong features when YP is merged with the projection of a different head. In the constellation we are considering, ΔP contains such a strong feature that is only checked when Δ moves in front of H, i.e., the feature was still unchecked when ΔP was merged with R. This is incompatible with the strict cycle. Summing up, we have seen that there can be no movement of a head Δ if the strong feature attracting Δ sits on a category different from Δ. Consider now the situation in which the strong feature f of H is matched by a feature f+ sitting on H itself. External merger cannot check f for obvious reasons. Consequently, (27) is the only constellation in which f can be checked, licensing head movement under the narrow constraints imposed by (24). Head movement is always bootstrapping: the feature triggering the movement of the head sits on the head itself. (27) [XP … X … ] → X [XP … tX …]

On obvious grounds, (27) does not violate the strict cycle, i.e., the requirement that strong features of X must be checked before the projection of X is embedded in a projection of a different element. (27) also satisfies the Chain Uniformity Condition. (27) thus turns out to be the only legal way in which a head can move. In a constellation leading to the head movement of X, the relevant feature must not be too deeply embedded. (27) implies that the head cannot move too far. In fact, it can only place itself immediately above (one of) its own projection(s). In its net effect, this is equivalent

14.  This is not meant as a claim that multiple or mutual checking of strong features is actually possible.

Bootstrapping verb movement 

to the Head Movement Generalization uncovered by Pesetsky & Torrego (2000). When X undergoes head movement as in (27), there is no attractor different from X present in the structure. In terms of the output, this consequence is identical to the effects of the generalization that the attracting category must always be invisible in head movement. Our restrictive15 theory of head movement thus meets all requirements a model of head movement should fulfil.

4.2  V-to-tense and V-to-V as examples of bootstrapping movement The preceding subsection has established that there is head movement only if the feature f attracting H is located on H itself. We now have to show that this constellation is indeed realized by lexical entries. Inflected verbs are obvious candidates.

15.  Typologies of head movement such as Roberts (1994) and van Riemsdijk (1998) distinguish at least three different types of head movement. V2 movement and the movement of the verb to Tense belong to a first type A, in which the head H does not adjoin to another overt word. Type B involves the adjunction of a head H to another independent word, as often assumed for ‘restructuring’ contexts such as transparent infinitives in German, see (i). (i)

dass er [TP PRO Kira ti] [V [V zu küssen]i wagt]] that he Kira to kiss dares “that he dares to kiss Kira”

Type C is “long head movement” constructions as we find them in, e.g., Croatian (ii). (ii) daoi mu ga je Ivan ti given him it is Ivan t “Ivan has given it to him” Type A can be reconstructed in terms of bootstrapping movement, but not type B. Riemsdijk (1998) argues that type B verb movement can be found in contexts only in which the attracted head and the attracting head are adjacent already before movement. Furthermore, the two verbal heads in clause-union or related transparent infinitival constructions do not function has a unit for further syntactic processes such as head movement of type A (e.g., the two verbs never go together to second position in a German or Dutch matrix clause), which rather argues against the idea that a syntactic unit is actually formed in such constellations. One major motivation for head movement then lies in the increased transparency of complements in clause union contexts, which can, however, be guaranteed otherwise. Another motivation for head movement by adjunction could lie in local inversion processes in verb-auxiliary clusters. Not all of them can convincingly be analysed in terms of head movement, however, and a purely PF-related solution may be superior (see Vogel 2003). Type B also figures in noun incorporation constructions (see Baker 1988, 1996), but a discussion of these is beyond the scope of the present paper. For long head movement (type C), we can assume it involves phrasal movement to a specifier position.

 Gisbert Fanselow

An inflected lexical item such as French aime “loves” combines features of both a verb and Tense, it has a representation such as (28). Syntactically, it is a verb selecting a D-feature, and a Tense element selecting a verb.16 We follow the minimalist assumption (Chomsky 2005) that two categories can be merged if that implies the checking of selectional features. Following Fanselow (2001, 2003) and Matushansky (2006) (see also Suranyi 2005), we argue that overt head movement is a consequence of the “strong” nature of selectional features. If the selectional feature (__V) of aime is strong, it must be checked by overt movement.17 Given the syntactic object in (29), the only way of checking this selectional feature is by adjoining aime to Σ. Σ is a projection of a verbal feature (because of the content of (28)), and can thus check the selectional requirements of aime in the creation of (30). (28) 〈PHON: aime, SYN: {Tense (_V), V (_D)}, SEM: 

(29) [Σ aime Marie] (30) [aime [aime Marie]]

Since aime already is a lexical item that has Tense features, we do not need to assume that aime moves from a V position and adjoins to a pre-existing Tense position. Rather, we can analyse V-to-T as shown in (30): when aime moves up, it adjoins to the root and reprojects. This is not a bad consequence given that aime is a Tense element. In other words, in languages that are traditionally analysed as allowing V-to-Tense movement, structures like (31) arise by a bootstrapping type of movement of the verb analysed as a V-Tense complex. The verb possesses the checking feature and feature to be checked at the same time (the probe and the goal are identical).18

16.  The set of syntactic features of inflectionally complex heads thus consists of several feature complexes that ideally correspond to the features of the morphemes the complex head was formed of in the numeration. 17.  Why could the strong selectional V-feature of aime not be checked by merging it with a further verb? This is excluded because the Tense-‘part’ and the verbal part have been already semantically composed in the lexicon. A further verb could not be semantically integrated into the structure. 18.  More precisely, in such structures, the two feature complexes of V and Tense co-project. Our proposal is thus reminiscent of the matching projection idea introduced into the generative discussion of German by Haider (1988). If the highest projection of aime in (31) derives its syntactic properties more from Tense than from V, we can express this property by assuming that categorical features become syntactically inert in terms of further projection once they have been checked.

Bootstrapping verb movement 


[Tense, V] subject

[Tense, V] [Tense, V]

[Tense, V] [Tense,V]




In contrast, in a language such as English, love is entered as a verb (and not as a V-Tense) into the syntactic representation. Therefore, a phonetically empty Tense element must be selected in the numeration if a TP is to be generated, but no V movement can apply (because such a movement would not be bootstrapping in nature in English). The resulting structure (32) is quite classical. (32)

[Tense] subject

[Tense] [Tense]

[V] [V]




Note that, just as in Matushansky’s approach, head movement never creates a morphologically complex word. Rather than assuming that a post-syntactic operation such as m-merge is responsible for the formation of the inflected verb (which presupposes the application of spellout after each derivational step), we follow Chomsky (2000) and have inflectional morphology apply within the numeration rather than postsyntactically. In other words, we assume a version of distributed morphology that applies before rather than after syntax19 (see Müller 2007a, Alexiadou & Müller 2008). In the context of the discussion of verb movement to Tense, it is worth being pointed

19.  A crucial difference between the combination of two morphemes in the morphology and the merge operation in the syntax lies in the absence of feature checking in the former process. Thus, even though the Tense morpheme possesses a V-feature, this feature is not checked when Tense and V come together morphologically.

 Gisbert Fanselow

out that any connection between the ‘richness of inflection’ and movement can only be expressed in a pre-syntax morphology, for obvious reasons (Müller 2007a, but see Alexiadou & Fanselow 2000). Bootstrapping head movement is also applicable in the context of Larsonian VP-shells. Larson (1988) showed that the verb moves up to a position c-commanding all arguments in the English VP, as shown in (33). We have already remarked that Holmberg (1991) proposed an implementation of this idea in terms of reprojection: gave adjoins to its own maximal projection, and reprojects after movement. (33) [VP gave [VP the book [t to Mary]]]

The movement in (33) can be triggered by the presence of a strong V-selecting feature in the lexical representation of verbs, which is added in English when a verb is entered into a numeration. This strong V-selection feature can only be checked by a bootstrapping movement of the verb. It should be pointed out that bootstrapping movement defines a mechanism by which verbs and other heads can move. It does not (and cannot) answer the question as to why such a movement applies from a functional perspective. Haider & Rosengren (2003) propose that Larsonian V-movement applies in English because of a conflict between the direction of licensing of arguments (V licenses an argument to its right) and the left branching nature of syntactic structure. Bootstrapping V-movement thus enables English verbs to play their role in licensing arguments. The mechanism and the function of verb movement must be distinguished. Strong features have to be checked in the derivation as early as possible, so the question arises why a verb such as aime cannot (does not have to) adjoin to itself immediately when it is entered into the syntactic derivation. This would lead to an unwanted structure such as (34). (34) [[aime [aime ]] Marie]

Such derivations can be blocked easily. Notice that the feature V of aime is checked in the very first step in (34). Suppose that the checking of the categorical feature of X inactivates X for the rest of the derivation, in the sense that X can no longer trigger any syntactic operations (serve as a probe). In the case of (34), this means that the D-selection of the feature V can no longer be satisfied, so that the derivation will crash because of the unchecked D-feature. In the context of the interaction of two syntactic objects that have a single syntactic feature complex only, the assumption just made implies the strict cycle, because it means that no operation within XP can be triggered by X once XP has been merged with Y, so that its categorical feature was checked by Y.  In the context of syntactic objects with several feature complexes on a single head (such as tensed verbs) formed in the course of the application of morphology in the numeration, it implies that

Bootstrapping verb movement 

the features of complex Y (say: V) must have been completely checked before Y can serve as the checking element (goal) in a bootstrapping movement triggered by Z (e.g., Tense), as required.

4.3  Bootstrapping V2 movement Above, we have argued that the finite verb does not move to any pre-existing head position in V2 constructions. Since head movement can be carried out in a bootstrapping way only, this property of V2 constructions is captured straightforwardly in the model proposed here. The element moving in V2 constructions is the finite verb, so that a strong feature checking Tense is a straightforward candidate for the element triggering V2 movement. It would be added to the feature complex of a finite verb once it is selected for the numeration. The addition of this feature can (and has to) be linked to pragmatic properties such as ‘independent assertion’,20 so that the basic distribution of V2 constructions (matrix clauses and certain embedded clauses) is described. In the most parsimonious approach, the strong feature checking Tense would be part of a feature complex that is a Tense itself. Ideally. there would even be only one Tense feature complex. The structure of a verb second clause would thus look like (35): the finite verb moves out of TP, adjoins to it, and reprojects. Koeneman (2000) postulates the same structure, but for different reasons. (35) [TP XP Tense [TP YP Tense vP]]

Note that the chain of Tense has two specifiers in (35). As mentioned above, in the Bantu language Kinande, Tense agrees with the upper specifier XP (yielding agreement with whatever immediately precedes the overt finite verb). In the Germanic languages, agreement goes with the lower specifier YP, yielding agreement with the (nominative) subject. This general property does not exclude that agreement may depend on the position of the nominative DP. In particular, it often seems relevant for agreement in the Germanic languages (and in other languages, see Samek-Lodovici 2002) whether the nominative DP is c-commanded by Tense, or c-commands it. Thus, we see different agreement patterns for nominative subjects and nominative objects in Icelandic (Sigurðsson 1996; Hrafnbjargarson 2001), and differences in the shape of agreement in several dialects of Dutch depending on whether the subject or some other category precedes the verb (see Zwart 1993). Relative to (35), the latter variation depends on whether Tense c-commands the nominative DP or not, so that we do not need to explain agreement differences in terms of the verb being in Tense or Comp.

20.  This link must be stipulated, just like, e.g., the presence (and properties) of a Force node as a syntactic reflex of independent assertion in other approaches.

 Gisbert Fanselow

As remarked above, for V2-languages other than Dutch and German, the link between the pragmatic property ‘independent assertion’ and the addition of the Tense selecting feature in the numeration captures the basic distribution of V2 clauses. More must be said for Dutch and German, two languages in which the presence of an overt subordinating complementizer excludes verb second constructions even under favourable pragmatic circumstances. Apparently, these complementizers select a feature of their complement that is incompatible with V2 movement – an unchecked Tense feature in the ideal case.21 The presence of a feature triggering bootstrapping movement may very well be functionally motivated on grounds of clause-internal dependency marking (as suggested by Zwart 2005, 2006), or on grounds of scope requirements on Tense (Koeneman 2000), just as overt phrasal movement may exist in order to make “clausal typing” possible and mark distinctions of information structure. In a minimalist syntax, one must not, however, mix up the various functions a construction22 may have with the formal means that guarantee that a construction can be formed. From a formal perspective, the V2 property corresponds to the endowment of verbs with the extra bootstrapping feature when they enter the numeration for matrix clauses.

5.  Filling the first position 5.1  (Why) Is the verb always mapped to second position? Theories of V2 usually tacitly assume a number of principles that guarantee that the finite verb ends up in second position. In the context of the CPA, we would have to assume that

21.  If the checking of the Tense feature in the context of V2 movement makes this feature syntactically inert, it could be invisible to the selection requirements of German and Dutch complementizers. Complementizers compatible with V2 movement as we find them in Scandinavian languages would then have to be sensitive to a different feature such as finiteness or agreement present in the Tense-complex. Alternatively, one could assume that the strong Tense-selecting feature is embedded in a feature complex that is not a Tense itself (say, a Force) and that would be incompatible with the c-selection of the complementizers in Dutch and German. 22.  The situation in V2 movement contexts is, however, different from V-movement in English VP-shells. The need for licensing argumental DPs in VPs has been established independently of attempts of functionally motivating V-movement, so their integration into a theory of VP shells is welcome. In contrast, it is not clear whether the proposals for the function of V2 movement reflect requirements that must be assumed independently of attempts of understanding V2 movement.

(36) a. b. c. d. and e.

Bootstrapping verb movement 

a CP with a verb in Comp must have a specifier this specifier must be filled by overt material CP has no multiple specifiers there are no adjuncts to CP there are no adjuncts to C′.

In a cartographic approach, corresponding assumptions need to be made for the head (e.g., Force) to which the finite verb moves. The EPC (see section 3.1) essentially stipulates some of these properties in a remnant movement account, but for properties such as b. or d., extra principles are required. For the V2 languages under consideration, there is no motivation for multiple specifiers of the same head for any category, so we will assume a counterpart of (36c), too. More precisely, we will assume that there is at most one specifier each time a head builds up a projection. In the case of V2 constructions, there will thus be at most two specifiers of Tense, viz. XP and YP of (35), as required. As for (36a–b), we agree with Zwart (2005) that topic drop constructions such as (37) refute (36b), and we also take over his view that the postulation of empty operators in Spec,CP for constructions such as yes-no-questions (38a), imperatives (38b), conditionals (38c) and narrative initial-declaratives is not really motivated but as an attempt to maintain a verb-second requirement in the light of clear counter-evidence. In other words, there is no intrinsic, grammatically necessary relation between the presence of a strong Tense-selecting feature (triggering verb fronting) and of a strong edge feature responsible for the filling of the prefield position preceding the finite verb. This relation exists only for certain illocutionary forces/moods: optionally for imperative mood (see 38b, 39), and obligatorily for assertions. (37) OP hab ich schon gesehen have I already seen “I’ve already seen that” (38) a. Hast Du Michele gesehen? Have you Michele seen? “Have you seen Michele?” b. Gib Du mir mal sofort mein Buch zurück! Give you me once at once my book back “Give back my book at once! c. Schläfst Du so sündigst Du nicht Sleep you so sin you not “While sleeping you do not sin” (39) Das Buch gib mir morgen zurück the book give me tomorrow back “Give me back the book tomorrow”

 Gisbert Fanselow

(36d–e) seem to distinguish CPs (40b) from TPs (40a) and might constitute a problem for our proposal analysing V2 clauses as TPs, too. Note, however, that there is exactly one A-bar-position in a V2 clause, viz. the highest specifier. If A-bar-positions cannot be c-commanded by A-positions in the same derivational phase, (40b) is already excluded. (40) a.

Yesterday, John came

b. *Gestern John kam Yesterday John came

5.2  The nature of the first position Does the verb “squeeze” itself in between the first constituent and the rest of the clause (as would be in line with the ideas put forth in Bierwisch 1963)? Are there two movements of the verb and some XP to the root of TP, or does the first constituent in a V2 construction undergo standard movement to Spec,CP (=‘COMP’, in the terminology of Thiersch 1978)? Note that the correctness of the last option would somewhat reduce the appeal of BTM, because it would force us to postulate an additional CP layer just above the position the verb moves to. Yet, the (at least superficial) resemblance of German and English constituent questions suggests that wh-phrases go to Spec,CP in German, too. That appears to undermine our analysis of V2 constructions. (41) wen hat Anna gesehen? who.acc has Anna seen


(42) who has Anna seen?

However, the analysis of German wh-questions is not that simple. Data that resists an analysis of wh-questions in terms of the establishment of an agreement relation between the wh-phrase and a [+wh]-Comp can be found easily. Wh-phrases can both move beyond their scope (as in (43), see, e.g., Müller & Sternefeld 1996) and stop below their scope position (as in (44), see., e.g., Fanselow 2005). (43) wen sag mal dass sie einlädt? who.acc say once that she invites “tell me who she invites” (44) was denkst Du wen sie einlädt? what think you who she invites “who do you think that she invites?

The constructions in (43) and (44) support the view that the head attracting whphrases to the left periphery of a German clause is not involved in wh-feature

Bootstrapping verb movement 

agreement.23 German wh-phrases thus do not move to Spec,CP. No Comp-layer above TP needs to be assumed for V2 clauses. However, the data is still compatible with a model in which phrases occupy the left periphery in German because they have undergone operator movement involving [+topic] and [+focus] features (the latter being responsible for the fronting of wh-phrases, too), i.e., in which the phrases in the left periphery are specifiers of Topic and Focus heads. However, the specifier positions of these two heads would not suffice to plausibly host all material that can appear in front of the finite verb in V2 constructions, as Fanselow & Lenertovà (accepted) show in detail. First, as already observed by Travis (1984), subjects can appear in the preverbal position without any particular pragmatic function, e.g., in all-focus constructions (45) or when they are meaningless expletives (and cannot, thus, be topics or foci), as in (46). (45) What happened? Ein Eisbär hat einen Zoobesucher verletzt a polar bear has a zoo visitor hurt “a polar bear hurt a zoo visitor” (46) es hat geschneit it has snowed

The privilege of being allowed to go to the first position of declaratives without any special pragmatic force is not confined to subjects, however. Rather, there is a strong correlation between the position preceding and following the finite verb in pragmatically unmarked clauses. As initially observed by Hubert Haider in a GGS talk of 1989, and as shown in detail in Fanselow (2002) and Müller (2004), all elements (with a few exceptions such as weak object pronouns) that are able to occupy the left edge of the TP behind the finite verb in matrix clauses and behind the complementizer in embedded clauses can also appear in front of the finite verb without any special pragmatic marking. This holds for the dative (non-subject) arguments of passive and unaccusative clauses, for sentence adverbs, and for temporal adverbs (see (47)). (47) What happened? a. Gestern Nacht hat ein Eisbär einen Touristen verletzt yesterday night has a polar bear a tourist hurt “Last night, a polar bear hurt a tourist”

23.  That wh-phrases (rather than other XPs) have to be fronted in questions is not guaranteed by the attracting feature, then. Rather, it is a consequence of the requirement of having clausal scope that wh-phrases must fulfil. They can take scope by movement to the specifier of the clause they quantify over, or by linking to a “scope marker”.

 Gisbert Fanselow

b. Man hört, dass gestern Nacht ein Eisbär einen Touristen one hears that yesterday night a polarbear a tourist

verletzt hat hurt has

“One hears that a polar bear hurt a tourist last night”

Furthermore, Fanselow & Lenertovà (accepted) present ample evidence for the claim that what goes to the left periphery may also be only part of the focus (and part of the topic), as in (48), where the context sentence induces VP-focus. (48) What have you done this morning? Ein Buch habe ich rezensiert a book have I reviewed “I have reviewed a book”

Furthermore, (49) illustrates that German allows the fronting of parts of idiomatic expression (underlined) to a much greater extent than English (as noted by Horne 2003). (49) a. Mit den Hühnern bin ich ins Bett gegangen with the chicken am I into-the bed gone “I went to bed quite early” b. Alle Tassen hat der aber nicht mehr im Schrank all cups has he but not more in-the cupboard “He is quite crazy”

Fanselow & Lenertová (submitted) argue that such data illustrate exactly the same point:  movement to first position is not tied to any pragmatic function whatsoever in German.24 (45) – (49) show that movement to the left periphery is triggered by an unrestricted edge feature,25 and Fanselow & Lenertovà develop a locality concept for


The same is true for Czech and several other languages.

25.  For obvious reasons, this unrestricted edge feature may happen to pick a category that is part of the focus of the utterance, larger than the focus, or coincide with the focus. As (i) shows, all three options are well formed, e.g., in a context in which a question such as the one in (i) introduces a narrow focus on the object. See Fanselow & Lenertovà (accepted) for more details on focus and topic constructions. (i) What have you bought in town? a. Bücher hab ich mir ein paar gekauft books have I me a few bought b.

Ein paar Bücher hab ich mir gekauft “I have bought a few books”


Ein Buch gekauft hab ich  mir a book bought have I me “I have bought a book”

Bootstrapping verb movement 

this movement linked to the linearization theory of Fox & Pesetsky (2003, 2005) and Müller (2007b). The evidence for the first position is thus in line with the situation concerning the second position: empirical facts do not support the view that movement to the first position is triggered in the context of agreement for features such as wh, focus or topic linked to Comp or some of the heads postulated in the cartographic approach. It seems as if the head moved to second position has an unrestricted edge feature that allows any category to move to the left periphery.

5.3  The order of attraction: Does the verb ‘squeeze’ itself in? As mentioned above, Hubert Haider made the important observation that whatever can appear in left periphery of the clausal projection behind the complementizer in pragmatically unmarked sentences (except deaccented object pronouns) can also show up in the left periphery of V2 clauses before the finite verb. Fanselow (2001, 2003) and Müller (2004) extend this observation by noting that there are a number of operations such as scrambling (50), contrastive focus movement, and sentence internal movement of topics that can place nearly any category at the left edge of the clausal projection below the complementizer: (50) dass den Fritz niemand t mag that the.acc Fritz nobody likes “that nobody likes Fritz”

The sets of elements that may appear in the left periphery of matrix clauses and in the left periphery of embedded clauses (behind Comp) are nearly identical. This is predicted by Bierwisch’s version of the BTM, because he “squeezes” the verb in between the leftmost XP and the rest of the clause. The leftmost XP is always in the same position in V2 clauses and in clauses introduced by a complementizer. The two structures differ only in terms of the position of the verb. If movement always targets the root of a phrase marker, the Bierwisch model cannot be translated literally into a minimalist approach.26 Its most important effects are, however, retained if the bootstrapping movement of Tense is due to the prefinal feature checking operation of Tense. Recall that both (external) merger and movement (internal merger) are operations linked to the checking of features. The feature of Tense that is checked last corresponds to the leftmost element in the clause; if the feature triggering head movement is checked immediately before this last feature, the verb will in effect be “squeezed” in. In such a model, no edge feature is added to main

26.  However, if V2 movement were a PF-process, it would have to be carried out exactly along the lines envisaged by Bierwisch. Thanks to Olaf Koeneman for pointing this out.

 Gisbert Fanselow

clause Tense in order to make sure that the verb ends up second. Alternatively, one could assume an additional head (Comp, Fanselow 2002; Frey 2005, or Force, Julien 2007) above TenseP that attracts the uppermost XP in TenseP to its specifier, or (as we have done above) an extra edge feature of Tense in V2 clauses moving the XP that was merged immediately below the verb in second position to the left periphery. A full-fledged version of the squeezing-in approach has to explain why unstressed object pronouns can go behind complementizers while they cannot precede the verb in second position (51–52). Empty topic operators, on the other hand, are confined to the left periphery of V2 clauses (53). (51) dass ’n keiner mag that ‘him noone likes “that noone likes him” (52) *n’ mag keiner him’ likes nobody “nobody likes him” (53) a. OP hab ich schon gesehen have I already seen “I’ve already seen that” b. *dass OP ich schon gesehen hab that I already seen have

Negative adverbs such as nicht and kaum can be leftmost in embedded clauses, but must not show up in the first position of matrix clauses (54). The same is true for focus sensitive operators such as nur “only”, and certain discourse particles such as ja (55).27 (54) a. dass nicht/kaum getanzt wurde that not/nearly not danced was “that one did not dance” b. *nicht/kaum wurde getanzt (55) a. dass ja wohl getanzt wurde that yes probably danced was “that probably one danced” b. *ja wurde wohl getanzt yes was probably danced c. *ja wohl wurde getanzt

27.  Kaum/nur have additional interpretations (kaum as a temporal conjunction, nur as a concessive) that are irrelevant in the present context and that may make the surface strings cited above grammatical.

Bootstrapping verb movement 

Data such as (51) – (55) show that formulating a squeezing-in account in the narrow sense is not a trivial task (cf. also Fanselow & Lenertovà, accepted, for further potential counterevidence). The alternatives involving an additional movement step fare better in this respect, for the simple reason that conditions on this further movement can, in principle, filter out the ungrammatical sentences. It is also unclear if the squeezing-in idea is helpful for other V2 languages. DP objects can come first in Scandinavian V2 sentences, in spite of the fact that these languages do not allow the fronting of objects across subjects in (non-V2) declaratives headed by a complementizer. In other words, a squeezing-in analysis of XP1 Vfin Z would be tenable for German because XP1 can always be fronted in Comp XP1Z whenever the corresponding V2 structure is grammatical, but the latter generalization is not true for Scandinavian. Object-initial V2 clauses would find no source in a squeezing-in analysis of Scandinavian.28 These considerations imply that the V2 property is a property of two movement processes, not a single one. The verb goes to the front as an instance of bootstrapping head movement, and an XP moves to the left periphery because of an edge feature that may be present in embedded clauses, while it must be so in matrix declaratives and constituent questions.

References Ackema P., Neeleman, A. & Weerman, F. 1993. Deriving functional projections. Paper, presented at NELS 23. Alexiadou, A. & Fanselow, G. 2000. On the correlation between morphology and syntax: The case of V-to-I movement. In Proceedings from the 15th Comparative Germanic Syntax Workshop, W. Abraham & J.-W. Zwart (Eds), 219–242. Amsterdam: John Benjamins: Alexiadou, A. & Müller, G. 2008. Class features as probes. In The Bases of Inflectional Identity, A. Bachrach & A. Nevins (Eds). Oxford: OUP. Baker, M. 1988. Incorporation. Chicago IL: The University of Chicago Press. Baker, M. 1996. The Polysynthesis Parameter. Oxford: OUP. Bayer, J. 2001. Asymmetry in emphatic topicalization. In Audiatur Vox Sapientiae, C. Féry & W. Sternefeld (Eds), 15–47. Berlin: Akademie Verlag. Bayer, J. 2004. Decomposing the left periphery: Dialectal and cross-linguistic evidence. In The Syntax and Semantics of the Left Periphery, H. Lohnstein & S. Trissler (Eds), 59–95. Berlin: Mouton de Gruyter. Bentzen, K., Hrafnbjargarson, G.H., Hróarsdóttir, Þ. & Wiklund, A.-L. 2007. The Tromsø guide to the force behind V2. Working Papers in Scandinavian Syntax 79: 93–118.

28.  Technically, squeezing-in might be saved by the assumption that squeezing in is mandatory in Scandinavian embedded clauses whenever DPs are moved in front of the subject, while V2 constructions are optional for fronted adverbs.

 Gisbert Fanselow den Besten, H. 1977. On the interaction of root transformations and lexical deletive rules. Ms, MIT and University of Amsterdam. (Published in H. den Besten, 1989). den Besten, H. 1983/1989. Studies in West Germanic Syntax. Amsterdam: Rodopi. den Besten, H. & Webelhuth, G. 1990. Stranding. In Scrambling and Barriers, G. Grewendorf & W. Sternefeld (Eds), 77–92, Amsterdam: John Benjamins. Bhatt, R.M. 1999. Verb Movement and the Syntax of Kashmiri. Dordrecht: Kluwer. Biberauer, T. & Roberts, I. 2004. Evidence that V2 involves two movements: A reply to Müller. Cambridge Occasional Papers in Linguistics 1: 41–62. Bierwisch, M. 1963. Grammatik des deutschen Verbs. Berlin: Akademie Verlag. Bródy, M. 1990. Some remarks on the focus field in Hungarian. UCL Working Papers in Linguistics 2: 201–225. Bury, D. 2000. Particles, V2, and the ungrammaticality of verb-initial structures. UCL Working Papers in Linguistics 12: 367–386. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, R. Martin, D. Michaels & J. Uriagereka (Eds), 89–155. Cambridge MA: The MIT Press. Chomsky, N. 2005. On phases. Ms, MIT. Collins, C. & Thráinsson, H. 1996. VP-internal structure and object shift in Icelandic. Linguistic Inquiry 27: 391–444. Donati, C. 2005. On Wh-head-movement. In Wh-movement on the Move, L. Cheng & N. Corver (Eds), Cambridge MA: The MIT Press. Fanselow, G. 2001. Münchhausen-style head movement. Paper, presented at the 2001 UCLA Workshop on Head Movement. Ms, Potsdam. Fanselow, G. 2002. Quirky subjects and other specifiers. In More than Words, I. Kaufmann & B. Stiebels (Eds), 227–250. Berlin: Akademie Verlag. Fanselow, G. 2003. Münchhausen-style head movement and the analysis of verb second. In Head Movement and Syntactic Theory, A. Mahajan (Ed.), Vol. 10 of UCLA/Potsdam Working Papers in Linguistics. Fanselow, G. 2005. Partial movement. In Blackwell Companion to Syntax. Vol. III, M. Everaert & H. van Riemsdijk (Eds), 437–492. Oxford: Blackwell. Fanselow, G. & Lenertová. D. Forthcoming. Left peripheral focus: Mismatches between syntax and information structure. Natural Language and Linguistic Theory. Fox, D. & Pesetsky, D. 2003. Cyclic linearization and the typology of movement. Ms, MIT. Fox, D. & Pesetsky, D. 2005. Cyclic linearization and its interaction with other aspects of grammar. Theoretical Linguistics 31: 235–262. Frey, W. 2004. A medial topic position for German. Linguistische Berichte 198: 153–190. Frey, W. 2005. The grammar pragmatics interface and the German prefield. Ms, Berlin. Frey, W. 2006. Zur Syntax der linken Peripherie im Deutschen. In Deutsche Syntax: Empirie und Theorie, F. d’Avis (Ed.), Göteborger Germanistische Forschungen. Freywald, U. 2007. Zur Syntax und Funktion von dass-Sätzen mit Verbzweitstellung. Paper, presented at the Workshop “Hauptsätze als Nebensätze?”. University of Potsdam, December 2007. Freywald, U. In preparation. Syntax und Pragmatik subordinierter V2-Sätze im Deutschen. Ph.D. dissertation, University of Potsdam. Haider, H. 1988. Matching projections. In Constituent Structure. Papers from the 1987 GLOW Conference, A. Cardinaletti, G. Cinque & G. Guisti (Eds), 101–121. Venice: Annali di Ca’Foscari 27.

Bootstrapping verb movement 

Haider, H. & I. Rosengren. 2003. Scrambling: Non-triggered chain formation in OV languages. Journal of Germanic Linguistics 15: 203–267. Holmberg, A. 1991. Head scrambling. Paper, presented at the 1991 GLOW Colloquium. Leiden. Horne, G. 2003. Idioms, metaphors and syntactic mobility. Journal of Linguistics 39, 245–273. Hrafnbjargarson, G.H. 2001. An Optimality Theory analysis of agreement in Icelandic. Ms, University of Stuttgart. Julien, M. 2007. Embedded V2 in Norwegian and Swedish. Working Papers in Scandinavian Syntax 80: 103–161. Kiparsky, P. & Kiparsky, C. 1970. Fact. In Semantics. An Interdisciplinary Reader in Philosophy, Linguistics, and Psychology, D. Steinberg & L. Jakobovits (Eds), 345–369. Cambridge: CUP. Kleemann-Krämer, A. Forthcoming. On apparent NP-internal focus particles in German. Journal of Comparative Germanic Linguistics. Koeneman, O. 1995. Flexible Phrase Structure. MA thesis, Utrecht. Koeneman, O. 2000. The Flexible Nature of Verb Movement. Utrecht: LOT Kuroda, S.-Y. 2005. Focusing on the matter of Topic: A Study of wa and ga in Japanese. Journal of East Asian Linguistics 14: 1–58. Lahne, A. 2008. A multiple specifier approach to the architecture of the left periphery. Linguistic Analysis 35. Larson, R.K. 1988. On the double object construction. Linguistic Inquiry 18: 239–266. Lechner, W. 2005. Interpretive effects of head movement. Ms, Universität Tübingen. Lechner, W. 2007. A puzzle for remnant movement analyses of Verb Second. Ms, University of Stuttgart. Legate, J. 2002. Warlpiri: Theoretical Implications. Ph.D. dissertation, MIT. Legendre, G. 2001. Masked V2 effects and the linearization of functional features. In OT Syntax, G. Legendre, S.Vikner & J. Grimshaw (Eds), 241–277. Cambridge MA: The MIT Press. Mahajan, A. 2003. Word order and (remnant) VP movement. In Word Order and Scrambling, S. Karimi (Ed.), London: Blackwell. Matushansky, O. 2006. Head movement in linguistic theory. Linguistic Inquiry 37: 69–109. Meinunger, A. 2006. On the discourse impact of subordinate clauses. In The Architecture of Focus, V. Molnár & S. Winkler (Eds). Berlin: Mouton de Gruyter. Müller, G. 1995. A-bar Syntax. A Study in Movement Types. Berlin: Mouton de Gruyter. Müller, G. 1998. Incomplete category fronting. Dordrecht: Kluwer. Müller, G. 2002. Two types of remnant movement. In Dimensions of Movement. A. Alexiadou, S. Barbiers & H.-M. Gärtner (Eds), 209–241. Amsterdam: John Benjamins. Müller, G. 2004. Verb-second as vP-first. Journal of Comparative Germanic Linguistics 7: 179–234. Müller, G. 2007a. Inflectional morphology in a minimalist grammar. Course Material. Leipzig. Müller, G. 2007b. Towards a relativized concept of cyclic linearization. Ms, Leipzig. Müller, G. & Sternefeld, W. 1996. A-bar chain formation and the economy of derivation. Linguistic Inquiry 27: 480–511. Nilsen, Ø. 2003. Eliminating Positions: Syntax and Semantics of Sentential Modification. Ph.D. dissertation, Utrecht University. Pesetsky, D. & Torrego, E. 2001. T-to-C movement: Causes and consequences. In Ken Hale: A Life in Language, M. Kenstowicz (Ed.), 355–426. Cambridge MA: The MIT Press. Pili, D. 2000. On A and A’ Dislocation in the Left Periphery. Ph.D. dissertation, University of Potsdam. Poletto, C. 1995. Complementizer deletion and verb movement in Italian. Working papers in linguistics, University of Venice.

 Gisbert Fanselow Poletto, C. 2002. The left-periphery of V2 - Rhaetoromance dialects: A new view on V2 and V3. In Syntactic Microvariation, S. Barbiers, L. Cornips & S. van der Kleij (Eds), 214–242. (Electronic Publication: meertens.nl/books/synmic). Richards, N. 2001. Movement in Language. Interactions and Architectures. Oxford: OUP. van Riemsdijk, H. 1989. Movement and regeneration. In Dialectal Variation and the Theory of Grammar, P. Benincà (Ed.), 105–136. Dordrecht: Foris. van Riemsdijk, H. 1998. Head movement and adjacency. Natural Language and Linguistic Theory 16: 633–678. Rizzi, L. 1997. The fine structure of the left periphery. In Elements of Grammar, L. Haegeman (ed.), 281–337. Dordrecht: Kluwer. Roberts, I. 1994. Two types of head movement in Romance. In Verb Movement, N. Hornstein & D. Lightfoot (Eds), 207–242. Cambridge: CUP. Samek-Lodovici, V. 2002. Agreement impoverishment under subject inversion. A crosslinguistic analysis. Linguistische Berichte Special Issue 11: 49–82. Samek-Lodovici, V. 2006. When right dislocation meets the left periphery. Lingua 116: 836–873. Schneider-Zioga, P. 2007. Anti-agreement, anti-locality and minimality. Natural Language and Linguistic Theory 25: 403–446. Sigurðsson, H. 1996. Icelandic finite verb agreement. Working Papers in Scandinavian Syntax 57: 1–46. Speas, M. 1990. Phrase structure in natural language. Dordrecht: Kluwer. Sportiche, D. 1998. Reconstruction and constituent structure. MIT Linguistic colloquium talk. Suranyi, B. 2005. Head movement and reprojection. Ms, Budapest. Thiersch, C. 1978. Topics in German Syntax. Ph.D. dissertation, MIT. Thiersch, C. 1985. VP and scrambling in the German Mittelfeld. Ms, University of Tilburg. Travis, L. 1984. Parameters and Effects of Word Order Variation. Ph.D. dissertation, MIT. Tsimpli, I.M. 1995. Focusing in Modern Greek. In Discourse Configurational Languages. É.K. Kiss (Ed.), Oxford: OUP. Van Craenenbroeck, J. 2006. Transitivity failures in the left periphery and foot-driven movement operations. Linguistics in The Netherlands 23: 52–64. Vikner, S. 1995. Verb Movement and Expletive Subjects in the Germanic Languages. Oxford: OUP. Vogel, R. 2003. Dialectal variation in German 3-Verb clusters. Looking for the best analysis. In Head Movement and Syntactic Theory, Vol. 10 of UCLA/Potsdam Working Papers in Linguistics, A. Mahajan (Ed.), 199–235. Wurmbrand, S. 2008. No TP-fronting meets nearly headless Nick. Ms (revised version to be published in Linguistic Inquiry). Zwart, C.J-W. 1993. Dutch Syntax. Ph.D. dissertation, University of Groningen. Zwart, C.J-W. 2005. Verb second as a function of Merge. In The Function of Function Words and Functional Categories. M.  den Dikken & C.M. Tortora (Eds), 11–40. Amsterdam: John Benjamins. Zwart, C.J-W. 2006. Uncharted territory? Towards a non-cartographic account of Germanic syntax. Ms, Groningen.

A conjunction conspiracy at the West Germanic left periphery John R. te Velde

Oklahoma State University In this analysis I consider one rather common coordinate construction and two less common ones from West Germanic that have two distinguishing properties in common: (i) all consist of conjoined verb-second (V2) clauses, and (ii) there is an ellipse at the left edge of the second conjunct. I propose that the conjunction c-commands the ellipse and that it is recovered in the semantic component through matching with a semantically parallel antecedent in a parallel syntactic position. This analysis utilizes Phase Theory to provide a derivational framework: each V2 clause must, as a phase, complete derivation before the next one is assembled. In this approach the Coordinate Structure Constraint is understood purely as a description of the semantic parallelisms required, and across-the-board movement is unnecessary; it is in fact incompatible with a phase-based approach. Finally, this approach requires three V2 positions in the functional domain of West Germanic; thus, the V2 phenomenon results from feature-checking requirements only (not positions available). Furthermore, it conspires with phase-based conjunction to create a position licensable for “deletion” (non-phonetic realization), thereby economizing the spoken form.

1.  Introduction* In the following I present an analysis of conjoined verb-second (V2) clauses in Dutch and German that occur with an elided element at the left periphery of the second

*Many thanks to Manuela Schönenberger for providing me with insightful comments and numerous corrections on the pre-final draft. Remaining errors are mine, of course. I also wish to thank participants at the CGSW-21 meeting in Santa Cruz, the GGS meeting in Stuttgart, and those present in the Zentrum für Allgemeine Sprachwissenschaft, Berlin, on June 9, 2006 for their comments. Thanks also to ZAS and Oklahoma State University for travel support.

 John R. te Velde

(and all subsequent) clauses. Constructions of this type with an elided subject are very common: (1) a. Williei heeft dit boek gelezen en ei zal het zijn vrienden aanbevelen W has this book read and will it his friends recommend ‘Willie has read this book and will recommend it to his friends.’ b. Hansi zeigt seinem Onkel die Briefmarken und ei verkauft sie H shows his.dat uncle the stamps and sells them

seiner Tante his.dat aunt

‘Hans is showing his uncle the stamps and will sell them to his aunt.’

Much less common, but equally grammatical, are variants of (1a,b) with an elided direct object instead of an elided subject, which must be fronted by an additional Internal Merge operation (Chomsky 2008): (2) a. Dit boeki heeft Willie ti gelezen en ei zal hij (he) zijn vrienden ti aanbevelen b. Die Briefmarkeni zeigt Hans seinem Onkel ti und ei verkauft er (he) seiner Tante ti

The same kind of construction is also possible, not surprisingly, with an elided indirect or prepositional object as in (3a) and (3b) respectively: (3) a. Seinem Onkeli schenkt Hans ti die Fotos und ei verkauft er ti die Münzen his.dat uncle gives H the photos and sells he the coins ‘To his uncle, Hans is giving the photos and selling the coins.’ b. [Für seine Schwester]i kauft Hans nichts ti und ei tut er nie etwas ti for his sister buys H nothing and does he never something ‘For his sister Hans doesn’t buy anything nor does he ever do anything.’

A number of proposals have been made to account for the elided subject in constructions like (1a,b); these will be reviewed in §2. No minimalist proposal can be found in the literature for constructions like those in (2) and (3). The objective here is to propose a unified account of all these constructions using an approach that combines Phase Theory, as outlined in Chomsky (2000) and (2001), with elements of a proposal for coordinate ellipsis outlined in te Velde (2005a). I argue that all such constructions have a gap at the left edge of all but the first conjunct; I call this form of coordinate ellipsis Left-Edge Ellipsis (LEE).1 The constructions in (1) differ structurally

1.  The other two common forms of coordinate ellipsis are Gapping, involving a finite verb (and optionally its complement), and Right Node Raising, which always involves a gap at the right edge of the first (and all but the last) conjunct. These are discussed in detail in te Velde (2005a), Chapter Four, and in literature cited there.

A conjunction conspiracy at the West Germanic left periphery 

and thus derivationally from those in (2) and (3): the latter require one additional Internal Merge operation for the fronting of the verbal complement, whether a direct, indirect or prepositional object. The structural and derivational differences do not affect the ellipsis operation LEE, however, because in all of these constructions the ellipse can be licensed by [&] c-commanding the empty position, and the recovery of the ellipse, which lies outside of the narrow syntax, utilizes Match in LF in all cases. A central argument of this proposal, presented in §3, is that clausal conjuncts are phases and thus must complete syntactic derivation before a second conjunct is selected, merged and derived. In this way the first conjunct provides a Copy template for the derivation of the remaining conjuncts and determines, both structurally and semantically, what element can be licensed for deletion in the next conjunct. I argue that the licensing requirement on ellipsis in these constructions manifests itself as a left-edge requirement because the coordinating conjunction must license the ellipse in a c-command relation. Recovery of the ellipse does not necessarily require the antecedent to be at the left edge, but rather that antecedent and ellipse are in parallel syntactic positions: [C-1 XP1 … [ & [C-2 e1 …]]] (C = conjunct). The result of this licensing requirement in V2 languages like Dutch and German has the form of a syntactic conspiracy: the left-edge requirement on licensing conspires with the syntax of V2 such that any argument (including prepositional objects) fronted to the left edge of the second (and all subsequent) conjunct(s) is eligible for ellipsis so long as a matching argument occurs in a parallel position in the first conjunct. Scopal elements such as adverbs do not need to be fronted and licensed for deletion in this way, since a scopal element in the first conjunct can satisfy the requirements of LF for parallel interpretation without the presence of an elided element in the following conjunct. By contrast, the elided elements in (1) – (3) have all the properties of the spoken equivalent except its phonetic features. A central claim of my proposal is that coordinate constructions of the type in (1) – (3) have to meet the Parallelism Requirement (cf. Chomsky 1995: 125–126, 203; Hornstein & Nunes 2002:  41) that has been noted to exist in all coordinate structures; it is particularly inviolable in certain respects when ellipsis occurs in ways that will be made more precise here. The explanation for this striking property of elliptical conjoined clauses, I argue, has both a syntactic and a semantic basis: the former is the c-command requirement for the licensing of the ellipse, and the latter is the requirement of Match in LF. The details of this part of the proposal are outlined in §3. In §4 some extensions of the proposal are outlined, and in §5 there is a brief discussion of a scope-based approach to the data. A conclusion follows in §6. In the next section I outline and comment on three other accounts of LEE with subject gaps.

 John R. te Velde

2.  Earlier accounts of LEE As stated earlier, the following accounts of LEE all have one major limitation: their empirical coverage is limited because for one reason or another none of them considers any type of gap other than a subject gap. Furthermore, none of them is fully compatible with assumptions of the Minimalist Program. For these two reasons an alternate account is presented in §3.

2.1  Heycock & Kroch (1994) Heycock and Kroch’s (1994) (H&K) analysis of subject gaps in LEE, as evident in (4), presupposes that 1) intermediate categories can be conjoined, and 2) that there really is no gap in LEE; rather, the subject of the second conjunct is interpreted on the basis of a sharing relation with the first conjunct: (4) [CP Hans [C′ zeigt seinem Onkel die Briefmarken] und [C′ verkauft sie Hans shows his.dat uncle the stamps and sells them seiner Tante]] his.dat aunt

I will not discuss any problems involved with the conjunction of intermediate projections, as they are not problems of conjunction per se, but of the role of intermediate projections in minimalist theory. The implications of subject sharing in LEE as required in the H&K analysis must be addressed here. One is that, along with subject sharing, we must assume that the V2 requirement of Dutch and German can also be satisfied by sharing. This is possible only if non-binary phrase structure is employed for these constructions so that the Spec position containing the subject somehow dominates both conjuncts equally; in other words, the phrase structure of (4) is not binary but rather looks like (5):


CP (Spec)

C′ C′


Phrase structures like (5) are incompatible with the minimalist assumption that all syntactic relations are asymmetric. Whether an exception can or should be made for coordinate structures in order to account for their parallelisms is a question that will be left to further research. In the remainder of the present analysis I will argue that a more unified approach is possible if no such exception is made in the way that structures are built up in syntactic derivation.

A conjunction conspiracy at the West Germanic left periphery 

2.2  Büring & Hartmann (1998) Büring & Hartmann (1998) (B&H) modify the H&K analysis using an adjunction account that requires an empty operator in the CP projection of the second conjunct:

(6) [CP Hans zeigt seinem Onkel die Briefmarken und [CP OP verkauft sie seiner Tante]]

In this analysis B&H avoid the problem with the H&K analysis just discussed, as the second conjunct is adjoined rather than conjoined and thus stands in an asymmetric relation to the first conjunct. This phrase structure takes us one step closer to the present proposal: coordinate structures are syntactically (phrase-structurally) asymmetric like all other syntactic structures. A question that the B&H proposal raises is: What is the relation of a second OP to the first OP and the controller antecedent when there are three conjuncts in a LEE construction? An example of this would be:2 (7) [CP Hans zeigt seinem Onkel die Briefmarken [CP OP verkauft manche seiner Tante und [CP OP schenkt einige sells some his.dat aunt and gives a-few seiner Oma]]] his.dat grandma

In coordinate structures of this sort, the Parallelism Requirement demands that all conjuncts be equal in certain crucial ways. This kind of parallelism is not obtainable under the analysis in (7) because the relation of the controller-antecedent Hans to the two OPs is different than the relation of the first OP to the second, or vice versa. This kind of asymmetry is expected in an analysis that involves adjunction because with the adjunction of a conjunct, an asymmetric relation is established between the conjuncts. A further concern for a minimalist account is the use of an empty operator in the B&H analysis. Current minimalist approaches have done away with operators in

2.  The relation of the traces to the fronted wh-element in (i) is similar to the relations in (7): (i)

Weri (who) zeigt seinem Onkel die Briefmarken ti verkauft seiner Tante manche (some) und ti schenkt seiner Oma andere (others)?

A difference, however, is that (7) does not presuppose any movement but instead requires a controller antecedent, i.e., one is a head-foot relation in a wh-chain and the other an antecedent-operator relation. Which one better captures the symmetries of the constructions under investigation will be left to other research. The assumption here is that a phase-based approach rules out a movement analysis such as (i) cf. §3; see above for further comments on the B&H analysis and why a phase-based approach is preferred.

 John R. te Velde

favor of Copy-movement.3 My own proposal does not require an operator, and it rules out as a violation of the Phase Impenetrability Condition (PIC, Chomsky 2000: 108) movement of the sort that has traditionally been assumed necessary for coordinate structures, namely across-the-board (ATB) movement. Another concern with the B&H analysis is the uniform V-to-C movement that it requires for the derivation of V2 clauses. This analysis of V2, I will argue, is too rigid for the asymmetries between subject-initial and other V2 clauses. These points will be outlined in more detail in §3.

2.3  Johnson (2002) Johnson identifies two problems with the H&K analysis:4 1) There is a violation of the Coordinate Structure Constraint (CSC) (see §3). 2) The scope of the subject Hans does not extend into the second conjunct, if defined syntactically, as indicated by the phrase markers in (4). He proposes a verb projection raising (VPR) account as a way to meet the CSC, using a construction with a fronted direct object in the first clause and a subject gap in the second clause to illustrate his approach: (8) Johnson’s VPR account of LEE in: Den Hund hat einer gefüttert und hat ihn geschlagen (from Höhle 1983) the.acc dog has one.nom fed and has him beaten ‘Someone fed the dog and (then) beat him.’ a. initial structure with coordinate F′-projections: [IP einer [FP[F′[VP den Hund gefüttert hat]] und [F′[VP ihn geschlagen hat]]]] b. raising of hat to F0 in second conjunct: [IP einer [FP[F′[VP den Hund gefüttert hat]] und [F′ hat1 [VP ihn geschlagen t1]]]] c. VP-raising out of first conjunct into Spec,FP: [IP einer [FP[VP den Hund gefüttert hat]2 [F′ t2] und [F′ hat1 [VP ihn geschlagen t1]]]] d. verb-raising to C from the raised VP in Spec,FP: [CP hat3 [IP einer [FP[VP den Hund gefüttert t3]2 [F′ t2] und [F′ hat1 [VP ihn geschlagen t1]]]]

3.  Empty operators, like empty categories of any sort, are a representational, rather than a derivational, tool for explaining ellipsis. In the minimalist framework the attempt is made to keep syntactic theory as free from representational elements like ECs as much as possible, as they reduce the degree of deductiveness in grammar. 4.  Because Johnson assumes that the B&H analysis does not involve ATB movement in SLF (Subjektlücke in finiten) sentences (Höhle 1983), he considers their solution inappropriate for the “exotic coordinations” that he analyzes. See also Schwarz (1998) and Zwart (1991) for earlier accounts.

A conjunction conspiracy at the West Germanic left periphery 

e. DP-fronting to Spec,CP from the raised VP [CP den H4 hat3 [IP einer [FP[VP t4 gefüttert t3]2 [F′ t2] und [F′ hat1 [VP ihn geschlagen t1]]]]

In evaluating Johnson’s derivation, we note that even though it is motivated by the need to satisfy the CSC, two exceptions are allowed. One exception is the asymmetric movement of the finite verb:  in conjunct 1 it raises completely out of the first conjunct but in conjunct 2 only to the edge (not ATB), in violation of the CSC.5 The second CSC violation occurs with subject raising: it is also asymmetric because only the subject in conjunct 1 raises to a Spec position in the functional domain. This kind of non-ATB raising also violates the CSC; it is justified, argues Johnson, because it is A-movement.6 No subject is generated in the second conjunct as the coordinate structure is assumed to be a conjunction of F′-projections in a non-binary phrase structure that is dominated by IP whose Spec position the subject occupies, cf. (5). With this structure and derivation Johnson assumes that subject scope eliminates the need for a left-edge subject gap. It is important to note that the phrase structure of the conjunction in (8) requires branching of the sort in (9):




This kind of structure is incompatible with the minimalist assumption that all relations in phrase structure are asymmetric. (9) could possibly be allowable if sufficient justification could be provided for making an exception to asymmetric phrase structure in the case of coordinate structures. My proposal in §3 rejects the need for this kind of exception. 5.  The CSC is essentially a Parallelism Requirement applying to movements out of coordinate structures (cf. Chomsky 1995: 125–126, 203; Hornstein & Nunes 2002 for other types of parallelism requirements). In the generative literature it has often been expressed as an ATB requirement on movement:  the same element must be moved from all conjuncts. Johnson justifies this violation, arguing that the strong features of the raised finite verb must be checked before Spell-Out. I will leave for further research whether this kind of feature checking adequately motivates the movement that Johnson proposes, specifically whether the parallelism requirement underlying this construction should be violated in this way and for this reason. In my proposal no such violation is necessary. 6.  The second conjunct has a subject via a sharing relation with the first conjunct. W.r.t. to the CSC violation, Johnson states, “But secure demonstrations of the [CSC] holding of A-movement, or of other forms of movement, are not available…I am also going to follow Ross (1967) in taking the [CSC] to be a purely geometrical condition, one that defines the configurations that block extractions in terms of the graphs that phrase markers are.”

 John R. te Velde

A second problem with Johnson’s proposal is the assumption that VPR occurs in Standard German. The empirical support that Johnson presents for VPR comes from West Flemish, which has been shown in e.g., Haegeman (1991, 1998) to have word order parameters not attested in Standard German that pertain to clause-final verb clusters, with the possible exception of double-infinitive constructions in perfective tense embedded clauses. Thus, a VPR approach to German is empirically weak.7 The most serious problem with Johnson’s approach, because it is an empirical one, is that it cannot account for left-edge object gaps of the kind in (10): (10) Die Briefmarkeni zeigt Karl der Tante ti und ei verkauft er the stamps shows K the.dat aunt and sells he dem Onkel ti the.dat uncle

Johnson argues that left-edge object gaps are ungrammatical, using the example in (11): (11) *Den Hundi [C′ hat keiner ti gefüttert] und [C′ hat er ti geschlagen] the.acc dog has no-one fed and has he beaten

The ungrammaticality of this construction is due to the scopal properties of the negative quantifier keiner. We note the grammatical (12), identical to (11) except for the subjects: (12) Den Hund hat Heinz gefüttert und hat Karl geschlagen

In sum, we can list the following reasons for seeking an alternative to Johnson’s proposal: 1) The methods used to satisfy the CSC are either incompatible with minimalist assumptions or lacking in empirical support; they are: a non-binary, “symmetric” phrase structure and a derivation that allows exceptions to the CSC. Furthermore, the proposal cannot account for all of the data, e.g., those with left-edge object gaps.

2.4  Section summary and conclusion The above overview of the proposals available in the generative literature on LEE constructions has shown that no account can be found of these constructions with gaps

7.  An example of a double-infinitive construction in a perfective tense embedded clause that could be considered support for using VPR is: (i) (?) … dass Hans noch ein Bier hat wirklich trinken wollen that H yet a beer has really drink.inf want.inf ‘that Hans really wanted to drink another beer’ Given the rarity of such constructions and the availability of an alternate analysis, they do not offer a good basis for the analysis of the relatively common constructions that Johnson considers.

A conjunction conspiracy at the West Germanic left periphery 

other than subject gaps. This inadequate empirical coverage is the most serious problem with investigations to date. Secondly, the accounts of subject gaps we have seen are not compatible with minimalist assumptions and for that reason (among others) do not offer a good basis for a phase-based account of other types of left-edge gaps. In the next section we turn to a proposal that combines Phase Theory with an approach to coordinate structures that does not require making exceptions with regard to the CSC or asymmetric phrase structure in order to accommodate the symmetry or parallelism requirements of conjunction. The CSC will be redefined as a semantic requirement on parallelism in coordination, a revision resulting from the elimination of ATB movement, as traditionally assumed, because of its incompatibility with Phase Theory.

3.  LEE and Phase Theory As an introduction to this section, we consider in §3.1 the implications of Phase Theory for conjunction. Then in §3.2 I outline my proposal for LEE that utilizes Phase Theory. In §3.3 the advantages of this proposal over earlier ones are outlined, followed in §3.4 by a discussion of some key assumptions regarding coordination, Copy and coordinate ellipsis. Chomsky (2000, 2001) makes no direct statements about conjunction in his proposals for phase-based derivation; thus, what is proposed here is an extension of the principles and concepts of his proposals to coordinate structures as I understand them. Of direct relevance here is the PIC.

3.1  Conjunction and Phase Theory Chomsky (2000: 108) formulates the PIC as follows: (13) In phase α with head H, the domain of H is not accessible to operations outside α, but only H and its edge.

In applying this condition to conjunction, we note that a clausal conjunct has undergone at least one phase. Therefore the PIC disallows any syntactic operation that extends from one clausal conjunct to another, unless it has a left-edge element of a conjunct or its head as a goal. What is not stated in (13) because it is part of movement theory in general in the Minimalist Program is that the target of movement must be the edge of the phase. We will return to this point in connection with the analysis in §3.2. It is useful at this point to consider why Chomsky proposes derivation by phase. As this notion is understood here, a phase is intended for managing the derivation of sentences in a way that meets the requirements of the semantic and phonetic components of sentence generation (or as often stated: to meet the interface conditions).

 John R. te Velde

The same reasoning applies directly to coordinate structures:  their computation in narrow syntax must proceed in a way that is manageable – a particularly significant point for LEE constructions, since they are made up of conjoined clauses, each one a phase – and this computational sequence must be so arranged as to explain how ellipses are handled semantically and phonetically, in particular how the left-edge gaps in LEE constructions are recovered on the perceptual side without phonetic realization. I see two consequences of the PIC for coordinate ellipsis: One, ATB movement is ruled out because it has a non-head, non-edge element as its goal and thus does not meet the requirements of the PIC. In effect, it requires both look-ahead and look-back, cf. (14a) and the related discussion. In (14b) is illustrated the manner in which the derivation of a coordinate structure proceeds, if it is phase-based:8 (14) a. ATB movement:

[C-1[ ] ([C-1)... [C-2 ... ([C-n ...)]]

b. Derivation by phase:8 CP



C = conjunct

Phase 1 VP





Phase 2 VP


(14) illustrates that ATB movement does not have an element at the edge of a phase as the goal of movement; rather, this goal is within the domain of each conjunct, each of which is a phase. In a phase-based approach, each conjunct is derived independently, and no movement from one to another is assumed. (14b) thus requires some other syntactic or semantic means to achieve the results of ATB movement in LEE constructions. The next subsection will outline what these are. The second implication is that the Coordinate Structure Constraint (CSC) is unnecessary in narrow syntax as a constraint on movement, its original intent. The CSC is formulated by Ross (1967: 161) as follows: (15) In a coordinate structure, no conjunct may be moved, nor may any element contained in a conjunct be moved out of that conjunct.

.  The assumption here is that the derivation of a coordinate structure consisting of conjoined clauses proceeds in a linear fashion, i.e. it starts with the linearly left-most conjunct and proceeds to the next one on the right, etc. I am assuming that this linear procedure manifests a computational limitation: narrow syntax is limited to the computation of one clause at a time; furthermore, the first clause generated is used as a template (for Copy) for the derivation of the next conjunct to whatever extent is possible (the more this occurs, the more symmetry results).

A conjunction conspiracy at the West Germanic left periphery 

In the literature on conjunction it is assumed that ATB movement, because it has an element in all of the conjuncts as a goal and a single position as a target, is allowable by the CSC. The CSC is intended by Ross as a condition on parallelism across conjoined clauses in which ellipsis occurs. Because the constraint is formulated in terms of movement from one conjunct to another, it overlaps with Phase Theory, which formulates constraints on movement in terms of the PIC. A central question that must be addressed is: Can Phase Theory account for the same properties of coordinate ellipsis – the parallelisms – that the CSC accounts for? In the next subsection we turn to this question in the context of the application of Phase Theory to LEE.

3.2  LEE and Phase Theory In a phase-based derivation of LEE, the leading conjunct (clause) completes syntactic derivation before the second conjunct is selected, merged and derived. In (16) we consider the derivation of a LEE construction with a fronted direct object in the first conjunct, and a left-edge subject gap in the second conjunct (assumptions underlying TopP are discussed in §3.4): (16) Phase-based derivation of:9 Die Briefmarken zeigt Karli dem Onkel und ei bietet sie ihm the stamps shows K the.dat uncle and offers them him.dat zum Verkauf an for sale part ‘The stamps Karl is showing his uncle and offering them to him for sale.’

9.  A close look at this and several of the derivations that follow will reveal that I assume German and Dutch have head-initial phrases in the functional domain (IP and higher), while in the lexical domain phrases are head-final. This apparently peculiar arrangement can be justified on the basis of some simple observations about German (and Dutch):  non-finite phrases such as found in indirect commands (Bitte herkommen!) or in notes found in calendars or shopping lists (um zwei den Aufsatz mit P besprechen; Kartoffeln fürs Abendessen einkaufen) are head-final, while any finite version of these are head-initial (Kommen Sie bitte her! Ich bespreche den Aufsatz um zwei mit P; Mein Mann kauft Kartoffeln fürs Abendessen ein). It is a standard assumption that all finite clauses involve a verb position in the functional domain. Assuming that the verb is internally merged from a clause-final position where it is the head of the VP – which is head-final – into a position in the functional domain where it sits in a head-initial phrase creates a structure of West Germanic syntax that allows a simpler account of many of its properties, and a one that is more easily unifiable with the syntax of all Germanic languages.

 John R. te Velde

a. Select and merge lexical items for the first conjunct: [VP Karl [V′[DP dem Onkel] [V′[DP die Briefmarken] zeigt]]] b. vP phase: [TP Karl [T′ zeigti [vP[v′[DP dem Onkel]j [v′[DP die Briefmarken]k[VP tj tk ti]]]]]] c. CP phase: DP-object fronting, V → C: [TopP [DP die B-marken]k [Top′ zeigti [TP K ti [vP[v′[DP dem O]j [v′ tk [VP tj tk ti]]]]]]] (conjunct 1 is held in active memory while the derivation of conjunct 2 proceeds:) d. Select and merge LA for the second conjunct, cf. (a): [VP Karl [V′[DP ihm][V′[DP sie][V′[Adv zum Verkauf][VP anbietet]]]]] e. vP Phase in conjunct 2: [TP Karl [T′ bieteti [vP[v′[DP sie]j [v′[DP ihm]k[VP[Adv zum Verkauf] tk tj ti an]]]]]] f. Merger of und and conjuncts; copying of subject features: [TopP[TP[v′[VP (Die Briefmarkenk zeigti Karl dem Onkelj tj tk ti) ↑ [TP und ← [TP K bietetl siem ihmn zum V tn tm tl an]]]]]] g. Ellipsis: und licenses the “deletion” (note the TP-conjuncts): [TopP[DP die B]k[Top′ zeigti [TP Karl ti [v′ [DP dem O]j ti [VP tj tk ti [TP und [TP Karl bietetl [v′ siem [v′ ihmn [VP z V tn tm tl an]]]]]]]]]] h. PF-realization with ellipse (recovered in LF): Die Briefmarken zeigt Karl dem Onkel und Karl bietet sie ihm zum Verkauf an

The primary syntactic feature of this derivation is that it proceeds by phase, i.e., one conjunct at a time. This manner of derivation disallows, in accordance with the PIC, any movement that extends from one conjunct to another, unless the goal of movement were to occur at the edge of a conjunct and have the left edge of the previous phase as its target. With these restrictions on movement it is not possible to derive the construction in (16) by moving Karl from the edge of the second conjunct into the first conjunct because the target of this movement is no longer at the edge of the first conjunct after the first conjunct has completed the CP phase, during which the object die Briefmarken is fronted to the edge. It might seem possible for LEE constructions to meet the requirements of the PIC in a language like German where fronting to the left edge of the CP phase occurs frequently; in (16) it occurs in the first conjunct. The problem with this assumption is that the movement operation in the second (and all subsequent) conjuncts has no target, given that the logical target, the Spec,TopP position of the first conjunct, is already occupied by a DP in a structure like (17), when the derivation proceeds by phase, cf. (16): (17) [TopP [Die Briefmarken]i zeigt Karl dem Onkel ti

und [TopP [die Briefmarken]j verkauft er der Tante tj]]

A conjunction conspiracy at the West Germanic left periphery 

This kind of movement, in which an element from one conjunct targets a position in another, was proposed by Williams (1977, 1978) for the derivation of conjoined clauses that shared a single, left-edge wh-element, as in (18): (18) Weni liebt Hans ti, heiratet Fritz ti und ignoriert Karl ti ? whom loves H.nom marries F.nom and ignores K.nom ‘Whom does Hans love, is Fritz marrying, and does Karl ignore?’

Such derivations do not proceed by phase, i.e., are theoretically possible only if the movement occurs in all the conjuncts at the same time, and all occurrences of wen are “collapsed” into one phonetic realization in the position indicated. Phase Theory, as interpreted for coordinate structures here, does not allow syntactic operations of this sort, however, because they must proceed in either one of two ways, both of which are ruled out in a phase-based approach. The one option is to require the derivation of one clause to be put on hold while the derivation of the next clause is completed up to the point where the derivation of the first conjunct was put on hold. At this point the derivation would need to look back into the first conjunct to see if a target is available. A general assumption of Phase Theory is that look-back of this sort is not allowed.10 The other option is for the derivation of all clauses to proceed simultaneously. This violates the basic principle of derivation by phase, as discussed earlier. I make the assumption here for coordinate structures that neither option should be allowed as a necessary exception to account for LEE constructions. This kind of exception would be no improvement over the exceptions to the CSC and phase structure allowed in Johnson (2002). So we see, as stated earlier, that the challenge for a phased-based approach is to find a substitute for ATB movement, required for (18). The solution I will propose involves dividing the task between narrow syntax and LF:  in narrow syntax each conjunct is derived in the manner outlined in (16); when the derivation involves the fronting of a DP to the left edge, i.e., to within the c-command domain of [&], then this element becomes eligible for avoiding phonetic realization (“deletion”), if at the interface with LF it is determined that this DP is redundant in such a way that its lack of phonetic realization does not present a problem for the perceptual side, i.e., the gap can be recovered. The recovery of a left-edge gap occurs in my proposal via Match at the LF interface: I assume that structural parallelism combined with lexical redundancy (as reflected in phonetic identity) are the requirements of LF-Match by which a left-edge gap is

10.  Stopping the derivation of the first conjunct at the necessary point – before DP fronting – requires the inverse: looking ahead into the next conjunct. Both look-ahead and look-back of this sort are, in my interpretation of Phase Theory, not allowed.

 John R. te Velde

recovered.11 That structural parallelism is a requirement can easily be proven. In (19) we note that an antecedent in any position other than the same position as the one occupied by the gap is not a good antecedent for recovery of the gap: (19) a. [TopP Die Briefmarkeni zeigt Hans seinem Onkel ti und the stamps shows H his.dat uncle and [TopP ei verkauft er seiner Tante ti]] sells he his.dat aunt a′. *Hans zeigt seinem Onkel [vP die Briefmarkeni und [TopP ei verkauft er seiner Tante ti]] b. [TopP Dit boeki heeft Willie ti gelezen en [TopP ei zal hij zijn this book has W read and will he his

vrienden ti aanbevelen]] friends recommend

b′. *Willie heeft [vP dit boeki gelezen en [CP ei zal hij zijn vrienden ti aanbevelen]] c. [TopP Seinem Onkeli schenkt Hans ti die Fotos und [CP ei verkauft his.dat uncle gives H the photos and sells

er ti die Münzen]] he the coins

c′. *Hans schenkt [vP seinem Onkeli die Fotos und [CP ei verkauft er ti die Münzen]]

The operations Copy and Match minimize the derivation of LEE constructions; Copy utilizes the working space of active memory for “transferring” formal features from a derived conjunct to the derivation of a second (and all subsequent) matching conjuncts. Whether a conjunct matches a preceding one or not is determined largely at Select; Match at the LF interface is the interpretive mechanism that guarantees the recovery of those elements that have been licensed for deletion, according to the requirements outlined earlier.12 In this approach, the CSC is stated as a condition on parallelism in coordinate ellipsis (only), not as a condition on movement. The Parallelism Requirement, earlier formalized as an ATB requirement on movement (Williams 1977, 1978), has more recently been understood as a requirement on interpretation in coordination

11.  Phonetic identity is of course not a guarantee of lexical redundancy in languages like English which has a relatively large number of homonyms. However, phonetic identity is nevertheless always a requirement of Match. 12.  See te Velde (2005a) Chapter Four for further details.

A conjunction conspiracy at the West Germanic left periphery 

(e.g., Chomsky 1995:  125–126, 20313) or as a condition on Copy applying locally (Hornstein & Nunes 2002:  41). The present proposal combines interpretation (recovery of gaps) and syntactic local licensing of a parallel position for ellipsis, for defining the specifics of the Parallelism Requirement. Thus, while Phase Theory eliminates ATB movement, it predicts the key properties of coordinate ellipsis, Edgeness and Parallelism. In the next section we consider these claims more closely.

3.3  Advantages of a phase-based approach The central claim of my proposal is that the CSC is captured as a condition on parallelism (symmetry) in coordinate ellipsis that results from the interaction between the fronting of a DP to a Spec position at the edge of the second (or later) conjunct (cf. §3.4 for specifics) and the syntactic licensing for the “deletion” of this DP, accomplished by [&] at the right edge of the leading conjunct (cf. previous section). A key aspect of the present analysis is that the perceived violations of the CSC in LEE-constructions described in Johnson (2002) are not CSC violations but rather the result of stylistic fronting to a Spec position left of the initial conjunct, cf. (8) and (16): (20) [TopP [Den Hund]i hatj [TP einerk tj ti gefüttert und [TP ek hat ihni geschlagen]]]

(phase one) (phase two)

The fronting of den Hund is not a CSC violation because it occurs within the domain of a single conjunct, which requires the CP phase for this operation and in that sense is asymmetric to the second conjunct. The resulting structure nevertheless satisfies the Parallelism Requirement because it does not alter the coordinate symmetry (required for licensing and recovery of the gap), which is TP-based. Such cases of stylistic fronting are predicted in a phase-based approach to coordinate structures because movement is not ATB, but is limited rather to a given phase, thus lacking a look-ahead capability. We can therefore restate the CSC in terms of Phase Theory as follows: (21) The CSC is an LF-interface condition on symmetry across conjoined clauses in which coordinate ellipsis can be syntactically licensed.

13.  Chomsky (1995: 203) illustrates the Parallelism Requirement with (i):


John said that he was looking for a cat, and so did Bill

He states the following: “In the elliptical case [i], a parallelism requirement of some kind (call it PR) requires that the second conjunct must be interpreted the same way – in this case, with he referring to Tom and a cat understood nonspecifically…. PR surely applies at LF…. There would be no need, then, for special mechanisms to account for the parallelism properties of [i].” Following this assumption, I do not propose any specific mechanism for either the licensing or recovery of ellipsis in the constructions discussed here, since c-command and Match are also used in non-coordinate structures of many kinds.

 John R. te Velde

There are several types of coordinate ellipsis, all of which have their own symmetry requirements.14 In LEE constructions coordinate symmetry is required in the antecedent-gap relation. As we have seen, this symmetry has both a syntactic and a semantic aspect: the syntactic is seen in the requirement that the antecedent and gap occupy the same syntactic position, and the semantic that the antecedent and gap be identically interpretable, i.e., have the same referent, semantic content, etc. Stated in these terms, the CSC is not a condition on movement in coordinate structures, if this movement is independent of these requirements and creates an asymmetry that is unrelated to the required coordinate symmetry. The symmetry required in the antecedent-gap relation of LEE constructions can be schematized in the following way: (22) ([YP Y )[αP-1 [DP ]i-1 … [&P [αP-2 [DP e]i-2 … [&P [αP-n [DP e]i-n … ]]]]]] (‘DP’ includes Pobj) where α is any functional head (e.g., T, Top, C, etc.), and Y is an optional functional head superior to α, resulting from an additional Internal Merge operation in conjunct αP-1. 

Fronting of the kind in the first conjunct of (23a), a typical LEE construction with a subject gap, can produce a structural asymmetry that is allowed by the CSC as outlined in (21). Interestingly, a similar structural asymmetry with wh-fronting is not allowed:15 14.  In the literature it is generally assumed that coordinate ellipsis comes in three forms: LEE (discussed here), Gapping (of the finite verb) and “Right Node Raising” which affects the right edge of all but the last conjunct. 15.  Although some speakers of both Dutch and German find constructions like (23a) somewhat degraded as compared to the same construction without the object fronted, or with an object gap instead of a subject gap, all speakers find constructions like (23b) ungrammatical in both languages. Interestingly, as pointed out by Manuela Schönenberger, (i) is not any better than (23b):


*Diese Frau liebt Hans vor allem, heiratet Marie und ignoriert Ute

It is curious that (i) is out, even though it is structurally the same in fundamental ways to (16) and (20): all three have a fronted DP in the first conjunct and a subject gap in the second conjunct. However, (i) is different in at least two crucial ways: it has three conjuncts, each with a subject gap and therefore a bit more complex. Most critical to the ungrammaticality, however, is that it has a certain ambiguity due to inadequate Case morphology combined with the fact that the objects are animate like the subject. There is a tendency to interpret Hans as the subject of the verbs heiratet and ignoriert because both have a female as an object. But another possible interpretation of the first conjunct exists in which diese Frau is the subject and therefore the antecedent of the subject gaps in the other conjuncts, but this leads to an unexpected interpretation in the second and third conjuncts.

A conjunction conspiracy at the West Germanic left periphery 

(23) a. Die Briefmarkeni zeigt Hansj seinem Onkel ti und ej verkauft sie the stamps shows H his.dat uncle and sells them

seiner Tante his.dat aunt

‘The stamps Hans is showing to his uncle and will sell them to his aunt.’ b. *Weni liebt Hansj vor allem ti, ej heiratet Marie und ej ignoriert Ute? whom loves H above all marries M and ignores U b′. Weni liebt Hans ti, ei heiratet Fritz ti und ei ignoriert Karl ti? whom loves H marries F and ignores K ‘Who does Hans love, is Fritz marrying and does Karl ignore?’

In (23a) but not in (23b) the antecedent of the gap can occur in a position other than the leftmost Spec-position. This indicates that the symmetry requirement on LEE constructions cannot be stated in terms of edgeness, but must be stated in terms of syntactic position, as in (21). That is, the antecedent of the gap must be in the same Spec-position as the gap, but this Spec-position can be preceded by another Spec-position to its left, i.e., the structural position of gap and antecedent determines coordinate symmetry; edgeness is required only for the licensing of the gap by [&]. Thus, the gap and antecedent must be at the left edge of the same projection, but the antecedent can be preceded by another projection resulting from an additional Internal Merge operation in that conjunct. A survey of the constructions we have seen so far would show that this kind of asymmetry, where the gap, but not the antecedent, must be at the left edge of its conjunct, occurs only in LEE constructions with subject gaps; it is impossible with object gaps. This subject-object asymmetry follows from the restatement of the CSC in (21) and the specific formulation regarding LEE constructions in (22). We can make the difference between subject- versus object-gap constructions more obvious with the use of the projection TopP, with Spec,TopP as the target of movement in object-gap constructions, whereas in subject-gap constructions the target is Spec,TP. Consider the comparisons between conjoined TPs with Spec,TP as the target of subject fronting in (24a,b), and conjoined TopPs with Spec,TopP as the target of object fronting in (24a′,b′):16 (24) Structure of LEE-constructions: (a) and (b) with subject gap, (a′) and (b′) with object gap: a. [TopP Die Briefmarkeni [Top0 zeigtj] [TP Hansk tj seinem Onkel ti und [TP ek verkauft sie seiner Tante]]]] 16.  Another asymmetry exists between object-gap constructions as in (24) and subject-gap constructions as in (1): the difference in the prosody required for each. A subject-gap construction requires no “prosodic support” establishing the required symmetries, while object-gap constructions depend crucially on prosody of this kind.

 John R. te Velde

b. [TopP Dit boeki [Top0 heeftj] [TP Willik tj ti gelezen en [TP ek zal het zijn vrienden ti aanbevelen]]]] a′. [TopP Die Briefmarkeni [Top˚ zeigtj] [TP Hans tj seinem Onkel ti und [TopP ei [Top˚ verkauftk] [TP er tk seiner Tante ti]]]]] b′. [TopP Dit boeki [Top0 heeftj] [TP Willi tj ti gelezen en [TopP ei [Top0 zalk] hij tk zijn vrienden ti aanbevelen]]]]

According to the analysis in (24), DP-object fronting targets a different Spec-position than subject raising. The comparison indicates that the object DP-fronting in (24a,b) involves an element that does not serve as an antecedent for a parallel gap; this element may therefore occupy a position non-parallel to the gap and for this reason it appears to violate the CSC. By contrast, in (24a′,b′) DP-fronting does involve such an element and therefore the gap and antecedent must both occupy Spec,TopP. For this reason, (24a′,b′) appear to meet the CSC better than (24a,b), but under the present analysis with the reformulation of the CSC in (21), all structures in (24) meet the CSC equally well. In (23) we noted that wh-constructions appear to rule out the kind of asymmetry apparent in LEE-constructions with subject gaps like (24a,b). A syntactic explanation of this difference is available, if we assume that wh-movement in (23b′) targets a different Spec-position than DP-object fronting in (24a′,b′):  wh-fronting targets Spec,CP, while object-fronting targets Spec,TopP. This means that there is more than one target available for fronting to the left of TP. Other evidence supporting this assumption, specifically that a position like Spec,TopP exists in WGmc., is available from non-wh-constructions like the V3-constructions in (25): (25) V3-constructions in German and Dutch: 1 2 3 a. [Hätten die Terroristen sich mit ihm beraten] er hätte ihnen ein had the terrorists refl with him conferred he had them a neues Konzept verordnet17 new plan prescribed ‘Had the terrorists conferrred with him, he would have given them a new plan.’

17.  Peter Schneider, Paarungen, p. 135 (Rowohlt Taschenbuch Verlag, Reinbek bei Hamburg, 1994).

A conjunction conspiracy at the West Germanic left periphery 

1 2 3 b. [Gingen in 2003 nog maar 91 Afghanen terug] vorig jaar waren went in 2003 even yet 91 Afghans back previous year were dat er 24818 that expl 248 ‘While even in 2003 91 Afghans returned, the previous year the total was 248.’ 1 2 3 c. [Wer mitspielt] der wird gewinnen whoever with-plays he will win ‘Whoever plays along will win.’ 1 2 3 d. [Wie meespeelt] die zal winnen

(see gloss and translation of 25c)

Because wh-fronting targets the farthest Spec available at the left periphery of WGmc., no other element may be fronted to its left, unlike constructions with DP-fronting to the left of a subject that is the antecedent of a gap in the second conjunct, as in (20). Thus, only in constructions like (20), but never in LEE-constructions with a wh-element as an antecedent, can the kind of asymmetry caused by DP-fronting over the antecedent occur, since even nominative (subject) wh-elements must target this position at the very left edge of the WGmc. left periphery.19 To sum up the findings of (23) – (25), we have seen that an analysis of Dutch and German in which subject, object and wh-fronting all target different positions accounts for the data and supports the assumption about the CSC formulated in (21). Furthermore, it supports a phase-based approach to LEE constructions. An approach to fronting that proceeds by phase and has multiple targets predicts that some conjuncts will have a different structure than others, that absolute symmetry is not required. Another construction with asymmetries similar to those in (24a,b) is given in (26); this one involves a fronted adverbial (PP) and a subject gap: (26) [TopP [PP In den Wald] ging [TP der Jägeri und [TP ei fing einen Hasen]]] into the wood went the hunter and caught a rabbit

18.  “Illegalen vertrekken vaker vrijwillig”, De Volkskrant (10 januari 2005, www.volkskrant.nl) 19.  For further discussion see te Velde (2005b). In this approach to V2, the “constraint” that positions the finite verb in second position only is driven by feature checking, not by a limitation on the positions available. This is in keeping with the minimalist strategy in which Merge is always triggered by the need to check a feature.

 John R. te Velde

Asymmetries apparent in (26) are: 1) Conjunct 1 has an intransitive verb and a TopP projection with a PP in its Spec position; conjunct 2 has a transitive verb and a leftedge subject gap with no TopP projection. The fronting of the adverbial in den Wald occurs in the same manner as the fronting of the DP-object in (24a,b); it is not ATB. Support for this assumption comes from the fact that the PP in den Wald cannot be associated with the verb fangen ‘to catch’ from which comes fing in (26). The problem is the use of the accusative den; only the dative determiner dem, indicating location rather than the destination indicated with den, is grammatical:20 (27) Der Jäger fing einen Hasen in *den / dem Wald the hunter caught a rabbit in   the.acc / the.dat wood

In the next section we turn to [&] as a licensing element for LEE as further justification for this assumption about the CSC and a phase-based approach to LEE. We recall the claim that a c-command relation from [&] to the gap at the edge of the next conjunct is required in LEE constructions for the licensing of this left-edge gap. A phase-based approach to LEE predicts that this relation is the only syntactic requirement on LEE, since ATB movement is no longer possible and therefore no requirement such as Ross’s formulation of the CSC in (15) is necessary. Rather, the CSC is an LF-interface condition on parallelism as stated in (21). This restated CSC hinges on properties of [&] and assumptions about Copy and coordinate ellipsis.

3.3  Assumptions about [&], Copy and coordinate ellipsis For [&] to license the gap in LEE constructions, it must be a functional element, or at the very least have properties of a functional element. I assume that its primary syntactic functions are: 1) inducing the operation Conjunction (a right-edge merge operation), and 2) licensing a redundant element in its minimal c-command domain

20.  The fact that adverbial scope can sometimes extend over an entire coordinate structure is not related to movement, but only to the fact that scopal elements in coordinate stuctures typically have scope over the entire structure, regardless of syntactic position, i.e., fronting to the edge of the first conjunct is not a requirement. This property of adverbial scope is apparent in (i): (i)

Der Jäger fing einen Hasen im Wald und sah auch viele Hirsche the hunter caught a rabbit in.the wood and saw also many deer

It is impossible to assume any location for the second conjunct other than im Wald. This interpretation is possible, and, given the lack of any other adverbial, essentially required because of adverbial scope.

A conjunction conspiracy at the West Germanic left periphery 

for ellipsis.21 Arguments and evidence supporting the first function are presented in te Velde (2005a). Evidence supporting a minimal c-command relation for the licensing of LEE gaps comes from constructions like those in (28): (28) a. Hansi zeigt seinem Onkel die Briefmarken und dann verkauft er/*ei H shows his.dat uncle the stamps and then sells he

sie seiner Tante them his.dat aunt

a′. Hansi zeigt seinem Onkel die Briefmarken und er/ei verkauft sie seiner Tante b. Hans zeigt seinem Onkel die Briefmarkeni und er verkauft sie/*ei seiner Tante b′. Die Briefmarkeni zeigt Hans seinem Onkel ti und ei verkauft er seiner Tante ti c.

Billi has shown his uncle the stampsj and he/ei will now sell them/*ej to his aunt

c′. The stampsi Billj has shown to his uncle and ei ej will sell ti to his aunt c″. *Die Briefmarkeni zeigt Hansj seinem Onkel ti und ei verkauft ej seiner Tante ti

In (28a) a subject gap is ungrammatical because it cannot be licensed by [&] in a c-command relation; dann blocks this relation, whereas in (28a′) no blocking occurs. The same blocking occurs in (28b), this time with the gap of a direct object, because the subject intervenes. If the direct object is fronted as in (28b′), the c-command relation can be established. The same facts are obtainable from the English data in (28c,c′) with (28c′) indicating that two gaps can be licensed at the left periphery in English because there is no V2 requirement (but note the restriction in English illustrated in (30)). The closest equivalent to (28c′) in German is ungrammatical because the second gap, which must be in a post-verbal position because of V2, cannot be licensed. The same is true, of course, for the Dutch equivalent. A supporting argument for the assumption that [&] licenses the gap in LEE is the following: Conjunction is an operation that is associated with two syntactic properties: 1) redundancy (that occurs through reinteration resulting from conjunction) to maintain nonambiguity, and 2) the elimination of unnecessary and ungrammatical redundancy to achieve minimality and avoid ambiguity. A construction in which both properties are evident is (29): (29) Sam gave his money to Sally and George (gave) his (money) *(to) Jane

21.  Minimal (asymmetric) c-command is generally defined as follows: α minimally c-commands β iff (i) α c-commands β and (ii) there is no γ such that α asymmetrically c-commands γ and γ asymmetrically c-commands β

 John R. te Velde

The syntactic operation Copy functions in coordinate structures for economizing the derivation and is exploited by Select and Merge in the derivation of coordinate clauses, resulting in a redundancy (selection of matching lexical items) and ellipsis (elimination of a redundancy through movement of an element to the left edge where it can be licensed for deletion). In (29) Copy supplies the formal features needed for the interpretation of the elements that can be licensed for deletion (no phonetic realization). The finite verb ‘gave’ can be licensed prosodically and matched at the LF-interface with the identical element in the first conjunct for interpretation. The element ‘money’ can be licensed anaphorically by ‘his’ and matched at the LF-interface in the same way as ‘gave’. The preposition ‘to’ cannot be deleted (it must have phonetic realization) even though there is a match for it. I will not go into the details of the reasons why it cannot be licensed for deletion, as this would take us too far afield, but in brief the reasons are directly related to the impoverishment of English Case morphology and the lack of V2 in most declarative constructions.22 The last assumption that must be addressed in this subsection that relates to properties of [&] and coordinate ellipsis is a standard one in the literature that bears repeating here because of its significance to the present account. This assumption is that ellipsis is manifested phonetically only (resulting in unspoken elements), but all other features of the deleted element occur in the derivation and are part of the recovery/interpretation process. Under this assumption the V2 requirement of Dutch and German is manifested in LEE in parallel with other V2 constructions (no sharing relation requiring non-binary phrase structure is required, cf. discussion in §2.3). Furthermore, because of the V2 property the leftmost Spec-position (‘edge’) of a V2 clause is an ideal ellipsis site in coordinate structures with the proper redundancy, i.e., the presence of an antecedent in the preceding conjunct. It is an ideal ellipsis site for three reasons: 1) this position can be licensed by [&], 2) it’s a target of Internal Merge in WGmc., and 3) the ellipse can be recovered without any ambiguity in symmetric

22.  In German it is possible to have Gapping in a construction with both a direct and an indirect object, whereas the equivalent in English is ungrammatical:

(i) Hans gabi Marie Blumen und Karl ei Erika eine CD (ii) *John gave Mary flowers and Carl Erica a CD

Even though the German DPs do not have unambiguous Case morphology, the structure of German, characterized by the V2 requirement, combined with the default positions or sequence associated with DPs, i.e., Subject – IO – DO, renders this kind of construction unambiguously interpretable. The same construction is fine in English if the marker for IO is added:


John gave Mary flowers and Carl a CD to Erica

For this reason, the IO-marker in (29) cannot be licensed for deletion.

A conjunction conspiracy at the West Germanic left periphery 

coordinate structures. The same construction is not possible in English declarative clauses lacking the V2 requirement; the closest equivalent is ungrammatical, while the variant in (30b) with V2 is grammatical:23 (30) a. *The stamps showed John his uncle and sold he his aunt b. These stamps John has never shown his uncle nor would he consider selling to his aunt

Thus, in a rather particular way conjunction conspires with WGmc. fronting and V2 to minimize phonetic realization. [&] is able to license this phonetic minimization, with the results evident in LEE, because it is a functional element that induces conjunction, and conjunction is associated with two properties, redundancy and ellipsis.

4.  Related coordinate constructions without ellipsis In this section we consider two other constructions that Johnson accounts for using his VPR proposal. I will show that a phase-based approach again provides a better account; in this case the improvement occurs not only in the avoidance of the CSC issue – no CSC violation must be accounted for and therefore no ATB movement is necessary – but also in the over-all simplification of the derivations.

4.1  Conjoined vPs without subject ellipsis Johnson’s (2002) analysis of the following construction again utilizes VPR and the projection FP for meeting the CSC. His analysis again requires non-binary phrase structure, cf. (9), and movement out of both conjuncts, but not always in ATB fashion: (31) Johnson’s derivation of: Die Suppe wird der Hans essen und sich hinlegen the soup will the H eat and refl down-lay ‘Hans will eat the soup and (then) lie down.’

23.  A similar construction without ellipsis is grammatical:


[TopP The stamps [TP John [vP showed his uncle and [vP sold his aunt]]]]

I assume there is no need for ellipsis in this analysis in which the conjuncts are vPs that are both dominated by the TP in whose Spec the subject for both conjuncts is located. In this phrase structure the asymmetric relations establish the necessary syntactic agreement and the basis for interpretation in LF. Much more that goes beyond the scope of the immediate analysis must be said about the derivation of this structure and the assumptions behind it for this analysis to be convincing. I refer the reader to te Velde (2005a).

 John R. te Velde

a. [IP der Hans [I′[FP[F′[F′[VP die Suppe essen wird]] und [F′[VP sich hinlegen wird]]]]]]

b. [IP der H [I′[FP[F′[F′ wirdi [VP die Suppe essen ti]] und [F′ wirdi [VP sich hinlegen ti]]]]]]

c. [IP der H [I′[FP[VP2 [VP die S essen ti][F′[F′ wirdi tj] und [F′ wirdi [VP sich hinlegen ti]]]]]]

d. [CP wirdi [IP der H [I′[FP[VP2 [VP die S essen ti][F′[F′ ti tj ] und [F′ ti [VP sich hinlegen ti]]]]]]

e. [CP die Sk wirdi [IP der H [I′[FP[VP tk essen ti][F′[F′ ti tj ] und [F′ ti [VP sich hinlegen ti]]]]]]

In summary, Johnson’s derivation of this construction comes closer than his derivation in (9) to what one expects in an approach that seeks to preserve the CSC using ATB movement. Here wird moves ATB (whereas the finite verb hat in (9) does not). The derivation leaves at least two questions open, however (besides justification for this kind of phrase structure):  First, how do the two occurrences of wird become one in step d? Second, what is the goal of VP-raising in step c? It appears to include the VP dominating the VP die Suppe essen and the trace of wird as follows: [VP [VP die Suppe essen] [V twird]], as is necessary for the trace to be raised along with the sister VP. However, this upper VP must remain intact so that the trace of VPraising has a projection. Therefore, a new VP, labeled VP2, must be generated at the target, a position dominated directly by FP. It appears to be an ad hoc solution to a problem created by Johnson’s VPR approach and the exceptions to the CSC that he allows:  the VP can be raised separately from wird in just the second conjunct (in non-ATB fashion) while wird in both conjuncts is raised ATB in accordance with the CSC. The formal details of these operations are not all clear and the exceptions to the CSC not convincing. A phase-based derivation following the assumptions about phrase structure already outlined would proceed as in (32): (32) Phase-based derivation of (1a) from Johnson (2002): Die Suppei wird der Hans ti essen und sich hinlegen the soup will the H eat and refl down-lay a.

Select lexical array: Hansi die Suppe essen Hansi hinlegen

b. Merge lexical items for the first conjunct: [VP Hans [V′[DP die Suppe] essen]] c. vP phase generating a TP (with merger of aux-fut in T0) [TP Hans [T′ wird [vP[v′[DP die Suppe]i [VP ti essen]]]]] d. CP phase with fronting of the DP-object and merger of der: [CP [DP die Suppe]i wird [TP der Hans [vP ti [VP ti essen]]]] e. Extract and merge lexical items for C-2 (subarray): [VP[V′[DP Hans] hinlegen]] (Note: Hans is DO in C-2) f. Derive conjunct 2 (as subphase of the vP Phase in c): [vP[v′[DP Hansj][VP tj hinlegen]]]

A conjunction conspiracy at the West Germanic left periphery 

g. Merge conjuncts (with merger of und and reflexivization of Hans):24 [CP d Si wk [TP d H tk [vP[v′ ti [VP ti essen] ]]]]] (die Suppe wird der Hans essen) ↑ [vP und ← [vP[v′[DP sichi][VP ti hinlegen]]]]]

I assume that conjoined vPs do not require subject ellipsis since the subject in Spec,TP dominates both vPs. This phrase structure, although it requires a type of sharing, stands in constrast to the structure in (4), here as (33), for which the sharing relation assumed requires intermediate projections as conjuncts and non-binary phrase structure: (33) [CP Hans [C′ zeigt seinem Onkel die Briefmarken] und [C′ verkauft sie seiner Tante]]

In (32) there is no violation of the CSC because the fronting of die Suppe does nothing to make the construction asymmetric in a way that violates the Parallelism Requirement. The derivation is more minimalistic in that only one movement operation is required. To reduce movement, I follow Chomsky (2001) where assumptions about the extraction of subarrays are outlined. I apply these here to the derivation of the conjunct sich hinlegen, a vP, which quite clearly does not have the status of a “full” phase.

4.2  Conjoined, split DPs In this subsection we turn to (34) and (35), (1b) and (47) from Johnson (2002) respectively, that have conjoined DPs in which one has its NP extracted and fronted: (34) Äpfeli isst der Hans drei ti und zwei Bananen apples eats the H three and two bananas ‘Of the apples Hans is eating three, and two bananas.’

Johnson points out that (35), identical to (34) except for the tense, is ungrammatical: (35) *Äpfeli wird der Hans drei ti und zwei Bananen essen apples will the H three and two bananas eat

Johnson proposes the derivation in (36) for (35), again using VPR, in which the fronting operation in step d violates the CSC, the reason he gives for its ungrammaticality: (36) a. [IP der Hans [I′[FP[F′[F′[VP[VP[DP[drei Äpfel] und [zwei Bananen]] essen] wird]]]]]] b. [IP der Hans [I′[FP[VP[DP[drei Äpfel] und [zwei Bananen]] essen]i][F′[F′[VP[VP ti] wird]]]]]

24.  An alternate analysis would be to assume that sich is part of the lexical array instead of Hans, and that there is no operation ‘reflexivization’. I leave this question to further research, as it doesn’t bear on the present analysis.

 John R. te Velde

c. [CP wirdj [IP der H [I′[FP[VP[DP [drei Äpfel] und [zwei Bananen]] essen]i] [F′[F′[VP ti] tj]]]]] d. [CP [drei Äpfel]k wirdj [IP der H [I′[FP[VP essen [DP tk und [zwei B]]]i] [F′[F′[VP ti] tj]]]]]

The argument that a CSC violation occurs in (35) but not in (34) seems weak, given that the operation is the same in both derivations, with the only difference between the two the use of the future auxiliary in (35), a difference that should be unrelated to DP-fronting. We recall the very similar construction in (32) in which a future auxiliary is used and DP-fronting is grammatical. In other construction pairs of this sort, no prohibition against fronting occurs in the construction with a future auxiliary. We take (37a,b) as one example: (37) a. Die Briefmarken wird Hans dem Onkel zeigen und der the stamps will H the.dat uncle show and the.dat

Tante verkaufen aunt sell

b. Die Briefmarken zeigt Hans dem Onkel und verkauft er (he) der Tante

Furthermore, the variant of (35) in (38) is perfectly grammatical, just stylistically a bit odd: (38) Äpfeli wird der Hans drei ti essen, und zwei Bananen

Again, we have reason to doubt the argument that fronting causes the ungrammaticality in (35). Let’s compare (35) and (38). In a phase-based derivation of (35), no CSC violation occurs when Äpfel is fronted, for reasons already outlined. The ungrammaticality of the construction is due to the unlicensed trace of Äpfel; this NP-trace after drei cannot be licensed by [&], even if it were a LEE-type gap. Another key difference is that this trace, in contrast to a gap in a LEE-construction, cannot be recovered by Match. Because the trace is in a conjunct-final position, the type of licensing required is the kind used in Right Node Raising (RNR) constructions. RNR is characterized by a gap in the clause-final position of all but the last conjunct. Hartmann (2000) shows convincingly that RNR requires prosodic licensing. The appropriate prosody needed for licensing the gap after drei is not felicitious in this construction, however, following the arguments of Féry and Hartmann (2005). In contrast to (35) the trace in (38) can be licensed anaphorically and prosodically. The derivation of (38) using a phase-based approach is outlined in (39): (39) Phase-based derivation of (38): a. select lexical array; merge the VP with first DP-conjunct only: [VP Hans [V′ drei Äpfel essen]]

A conjunction conspiracy at the West Germanic left periphery 

b. vP-phase; merge wird: [TP Hans wird [vP [DP drei Äpfel]i [VP ti essen]] c. front the NP Äpfel, with V-to-Top0; merge der: [TopP Äpfelj wird [TP der Hans [vP[DP drei tj]j [VP ti essen]] d. extract second DP-conjunct and conjoin/merge with first DP-conjunct: [TopP Äpfelj wird [TP der Hans [vP[DP drei tj]i [VP ti essen] ]]] ↑ [DP und ← [DP zwei Bananen]]

The licensing of the trace in (39) proceeds much as in VP Ellipsis, which has been shown by Merchant (2001) and others to utilize an anaphor, in this case drei; it also requires a certain prosody. Space considerations do not allow a discussion of anaphoric licensing outlined by Merchant. As with the derivation in (32), I follow the assumption that a subarray, in this case a DP conjunct, can be extracted later from the lexical array after it has completed its derivation in narrow syntax. At this point the subarray is extracted, merged and derived as necessary. The merge position is on the right branch of V′, thus having the structure in (40):25 (40) structure of the conjoined DPs in (38/39) (ignoring traces in the VP): vP v′ drei tÄpfel

VP V′ V | essen

vP v′

& DP


zwei Bananen

It is readily apparent from (40) that conjunction in my proposal, as a right merge operation, can apply rather freely as the phrase structure permits, and according to 25.  An analysis could be constructed in which there is a VP on the right branch of the lowest v′ (where … occurs) and that this VP has an ellliptical form of essen. I will leave this possibility to further research.

 John R. te Velde

where features provided by Copy can be inserted in the merged conjunct. Whether the result is grammatical is determined at the LF-interface where Match applies. The structure in (40) meets the requirements of Match because the DP drei Äpfel preceding it has many features in common with it. Copy applies at conjunction for transferring the formal features to the second DP conjunct, thus simplifying the syntactic derivation and assuring a structure that will Match with the first conjunct at the LF-interface. The semantic symmetries of the two DP-conjuncts are obvious. In the next section we consider an alternate way to account for LEE that has not to my knowledge been explored in the literature.26

5.  LEE and scope in asymmetric phrase structure It has been suggested that in an approach to coordinate structures with asymmetric phrase structure throughout, scopal relations should account for the “missing” lexical items that in my proposal occur in the form of gaps representing unspoken words. A possible advantage of a scopal approach is that no gaps are necessary, and thus the derivation is minimalized in certain respects. A problem with this approach is that there is evidence that DPs behave differently than adverbs w.r.t. to scope. Thus, in (41a) it is not possible to have a subject gap, even though Hans, the matching element, is in a position where it should have scope over the gap, if it had the same scopal properties as heute in (41b). A subject gap is permitted in (41a’) for reasons explored earlier: (41) a. Hans kauft den Wagen im Stadtzentrum und dann fährt H buys the car in-the city-center and then drives

er/*e damit in die Berge he it-with into the mountains

a′. Hansi kauft den Wagen im Stadtzentrum und er/ei  fährt dann damit in die Berge b. Heute kauft [TP Hansi den Wagen und [TP Karl/ei  fährt (heute) damit in die Berge]] b′. Hansi kauft den Wagen heute und Karl/ei fährt (heute) damit in die Berge c. Heute kauft [TP Hansi den Wagen und [TP Karl/ei  fährt morgen damit in die Berge]] c′. Hansi kauft den Wagen heute und Karl/ei fährt morgen damit in die Berge

26.  This alternate analysis was suggested to me by an anonymous reviewer, and it is inherent in Johnson’s (2002) analysis.

A conjunction conspiracy at the West Germanic left periphery 

Objects do not behave any differently than subjects; both can be licensed for deletion when they occur at the left edge  – but only at the left edge  – and both can be recovered: (42) a. Den neuen (new) Wagen kauft Hans im Stadtzentrum und ei fährt er in die Berge a′. Den neuen Wagen kauft Hans im Stadtzentrum und Karl fährt ihn/*ei in die Berge a″. Hans kauft den Wageni im Stadtzentrum und Karl fährt *ei/ihn in die Berge b. Den Wagen kauft [TP Hansi heute und [TP Karl/ei fährt ihn (heute) in die Berge]] b′. Heute kauft [TP Hansi den Wagen und [TP Karl/ei fährt ihn (heute) in die Berge]] b″. Hansi kauft den Wagen heute und Karl/ei fährt ihn (heute) in die Berge c. Den Wagen kauft [TP Hansi heute und [TP Karl/ei fährt ihn morgen in die Berge]] c′. Heute kauft [TP Hansi den Wagen und [TP Karl/ei fährt ihn morgen in die Berge]] c″. Hansi kauft den Wagen heute und Karl/ei fährt ihn morgen in die Berge

As the data indicate, an adverb in a first conjunct has scope over the entire coordinate structure regardless of position, as long as another adverb is not introduced in the second conjunct. A DP, on the other hand, must be in a position parallel to that of the gap in order to be interpretable in the second conjunct. The reason for this, as maintained in my analysis, is that a DP is an antecedent, does not have scopal properties like an adverb(ial), and can be matched at the LF-interface with a gap in a parallel position for rendering the interpretation. A DP can appear to have scopal properties because of its dominance over other positions in asymmetric phrase structure, but in no case is this dominance relation equatable with adverbial scope.

6.  Summary and conclusion The phase-based approach to elliptical V2 coordinate structures in Dutch and German proposed here accounts for the asymmetries of certain of these structures without resort to ad hoc solutions for CSC violations that result from other analyses. In this analysis the CSC is understood as a statement on the parallelism required on the semantic side, not as a syntactic constraint, since no ATB movement occurs (prohibited by the PIC). This phase-based approach derives coordinate symmetries using Select, Merge and Copy, and recovers elliptical elements with Match. It predicts the edge requirement on gaps in LEE-constructions, if it is assumed that the PIC has validity for semantic interpretation. This constraint can be unified with the syntax of conjunction, a merge operation, for the analysis of LEE-constructions by positing a licensing requirement on left-edge gaps, satisfied by the c-command relation of [&] to the gap. When the edge of a V2 clause is defined according to the well-documented asymmetries between

 John R. te Velde

subject-initial, wh-, and all other V2 structures, and at least two head positions are used to the left of T0, many of the problems that the derivation of these coordinate structures have caused in previous accounts can be eliminated.

References Büring, D. & Hartmann, K. 1998. Asymmetrische Koordination. Linguistische Berichte 174: 172–201. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by Step, R. Martin, D. Michaels & J. Uriagereka (Eds), 89–155. Cambridge MA: The MIT Press. Chomsky, N. 2001. Derivation by phase. In Ken Hale. A Life in Language, M. Kenstowicz (Ed.), 1–52. Cambridge MA: The MIT Press. Chomsky, N. 2008. On phases. In Foundational Issues in Linguistic Theory, Essays is Honor of JeanRoger Vergnaud, R. Freidin, C.P. Otero & M.-L. Zubizaretta (Eds), 133–166. Cambridge MA: The MIT Press. Féry, C. & Hartmann, K. 2005. The focus and prosodic structure of German right node raising and gapping. The Linguistic Review 22: 69–116. Haegeman, L. 1991. On the relevance of clitic placement for the analysis of subject-initial verb second in West Flemish. Groninger Arbeiten zur Germanistischen Linguistik (GAGL) 34: 26–66. Haegeman, L. 1998. Verb movement in embedded clauses of West Flemish. Linguistic Inquiry 29: 631–656. Hartmann, K. 2000. Right Node Raising and Gapping. Interface Conditions on Prosodic Deletion. Amsterdam: John Benjamins. Heycock, C. & Kroch, A. 1994. Verb movement and coordination in a dynamic theory of licensing. The Linguistic Review 11: 257–284. Höhle, T.N. 1983. Subjektlücken in Koordinationen. Ms, Universität Köln. Hornstein, N. & Nunes, J. 2002. On asymmetries between Parasitic Gap and Across-the-Board constructions. Syntax 5: 26–54. Johnson, K. 2002. Restoring exotic coordinations to normalcy. Linguistic Inquiry 33: 97–156. Merchant, J. 2001. The Syntax of Silence: Sluicing, Islands and Identity in Ellipsis. Oxford: OUP. Ross, J.R. 1967. Constraints on Variables in Syntax. Ph.D. dissertation, MIT. Schwarz, B. 1998. On odd coordinations in German. Journal of Comparative Germanic Linguistics 2: 191–219. te Velde, J.R. 2005a. Deriving Coordinate Symmetries. A Phase-based Approach Integrating Select, Merge, Copy and Match [Linguistik Aktuell 89]. Amsterdam: John Benjamins. te Velde, J.R. 2005b. New evidence for relativized V2 in West Germanic. Paper presented at the Comparative Germanic Syntax Workshop 20 (Tilburg). Williams, E. 1977. Across-the-Board application of rules. Linguistic Inquiry 8: 419–423. Williams, E. 1978. Across the-Board rule application. Linguistic Inquiry 9: 31–43. Zwart, C.J.-W. 1991. Subject deletion in Dutch: A difference between subjects and topics. In Language and Cognition 1. Research Group for Linguistic Theory and Knowledge Representation of the University of Groningen, M.  Kas, E.  Reuland & C.  Vet (Eds), 333–350. Groningen: University of Groningen.

part ii

Word order and movement

Reconsidering odd coordination in German* Hironobu Kasai

University of Kitakyushu This paper investigates why the Coordinate Structure Constraint (CSC) is violable in odd coordination in German. Schwarz’s (1998) analysis successfully handles nominal odd coordination but some cases of verbal odd coordination are problematic for his analysis. This paper offers an alternative analysis of verbal odd coordination, under which the CSC is subject to the Principle of Minimal Compliance, proposed by Richards (1998) on independent grounds. Verbal odd coordination involves V-to-C movement in an ATB-way, which allows the computational system to ignore a violation of the CSC. This paper has two theoretical implications. One is that head movement takes place in the narrow syntax. The other one is that the CSC is defended as a syntactic constraint on overt movement.

1.  Introduction The Coordinate Structure Constraint (henceforth, CSC) is one of the syntactic constraints on locality, formulated by Ross (1967, 1986), which is given in (1).

(1) In a coordinate structure, no conjunct may be moved, nor may any element contained in a conjunct be moved out of that conjunct. (Ross 1986: 98–99)

The constraint given in (1) captures the ungrammaticality of (2). (2) a. *Who1 do you think Mary likes t1 and Bill hates Sue? b. *Who1 do you think Mary hates t1 and Bill?

In (2a), who moves out of the first conjunct. On the other hand, in (2b), the first conjunct itself undergoes movement. The CSC has a well-known exception. If movement is applied to both of the conjuncts, as shown in (3), then the result of the application

*An earlier version of this paper was presented at the 22nd Comparative Germanic Syntax Workshop. I thank the audience at the conference for useful comments and questions. I thank Clemens Mayr and Dennis Ott for helpful discussion. I am grateful to two anonymous reviewers for valuable comments and suggestions. Thanks also go to Adam Stott for suggesting stylistic improvements. All remaining errors are my own.

 Hironobu Kasai

is well-formed. This type of movement has been referred to as across-the-board (henceforth, ATB) movement.

(3) Who1 do you think Mary likes t1 and Bill hates t1?

Interestingly, the following grammatical examples from German seem to violate the CSC:1 (4) a. Die Suppe wird der Hans essen und sich hinlegen. the soup will the Hans eat and self down-lie ‘Hans will eat the soup and lie down.’ (Schwarz 1998: 191) b. Das Gepäck sollst du sofort fallen lassen und zum the luggage must you immediately fall let and to-the

Ausgang rennen. exit run

‘You should immediately drop the luggage and run to the exit.’ (Heycock & Kroch 1994: 272)

On the assumption that the same syntactic categories are conjoined, the examples above would involve coordination of verbal categories and violate the CSC. For example, in (4a), die Suppe ‘the soup’ undergoes movement out of the first conjunct, as illustrated below:

(5) Die Suppe1 wird der Hans [[t1 essen] und [sich hinlegen]]

Schwarz (1998) provides similar apparent CSC violations, where movement out of NP-coordination appears to violate the CSC. The following example is one of them: (6) Äpfel ißt der Hans drei und zwei Bananen. apples eats the Hans three and two bananas ‘Hans eats three apples and two bananas.’ (Schwarz 1998: 195)

1.  In Dutch, the same type of asymmetric extraction exhibits less grammaticality. (i) ??Dit this

voorstel wil de commissie volgen en een suggestion wants the committee follow and a

nieuwe subcommissie vormen. new subcommittee set-up

‘The committee wants to follow this suggestion and set up a new subcommittee.’ (Heycock & Kroch 1994: 273)

This paper has no explanation for the difference between German and Dutch. The issue is left for future research.

Reconsidering odd coordination in German 

As illustrated in (7), äpfel ‘apples’ apparently moves out of the first conjunct of the conjoined NPs.

(7) Äpfel1 ißt der Hans [[ drei t1 ] und [zwei Bananen]]

In this paper, following Schwarz (1998), (4) and (6) are referred to as verbal odd coordination and nominal odd coordination, respectively. In fact, Schwarz argues that the analyses given in (5) and (7) are incorrect and odd coordination involves no extraction out of a coordinate structure (see section 2 for details).2 One might say that the CSC is not operative in German. However, this is not the case. Let us consider (8). (8) *Die Suppe wird der Hans essen und wird der Peter sich hinlegen. the soup will the Hans eat and will the Peter self down-lie ‘Hans will eat the soup and Peter will lie down.’ (Schwarz 1998: 193)

In (8), die Suppe moves out of the first conjunct, as illustrated in (9), which leads to a violation of the CSC.

(9) Die Suppe1 [[C′ wird der Hans t1 essen] und [C′ wird der Peter sich hinlegen]]

If the CSC were not operative in German, then it would be unclear why (8) is ungrammatical. It is not plausible to simply weaken the CSC in German. The aim of this paper is to investigate why the CSC is violable in odd coordination in German. The paper is organized as follows. Section 2 reviews Schwarz’s analysis of odd coordination and points out that his analysis of verbal odd coordination needs to be reconsidered. Section 3 offers an alternative analysis of verbal odd coordination, under which the CSC is subject to the Principle of Minimal Compliance, which is

2.  There is another interesting construction in German in terms of the CSC, which is referred to as “Subject gap in finite sentences.” The relevant example is given below: (i)

Den Hund hat einer gefüttert und hat ihn geschlagen. the dog has someone fed and has it beaten ‘Someone fed the dog and beat it.’ (Schwarz 1998: 213)

In a similar way to odd coordination, den Hund ‘the dog’ seems to violate the CSC, as illustrated in (ii).


Den Hund1 [[C′hat einer t1 gefüttert] und [C′ hat ihn geschlagen]]

This construction is also problematic in the sense that the second conjunct lacks a subject. Recall that German is not a null subject language. This construction is beyond scope of this paper. See Höhle 1990; Zwart 1991; Heycock and Kroch 1994; Büring and Hartmann 1998; Johnson 2002 and Hallman 2004, and references therein, for their detailed analyses.

 Hironobu Kasai

proposed by Richards (1998) on independent grounds. Under the proposed analysis, a violation of the CSC can be saved by the presence of ATB movement, which is involved in verbal odd coordination in German. Section 4 discusses an implication for the theoretical status of the CSC. Section 5 concludes the paper.

2.  Schwarz (1998) 2.1  Ellipsis analysis This section will review Schwarz’s (1998) analysis of odd coordination. He argues that the relevant construction involves no asymmetric movement out of the first conjunct, hence the CSC is not violated. The apparent CSC violation is caused by ellipsis in the second conjunct. First, let us consider nominal odd coordination, exemplified by (6), repeated as (10a). (10) a.


Äpfel1 ißt der Hans drei t1 und zwei Bananen. apples eats the Hans three and two bananas ‘Hans eats three apples and two bananas.’ (Schwarz 1998: 195) [Äpfel1 ißt der Hans drei t1] und [der Hans ißt zwei Bananen]

He argues that (10a) involves CP-coordination with ellipsis in the second conjunct, instead of NP-coordination with a violation of the CSC. Äpfel ‘apples’ does not move out of the coordinate structure but rather undergoes movement within the first conjunct, as shown in (10b). In the second conjunct, only the object is left behind, as a result of the ellipsis of the subject and the verb. Schwarz claims that the relevant ellipsis operation is gapping, which takes place in a forward way. It is expected that the first conjunct of odd coordination should be fullfledged in isolation because ellipsis takes place only in the second conjunct, not in the first conjunct in the relevant construction. The first conjunct of (10a) is grammatical in isolation. The ungrammaticality of the following example also confirms Schwarz’s analysis: (11) *Äpfel wird der Hans drei und zwei Bananen essen. apples will the Hans three and two bananas eat ‘Hans will eat three apples and two bananas.’ (Schwarz 1998: 197)

The ungrammaticality of (11) has the same source as (12), which misses a verb. (12) *Äpfel wird der Hans drei. apples will the Hand three (Schwarz 1998: 199)

In order to derive (11), it is necessary to elide essen ‘eat’ in the first conjunct, in addition to the ellipsis of der Hans ‘the Hans’ and wird ‘will’ in the second conjunct. However, gapping does not allow essen to be elided because gapping cannot take place in the first

Reconsidering odd coordination in German 

conjunct. If (10a) and (11) are analyzed in terms of NP-coordination, as given in (13a) and (13b), respectively, the contrast between them fails to be captured. Both of them are supposed to violate the CSC. (13)


Äpfel1 ißt der Hans [[NP drei t1] und [NP zwei Bananen]]


Äpfel1 wird der Hans [[NP drei t1] und [NP zwei Bananen]] essen

The availability of the ellipsis in the second conjunct explains not only nominal odd coordination such as (10a) but also the following example: (14) a. Äpfel wird der Hans drei essen und zwei Bananen. apples will the Hans three eat and two bananas ‘Hans will eat three apples and two bananas.’ b. [Äpfel wird der Hans drei essen] und [der Hans wird zwei Bananen essen]

(Schwarz 1998: 204)

It appears that CP and NP are conjoined on the surface in (14a). On the assumption that the same syntactic categories are conjoined, the full clause is supposed to be reduced in the second conjunct, as illustrated in (14b). Thus, Schwarz’s analysis does not have to postulate any additional mechanism for nominal odd coordination. Verbal odd coordination such as (15a) also receives a similar explanation. The subject and the auxiliary are elided in the second conjunct, as shown in (15b). (15) a. Die Suppe wird der Hans essen und sich hinlegen. the soup will the Hans eat and self down-lie ‘Hans will eat the soup and lie down.’ (Schwarz 1998: 191) b. [Die Suppe1 wird der Hans t1 essen] und [der Hans wird sich hinlegen]

In (15b), the movement of die Suppe ‘the soup’ takes place within the first conjunct, which involves no CSC violation, again. The ellipsis involved in verbal odd coordination is also gapping. Since the relevant ellipsis cannot apply in the first conjunct, the first conjunct of verbal odd coordination should be full-fledged in isolation in a parallel way to nominal odd coordination. The ungrammaticality of (16a) confirms this point. (16) a. *Die Suppe soll der Hans zu essen und sich hinzulegen versuchen. the soup should the Hans to eat and self down-to-lie try ‘Hans should try to eat the soup and lie down.’ (Schwarz 1998: 208) b. *Die Suppe soll der Hans zu essen the soup should the Hans to eat (Schwarz 1998: 209)

(16a) is excluded in the same way as (16b), which lacks a verb which takes the infinitival clause.

 Hironobu Kasai

2.2  Reconsidering verbal odd coordination Schwarz’s analysis of nominal coordination is plausible and maintainable. On the other hand, there are two problems with his treatment of verbal odd coordination. First, some instances of verbal odd coordination cannot be handled readily under the ellipsis analysis, which is noted by Schwarz himself. The relevant example is given below: (17) Den Hund hat sie nicht gefüttert und ihn geschlagen. (neg> &, &>neg) the dog has she not fed and it beaten ‘She did not feed the dog and beat it.’ (Schwarz 1998: 212)

Under the ellipsis analysis, (17) would be analyzed as (18), where the subject and the auxiliary undergo ellipsis in the second conjunct. (18) [Den Hund1 hat sie nicht t1 gefüttert] und [sie hat ihn geschlagen]

(17) is ambiguous with respect to scope interaction between negation and conjunction. The narrow scope interpretation of negation is expected under the structure presented in (18). However, the wide scope reading of negation is also available for (17), which cannot be captured under the structure in (18). In order to capture the wide scope interpretation of negation, negation should be base-generated outside of the conjoined phrases, as illustrated in (19).

(19) Den Hund1 hat sie nicht [[t1 gefüttert] und [ihn geschlagen]]

Under the structure in (19), however, den Hund ‘the dog’ moves out of the first conjunct, obviously. If the analysis given in (19) is correct, then it is necessary to investigate why the CSC is violable in (19). Second, recall that under Schwarz’s analysis, verbal odd coordination such as (15a) is derived from the conjoined sentences, as illustrated in (15b). However, as has been pointed out in the literature, it would be suspicious that the apparent VP-coordination is derived from coordination of full clauses via ellipsis (cf. Partee and Rooth 1983). If (20a) were analyzed as (20b), (20a) would have the same interpretation as (21). (20) a. No student got an F and passed the course. (Munn 1993: 144) b. No student got an F and no student passed the course. (21) No student got an F and no student passed the course. (ibid.: 144)

However, it is crucial that (20a) does not have the same interpretation as (21). Thus, the analysis given in (15b) needs to be reconsidered. The analysis presented in the next section, on the other hand, argues that verbal odd coordination involves no ellipsis, in contrast to Schwarz’s analysis. Rather, verbal odd coordination just involves coordination of verbal categories, not finite CPs and no ellipsis takes place in the second conjunct. If so, whatever the ultimate analysis of the interpretive difference between (20a) and (21)

Reconsidering odd coordination in German 

may be, under the alternative analysis it would be possible to explain the difference in question, because they have different structures. To sum up, in this section, Schwarz’s analysis of odd coordination has been carefully examined. His ellipsis analysis of nominal odd coordination is maintainable, but that of verbal odd coordination needs to be reconsidered. As has been discussed in this section, the scope interaction between negation and conjunction strongly indicates that verbal odd coordination indeed involves movement out of the first conjunct. The next section will give an answer to the question as to why the CSC is violable in the construction under investigation.

3.  An alternative analysis 3.1  The Principle of Minimal Compliance The aim of this section is to provide an alternative analysis. Before that, let us introduce Richards’ (1998) Principle of Minimal Compliance (henceforth, PMC), which plays a crucial role for the alternative analysis. (22) Principle of Minimal Compliance For any dependency D that obeys constraint C, any elements that are relevant for determining whether D obeys C can be ignored for the rest of the derivation for purposes of determining whether any other dependency D’ obeys C. (Richards 1998: 601) (23) An element X is relevant to determining whether a dependency D with head A and tail B obeys constraint C if a. X is along the path of D (that is, X=A, X=B, or A c-commands X and X c-commands B); and b. X is a member of the class of elements to which C makes reference. (ibid.: 601)

The basic idea of the PMC is that an ill-formed dependency in isolation can be saved by the presence of a well-formed dependency with respect to a particular constraint. Let us consider the following examples: (24) a. *Who1 did his1 mother introduce t1 to Mary? b. Who1 did John introduce t1 to his1 teacher? c. ?Who1 did his1 mother introduce t1 to his1 teacher? (Richards 1998: 600, adapted from Hornstein 1995)

Suppose the dependency between a moved wh-phrase and its bound pronoun is subject to the constraint given in (25). (25) All pronouns bound by a wh-phrase must also be bound by a trace of that wh-phrase in an A-position. (Richards 1998: 603)

 Hironobu Kasai

(24b) satisfies the constraint (25) because the bound pronoun is bound by the trace of who. On the other hand, in (24a), the bound pronoun is not bound by the trace of who. (24c) also involves this ill-formed dependency, but surprisingly (24c) is significantly better than (24a). This otherwise mysterious contrast between (24a) and (24c) is expected under the PMC. The dependency between the second instance of his and who satisfies (25) because the former is bound by the trace of the latter. This well-formed dependency enables who to be ignored, for the purpose of determining whether the dependency between the first instance of his and who obeys the constraint in (25). Since the computational system ignores who, the ill-formed dependency between who and the first instance of his also ends up being ignored. Thus, the grammaticality of (24c) follows. Richards’ PMC also gives an explanation for the asymmetry between overt movement and covert movement with respect to subjacency. The overt movement exhibits the subjacency effect in (26a), in contrast to the covert movement of what in (26b). (26) a. *What1 do you wonder who bought t1? b. Who1 t1 wonders who bought what? (Richards 1998: 604)

Suppose wh-movement obeys the following constraint: (27) A [+wh] complementizer cannot be associated with a wh-phrase via movement in a way that crosses barriers of a certain kind (wh-islands, complex noun phrases, etc.). (Richards 1998: 605)

In (26b), the matrix [+wh] complementizer is associated with who, which satisfies (27). This well-formed dependency allows the [+wh] complementizer to be ignored for evaluating the dependency between the [+wh] complementizer and what, which is in the island. Since the [+wh] complementizer is ignored, the dependency between the [+wh] complementizer and what, which is ill-formed in isolation, is also ignored. See Richards (1998) for more empirical arguments for the PMC.

3.2  PMC and CSC The main proposal of this paper is that the CSC is also subject to the PMC. Suppose the movement out of a coordinate structure obeys the following constraint: (28) Suppose a functional category F c-commands a coordinate structure and triggers movement out of the coordinate structure. Then, the movement dependency with F must be established in an across-the-board way.

Under the proposed analysis, if F induces ATB-movement out of a coordinate structure, which is a well-formed dependency with respect to the CSC, then the computational

Reconsidering odd coordination in German 

system is capable of ignoring F for determining whether any other movement dependency with F obeys the CSC.3 This is schematically illustrated in (29). FP





….&P… YP α

&′ β &′

ZP β

ATB-movement --- >OK CSC

Suppose both α and β undergo movement, which is triggered by F. In (29), the existence of the ATB-movement of β enables F to be ignored for the purpose of evaluating whether the movement of α satisfies the CSC. This is indeed involved in verbal odd coordination in German, where F in (29) is C. Now let us return to the derivation of (4a), repeated as (30). (30) Die Suppe1 wird der Hans t1 essen und sich hinlegen. the soup will the Hans eat and self down-lie ‘Hans will eat the soup and lie down.’ (Schwarz 1998: 191)

Nothing prevents (30) from being analyzed as (31), where wird ‘will’ is base-generated within the coordinate structure and moves to C out of each conjunct in an ATB way.

(31) Die Suppe2 wird1 der Hans [[[t2 essen] t1] und [[sich hinlegen]t1]]

Then, die Suppe ‘the soup’ moves to [Spec, CP]. The movement of die Suppe violates the CSC, but it is saved by the existence of the ATB-movement of wird in the way illustrated in (29).

3.  Various approaches have been proposed for ATB-movement in the literature (see Williams 1978; Munn 1992; Hornstein & Nunes 2002; Citko 2005; Bachrach & Kazir 2007 and references therein). The issue as to what mechanism is involved in ATB-movement is beyond the scope of this paper.

 Hironobu Kasai

The proposed analysis also gives a natural explanation for the ungrammaticality of (16a), repeated as (32a). (32) a. *Die Suppe soll der Hans zu essen und sich hinzulegen versuchen. the soup should the Hans to eat and self down-to-lie try ‘Hans should try to eat the soup and lie down.’ (Schwarz 1998: 208) b.

Die Suppe1 soll2 der Hans [[[zu t1 essen und sich hinzulegen] versuchen]t2]

In contrast to (30), the V-to-C movement does not take place in an ATB way in (32a). As illustrated in (32b), soll ‘should’ is base-generated outside of the coordinate structure and moves to C. The complementizer in (32) cannot be ignored for the purpose of evaluating whether the movement dependency between the complementizer and die Suppe ‘the soup’ satisfies the CSC, because of the absence of ATB-movement. The example is excluded in terms of the CSC violation induced by the movement of die Suppe. One might say that the movement of soll to C vacuously satisfies the CSC, but it is naturally assumed that the computational system does not take into consideration any movement outside of a coordinate structure for evaluating whether the movement in question satisfies the CSC. The ungrammaticality of (8), repeated as (33a), confirms the claim that the presence of ATB-movement is needed to save a CSC violation. (33) a. *Die Suppe wird der Hans essen und wird der Peter sich hinlegen. the soup will the Hans eat and will the Peter self down-lie ‘Hans will eat the soup and Peter will lie down.’ (Schwarz 1998: 193) b.

Die Suppe1 [[C′ wird2 der Hans t1 essen t2]

und [C′ wird3 der Peter sich hinlegen t3]]

The V-to-C movement in (33a) takes place within each conjunct, not in an ATB way, as illustrated in (33b). (33a) is simply excluded in terms of the CSC. (34a), which involves a subject in the second conjunct, also falls under the proposed analysis.4 (34) a. Die Suppe wird der Hans essen und der Peter sich hinlegen. the soup will the Hans eat and the Peter self down-lie ‘Hans will eat the soup and Peter will lie down.’ (Schwarz 1998: 207)


Die Suppe2 wird1 [[der Hans t2 essen t1] und [der Peter sich hinlegen t1]]

4.  An anonymous reviewer points out that (34a) is ungrammatical to the reviewer. This speaker variation is left for future research.

Reconsidering odd coordination in German 

As shown in (34b), wird ‘will’ undergoes ATB-movement, which saves a CSC violation induced by the movement of die Suppe ‘the soup’ out of the first conjunct.5 Let us turn to the issue of the interaction between negation and conjunction. As reviewed in the last section, (35) is ambiguous with respect to scope interaction between negation and conjunction. Recall that the interpretation where negation takes wide scope over conjunction is problematic for Schwarz’s approach. (35) Den Hund hat sie nicht gefüttert und ihn geschlagen. (neg> &, &>neg) the dog has she not fed and it beaten ‘She did not feed the dog and beat it.’ (Schwarz 1998: 212)

Under the proposed analysis, neither interpretation is problematic for the proposed analysis. For the wide scope reading of negation, (35) has the structure given in (36a), where negation is base-generated outside of the coordinate structure. On the other hand, for the wide scope of conjunction, nicht ‘not’ is embedded in the first conjunct, as shown in (36b). (36) a. Den Hund1 hat2 sie nicht [[[t1 gefüttert] t2] und [[ihn geschlagen] t2]] b. Den Hund1 hat2 sie [[[nicht t1 gefüttert] t2] und [[ihn geschlagen] t2]]

In either of them, den Hund moves out of the first conjunct, which is made possible by the existence of the ATB-movement of hat. The proposed analysis defends the position that head movement takes place in the narrow syntax, not at PF, contrary to Boeckx and Stjepanović (1999), Chomsky (2000) and Harley (2005). As has been argued so far, in the case of verbal odd coordination, the presence of ATB head movement makes an ill-formed dependency in

5.  The following ungrammatical example is problematic for the proposed analysis:


*Who1 did2 [TP Susan t2 hit t1] and [TP Mary t2 kick his friend]?

The presence of ATB movement of T should save the CSC violation induced by the whmovement. Under the proposed analysis, it is not clear why T-to-C movement is not helpful in the relevant sense, in contrast to V-to-C movement in verbal odd coordination. It is speculated that Subject-Aux inversion in English is triggered not by C but by some functional category between C and T. This functional category is tentatively referred to as X. X is available only in the matrix clause because Subject-Aux inversion does not take place in the embedded clause in the language. If this speculation is correct, did occupies the head of XP, not CP, as shown in (ii).


*[CP Who1 C [XP did2 [TP Susan t2 hit t1] and [TP Mary t2 kick his friend]]]?

The ATB movement of did enables the computational system to ignore X with respect to the CSC, but C cannot be ignored. Thus, the CSC violation induced by who cannot be saved. Independent evidence for the existence of X is needed but the issue is left for future research.

 Hironobu Kasai

isolation ignorable to the computational system. To the extent that the PMC is a principle of the narrow syntax, head movement is supposed to be an operation of the narrow syntax. Otherwise, the relevant ill-formed dependency could not be saved. On independent grounds, Lechner (2006) empirically argues that some instances of head movement affect semantic interpretation, which indicates that they take place in the narrow syntax. Before moving to the next section, let us consider some asymmetry between the first conjunct and the second conjunct, which is problematic for the proposed analysis. The ungrammaticality of (37a) cannot be expected under the proposed analysis so far. Since the example involves the ATB-movement of wird ‘will’, nothing prevents extraction out of the second conjunct. However, the movement of die Suppe ‘the soup’ leads to the ungrammaticality of (37a). (37) a. *Die Suppe wird der Peter sich hinlegen und essen. the soup will the Peter self down-lie and eat ‘Peter will lie down and eat the soup.’ (Schwarz 1998: 210)


Die Suppe1 wird2 der Peter [[[sich hinlegen]t2] und [[t1 essen]t2]]

Schwarz’s analysis explains the ungrammaticality of (37a) successfully, on the other hand. His analysis simply reduces the ungrammaticality to a violation of the CSC, as illustrated in (38).

(38) Die Suppe1 wird der Peter [[sich hinlegen] und [t1 essen]]

This paper assumes with Chomsky (2000) that Move is decomposed into Agree and Merge and that Agree is a prerequisite of Move. In order for die Suppe ‘the soup’ to move to [Spec, CP] in (37), C has to undergo Agree with die Suppe in the first place. The solution presented in this paper is that a constituent inside the second conjunct is prevented from undergoing Agree with a probe outside of the coordinate structure somehow. Thus, die Suppe fails to undergo Agree with C in (37), which leads to the immovability of it. There is an independent piece of evidence for the opaqueness of the second conjunct with respect to the applicability of Agree. Let us consider the contrast between (39a) and (39b). (39) a. There was a man in the bathroom and two cats in the kitchen. b. *There were a man in the bathroom and two cats in the kitchen. (Niinuma & Park 2004: 353)

The auxiliary verb exhibits agreement with the associate NP in the first conjunct in (39a), not in the second conjunct in (39b). The ungrammaticality of (39b) suggests that T fails to undergo Agree with two cats, which is in the second conjunct (see Niinuma & Park 2004 for relevant discussion). Under the proposed analysis, the ungrammaticality of

Reconsidering odd coordination in German 

(37a) and (39b) falls under the generalization that the second conjunct is not accessible with respect to Agree.

3.3  Gapping and CSC Johnson (1996) argues that gapping of a verb is derived from ATB-movement of the verb, not from ellipsis. Thus, (40a) is analyzed in (40b) under his analysis. (40) a. Some ate a green banana and others a ripe one. b. [TP some1 [T + eat2 [[vP t1 t2 a green banana] and [vP others t2 a ripe one]]]]

In addition to the ATB-movement of eat to T, some also moves to [Spec, TP], which violates the CSC. In contrast to some, others, which is a subject of the second conjunct, remains in-situ. The movement of the subject out of the first conjunct, violating the CSC, is empirically supported on the basis of the following contrast: (41) a. No boy1 joined the navy and his1 mother the army. (Johnson 2003: 19) b. *No boy1 joined the navy and his1 mother joined the army.

The availability of variable binding in (41a) suggests that no boy c-commands the second conjunct. Without gapping, the variable binding is unavailable, as shown in (41b). The contrast above is readily captured, on the assumption that the subject moves out of the first conjunct under gapping. Under the proposed analysis, the relevant CSC violation under gapping is expected because gapping involves ATB-movement of V to T. It would be difficult to claim that A-movement is not subject to the CSC at all. If the CSC were freely violable with A-movement, the following ungrammatical example would be grammatical, contrary to fact: (42) *John likes coffee and Mary like tea.

(42) would have the following derivation, where John just undergoes A-movement out of the first conjunct and the tense morpheme is attached to the verb like:

(43) [TP John1 [T [[vP t1 like coffee] and [vP Mary like tea]]]]

In (43) there is no ATB-movement of V, in contrast to (40). The ungrammaticality of (42) suggests that the CSC does constrain A-movement. Under the present analysis, the contrast between (40a) and (42) with respect to the CSC is correctly captured because ATB-movement takes place only in the former.

3.4  Summary To sum up, in this section, it has been proposed that the CSC is subject to the PMC. The presence of ATB-movement plays a crucial role for saving a violation of the CSC.

 Hironobu Kasai

This is what is involved in verbal odd coordination. The proposed analysis is compatible with the claim that head movement takes place in the narrow syntax, contrary to Boeckx and Stjepanović (1999), Chomsky (2000) and Harley (2005). Finally, the question is also addressed as to why the CSC is violable under gapping.

4.  On the theoretical status of the CSC The discussion so far has been based on the assumption that the CSC is operative as a syntactic constraint on movement. The theoretical status of the CSC has been controversial, however. Several researchers have argued that the CSC should not be stated as a constraint on movement operations themselves, but rather as a constraint on the result of movement operations, that is, representations (Goodall 1987; Muadz 1991; Moltmann 1992; Munn 1993; Ruys 1992; Fox 2000; Lin 2001, 2002; Kato 2006). Under the latter view, the derivational step itself does not lead to any ungrammaticality. Following Kato (2006), this paper refers to the former approach to the CSC as the derivational approach to the CSC and the latter approach as the representational approach to the CSC. The aim of this section is to show that the facts on verbal odd coordination in German support the derivational approach to the CSC and argue against the representational approach to the CSC. However, given that some pieces of evidence favor the representational approach to the CSC over the derivational approach to the CSC, this paper will suggest that both of the approaches are correct. To make the following discussion explicit, this paper adopts a version of the representational approach to the CSC, presented by Fox (2000). Under the representational approach to the CSC, the CSC is not an independent constraint but the CSC effects are due to (44). (44) a. Extraction out of a coordinate structure is possible only when the structure consists of two independent substructures, each composed of one of the coordinates together with material above it up to the landing site (henceforth, component structures). b. Grammatical constraints are checked independently in each of the component structures. (Fox 2000: 50)

Given (44), (45) has the component structures given in (46). (45) *Who1 do you think Mary likes t1 and Bill hates Sue? (46) a. Who1 do you think Mary likes t1 b. Who1 do you think Bill hates Sue

While there is nothing wrong with (46a), (46b) induces a violation of the Theta Criterion because who does not receive any theta-role. In addition, the ban on vacuous

Reconsidering odd coordination in German 

quantification is also violated because who has nothing to bind. In contrast, (47) has the component structures given in (48). (47) Who1 do you think Mary likes t1 and Bill hates t1? (48) a. Who1 do you think Mary likes t1 b. Who1 do you think Bill hates t1

The ATB-movement in (47) has no problem because each component structure satisfies the Theta Criterion and the ban on vacuous quantification. There are pieces of evidence in favor of the representational approach to the CSC. Let us consider (49a), first. (49) a. A (#different) student likes every professor and hates the dean. (a > every,*every > a) (Fox 2000: 51) [every professor1 [a (different) student [likes t1] and [hates the dean]]]


In (49a), every professor cannot take wide scope over the subject. In order to take wide scope over the subject, every professor must undergo QR, as illustrated in (49b). The absence of the relevant wide scope reading indicates that QR exhibits the CSC effect (Rodman 1976: 171, May 1985: 59, and Ruys 1992). Interestingly, Ruys (1992) observes that if the second conjunct involves a pronoun to be bound by a quantifier in the first conjunct, the object quantifier can take wide scope over the subject QP. The relevant example is given in (50). (50) A (different) student [[likes every professor1] and [wants him1 to be on his committee]]. (a > every, every> a) (ibid.: 52)

For the wide scope interpretation of every professor, it moves out of the first conjunct, in the following way: (51)

[every professor1 [a (different) student [[likes t1] and [wants him1 to be on his


The contrast between (49a) and (50) is captured under the representational approach to the CSC. For the wide scope interpretation of every professor, (49a) and (50) have the following component structures, respectively: (52) a. every professor1 a (different) student likes t1 b. every professor1 a (different) student hates the dean (53) a. every professor1 a (different) student [likes t1] b. every professor1 a (different) student [wants him1 to be on his committee]

In (52a), every professor binds its trace in the first conjunct, but there is nothing to be bound by every professor in the second conjunct, which violates the ban on vacuous

 Hironobu Kasai

quantification. In contrast, (53b) does not have this problem because every professor binds the bound pronoun him in the second conjunct. The contrast fails to be captured under the view that the CSC is a constraint on movement, on the other hand. The wide scope interpretation of every professor is unexpected in (50) because it moves out of the first conjunct, violating the CSC. One might suggest that the CSC does not constrain covert movement in order to explain the wide scope reading of (50). However, this suggestion is unworkable because it fails to explain the absence of the wide scope reading in (49a). Another argument for the representational approach to the CSC comes from some obligatory reconstruction effect under gapping, discussed by Lin (2001, 2002). Recall that under Johnson’s analysis of gapping, the subject of the first conjunct, some in (40), repeated as (54), moves out of the first conjunct. (54) a. Some ate a green banana and others a ripe one. b. [TP some1 [T + eat2 [[vP t1 t2 a green banana] and [vP others t2 a ripe one]]]]

Lin (2001, 2002) raises a question as to how the gapping example such as (54) satisfies grammatical constraints at each of the component structures under (44). Suppose (54a) would have the following component structures: (55) a. some1 T+eat2 [t1 t2 a green banana] b. some1 T+eat2 [others t2 a ripe one]

(55b) does not satisfy the Theta Criterion because some cannot receive any theta role. Lin suggests that the subject of the first conjunct undergoes reconstruction in such a way that the Theta Criterion can be satisfied. If some undergoes reconstruction at LF, the component structures of (54a) would be (56), rather than (55).6 (56) a. T+eat2 [some t2 a green banana] b. T+eat2 [others t2 a ripe one]

Lin (2001, 2002) provides independent pieces of evidence in favor of the obligatory reconstruction of the subject. Let us consider the interpretation of (57), first.

6.  The subject of the first conjunct under gapping does not necessarily undergo reconstruction. No boy is not required to undergo reconstruction in (41a), where there is a bound pronoun to be bound by no boy in the second conjunct. The component structures of (41a) are given below: (i)

a. b.

No boy1 t1 joined the navy No boy1 his1 mother joined the army

If his is interpreted as a kind of resumptive pronoun and incorporated into a chain with no boy, (ib) is correctly interpreted without appealing to reconstruction.

Reconsidering odd coordination in German 

(57) Many drummers can’t leave on Friday. (many>neg, neg>many) (Lin 2002: 72)

Lin (2002) provides the following scenario: (58) We are in charge of a music camp, and we have to worry about how many different types of musicians we will have in the camp at any one time. This entails knowing about when different groups of musicians will leave and arrive (ibid.: 72).

She observes that (57) is ambiguous with respect to the scope interaction between many and negation. When many takes wide scope, (57) can be paraphrased as:  for many different individual drummers, it’s the case that they are unable to leave on Friday. When many takes narrow scope, (57) can be roughly paraphrased as: it is not allowed that a large group of drummers leave on Friday. In contrast to (57), many has to take narrow scope in (59), which involves gapping of can’t. (59) Many drummers can’t leave on Friday, and many guitarists arrive on Saturday. (*many>neg, neg>many) (ibid.: 72)

(59) is paraphrased as: it can’t be the case that many drummers leave on Friday and many guitarists arrive on Sunday. The absence of the wide scope reading of many indicates that the subject has to undergo reconstruction, as illustrated in (60). (60)

[TP Many drummers1 can’t [&P [vP t1 leave on Friday] obligatory reconstruction

and [vP many guitarists arrive on Saturday]]]

This obligatory reconstruction effect is totally unexpected under the derivational approach to the CSC, in contrast to the representational approach to the CSC. However, it will be shown that the facts on verbal odd coordination in German cannot be captured under the representational approach to the CSC.7 It is difficult to capture the contrast between (30) and (32a), which are repeated below, respectively: (61) Die Suppe2 wird1 der Hans [[[t2 essen] t1] und [[sich hinlegen]t1]]. the soup will the Hans eat and self down-lie ‘Hans will eat the soup and lie down.’ (Schwarz 1998: 191) (62) *Die the

Suppe1 soll2 der Hans [[[zu t1 essen und sich soup should the Hans to eat and self

hinzulegen] versuchen]t2]. down-to-lie try ‘Hans should try to eat the soup and lie down.’ (Schwarz 1998: 208)

7.  See also Kato (2005) for arguments against the representational CSC based on Japanese data.

 Hironobu Kasai

On the assumption that die Suppe undergoes reconstruction to the base-position at LF in such a way that the Theta Criterion can be satisfied, similarly to the derivation of gapping, (61) and (62) would have the component structures given in (63) and (64), respectively. (63) a. wird1 der Hans die Suppe essen t1 b. wird1 der Hans sich hinlegen t1 (64) a. soll2 der Hans zu die Suppe essen versuchen t2 b. soll2 der Hans sich hinzulegen versuchen t2

There is nothing wrong with (63) and (64) with respect to the Theta Criterion. However, (62) is ungrammatical. Thus, the contrast between (61) and (62) is problematic for the representational approach to the CSC.8 On the other hand, the contrast can be captured under the derivational approach to the CSC, together with the proposal that the CSC obeys the PMC, as discussed in section 3. Summing up the discussion so far, verbal odd coordination defends the CSC as a constraint on overt movement.9 On the other hand, the representational approach to the CSC is also motivated on empirical grounds, as has been reviewed in this section. This paper suggests that both of the approaches are correct, which is not implausible because both of them do not exclude each other in principle.

5.  Concluding remarks This paper has reconsidered Schwarz’s (1998) analysis of odd coordination in German. His analysis of nominal odd coordination is successful, but there are problems with his treatment of verbal odd coordination. The upshot of the proposal is that the CSC obeys

8.  te Velde (2006: 277–278) presents an alternative analysis of (61), but his analysis is not clear to me, unfortunately. As far as I understand correctly, under his alternative derivation, (61) involves vP-coordination and die Suppe ‘the soup’ moves out of the first conjunct, violating the CSC, as shown in (i).

(i) Die Suppe1 wird der Hans [[vP t1 essen] und [vP sich hinlegen]]

He assumes that “the CSC is analyzed not as an independent condition or constraint, but rather as an epiphenomenon of underlying principles. These principles apply to syntactic requirements of Case, Agree, Move and Merge (te Velde 2005: 271).” It seems that he assumes a sort of the representational approach to the CSC. However, even under his analysis, it is not clear how to capture the contrast between (61) and (62). 9.  Recall that the wide scope reading of every professor in (50) indicates that the CSC does not constrain covert movement.

Reconsidering odd coordination in German 

the PMC. The presence of ATB-movement allows the computational system to ignore a violation of the CSC. This is indeed what is involved in verbal odd coordination. As argued by Johnson (1996), the CSC is violable with the movement of the subject out of the first conjunct under gapping. The proposed analysis gives a natural explanation for it. Gapping involves V-to-T movement in an ATB way, which nullifies the CSC violation induced by the movement of the subject out of the first conjunct. There are two theoretical implications of the analysis. One is that head movement takes place in the narrow syntax, not at PF, contrary to Boeckx and Stjepanović (1999), Chomsky (2000), and Harley (2005). The other one concerns the theoretical status of the CSC, which has been controversial in the literature. The position of this paper is that the CSC should be defended as the constraint on overt movement while each of the conjuncts needs to be evaluated with respect to grammatical constraints.

References Bachrach, A. & Katzir, R. 2007. Right-Node Raising and Delayed Spell-Out. Ms., MIT. Boeckx, C. & Stjepanović, S. 1999. Head-ing toward PF. Linguistic Inquiry 32: 345–355. Büring, D. & Hartmann, K. 1998. Asymmetrische Koordination. Linguistische Berichte 174: 172–201. Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik. R. Martin, D. Michaels, & J. Uriagereka (Eds), 89–155. Cambridge MA: The MIT Press. Citko, B. 2005. On the nature of Merge: External Merge, Internal Merge, and Parallel Merge. Linguistic Inquiry 36: 475–496. Fox, D. 2000. Economy and Semantic Interpretation. Cambridge MA: The MIT Press. Goodall, G. 1987. Parallel Structures in Syntax. Cambridge: CUP. Hallman, P. 2004. On the derivation of verb-final and its relation to verb-second. Ms, McGill University. Harley, H. 2005. Merge, conflation and head movement: The First Sister Principle revisited. In Proceedings of NELS 34, K. Moulton & M. Wolf (Eds), 239–254. Amherst MA: GLSA. Heycock, C. & Kroch, A. 1994. Verb movement and coordination in a dynamic theory of licensing. The Linguistic Review 11: 257–283. Höhle, T. 1990. Assumptions about asymmetric coordination in German. In Grammar in Progress. GLOW Essays for H. van Riemsdijk, F.J. Mascaró & M. Nespor (Eds), 221–235. Dordrecht: Foris. Hornstein, N. 1995. Logical Form: From GB to Minimalism. Cambridge MA: Blackwell. Hornstein, N. & Nunes, J. 2002. On asymmetries between parasitic gap and across-the-board constructions. Syntax 5: 26–54. Johnson, K. 1996/2003. In search of the English middle field. Ms, University of Massachusetts, Amherst. Johnson, K. 2002. Restoring exotic coordinations to normalcy. Linguistic Inquiry 33: 97–156. Kato, T. 2005. A case against the representational approach to the Coordinate Structure Constraint. In Proceedings of NELS 35, L. Bateman & C. Ussery (Eds), 307–321. Amherst MA: GLSA.

 Hironobu Kasai Kato, T. 2006. Symmetry in Coordination. Ph.D. dissertation, Harvard University. Lechner, W. 2006. An interpretive effect of head movement. In Phases of Interpretation, M. Frascarelli (Ed.), 45–70. Berlin: Mouton de Gruyter. Lin, V. 2001. A way to undo A-movement. In Proceedings of WCCFL 20, K. Megerdoomian & L.A. Barel (Eds), 358–371. Somerville MA: Cascadilla Press. Lin, V. 2002. Coordination and Sharing at the Interfaces, Ph.D. dissertation, MIT. May, R. 1985. Logical Form: Its Structure and Derivation. Cambridge MA: The MIT Press. Moltmann, F. 1992. Coordination and Comparatives. Ph.D. dissertation, MIT. Muadz, H. 1991. A Planar Theory of Coordination. Ph.D. dissertation, University of Arizona. Munn, A. 1992. A null operator analysis of ATB gaps. The Linguistic Review 9: 1–26. Munn, A. 1993. Topics in the Syntax and Semantics of Coordinate Structures. Ph.D. dissertation, University of Maryland, College Park. Niinuma, F. & Park, M.-K. 2004. Moved elements undergo Agree after Move. In Generative Grammar in a Broader Perspective, Hang-Jin Yoon (Ed.), 351–368. Seoul: Hankook. Partee, B. & Rooth, M. 1983. Generalized conjunction and type ambiguity. In Meaning, Use and Interpretation of Language, R. Bäuerle, C. Schwarze & A. von Stechow (Eds), 361–383. Berlin: Walter de Gruyter. Richards, N. 1998. The principle of minimal compliance. Linguistic Inquiry 29: 599–629. Rodman, R. 1976. Scope phenomena, ‘movement transformations’, and relative clauses. In Montague Grammar, B.H. Partee (Ed.), 165–176. New York NY: Academic Press. Ross, J.R. 1967. Constraints on Variables in Syntax. Ph.D. dissertation, MIT. Ross, J.R. 1986. Infinite Syntax! Norwood NJ: Ablex. Ruys, E. 1992. The Scope of Indefinites. Ph.D. dissertation, University of Utrecht. Schwarz, B. 1998. Odd coordination. Journal of Comparative Germanic Linguistics 2: 191–219. te Velde, J.R. 2006. Deriving Coordinate Symmetries: A Phase-Based Approach Integrating Select, Merge, Copy and Match. Amsterdam: John Benjamins. Williams, E. 1978. Across-the-board rule application. Linguistic Inquiry 9: 31–43. Zwart, J.-W. 1991. Subject deletion in Dutch:  A difference between subjects and topics. In Language and Cognition 1, M.  Kas, E.  Reuland & C.  Vet (Eds), 333–349. Groningen: University of Groningen.

The syntax and semantics of the temporal anaphor “then” in Old and Middle English Carola Trips & Eric Fuß

Universities of Mannheim and Frankfurt The fact that þa/þonne ‘then’ trigger V2 in OE is commonly accounted for by assuming that these adverbs are operators that trigger V-to-C movement. This paper presents an alternative analysis based on the observation that þa/þonne and pronouns are in complementary distribution in preverbal position. We identify this position as SpecTP, arguing that OE was a discourseconfigurational language where SpecTP was linked to the discourse anchoring of anaphoric/deictic expressions, including pronouns and temporal anaphora such as þa/þonne. Under these assumptions, V2 with these temporal adverbs results from a spec-head relationship in TP. The loss of V2 in the ME period is then attributed to the independent development of a (subject-oriented) EPP-feature in TP and the overall loss of discourse-configurationality.

1.  Introduction The early Germanic languages exhibit a class of temporal adverbs that originated from former demonstratives: Gothic (G) þan(uh), Old English (OE) þa, þonne, Old High German (OHG) thô, thanne, Old Saxon (OS) tha, thanna, all roughly meaning ‘then’. These adverbs often show a peculiar syntactic behaviour that sets them apart from other adverbs:  (i) In OE, þa and þonne consistently trigger inversion (examples (1) and (2)); (ii) they can assume the role of conjunctions in all early Germanic languages (example (3)):1


þa for he norþryhte be þæm lande; then went he northwards to that land ‘Then he went northwards to that land.’ (Orosius:

1.  The diachronic data cited are from the following corpora:  the York-Toronto-Helsinki Parsed Corpus of Old English Prose (Taylor et al. 2003) and the Penn-Helsinki Parsed Corpus of Middle English 2 (Kroch & Taylor 2000).

 Carola Trips & Eric Fuß

þonne ærnað hy ealle toweard þæm feo; then run-to they all towards the treasure ‘Then they all ran towards the treasure.’ (Orosius:


(3) Ða se wisdom þa ðis spell asæd hæfde, þa ongan he when the wisdom then this story said had then began he

eft sing again singan

‘When wisdom then had told this story, he began to sing again.’ (Fischer et al.: 2000, 57)

If we consider the etymology of these elements, we find that all early Germanic forms once were demonstratives, so OE ða, þa (ModE then, when) used to be a case-form (accusative singular) of the demonstrative stem þa- ‘the, that’ (Proto-Germanic demonstrative *TO-). The OE forms þanne (þonne), þænne, þenne that occur in ME as þenne, þan, þen (OHG danne, denne, MHG danne, denne, Modern German (ModG) dann, Gothic (G) þan) derive from the demonstrative that acquired the nasal suffix -n (see below for the contrast between þa and þonne). This also means that both the ModE adverb then and the conjunction than originate in the same word, both functions of which varied in ME2 and the 16th century between the forms þenne/then and þan/than (ModG has dann (adverb) ‘then’, denn (conjunction) ‘than’ (from the Old English Dictionary (DOE), 2002). According to Ramat (1981) these elements have always had an anaphoric and a deictic function in relation to something previously mentioned. This property bears on the syntactic behaviour of OE þa/þonne and the changes that affected the distribution of these elements in the Middle English (ME) period. This paper presents a new approach to their syntactic behaviour which is based on the following assumptions: first, we assume that OE was a discourse-configurational language which we mean to imply that the word order was determined by discourse-related factors such as anaphoricity, or the distinction between old/new information (in contrast to ME and ModE where it primarily serves to discriminate syntactic functions, cf. e.g., Fischer et al. 2000; van Kemenade & Los 2006); second, since clause-initial þa/þonne have an anaphoric and a deictic function, they compete with subject pronouns for the same structural position if they occur in clause-initial position, which is linked to the discourse anchoring of anaphoric expressions in OE (and which we identify as SpecTP). Before we elaborate on our analysis in section 3, in the next section we will briefly deal with clauses with clause-initial þa/þonne in relation to verb and subject placement, and a number of analyses that tried to account for the syntactic properties of these clauses. 2.  It can be assumed that the function of then as conjunction developed during a regular grammaticalization process from adverb to conjunction.

The syntax and semantics of the temporal anaphor “then” in Old and Middle English 

2.  þa, þonne and Verb Second in Old English It is a well-known fact that OE main clauses exhibit word order patterns reminiscent of the Modern Germanic Verb Second languages, i.e., the finite verb occupies the second position after a fronted XP, leading to subject-verb inversion (examples taken from Trips (2002: 231)): Object-Vfin-subject


[Þæt hus] hæfdon Romane to ðæm anum tacne geworht … that house had Romans to the one sign made ‘The Romans had made that house to their sole sign.’ (Or_3:



[On þysse dune ufanweardre] bæd Sanctus Albanus fram Gode … on this hill higher up bade Saint Alban from God ‘On this hill higher up Saint Alban asked from God …’ (Bede_1:



[Uneaðe] mæg mon to geleafsuman gesecgan … Hardly may man to faithful speak ‘Hardly may man speak to the faithful …’ (Or_3:

In this type of clause, subject pronouns regularly intervene between the clause-initial XP and the finite verb, giving rise to Verb Third order:


[Æfter þysum worde] he wearð eall gehæled. after these words he was all healed ‘After these words, he was all healed.’ (ÆLS_[Sebastian]:299.1391)

However, subject-verb inversion is obligatory with both pronominal and nominal DP subjects if the fronted element is an operator such as a wh-phrase as in (8) or the negation ne as in (9): (8) a. Hwæt sculon we þæs nu ma secgan? what shall we afterwards now more say ‘What shall we afterwards say now more?’ (Bede_2:

b. hu wurð he elles gelæred? how was he otherwise taught ‘How was he taught otherwise?’ (BedePref:2.11.153)

 Carola Trips & Eric Fuß

(9) a. ne bið he lengra þonne syfan elna lang. neg is he longer than seven ells long ‘He is not taller than seven ells.’ (Orosius: b. Ne meaht þu deman Gallia biscopas buton heora agenre neg might you judge Gaul’s bishops but their own

aldorlicnesse, … authority

‘You might not judge the Gaul’s bishops but their own authority.’ (Bede_1:

Moreover, subject-verb inversion is obligatory with all kinds of subjects in cases where the clause-initial position is occupied by the temporal adverbs þa/þonne ‘then’ (cf. Mitchell 1985; van Kemenade 1987; Kroch & Taylor 1997; Pintzuk 1999) 3: (10) Þa for he norþryhte be þæm lande; then went he northwards to that land ‘Then he went northwards to that land.’ (Orosius: (11) Þonne ærnað hy ealle toweard þæm feo; then run-to they all towards the treasure ‘Then they all run towards the treasure.’ (Orosius:

A wide-spread analysis of these word order facts is that the deviating behaviour of subject pronouns can be explained if it is assumed that they occupy a fixed position at the left edge of IP, preceded by fronted topics (cf. e.g., Cardinaletti & Roberts  2002, Pintzuk 1993, 1999; Hulk & van Kemenade 1995; van Kemenade 1997; van Kemenade 1999; Kroch & Taylor 1997; Haeberli 1999; Haeberli 2002; Fischer et al. 2000). In contrast, full DP subjects occupy a lower, presumably VPinternal, position which explains why in standard Verb Second clauses the finite verb precedes this type of subject (in more technical terms, Infl/T does not host an EPP feature in OE). The finite verb occupies a head position in the IP domain and

3.  Verb Second can also be observed with other temporal adverbs such as nu ‘now’: (i)

Nu hæbbe we ymb Affrica landgemæro gesæd. now have we about Africa’s boundary said “Now we have spoken about Africa’s boundary.” (Orosius,:

However, Verb Second is much less regular with nu than with þa/þonne (cf.  Mitchell & Robinson 2003). Therefore, this paper focuses on Verb Second patterns with þa/þonne.

The syntax and semantics of the temporal anaphor “then” in Old and Middle English 

moves further to C only in operator contexts (leading to obligatory subject-verb inversion). The analysis of the word order facts presented with the examples above is illustrated with the examples in (12) to (14): Verb Third with pronominal subjects (12) [CP Æfter þysum worde [IP he [I′ wearð [VP eall gehæled]]]]. after these words he was all healed ‘After these words, he was all healed.’ (ÆLS_[Sebastian]: 299.1391)

Verb Second with full DP subjects (13) [CP Þæt hus [IP Ø [I′ hæfdon [VP Romane to ðæm anum tacne geworht]]]] that house had Romans to the one sign made ‘The Romans had made that house to their sole sign.’ (Or_3:

Obligatory inversion with fronted operators (14) [CP hu [C′ wurði+C0[IP he [I′ t′i [VP elles gelæred ti ]]]]]? how was he otherwise taught ‘How was he taught otherwise?’ (BedePref:2.11.153)

The problem with this analysis is, however, that there is no satisfying and convincing explanation why inversion is obligatorily triggered in contexts where þa/þonne occur in clause-initial position. The traditional analysis claims that þa/þonne are syntactic operators, on a par with wh-phrases, negation etc. which all trigger verb movement to C0 (cf. e.g., van Kemenade 1987): (15) [CP þa/þonne [C′ Vfin [IP pron. [I′ tV [vP … ]]]]

The severe problem that arises here is that the lack of Verb Second effects with then in ModE cannot be explained at all. Although fronted operators such as wh-phrases and negation continue to trigger inversion in Modern English, then, the present-day equivalent of OE þa/þonne fails to do so:4 (16) a. *Then will Harry/he read that book. b. Then Harry/he will read that book.

4.  An anonymous reviewer raises the objection that this observation does not constitute a conclusive argument in favour of the claim that V2 with þa/þonne must be analysed in a way that substantially differs from the analysis of V2 with wh-phrases and negatives. He/She points out that the elements that trigger inversion in ModE hardly form a natural class, and that there is evidence from other Germanic languages that exceptions to V2 can be of a very specific

 Carola Trips & Eric Fuß

What van Kemenade and Los (2006) propose is that the clause-initial þa/þonne is a discourse operator (located in SpecCP) signalling discourse continuity and requiring the finite verb to occupy C. What remains completely unclear, however, is their notion of “discourse operator” and concomitantly the trigger for verb movement to C0. In the next section, we will show that with our analysis we can account for the hitherto unexplained facts.

3.  An alternative approach 3.1  The syntax-discourse interface and the structure of OE To continue a given discourse in a coherent way, a set of conditions concerning the syntax-discourse interface must be met. This is particularly obvious with respect to clausal typing, information-structural distinctions, and the interpretation of anaphoric expressions. For example, sentential mood must be coded in order to distinguish between questions and assertions, information-structural differences such as ‘topic’ and ‘focus’ must be properly marked (either via word order, or by assigning certain stress patterns), and anaphoric expressions must be anchored in the discourse to warrant a correct interpretation (e.g., pronouns must receive a referential index). Following proposals by Rizzi (1997), the first and second of these properties are directly implemented into the structure of the clause. More specifically, clausal typing is associated with properties of Force, which closes off the series of the projections in a split-CP and represents the interface to the discourse context (or a matrix clause), while the encoding of information-structural distinctions such as topic and focus is linked to specifier positions made available by the relevant functional heads in the left periphery of the clause (Topic and Focus, respectively). Of course, languages may differ with respect to the extent

type, in the sense that individual lexical items may fail to trigger V2 (cf. Westergaard 2007 on Norwegian). Thus, what he/she implies with this comment is that the ability to trigger inversion (i.e., V-to-C movement) reduces to an idiosyncratic lexical property of individual lexical items fronted to clause-initial position. Accordingly, both the ability of þa/þonne to trigger inversion in OE and the loss of that property in later historical stages would be completely accidental facts (lexical idiosyncrasies, in fact) in the history of English, which elude any systematic explanation (apart from stating that ‘then’ lost its relevant lexical feature at some point in the history of English). However, this view is at odds with the observation (see section 4 below) that there are chronological parallels between the loss of V2 in the context of þa/þonne and the rise of a structural subject position in the Middle English period. This suggests that the loss of the former property is not just an accidental fact (in terms of the loss of a lexical feature), calling for a formal account that establishes a link between these two changes.

The syntax and semantics of the temporal anaphor “then” in Old and Middle English 

they encode discourse properties by syntactic means such as word order (i.e., via overt movement to positions such as SpecTopP or SpecFocP). Languages where word order does not primarily serve to identify grammatical functions, but is rather used to signal the information-structural status of different elements of the clause are often called ‘discourse-configurational languages’. In the current literature on OE syntax, it is often pointed out that certain characteristics of OE can be taken to indicate that it was discourse-configurational as well (cf. Fischer et al. 2000; van Kemenade 2002; van Kemenade & Milicev 2005; van Kemenade & Los 2006). The basic observation is that topical material referring anaphorically to discourse referents figuring prominently in the discourse contexts (in particular personal pronouns, but also definite/specific nominal expressions marked e.g., by a demonstrative) occupies a position at the left edge of the inflectional domain (i.e., directly to the right of the complementiser in embedded clauses, or adjacent to a fronted finite verb in main clauses). In contrast, non-definite/non-specific noun phrases, which typically represent the focus of the clause (or, more generally, new information), occupy a lower position, for example directly to the left of non-finite verbs, quite similar to e.g., ModG. Van Kemenade & Los illustrate the syntactic effects of discourse-configurationality in OE with the following example in which the object pronoun (representing old information) is situated to the left of the temporal adverb þonne, while the non-definite subject NP (representing new information) occupies a lower position to the right of þonne: (17) Gif hine þonne [yfel mon] hæfð […] if him then  evil man has ‘If an evil man has him …’ (Bo:; Kemenade & Los 2006: 237f.)

In what follows, it is claimed that in OE, not only clausal typing and the distinction between old and new information were linked to certain structural positions/configurations in the syntax, but also the discourse anchoring of anaphoric expressions. Before we turn to the specifics of our proposal, let’s have a closer look at the temporal interpretation of ‘then’, and how these considerations carry over to the analysis of OE þa/þonne.

3.2  The temporal interpretation of clause-initial þa/þonne In the literature on the semantics of ‘then’, it is often analysed as a temporal anaphor that introduces a temporal relation between the events described by two successive sentences. It has been claimed that the anaphoric character of ‘then’ requires that it be linked to an anchor time given in the discourse context (cf. Smith 1981; Schiffrin 1992;

 Carola Trips & Eric Fuß

Glasbey 1993; Thompson 1999).5 More precisely, ‘then’ is usually taken to express that the event or state described by the ‘then’-clause is temporally situated after a time point/interval given in the immediate discourse context. Furthermore, it has been observed that the temporal interpretation of ‘then’ is determined by its syntactic position. It can be observed that there are systematic differences concerning the interpretation of clause-initial and clause-medial/final ‘then’ (cf. e.g., Schiffrin 1992; Thompson 1999; Roßdeutscher 2005a,b, on ModE then, Roßdeutscher & von Stutterheim 2006 on German dann). This is illustrated with examples (18) and (19) from ModE. When occurring in clause-initial position, then is interpreted as a sentence adverb (presumably associated with IP) that gives rise to an interpretation where the events described by two successive sentences are understood as temporally ordered –  in (18), the speaking event occurs after the visiting event, and there is no temporal overlap between these events (sequential/ordered reading, henceforth sequential ‘then’): (18) Mary visited the exhibition. Then she spoke to the reporters.

In contrast, clause-final placement of then (presumably a VP-adverb, henceforth cotemporal ‘then’) leads to an interpretation where the event described by the ‘then’clause is taken to overlap with the event described by the previous clause – in (19), Mary spoke to the reporters while she was visiting the exhibition (note that despite the overlap, then still seems to indicate that the speaking event began somewhat after the visiting event): (19) Mary visited the exhibition. She spoke to the reporters then.

We adopt the idea that the different readings of ‘then’ are not to be attributed to different instances of ‘then’. Rather, we assume that the lexicon contains only a single temporal anaphor ‘then’, the different interpretations of which are determined by the structural position it occupies in the structure of the clause (cf. e.g., Thompson 1999; Roßdeutscher & von Stutterheim 2006). Thompson (1999) argues that the different interpretations of anaphoric ‘then’ result from linking different times in tense structure with the relevant times given in the immediate discourse context (making use of a Neo-Reichenbachian model of tense structure, cf. Hornstein 1990). According to Thompson, cotemporal ‘then’ is attached to VP and serves to link the Event time (by assumption associated with VP) of two consecutive clauses, giving rise to an interpretation where the relevant events overlap temporally. In contrast, clause-initial then is taken to be adjoined to IP, linking the

5.  In somewhat more formal terms, we can say that the interpretation of temporal anaphora requires the assignment of a temporal index given in the discourse.

The syntax and semantics of the temporal anaphor “then” in Old and Middle English 

Reference time (commonly associated with IP) of its clause with the Reference time of the previous clause, which entails that the relevant events take place one after the other. Leaving aside further technicalities and questions raised by this approach, what is important to keep in mind is that we share with Thompson the intuition that the sequential reading of ‘then’ becomes available when ‘then’ is associated with IP, while lower attachment leads to the cotemporal reading illustrated with (19) above.6 Now let’s come back to the question of how this carries over to the analysis of fronted temporal anaphora in OE. It has repeatedly been pointed out that in OE, clause-initial þa/þonne are typically used to mark a sequence of foregrounded successive actions/events that do not overlap temporally (cf. e.g., Foster 1975; Enkvist & Wårwik 1987; Wårwik 1995). Compare the Ohthere interpolation in Alfred’s Orosius which is a reproduction of oral narrative (simple narrative structure): He sæde þæt he æt sumum cirre wolde fandian hu longe þæt land norþryhte læge, oþþe hwæðer ænig mon be norðan þæm westenne bude. Þa for he norþryhte be þæm lande; let him ealne weg þæt weste land on ðæt steorbord & þa widsæ on ðæt bæcbord þrie dagas. Þa wæs he swa feor norþ swa þa hwælhuntan firrest faraþ. Þa for he þa giet norþryhte swa feor swa he meahte on þæm oþrum þrim dagum gesiglan. Þa beag þæt land þær eastryhte, oþþe seo sæ in on ðæt lond, he nysse wæðer buton he wisse ðæt he ðær bad westanwindes & hwon norþan & siglde ða east be lande swa swa he meahte on feower dagum gesiglan. Þa sceolde he ðær bidan ryhtnorþanwindes, for ðæm þæt land beag þær suþryhte, oþþe seo sæ in on ðæt land, he nysse hwæþer. Þa siglde he þonan suðryhte be lande swa swa he mehte on fif dagum gesiglan. Ða læg þær an micel ea up in on þæt land. (Or_1:

6.  While the analysis put forward in Thompson (1999) captures the anaphoric character of ‘then’ in a more or less intuitive way, it suffers from a number of shortcomings. In particular, it fails to make explicit how the linking of times in tense structure actually leads to the relevant interpretative differences. For example, while it seems to be plausible to a certain extent to attribute the cotemporal reading to the linking of Event times, that assumption still fails to account for the fact that even cotemporal ‘then’ usually leads to an interpretation where the event described by the consequent clause sets in after the event described in the antecedent clause (i.e., the temporal settings overlap, but are not identical). In a similar vein, it is not clear to us how exactly the linking of Reference times results in an ordered reading (Thompson’s 1999 article does not give any clear clues here). Of course, one may come up with some additional (pragmatic) machinery e.g., one might assume that the event described by the first clause is (by default) interpreted as completed when a subsequent clause describes a second action or event that is viewed from the same Reference time (if there is no additional link associating the Event times of the two clauses), but it would certainly be more desirable if such major aspects followed directly from the central assumptions of the theory. See Fuß & Trips (2009) for an account based on a more elaborate (semantic) analysis of ‘then’ based on DRT-based approaches such as Roßdeutscher (2005a,b) and Roßdeutscher and von Stutterheim (2006).

 Carola Trips & Eric Fuß

‘He said that at one occasion he wanted to find out how far that land extended northwards, or whether any man lived north of the wilderness. Then he travelled northwards along the coast; keeping all the way the waste land on the starboard and the open sea on the portside for three days. Then he was as far north as the whalehunters go furthest. Then he travelled still northwards as far as he could sail in another three days. Then the land turned east, or the sea into the land, he didn’t know which, but he knew that he there waited for a wind from the west and somewhat from the north and sailed then east along the coast as far as he could sail in four days. Then he had to wait for a due north wind, because that land turned there directly to south, or the sea into the land, he didn’t know which. Then he sailed from there southwards along the coast as far as he could sail in five days. Then there was a large river reaching up into the land.’ (Enkvist & Wårvik 1987: 234)

A cursory look at the text shows that after a brief backgrounding introduction (‘He said that at one occasion …’), the main story line is carried forward by a series of clauses introduced by þa. These clauses describe a sequence of actions/events that take place one after the other. Furthermore, note that the discourse referent the subject pronoun refers to remains constant, while the clause describes a new action or a change affecting the state of the discourse referent. In general, it seems that in passages of that type fronted þa has the narrative function of marking foregrounded actions/events, while its temporal properties lead to an interpretation where these events are understood as taking place sequentially (i.e., one after the other, without temporal overlap). In other words, it appears that clause-initial þa/þonne triggering inversion are instances of sequential ‘then’. As noted above, this particular temporal interpretation is presumably associated with attaching ‘then’ to IP, the locus of Reference time. In the following sections, we will further explore the structural position of þa, þonne, focussing on the question of why these elements trigger inversion in OE.

3.3  The distribution of sequential þa/þonne and subject pronouns A closer look at the syntactic distribution of sequential ‘then’ in OE reveals that fronted þa/þonne may be preceded by a topicalised phrase, giving rise to Verb Third orders that are reminiscent of the kind of Verb Third typically occurring with subject pronouns (compare (7) above): (20) a. On þa ilcan tima þa comon hi to Medeshamstede… at the same time then came they to M. (ChronE_[Plummer]:870.5.1115)

b. Syððan þa com he to se cyng Eadgar, … afterwards then came he to the king E. (ChronE_[Plummer]:963.9.1396)

The syntax and semantics of the temporal anaphor “then” in Old and Middle English 

c. For þi þonne wacion we, … for that then stay-awake we ‘because then we stay awake …’ (ChrodR_1:14.6.277)

(21) a. Mid þam ða com þæt wif. with that then came that woman (ACHom_II,_8:67.14.1355)

b. Him þa andswarode se biscop. him then answered the bishop (GD_1_[C]:


On ðone sexteoðan dæg ðæs monðes þonne bið Sancte Marcelles on the sixteenth day of-the month then is Saint Marcel-gen tid ðæs papan. feast-day of-the pope-gen


The examples above show that fronting of þa/þonne requires subject-verb inversion with both pronominal (20) and full nominal subjects (21). Furthermore, note that while (20) a.,b. and (21) c. initially seem to suggest an analysis in terms of left dislocation, where a fronted adverbial expression co-occurs with an appropriate pronominal form (similar to e.g., ModG Am Samstag, da ging er ins Kino ‘On Saturday, he went to the movies’), examples like (20) c., and (21) a., b. clearly show that this analysis cannot be generalised to all cases of Verb Third with fronted þa/þonne.7 What should a proper analysis of examples like (20) and (21) take into account then? Recall that in section 2 above, we have argued that fronted þa/þonne are better analysed as non-operators. Accordingly, we are led to expect that the finite verb does not move into the C-domain in (20) and (21), but rather occupies Infl/T, as in all other clauses with fronted non-operators. The data in (20) and (21) can then be accounted for under the following set of assumptions:8 i. The fronted XP occupies SpecCP (or, a relevant spec in a split-C system). ii. þa/þonne occupy a specifier in the inflectional domain, presumably SpecTP (the specifier of the head associated with the encoding of Reference time). iii. The finite verb is located in T. iv. All subjects, including pronouns occupy a lower, vP-internal position.

7.  We found 55 cases which clearly do not have the status of left-dislocation structures, among them 14 cases with pronominal subjects. 8.  An anonymous reviewer claims that in contrast to our analysis, the standard analysis of inversion with þa/þonne predicts that there is a certain asymmetry between (20) and (21) in

 Carola Trips & Eric Fuß

Accordingly, examples such as (20) and (21) are analysed as in (22), where þa/þonne occupy the specifier of TP (directly adjacent to the finite verb in T), while subjects generally stay behind in their vP-internal theta-position. Verb Third with þa/þonne9 (22) [CP Topic [TP þa/þonne [T′ Vfin [vP subject (pronoun) …]]]]

If we compare the structure in (22) with the structure commonly posited for Verb Third orders with pronominal subjects in (23), it appears that the preverbal position is apparently either filled by the subject pronoun (giving rise to Verb Third without inversion) or with þa/þonne (leading to Verb Second/Verb Third and obligatory inversion): Verb Third orders with fronted non-operators (23) [CP Topic [TP subject pronoun [T′ Vfin [vP …]]]]

The similarities between the patterns in (22) and (23) suggest that þa/þonne and subject pronouns compete for the same structural position. As noted above, this position is presumably to be identified with SpecIP/TP, since fronted þa/þonne specify Reference time, which is commonly associated with IP/TP (Hornstein 1990; Stowell 1995; Thompson 1999). Further evidence for this hypothesis comes from the observation that þa seems to require that the finite verb be in the preterite indicative (cf. Mitchell 1985; Wårwik 1995), that is, þa has selectional properties linked to the inflectional domain of the clause (for a comprehensive description of the difference between þa and þonne see Mitchell 1985).

that “(20) must involve V-to-C movement (so that inversion with the subject pronoun can be derived), whereas (21) would not have to involve this type of movement (as inversion with a full DP subject can be derived without such movement).” What the reviewer seems to suggest is that under the (common) assumption that in OE, pronouns occupy a structural position (left edge of IP/TP) higher than the position of full DP subjects (SpecvP), inversion with full DP-subjects can be derived without V-to-C movement (i.e., with V-to-T movement). But note that this does not represent the ‘standard analysis’ of these word order facts; rather the ‘standard analysis’ assumes that þa/þonne occupy SpecCP and trigger obligatory V-to-C movement (van Kemenade 1987; Pintzuk 1999; Haeberli 2002 and many others). It is true, however, that the ‘standard analysis’ predicts an asymmetry between (20) and (21) that concerns the position of the subject: Due to the fact that full subject DPs may remain in a lower position in OE, it is expected that certain adverbs should be able to intervene between the fronted verb and the subject, while this should not be possible in the case of subject pronouns, which move to the left edge of IP/TP. In the interest of time and space, we leave the investigation of this issue for future research. 9.  The absence of a fronted topic leads to Verb Second orders as in the examples above.

The syntax and semantics of the temporal anaphor “then” in Old and Middle English 

Under these assumptions, the observation that þa/þonne seem to force subject pronouns to stay in a lower, post-verbal position can be analysed as an instance of Merge over Move (Chomsky 1995). Thus, we propose that sequential þa and þonne are merged directly in the specifier of TP, thereby blocking movement of the subject pronoun to this position.10 As a consequence, the pronoun has to stay behind in its theta-position (SpecvP), with Case and agreement checking being accomplished via an agree-relation initiated by T.11 In the next section, we will take a closer look at the properties of SpecTP in OE and the question of why both pronouns and certain temporal anaphora seem to be attracted by this position.

3.4  Discourse-configurationality and the nature of SpecTP in OE As noted above, in current work OE is often portrayed as a discourse-configurational language, in which structural positions did not primarily encode grammatical functions, but rather were linked to discourse-related distinctions such as information-structure (cf. Fischer et al. 2000; van Kemenade & Los 2006). The basic proposal we want to put forward here is that this particular property of OE was not limited to information-structural categories such as topic or focus, but also included the discourse-anchoring of anaphoric expressions, that is, we assume that the interpretation of anaphoric expressions was linked to a certain position in the clausestructure of OE. This hypothesis opens a new perspective on the complementary distribution of referential (subject) pronouns and þa/þonne in preverbal position if we further assume that in OE, the relevant functional specifier is to be identified as SpecTP, the position apparently targeted by the elements in question.12 Let’s now take a closer look at the relevant feature specifications shared by (subject) pronouns and the temporal anaphora þa/þonne that qualifies them as potential realisations of SpecTP in OE.

10.  Alternatively, the speaker may choose to merge þa/þonne in a lower position to achieve a different communicative effect (cotemporal ‘then’). In that case, the (subject) pronoun can freely move to SpecTP, giving rise to a word order option where the pronoun precedes þa/þonne. See Roßdeutscher (2005a,b), Roßdeutscher and von Stutterheim (2006), for an analysis that attributes the different readings of then to the relative ordering of subject pronoun and then. 11.  Under the assumption that discourse-relatedness is obligatorily associated with the C-domain, we might assume that þa/þonne move to SpecFinP after being merged in SpecTP. The shift from SpecTP to a higher position, which took place in the ME period, can then be analysed as an instance of a change in which a movement dependency is reanalysed as direct merge in the former target position (while the earlier first merge position SpecTP is obligatorily occupied by the subject/expletive in ME and ModE, see below for details). 12.  We found 119 cases showing the order subject pronoun –  þa/þonne – Vfin, a  finding which at first sight seems to weaken our claim that subject pronouns and þa/þonne are in

 Carola Trips & Eric Fuß

A property common to both pronouns and then seems to be that the interpretation of these elements involves a variable that must be bound by (or identified with) a topical element/referent in the given discourse (see e.g., Heim & Kratzer 1998 for (referential) pronouns and Webber et al. 2003 for a unified analysis of (referential) pronouns, then, and discourse adverbials such as otherwise or instead). In what follows, we will use the feature [+anaphoric] to refer to this property.13 Another feature shared by the elements under investigation is morphosyntactic in nature. Recall that þa/þonne developed from former demonstratives. As noted above, this aspect of the etymology of these elements is still transparent in OE. This suggests that þa/þonne are nominal in nature.14 Let’s now address the question of how this relates to the feature specification of T in OE. A different way of phrasing the idea that the interpretation of anaphoric expressions was linked to SpecTP in OE is that T was employed to implement the assignment of discourse-related (referential or temporal) indices to variables introduced by pronouns and temporal anaphora.15 Furthermore, it is commonly assumed that in general, SpecTP is reserved for nominal material only (i.e., it is specified as [+D]):

complementary distribution. However, this is not the case if we take into account that in OE pronouns can be topicalised: (ii)

Hig þa forlættan þone wall & heora burh they then left the wall and their fort (Bede_1:

Also note that if this order occurs in embedded clauses the temporal adverbs þa/þonne are clause-medial and non-sequential. 13.  We are aware of the fact that this practice seems to be at odds with Chomsky’s (1981, 1986) classification of pronouns as [+pronominal, –anaphoric]. But note that this proposal has been formulated for the particular purposes of Binding Theory. In recent theoretic work (following Reinhart & Reuland 1993), there is a tendency to dispense with this featural characterisation of pronouns and instead attribute their Binding properties to other feature specifications or the internal structure of pronouns. To avoid terminological confusion, it should be kept in mind that our use of the term differs from its use in classic GB theory. 14.  In fact, most other adverbials (see above) that frequently trigger inversion are also nominal in nature. This is less clear for OE nu. However, according to the Oxford English Dictionary (OED online), it can be used as a nominal denoting ‘the present time or the time spoken of or referred to’, as in for example “The here and now.” It is cognate with German Nu, nun where it can also be used as a noun: “Er kam im Nu” (he came at once). Interestingly, in Old Irish a present tense verbal prefix nu-/no- can be found which is further support for our hypothesis. 15.  Presumably, the syntactic component is blind to the exact semantic content of these indices. However, in a discourse-configurational language like OE, syntax may make available a

The syntax and semantics of the temporal anaphor “then” in Old and Middle English 

(24) SpecTP in OE: [+anaphoric, +D]16

The different syntactic distribution of (subject) pronouns, full referential DPs and þa/þonne can then be accounted for as follows: (subject) pronouns are [+D], [+anaphoric]. They are merged in SpecvP where they receive a thematic role and subsequently move to SpecTP (if possible), where they are assigned a referential index associated with a topical discourse referent. In contrast, full referential DP subjects are [+D], but crucially [–anaphoric]. Accordingly, they do not move to SpecTP and remain in situ (SpecvP) in OE (but they may move further up into the C-domain to be interpreted as topic, focus etc.). Now, turning to the primary subject matter of this paper, þa/þonne share the features [+D], [+anaphoric] with pronouns. If these temporal anaphora are to be interpreted as sentence adverbs (i.e., sequential ‘then’), they are directly merged in SpecTP (the locus of Reference time) to receive a temporal index (i.e., they are linked with a Reference time given in the discourse), blocking movement of subject pronouns to this position (Merge over Move, Chomsky 1995). In cases where þa/þonne occupy SpecTP, the assignment of a referential index to a lower subject pronoun proceeds via an Agree-relation between T and the pronoun (established during the syntactic derivation for independent reasons (Case and agreement)).17 Of course, this analysis raises a number of further questions which, due to space limitations, we cannot address in detail in this paper. For example, more has to be said about the fact that in embedded clauses, subject pronouns generally occur to the left of þa/þonne, directly adjacent to the complementiser: (25) Þa hi þa hine geornlice beheoldon … when they then them carefully beheld ‘when they then carefully beheld him…’ (eust, LS_8_[Eust]:270.286; van Kemenade & Los 2006: 236f.)

This is not expected under the analysis proposed because the presence of þa/þonne should block movement of subject pronouns. A possible answer to this problem is that the tense properties of embedded clauses differ significantly from the tense

structural variant of the relevant semantic assignment procedure that is necessary to interpret anaphoric expressions in a given clause. 16.  Recall that we assume that in OE, T lacked an EPP feature. The frequent presence of subject pronouns in SpecTP (due to their anaphoric nature) possibly supported the development of [+EPP] T in the ME period (see section 4 below). 17.  Note that this seems to be reminiscent of the relation between there and its associate DP in existential constructions. However, there are also major differences between these two constructions. For example, in contrast to there, þa/þonne are not expletives that lack semantic content. As a result, they may not be deleted (or substituted by the subject pronoun) at LF.

 Carola Trips & Eric Fuß

properties of main clauses. More precisely, embedded tense is dependent on the temporal anchoring of the matrix clause (cf. e.g., Enç 1987). Presumably, this dependency is mediated by the complementizer (cf. Travis 1984; Bennis & Hoekstra 1989). Therefore, it seems plausible to assume that in embedded clauses the interpretation of temporal anaphora such as þa, þonne is less dependent on an anchor time given in the discourse, but can rather be computed directly from the tense properties of the matrix clause transferred to embedded T. Hence, þa/þonne are not required to occupy SpecTP, giving rise to more word order options. Moreover, note that the typical rhetoric use of sequential ‘then’ in OE (marking of foregrounded successive actions/events, cf. Foster 1975; Enkvist & Wårwik 1987) is much less called for in embedded clauses, which are typically associated with backgrounded information. In contrast, the requirements for identifying the reference of pronominal elements do not differ much from the situation in main clauses. As a result, SpecTP is regularly occupied by subject pronouns in embedded clauses.18 A prediction of these assumptions is that þa/þonne may occur in SpecTP, directly adjacent to the complementizer, if the clause does not contain other anaphoric elements (in particular, no subject pronouns). This seems to be borne out: (26) a. Gif þonne swiðra wind aras, þonne tynde he his bec if then stronger wind arose then closed he his books ‘if a stronger wind then arose, then he closed his book’ (cobede,Bede_4:; van Kemenade & Los 2006: 238)


Gif þonne hwylc læsse þing sie to smeagenne, þonne … if then any less thing be to think on then ‘if there is any more minor thing to consider, then …’ (cobenrul,BenR:; van Kemenade & Los 2006: 238)

Furthermore, object pronouns may occur to the left of þa either alone (if there is no pronominal subject present), or together with the subject pronoun: (27) ðætte hie ðonne gemonnðwærige sio lufu & sio geferræden that them then may-humanize the love and the society

hiora niehstena … of-their neighbors

‘that love and the society of their neighbors may humanize them’ (cocura,CP:47.363.15.2461; van Kemenade & Los 2006: 236)

18.  For reasons of Relativized Minimality, object pronouns may occupy SpecTP only (i) if they form a cluster with the subject pronoun prior to movement to SpecTP, or (ii) if no subject pronoun is present.

The syntax and semantics of the temporal anaphor “then” in Old and Middle English 

(28) gif he hit him ðonne sellan mæge if he it him then give may ‘if he can give it him then’ (cocura,CP:44.323.24; van Kemenade & Los 2006: 235)

The presence of all pronouns in front of þa/þonne can perhaps be accounted for under the assumption that pronominal elements may form a cluster prior to movement to SpecTP. In addition, it is of course possible that some of the apparently problematic examples involve instances of cotemporal ‘then’, which occupies a lower, VP-adjoined position. Another obvious problem consists in the observation that contrary to what is expected under the analysis developed in this paper, a subject pronoun always immediately follows the fronted verb in root wh-questions. In other words, the pattern wh-Vfin-þa/þonne-pron.subj is apparently not attested in OE. Again, this might be due to independent reasons, for example a morphophonological requirement (or at least strong tendency) that the (weak) subject pronoun must be adjacent to the fronted finite verb (similar constraints hold in many present-day Germanic V2-languages). Moreover, a closer look reveals that in many of the relevant examples, þa/þonne should rather be interpreted as instances of cotemporal ‘then’ as in (29)b: (29) a.

and þonne gyt ne cymð se brydguma; Eac swilce þa six and then still not comes the bridegroom; also such the six

ðusend geara fram Adame beoð geendode and ðonne gyt thousand years from Adam is ended and then still

e lcað se brydguma. delays the bridegroom.

(ÆCHom_II,_44:330.117.7427-7430) b. [Hu mage we þonne witan hwænne he cymð?] How may we then know when he comes? (ÆCHom_II,_44:330.117.7427-7430)

Accordingly, one might speculate that the order wh-Vfin-þa/þonne-pron.subj is not attested in the corpus for the following reasons: (i) it would have been quite rare anyway, since it is confined to a special context (a wh-question concerning a foregrounded sequence of actions/events); (ii) for PF-reasons, subject pronouns are preferably adjacent to a fronted finite verb (the same reasons presumably require adjacency of C and subject pronouns in embedded clauses as shown in (25) above). A comprehensive view on these matters is elaborated in Fuß & Trips (2009), including an in-depth study of the distribution and interpretation of pronouns and þa/þonne in both main and embedded clauses. In the following section, it is shown that the analysis proposed in this section receives further support from observations on the loss of Verb Second in the ME period.

 Carola Trips & Eric Fuß

4.  The loss of then+inversion in the Middle English period In this section, we will address the question of how the loss of subject-verb inversion in clauses with clause-initial þa/þonne came about since it presents further evidence for our assumptions. There are chronological parallels between the loss of inversion with þenne/then, þan/than and changes that affected the status of the subject position in the ME period.19 In contrast to ModE, OE displays a number of subjectless constructions where neither a nominative subject nor an expletive element shows up in the subject position (SpecTP). Relevant examples include impersonal constructions with weather verbs, experiencer verbs and impersonal passives: (30) norþan sniwde [from] north snowed ‘it snowed from the north’ (Seafarer, 31; Kiparsky 1997:471 ) (31) him ofhreow þæs mannes him-dat pitied the man-gen ‘he pitied the man’ (AColl, 192.16; Allen 1995:68) (32) þæt eallum folce sy gedemed beforan ðe that all people-dat be judged before thee ‘that all the people be judged before you’ (Paris Ps. 9.18; van Kemenade 1997:335)

In Early ME, these constructions began to disappear, a development which is accompanied by the emergence of the expletive there. According to e.g., Breivik (1989), Breivik (1990), Allen (1995), van Kemenade (1997) and Haeberli (2002), the loss of subjectless constructions took place roughly between 1350 and the early 15th century. From a theoretical point of view, this change can be attributed to the development of an EPP feature that requires the subject position (here identified as SpecTP) to be overtly filled either by a nominal bearing nominative case or a semantically vacuous expletive element such as there. Interestingly, it seems that the loss of þenne/then, þan/than and Verb Second took place in the very same period. A survey over a set of ME texts in the PPCME2 shows that then loses its special status as a trigger of

19.  More generally, it seems that there are chronological parallels between the overall loss of verb second patterns and the rise of the requirement that the subject position be overtly filled (in the ME period, roughly from 1350 to 1425; cf. Hulk & van Kemenade 1995; van Kemenade 1997; Haeberli 2002; Fuß 2003).

The syntax and semantics of the temporal anaphor “then” in Old and Middle English 

Verb Second in the period from 1340–1475, cf. Fuß and Trips (2003), Fuß & Trips (2009). This period was characterised by a whole set of surface changes that contributed to the overall loss of discourse-configurationality (loss of case inflections, loss of subjectless constructions, general loss of word order variation, rise of structural passive constructions etc.). The correlation between the loss of inversion with þenne/then, þan/than and the development of the expletive there becomes particularly clear in the Ayenbite of Inwit (1340), a Kentish text which exhibits variation between inverted and non-inverted orders after clause-initial þanne ‘then’.20 However, all examples (10 cases) with the expletive þer ‘there’ display Verb Third order, with the expletive intervening between þanne and the finite verb: (33) Þanne þer nys prowesse ariht:  … then there not-is prowess proper ‘Then there is no nobleness.’ (AYENBI,87.1702)

The systematic absence of Verb Second orders in clauses in which þanne and þer co-occur supports the conjecture that there is a close connection between the loss of inversion with þenne/then, þan/than and the rise of an EPP feature in T: in cases where an expletive is inserted as SpecTP to satisfy T’s EPP feature, the adverb þanne must occupy another position (e.g., in the CP domain, or adjoined to TP).21 In the course of time, Verb Second patterns with then completely dropped out of the grammar, since SpecTP became a position reserved for subjects/expletives. It could no longer host adverbs:22 (34) [CP Ø [TP then [TP expl./subj. [T′ Vfin [vP …]]]]]

20.  In the Ayenbite of Inwit, we found 70% inversion with subject pronouns (16 of 23 cases), and 44% inversion with full subject DPs (14 of 32 cases), probably an instance of Grammar Competition (Kroch 1989). 21.  In a similar vein, Alexiadou (2000) assumes that SpecTP can host temporal adverbs only if there is no EPP feature in T. However, in languages where such a feature requires subjects to appear in SpecTP, temporal adverbs cannot occur in this position. 22.  A reviewer pointed out that it is not entirely true that SpecTP is reserved for subjects/ expletives in ModE, mentioning the phenomenon of locative inversion where, according to some analyses (cf. e.g., Hoekstra & Mulder 1990; Collins 1997), the fronted locative is located in SpecTP. From the observation that elements other than subjects/expletives can apparently also occupy SpecTP in ModE, he/she concludes that “it is not clear why then could not have maintained such a status beyond ME”. However, first of all it should be noted that there are also alternative analyses of locative inversion where it is assumed that the locative is a topic moving to SpecCP (Bowers 1976; Newmeyer 1987; Rochemont & Culicover 1990). Moreover, even if locative inversion targets SpecTP in ModE, it is important to note that this seems to be the only context where elements other than subjects/expletives can occupy

 Carola Trips & Eric Fuß

The connection between the two changes in question can be nicely captured under the assumption that OE þa/þonne occupy SpecTP, but is much more difficult to account for if these temporal adverbs are analysed as operators located in SpecCP. The development in question is also reflected by changes affecting the so-called ‘correlative construction’, in which an adverb at the beginning of an independent clause recapitulates or anticipates a temporal adverb clause. We repeat here the OE example from above: (35) Ða se wisdom þa ðis spell asæd hæfde, þa ongan he when the wisdom then this story said had then began he eft singan again sing ‘When wisdom then had told this story, he began to sing again.’ (Fischer et al.: 2000: 57)

In ME we find three patterns: the (chronologically) first pattern shows the order whenS-Vfin … then-Vfin-S. It is the OE order (34 cases) where the difference between the main clause and the embedded clause is signalled with word order differences (S-Vfin as a signal for the embedded clause and Vfin-S as a signal for the main clause), although in ME the pronoun whan has replaced the form thanne as conjunction: (36) for whan they knowen thy naked purpos, thanne haue thei no cause for when they know your naked purpose then have they no cause to repreue the, neither for pore folke ne for noo religious gystes. to repreive you, neither for poor folk nor for no religious gests (AELR4,3.68)

The second pattern displays the order then, when-S-Vfin … Ø/S-Vfin … (47 cases) and might indicate the change from the OE pattern þa-Vfin-S to the ModE pattern thenS-Vfin (see below): Under the assumption that þa triggers Verb Second for discourse reasons, when the main clause is interrupted by the when-clause (with þa preceding

SpecTP (leaving aside the analysis of quotative inversion poposed in Collins 1997); crucially, no other adverbs are allowed in SpecTP (cf. also Alexiadou 2000 for the notion that the presence of an EPP feature rules out adverb insertion into SpecTP). Furthermore, it has been noted that locative inversion is possible only under highly restricted syntactic and pragmatic conditions (for a discussion see Birner 1994). Thus, in light of the facts (unclear theoretical status of locative inversion, no other elements possible in SpecTP), it seems that the phenomenon of locative inversion does not constitute a conclusive argument against an analysis that links the loss of þa/þonne+V2 to the rise of an EPP-feature in T.

The syntax and semantics of the temporal anaphor “then” in Old and Middle English 

that clause), subject-verb inversion can no longer be triggered and thus S-Vfin orders start to occur. (37) Then aftyr, when scho was wened, þay broght hur to þe tempull, then after when she was accustomed they brought her to the temple (MIRK,16.471)

The third pattern is when-S-Vfin … then-S-Vfin (38 cases). Here, the word order differences have been lost, both the embedded and main clause show the order S-Vfin. This is the pattern we also find in ModE. (38) bote whanne he schal come hym-self to gyue rigtful dom ate but when he shall come himself to give rightful judgement at day of dome thanne he schal be knowe a verrey mygtful God. day of judgement then he shall be known a very mightyful God (AELR3,47.668)

5.  Conclusions In this paper, we presented a new approach to Verb Second patterns triggered by þa/þonne in clause-initial position in OE which does not make use of the problematic assumption that these adverbs are operator-like elements. Clause-initial þa/þonne were analysed as temporal anaphora that give rise to a reading where actions/events are temporally ordered. This particular interpretation was accounted for by the assumption that clause-initial þa/þonne link the Reference time of consecutive clauses. In narratives, þa/þonne are employed as discourse markers that mark a sequence of foregrounded actions/events. The relative distribution of þa/þonne and subject pronouns was taken to indicate that these elements compete for the same structural position, SpecTP. We argued that this position was linked to the discourse anchoring of anaphoric expressions in OE. The fact that þa/þonne trigger inversion with pronominal subjects was attributed to the assumption that these temporal anaphora are merged in SpecTP, thereby forcing subject pronouns to stay behind in a vP-internal position (with the finite verb in T). Further support for our analysis comes from the loss of Verb Second patterns with then in the ME period, which can be attributed to an independent change, namely the rise of an EPP feature in T (and the overall loss of discourse-configurationality). Of course, this analysis raises the question of whether we can observe similar facts in other Early Germanic languages. A brief glance over the relevant literature suggests that this is indeed the case (cf. Betten 1987 on OHG thô, Klein 1994 on the use of clause-initial þanuh in Gothic; see Fuß & Trips (2009) for a fuller discussion and a closer look on OHG, in particular), compare the following examples from the

 Carola Trips & Eric Fuß

Gothic Bible, where clause-initial þanuh and þaruh (according to Klein (1994: 262) “discourse-continuative foregrounding markers, carrying forward the discourse along the time-line of the main story”) trigger subject-verb inversion:23 (39) a. þaruh qaþ Iesus du þaim twalibim […] then said Jesus to the twelve b. þanuh andhof imma Seimon Patrius […] then answered him Simon Petrus (John 6.67-68; Klein 1994: 260)

Acknowledgments Thanks to the audiences at the LAGB 2003 in Oxford, at the university of Frankurt and at the CGSW 2007 in Stuttgart where versions of this paper have been presented. Special thanks to Eric Haeberli, Hans Kamp, Antje Roßdeutscher and Hans-Christian Schmitz.

References Alexiadou, A. 2000. On the syntax of temporal adverbs and the nature of Spec,TP. Rivista di Linguistiica 12: 53–73. Special issue on adverbs edited by Norbert Corver & Denis Delfitto. Allen, C.L. 1995. Case Marking and Reanalysis: Grammatical Relations from Old to Early Modern English. Oxford: Clarendon Press. Bennis, H. & T. Hoekstra. 1989. Why Kaatje was not heard sing a song. In Sentential Complementation and the Lexicon, D. Jaspers et al. (Eds), 21–40. Dordrecht: Foris. Betten, A. 1987. Zur Satzverknüpfung im althochdeutschen Tatian. Textsyntaktische Betrachtungen zum Konnektor tho und seinen lateinischen Entsprechungen. In Althochdeutsch, R. Bergmann et al. (Eds), volume I, Grammatik. Glossen und Texte, 395–407. Heidelberg: Carl Winter. Birner, B. 1994. Information status and word order: An analysis of English inversion. Language 70: 233–259. Bowers, J. 1976. On surface structure grammatical relations and the structure-preserving hypothesis. Linguistic Analysis 2: 225–242. Breivik, L. 1989. On the causes of syntactic change in English. In Language Change: Contribution to the Studies of its Causes, L. Breivik & E. Hakon (Eds), 29–70. Berlin: Mouton de Gruyter. Breivik, L. 1990. Existential ‘there’: A Synchronic and Diachronic Study. Oslo: Novus Press.

23.  Interestingly, Hirt (1929: 352f) observes that clause-initial ‘then’ triggers regular inversion in other early Indo-European languages as well (including Sanskrit and Old Greek). This might be taken to indicate that the phenomenon in question is actually of greater antiquity than originally assumed.

The syntax and semantics of the temporal anaphor “then” in Old and Middle English 

Cardinaletti, A. & I. Roberts. 2002. Clause structure and X-second. In Functional Structure in DP and IP: the Cartography of Syntactic Structures, G. Cinque (Ed.), volume 1, 123–166. Oxford: Oxford University Press. Chomsky, N. 1981. Lectures on Government and Binding. Studies in Generative Grammar 9. Dordrecht: Foris. Chomsky, N. 1986. Knowledge of Language. Its Nature, Origin, and Use. New York: Praeger. Chomsky, N. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Collins, C. 1997. Local Economy. Cambridge, Mass.: MIT Press. Enkvist, N. & B. Wårwik. 1987. Old English þa, temporal chains, and narrative structure. In Papers from the 7th International Conference on Historical Linguistics, A.G. Ramat et al. (Eds), 221–237. Amsterdam: John Benjamins. Enç, M. 1987. Anchoring conditions for tense. Linguistic Inquiry 18: 633–657. Fischer, O., A. van Kemenade, W. Koopman & W. van der Wurff. 2000. The Syntax of Early English. Cambridge: Cambridge University Press. Foster, R. 1975. The use of þa in Old and Middle English. Neuphilologische Mitteilungen 77: 404–414. Fuß, E. 2003. On the historical core of V2 in Germanic. Nordic Journal of Linguistics 26(2): 195–231. Fuß, E. & C. Trips. 2003. þa, þonne and V2 in Old and Middle English. Paper presented at the annual LAGB meeting, University of Oxford. Fuß, E. & C. Trips. 2009. The syntax of temporal anaphora in early Germanic. In prep. Glasbey, S. 1993. Distinguishing between events and times: Some evidence from the semantics of then. Natural Language Semantics 1(3): 285–312. Haeberli, E. 1999. Features, Categories and A-Positions:  Cross-Linguistic Variation in the Germanic Languages. Ph.D. dissertation, University of Geneva. Haeberli, E. 2002. Features, Categories and A-positions: Cross-Linguistic Variation in the Germanic Languages. Dordrecht: Kluwer. Heim, I. & A. Kratzer. 1998. Semantics in Generative Grammar. Oxford: Blackwell. Hirt, H. 1929. Indogermanische Grammatik, volume V: Der Akzent. Heidelberg: Carl Winter. Hoekstra, T. & R.  Mulder. 1990. Unergatives as copular verbs:  Locational and existential predication. The Linguistic Review 7: 1–79. Hornstein, N. 1990. As Time Goes by. Cambridge, Mass.: MIT Press. Hulk, A. & A. van Kemenade. 1995. Verb second, pro-drop, functional projections and language change. In Clause Structure and Language Change, A. Battye and I. Roberts (Eds), 227–256. Oxford: Oxford University Press. van Kemenade, A. 1987. Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris. van Kemenade, A. 1997. V2 and embedded topicalization in Old and Middle English. In Parameters of Morphosyntactic Change, A. van Kemenade & N. Vincent (Eds), 326–351. Cambridge: Cambridge University Press. van Kemenade, A. 1999. Sentential negation and clause structure in Old English. In Negation in the History of English, I.T.-B. van Ostade et al. (Eds), 147–165. Berlin: Mouton de Gruyter. van Kemenade, A. 2002. Word order in Old English prose and poetry:  The position of finite verbs and adverbs. In Studies in the History of the English Language: A Millennial Perspective, D. Minkova & R. Stockwell (Eds), 355–373. Berlin: Mouton de Gruyter. van Kemenade, A. & B. Los (Eds), 2006. Blackwell Handbook of the History of English. Oxford: Blackwell.

 Carola Trips & Eric Fuß van Kemenade, A. & T. Milicev. 2005. Syntax and discourse in Old and Middle English word order. In Articles from the 8th Diachronic Generative Syntax Conference, S.  Anderson & D. Jonas (Eds), Oxford University Press. Kiparsky, P. 1997. The rise of positional licensing. In Parameters of Morphosyntactic Change, A. van Kemenade & N. Vincent (Eds), 460–493. Cambridge: Cambridge University Press. Klein, J. 1994. Gothic þaruh, þanuh and -(u)þan. Indogermanische Forschungen 99: 253–276. Kroch, A. 1989. Reflexes of grammar in patterns of language change. Language Variation and Change 1: 199–244. Kroch, A. & A.  Taylor. 1997. Verb movement in Old and Middle English:  Dialect variation and language contact. In Parameters of Morphosyntactic Change, A.  van Kemenade & N. Vincent (Eds), 297–325. Cambridge: Cambridge University Press. Kroch, A. & A. Taylor (Eds), 2000. The Penn-Helsinki Parsed Corpus of Middle English, Second Edition (PPCME2). Philadelphia: University of Pennsylvania. Mitchell, B. 1985. Old English Syntax. Oxford: Clarendon. Mitchell, B. & F.C. Robinson. 2003. A Guide to Old English. Oxford: Blackwell, 6th Edn. Newmeyer, F. 1987. Presentational there-insertion and the notions ‘root transformation’ and ‘stylistic rule’. In Papers from the Twenty-Third Regional Meeting of the Chicago Linguistic Society, B. Need, E. Schiller & A. Bosch (Eds), 295–308. Chicago: Chicago Linguistic Society. OED online. 2008. Oxford English Dictionary Online. Oxford:  Oxford University Press, 2nd edition. Pintzuk, S. 1993. Verb seconding in Old English:  Verb movement to Infl. The Linguistic Review 5–35. Pintzuk, S. 1999. Phrase Structures in Competition:  Variation and Change in Old English. New York: Garland. Ramat, P. 1981. Einführung in das Germanische. Tübingen: Niemeyer. Reinhart, T. & E. Reuland. 1993. Reflexivity. Linguistic Inquiry 24: 657–720. Rizzi, L. 1997. The fine structure of the left periphery. In Elements of Grammar: Handbook in Generative Syntax, L. Haegeman (Ed.), 281–337. Dordrecht: Kluwer. Rochemont, M. & P. Culicover. 1990. English Focus Constructions and the Theory of Grammar. Cambridge: Cambridge University Press. Roßdeutscher, A. 2005a. ‘On-line’-inferences in the semantics of dann and then. In Proceedings of SuB9, E. Maier, C. Bary & J. Huitink (Eds), 325–339. Nijmegen Center for Semantics. http://www.ru.nl/ncs/sub9. Roßdeutscher, A. 2005b. ‘On-line’-interpretation and ‘cause’ in the semantics of temporal dann/ then and German da. Submitted to Journal of Semantics. Roßdeutscher, A. & C. von Stutterheim. 2006. Semantische und pragmatische Prinzipien der Positionierung von dann. Linguistische Berichte 205: 29–60. Schiffrin, D. 1992. Anaphoric then:  Aspectual, textual, and epistemic meaning. Linguistics  20: 753–792. Smith, C. 1981. Semantic and syntactic constraints on temporal interpretation. In Syntax and Semantics 14: Tense and Aspect, P.T.A. Zaenen (Ed.), 213–237. New York: Academic Press. Stowell, T. 1995. The phrase structure of tense. In Phrase Structure and the Lexicon, J. Rooryck & L. Zaring (Eds), 277–291. Dordrecht: Kluwer. Taylor, A., A.  Warner, S.  Pintzuk & F.  Beths (Eds), 2003. The York-Toronto-Helsinki Parsed Corpus of Old English Prose. University of York. Thompson, E. 1999. The temporal structure of discourse: The syntax and semantics of temporal then. Natural Language and Linguistic Theory 17: 123–160.

The syntax and semantics of the temporal anaphor “then” in Old and Middle English 

Travis, L. 1984. Parameters and Effects of Word Order Variation. Ph.D. dissertation, MIT. Trips, C. 2002. From OV to VO in Early Middle English. Amsterdam: John Benjamins. Webber, B., M. Stone et al. 2003. Anaphora and discourse structure. Computational Linguistics 29(4): 545–587. Westergaard, M. 2007. Microvariation as diachrony:  A view from diachrony. Ms. University of Tromsø. Wårwik, B. 1995. þa and þonne in Middle English. In Historical Pragmatics, A.  Jucker (Ed.), 345–357. Amsterdam: John Benjamins.

Jespersen’s Cycle and the issue of prosodic ‘weakness’* Patrizia Noel Aziz Hanna

Ludwig-Maximilians-Universität München Jespersen’s Cycle (1917), according to which a negation particle is weakened and therefore later strengthened through an additional element, is a generally accepted explanation in language change. The German data, however, reveals that the weakening of the negation particle preceded the initiation of Jespersen’s Cycle by centuries. The further weakening caused the inefficacy of the old negation particle and thus initiated Jespersen’s Cycle. The reduction is demonstrated to result from segmental, prosodic, and syntactic factors. The progressive weakening of the negation particle (PIE *ne > OHG ni > MHG ne > NHG Ø) is due to the fact that its chances of attracting rhythmic stress became increasingly rare.

1.  Jespersen’s Cycle (1917) in the history of German Otto Jespersen claimed in his study Negation in English and other languages: “The original negative adverb is first weakened, then found insufficient and therefore strengthened” (1917:  4). Today, this observation is well-known as Jespersen’s Cycle and serves as an explanation in language change. The weakening of the negation particle in Germanic languages is a fact. However, the motivation for this reduction is in need of an explanation, because the weakness of negation particles is not inherited from Proto-Indo-European: Im Germanischen steht die Negation + Verbum am Eingang des Satzes, so daß also dieses in Mittelstellung erscheint; das stammt offenbar aus dem Igm., vgl. lat. nequeo, nescio; es wird wohl ursprünglich die Negationspartikel betont gewesen sein, mindestens stärker als das Verbum. (Behaghel 1932: §1428) [‘In Germanic, the negation + verb are placed at the beginning of the sentence, which means that the verb appears in the middle of the sentence; this is evidently inherited from Indo-European, cf. Lat. ˈnequeo, ˈnescio. Originally the negation particle was probably stressed, and was at least more prominent than the verb.’]

*I would like to thank Theo Vennemann, Katrin Lindner, Peter-Arnold Mumm, Dagmar Hirschberg, and two anonymous reviewers for their helpful comments, Christian Riepl for creating the electronic database, and Sarah Catherine Clemitshaw for correcting my English. All translations are mine, unless otherwise indicated.

 Patrizia Noel Aziz Hanna

Since neither Proto-Indo-European nor Standard German prohibit stress on negation particles, the question of why the reduction of PIE *ne took place in the history of Germanic it not a trivial one. Table 1 represents the German version of Jespersen’s Cycle. The account is in a way a simplification, since there was the option of double negation in Old High German and since even today double negation is a feature of dialectal grammar (Donhauser 1996: 203). In dialects, too, the old sentence negation ni/ne is replaced by nicht and its dialectal variants. Table 1.  Jespersen’s Cycle (cf. Donhauser 1996: 202) Phase I Phase II Phase III Phase IV Phase V

ni ni en/ne (ne)

+ Vfin + Vfin + Vfin + Vfin + Vfin

+ (niwiht) + niht + niht + niht

Phase I gives the Early Old High German mononegation with ni, phase II the Late Old High German state with optional niwiht ‘not a thing’. Phase III illustrates the Middle High German situation with obligatory double negation. The Middle High German reduced forms ne and en with schwa instead of a full vowel confirm that the particle was unstressed. Phase IV represents the Early New High German situation with optional ne and obligatory nicht. In phase V, i.e., after the 16th century, mononegation with nicht becomes effective. The German data, which shows a gradual weakening of the negation particle (PIE *ne > OHG ni > MHG ne > NHG Ø), raises the following questions: 1. What does ‘weakness of the German negation particle’ mean in terms of stress categories (sentence stress, main stress, secondary stress)? 2. Why is the negation particle weakened? 3. What are the specific criteria which caused the increasing reduction of the negation particle? Although it is well understood that the negation cycle is not just a Germanic phenomenon, it is demonstrated in this article to be the consequence of a language-specific interaction between syntax and phonology.1 The language-specific hypotheses are:

1.  The French negation cycle is most frequently discussed (cf. Vennemann 1974: 366–368 and 1993: 334–342 for a typological analysis). Like in German, the old negation particle was strengthened and resulted in ne … pas as an intermediate state, which is still obligatory in

Jespersen’s Cycle and the issue of prosodic ‘weakness’ 

1. The reduction of the German negation particle is motivated neither by sentence stress restrictions nor by main stress restrictions. 2. The negation particle is weakened because of prosodic, segmental, and syntactic factors. 3. The negation particle was subject to a process of increasing reduction, because the chances for the negation particle to attract rhythmic stress decreased. The hypotheses were tested on the basis of two texts, the Old High German Evangelienbuch by Otfrid von Weissenburg and the Middle High German Nibelungenlied (manuscripts A, B, and C).

2.  S  tress categories: sentence stress, main stress, and secondary stress in German The argumentation is based on a typology of stress categories. Sentence stress, main stress, and secondary stress have different functions in sentences. While sentence stress is linked to information structure and main stress generalizations refer to word lists instead of sentences, secondary stress evolves in the sentence as sentence rhythm.

2.1  Sentence stress The unit which is considered most important by the speaker is prominent in the sentence (cf. e.g. Firbas 1989) and carries sentence stress.2 As Bolinger pointed out, sentence “accent should be viewed as independent, directly reflecting the speaker’s intent and only indirectly the syntax” (1972: 633). He also stressed: The distribution of sentence accents is not determined by syntactic structure but by semantic and emotional highlighting. Syntax is relevant indirectly in that some structures are more likely to be highlighted than others. (Bolinger 1972: 644)

written language. It is obvious that the prosodic system of French with its development of phrase-final stress (Kuryłowicz 1945; Dufter 2003) is different from the German system. Although the outcome of a new negation particle is comparable, prosodic weakness in French does not result from the same conditions as in German. In French, double negation became stable in written French in the 17th century, although pas was already used mononegative in spoken language (Ernst 1985:  85; Detges 2003:  214.). At this point, German had recovered mononegativity for about a century. 2.  For the purpose of this paper, I consider it sufficient to subsume rhematic stress, contrastive stress, focus stress, etc. under the label “sentence stress”.

 Patrizia Noel Aziz Hanna

In Modern Standard German, the negation particle nicht ‘not’ can bear sentence stress. Example (1) shows NHG nicht with contrastive stress:3 (1) Das meine ich aber ˈnicht. that mean I really not ‘That is really not what I mean.’

Sentence stress is the result of an interaction between syntax and information structure. In Modern Standard German, all syllables, even reduction syllables, can carry sentence stress if the speaker wishes to focus them. It will be demonstrated that sentence stress on negation particles has always been an option.

2.2  Main stress Unlike sentence stress, main stress is usually ascribed only to the phonological system. However, this view is controversial. Although there are rules and defaults concerning its distribution which refer to syllable structure (Giegerich 1985; Vennemann 1990; Eisenberg 2006), the realisation of main stresses in a sentence is not predictable from a purely phonological perspective. Though main stress is frequently understood to be a feature of an assumed prosodic word, it is really the rhythmic description of a one-word sentence, i.e., of a marked environment (cf. Saran 1907; Vennemann 1994). Whether main stress in a specific word is ever realized in a sentence depends, among other factors, on its information weight. For instance, the conjunction und ‘and’ quite naturally bears stress on its only syllable if analysed as a one-word sentence:  ˈund. In the phrase ˈÄpfel und ˈBirnen ‘apples and pears’ it is usually unstressed because of its relative information weight. Therefore, the question of whether the old German negation particle ni/ne bears main stress as an isolated word is irrelevant for an analysis which deals with stress distribution in sentences.

2.3  Secondary stress In the phrase Äpfel und auch ˈBirnen, the function word und carries secondary or ˈ ˈ rhythmic stress. Otherwise the sentence would not be well-formed in terms of language rhythm. It would violate rhythmic well-formedness, which does not permit lapses, i.e., more than two unstressed syllables in a foot:*(x̀xn>2). Äpfel und ˈBirnen would also be a violation of well-formed rhythm:  *(x̀x̀). ˈ ˈ Rhythmic well-formedness prohibits stresses on adjacent syllables, since Modern German is not a quantity language.

3.  “ˈ”: sentence stress.

Jespersen’s Cycle and the issue of prosodic ‘weakness’ 

With rhythm evolving in the sentence, secondary stress is the outcome of an interaction between phonology and syntax (cf. Hyman 1977; Vennemann 1986).4 In Standard German, well-formed prosodic feet are either quantity-insensitive dactyls (x́xx) or trochees (x́x). Secondary stress is sensitive to rhythmic context, to syllable structure, and to vowel height (cf. Noel Aziz Hanna 2003). In Modern German, the negation particle nicht ‘not’ can bear secondary stress (cf. 2):5 (2) Das ˈhabe ich ˈnicht geˈsagt. that have I not said ‘I didn’t say that.’

Example (2) displays nicht with rhythmic stress. The particle nicht is rhythmically prominent, because it follows the dactyl habe ich and precedes the unstressed prefix ge-. Therefore, according to the well-formedness condition which prohibits lapses, nicht must form the rhythmic head of the foot. Rhythmic well-formedness in this sentence demands the stressing of the negation particle. As will be demonstrated on the basis of the corpus, the weakness of the old negation particle results from the fact that its chances to attract secondary stress become increasingly rare.

3.  Th  e weakness of the negation particle in Old High German and Middle High German 3.1  Corpus The corpus consists of the Old High German Evangelienbuch by Otfrid von Weissenburg (9th century, South-Rhenish Franconian, approximately 66,000 words) and the Middle High German Nibelungenlied (13th cent.; High Alemannic copies of south Bavarian models, manuscripts A, B, and C; approximately 237,000 words). Both the Old High German and the Middle High German texts form part of an SQL-data base.6 4.  Within this framework, secondary stress is not a type of word stress, but refers to the strong part of the foot. Since the distribution of secondary stress in polysyllabic words can be manipulated by changing the rhythmic context of the word (e.g. by creating stress clash situations with adjacent words), secondary stress is part of the domain of the sentence (Noel Aziz Hanna 2003). .  “ˈ”: secondary/rhythmic stress. 6.  SQL (= Structured Query Language) is a language for relational database management systems. The data base with texts starting from the 8th century was built in order to investigate the serialisation of categories in German and the diachronic changes in serialisation in the left sentence periphery.

 Patrizia Noel Aziz Hanna

There were two reasons for choosing the Evangelienbuch. While commenting on double negation, Otfrid writes in his letter Ad Liutbertum that he follows everyday speech. The other reason is Otfrid’s diligent use of diacritics marking elided vowels and stress. This makes it possible to analyse his analysis of the prosodic weakness of OHG ni and thus his view of prosodic well-formedness. The three manuscripts A, B, and C of the Nibelungenlied show strict metrical organisation, which allows a comparison of sentence rhythm in parallel lines. Meters are not just a condition a poet has to meet. They evolve from everyday speech and stylise speech rhythm (cf. Miller 1902; Vennemann 1995; Noel Aziz Hanna 2006). The comparison of metrically strong positions offers the possibiliy to infer whether the negation particle can form the head of a foot or not, and if yes, under which conditions this is possible. Placement in this position is equivalent to the particle being stressed in a sentence. As will be demonstrated, MHG ne is avoided as the head of a rhythmic foot.

3.2  Th  e negation particle ni in Otfrid von Weissenburg’s Evangelienbuch (9th cent.) This section investigates Otfrid’s rhythmic assessment of the negation particle ni. It focuses both on information structural and on segmental characteristics of the Old High German negation particle.

3.2.1  Stressed ni in Otfrid’s Evangelienbuch In his Evangelienbuch, Otfrid von Weissenburg marks stress with an acute. He gives two lines with the stressed negation particle ni and 23 cases of stressed nist ‘is not’ (cf. 3). (3) Examples of stressed ní in Otfrid’s Evangelienbuch7 “Ní dúemes”, quádun se, “lés! wértisal thes wérkes; (Otfrid, Evb4.28.11) ‘They said: “Oh, let us not destroy this cloth here’ ní sie sih ginérien joh scóno giwerien. (Otfrid, Evb2.22.11-12) ‘[The birds do not lack anything], neither food nor clothes.’ mit mánagfalten éhtin níst iz bi unsen fréhtin. (Otfrid, Evb1.1.68) ‘[A country] with many treasures. It has not been our merit.’

Instances like this provide evidence that ni can be stressed in Old High German. Table 2 lists the instances of diacritically marked stressed and unstressed ni.

7.  For glosses see appendix.

Jespersen’s Cycle and the issue of prosodic ‘weakness’ 

Table 2.  Statistics: ni with and without acute in Otfrid’s Evangelienbuch word


ni ní níst nị

1111 2 23 6

The absolute numbers are self-explanatory. Stressed occurrences of ni are infrequent, which is unsurprising since the negation particle is a function word and thus only rarely carries sentence stress.


3.2.2  Elision of i in Otfrid’s Evangelienbuch Unlike the Proto-Indo-European negation particle *ne, the Old High German negation particle has i as a nucleus. This fact contributes to the further reduction of the negation particle. Otfrid placed subscript dots under elided vowels. Figure 1 gives the absolute numbers of words with elided 〈a, e, i, o, u〉.





i elided vowels



Figure 1.  Elided vowels: High frequency of elided i in Otfrid’s Evangelienbuch

The figure shows that i is the vowel which is most frequently elided. The vowel i is also the least frequently stressed vowel relative to its occurrence, cf. figure 2. Although a single text by a single author does not prove that segmental reasons support the reduction of the negation particle, both observations do fit a general tendency. Otfrid has the pronominal forms mo (instead of imo) and nan (instead of inan); the forms occur even in non-elision contexts (er nan L 55; 79; I. 4, 26; 5, 58; 25, 24; stunt nan V. 15, 23; lês nan IV. 27, 18; cf. Kelle 1963:  326, 327 and Braune/Eggers 1987:  §283.2). This is peculiar, since Old High German had initial stress, which should prevent reduction in initial syllables.

 Patrizia Noel Aziz Hanna 30000

20000 words

vowel graphemes: acute vowel graphemes: no diacritics




e i o vowel graphemes


Figure 2.  Orthography: Low frequency of stressed i in Otfrid’s Evangelienbuch

Another supporting argument for the weak status of the vowel i in Germanic is High Vowel Deletion (cf. Hogg 2000). Sievers, in his edition of the Old High German Tatian (1892: LXI–LXII), points out the syncopation of i, e.g. arougta (‘show:3sg.prt’) instead of arougita and gifulte (‘fill:ptcp.prt’) instead of gifullite. Even in Modern Standard German, the weakness of i is observable. Vennemann’s Naked Ultima Default proves the segmental share in stress distribution. Simplex words with a naked ultima and with a penult ending in a high vowel are usually not accented on the penult (Vennemann 1990: 408). The stressing of another syllable is preferred, e.g. ˈKaries ‘caries’, Idiˈot ‘idiot’, ˈJaguar ‘jaguar’ instead of *Kaˈries, *Iˈdiot, and *Jaˈguar.8 The avoidance of stressing open syllables ending in high vowels can also be found in secondary stress assignment. ˈChefgeˈrontoˈloge ‘chief gerontologist’ with secondary stress on -ron- is normal, while ˈCheforˈnitoˈloge with secondary stress on -ni- is unusual (Noel Aziz Hanna 2003).9 The vowel i is frequently elided and responsible for open syllables remaining unstressed. Phonetically, i is the vowel with the highest intrinsic pitch and, in this respect, the weakest vowel.10 Since the Old High German negation particle ni ends with this vowel, the segmental make-up of this monosyllable contributed to its reduction.

8.  The disprefered stress pattern occurs, for instance, in Maˈria, Thaˈlia, Kaˈlua. 9.  A similar effect was described by Kenstowicz (1994) for Mordwin, Mari, Kobon, Chukchee, and Aljutor. 10.  In general, high vowels are shorter than lower vowels (cf. Lehiste 1970). In addition, there is a relation between vowel height and the perception of stress accent (cf. e.g. Hitchcock & Greenberg 2001 for American English who demonstrate that high vowels are “far more likely to lack accent entirely”).

Jespersen’s Cycle and the issue of prosodic ‘weakness’ 

3.3  Th  e rhythmic embedding of ne in versions A, B, and C11 of the Nibelungenlied (13th cent.) Middle High German shows an even weaker negation particle with respect to the Old High German situation. The comparison of the Nibelungenlied manuscripts aims at revealing the conditions under which the negation particle can form the head of a metrical foot and thus also of a prosodic foot. The variants of the Nibelungenlied provide the possibility to check the scribes’ different evaluations of prosodic well-formedness in terms of Middle High German stressed ne.

3.3.1  Middle High German prosody Old High German and Middle High German are accent-based quantity languages. Like Classical Latin, they have long and short syllables, but unlike Latin, long (heavy) and short (light) syllables are distinguished only if the syllable is stressed (cf. Dresher/ Lahiri 1991; Vennemann 1995). All unstressed syllables are light, irrespective of their syllable structure. This is of direct consequence for the rhythmic embedding of the negation particle ne in sentences. Both in Middle High German prosody and in Middle High German metrics, a minimal foot (Fmin) consists of a stressed syllable comprising two moras. It is filled, for instance, by a stressed closed monosyllable: ˈwîp ‘dame’, a stressed syllable with a diphthong: ˈhie ‘here’, or a stressed open syllable with a long vowel ˈdâ ‘there’. A minimal foot can also consist of two syllables, i.e., a stressed short syllable and an adjacent unstressed syllable which together form the minimal foot (resolution), for instance ˈmu˘get ‘(you [2pl]) can’, ˈzü˘ge ‘(you [2sg]) drew, reared’. Besides minimal feet, there are extended feet (Fext) which are filled by three moras (i.e., they have an additional unstressed syllable): ˈlie.ber, ‘dear’, ˈnâ.hen ‘close’, ˈkü˘.ne.ge ‘kings’. 3.3.2  Metrical organisation of the Nibelungenlied The tetrametric organisation of the Nibelungenlied makes it possible to observe whether the negation particle ne ever forms the head of a foot. Each of the eight halflines consists of four metra (cf. 4). Each metron is either filled by a minimal or by an extended foot. There is also a metrical speciality in the Nibelungenlied – the last foot of

11.  A, Hohenems-Münchener Handschrift, Cod. germ. 34, Bayerische Staatsbibliothek München, second half of the 13th century; B, St. Galler Handschrift, Ms. 857, Stiftsbibliothek St. Gallen, second half of the 13th century; C, Hohenems-Laßbergische or Donaueschinger Handschrift, Ms. 63, Hofbibliothek zu Donaueschingen, probably first half of the 13th century (cf. Becker 1977).

 Patrizia Noel Aziz Hanna

the on-verse is frequently filled by a parasitic foot. Parasitic feet are monomoric and only occur at the endings of on-verses, e.g. (vin)den in the final on-verse in (4) (for a discussion of parasitic feet in terms of naturalness cf. Noel Aziz Hanna/Lindner/ Dufter 2002). Anacrusis is optional, and every second half-line but the fourth ends in a pause. (4) A strophe from the Nibelungenlied (A: 361): Si |sprach: “vil | lieber | pruo| der, | ir | mohtet | noch be|stân | Ø A Fext Fext Fmin Fpar A Fext Fext Fmin Ø vnd | wurbet | ander | frov| wen | - daz | hiez ich | wol ge| tân | - Ø A Fext Fext Fmin Fpar A Fext Fext Fmin Ø vnde | dâ iv | niht en| stvn|de | en |wâge | sô der | lîp. | Ø A Fext Fext Fmin Fpar A Fext Fext Fmin Ø ir | muget hie | nâhen | vin|den | |ein als |hôch ge|born |wîp.” | A Fext Fext Fmin Fpar Fext Fext Fmin Fmin ‘She spake: “O dearest brother, / still here thou mightest stay, And woo another woman— / that were the better way— Where so sore endangered / stood not thus thy life. Here nearer canst thou find thee / equally a high-born wife.”’  (Translation by Needler 1904)

Since all metrical and prosodic feet are left-headed, the investigation concentrates on ne as the first syllable of a metron. This metrical position is equivalent to ne bearing stress in a sentence. Otfrid’s treatment of the negation particle in his Evangelienbuch (cf. section 3.2. above) provided evidence that the negation particle does not regularly attract secondary stress in Old High German. This is true to an even higher degree for Middle High German.

3.3.3  The rhythmic problem of the negation particle ne Both in Old High German and in Middle High German, the negation particle in the most frequent constellation precedes a stressed verb-initial syllable. In this rhythmic context, ne can never form the head of the foot. This follows from the fact that ne does not form a foot on its own. Even if ne was stressed, it would still just be a monomoric syllable (cf. Table 3a). As a default, ne forms a minimal or an extendend foot in combination with another head (cf. Table 3b, c). In order to form the head of a foot, ne has to resolve with a right-adjoining unstressed syllable (cf. Table 3d). However, as a comparison of the manuscripts A, B, and C of the Nibelungenlied shows, there are considerable signs of caution in placing the negation particle in head position.

Jespersen’s Cycle and the issue of prosodic ‘weakness’ 

Table 3.  MHG ne in its rhythmic contexts12 (a)









moras foot

1 -

3 Fext

2 resolved Fmin

2 resolved Fmin


The question which poses itself is whether stressed ne is documented at all in the data. Example (5), manuscript C, gives a line with emphatic ne.13 In this case, ne forms the head of the foot and resolves with the (unstressed) verbal prefix ge-.

(5) Emphatically stressed ne (A: 1680.2, B: 1739.2, C: 1782.2)

A daz ich der Niblunge B daz ich hort der Nibelvnge

| hortes | nie ge|pflach. | Ø Fext Fext Fmin Ø | nie | nie ge|pflach. | Ø Fmin Fext Fmin Ø

C deich hort der Nibelunge | nie|ne ge|pflach. | Ø Fmin Fmin Fmin Ø ‘that I did not take care of the hoard of the Nibelungs anymore’

Unambiguous examples of stressed ne, however, are rare in the data. Whenever ne is placed in front of a verb which starts with a stressed syllable, ne is integrated in the preceding foot, cf. (6).14

(6) ne as part of the preceding foot (A: 1691.4, B: 1750.4, C: 1794.4)


ich lâze ivch daz beschowen,

daz | ich ge|logen | niene | hân.”

A Fext

Fmin Fext


12.  “ˈ”: stressed syllable; x: unstressed syllable (i.e. light syllable, 1 mora); ˈxL: stressed light syllable (1 mora); Fmin: minimal foot (2 moras), Fext: extended foot (3 moras). 13.  The emphatic use is supported by Manuscript B 1739, which has nie nie. 14.  The negation particle ne is not always enclitic, e.g. (a) Nibelungenlied A: 112 Ich ne wils niht erwinden, sprach der kuone man ‘I do not want to desist from it, said the brave man’. (b) Nibelungenlied C: 2381 owe daz vor leide niemen sterben ne mac ! ‘Dear me, how is it possible that no one can die of sorrow’. Encliticisation would be ineffective in a), because ne cannot be integrated in the preceding syllable for reasons of consonantal strength. In b) encliticisation does not take place in order to avoid the inconvenient assimilation of n.

 Patrizia Noel Aziz Hanna

B ich lâze ivch daz wol schowen, daz | ich ge|logen | nine | han.” A Fext Fmin Fext Fmin C ich lâze ivch daz wol schowen, daz | ich ge|logen | niene | han.” A Fext Fmin Fext Fmin ‘I will show you that I did not lie’

Another way of dealing with unstressed ne is its deletion. In (7), manuscript A, this option is chosen instead of double negation.

(7) Deletion of ne (A: 809.2, B: 863.2, C: 874.2)


|ia ge|diende | Sî|frit | Fmin Fext Fmin Fmin

nie alsolhen haz,

B |iane ge|diente | Sî|frit | nie alsolhen haz, Fext Fext Fmin Fmin C |iane ge|diente | Sî|vrit | nie alsolhen haz,”15 Fext Fext Fmin Fmin ‘Yes, Siegfried has never deserved such hatred’

If the verb following the negation particle begins with an unstressed prefix, ne can form the rhythmic head of a resolved foot (cf. Table 3d above). However, constellations in which ne forms the head of a foot are avoided in the Nibelungenlied. Solutions are in accordance with Middle High German prosody. Example (8) shows denuclearisation, i.e., the loss of the reduced vowel, instead of stressed ne. Denuclearisation is the most frequent and also the easiest way of avoiding stressed ne – an n is simply added to the preceding word. In (8) the denuclearization of the negation particle is mandatory. Otherwise, the foot |niene ge| would either contain four moras (option I), which violates rhythmic well-formedness in everyday speech, or the line would contain 5 metra (option II), which is unmetrical in the Nibelungenlied. Trisyllabic anacrusis (option III) is both unlikely because manuscripts A and B do not have it and because the anacrusis would contain more than three moras. Denuclearisation (option IV) gives a scansion which is both metrical and in line with the scansion of the other manuscripts. (8) Denuclearisation: loss of the reduced vowel (A: 131.3, B: 130.3, C: 133.3) A vnd ovch in ein diu frovwe, die | er noch | nie ge|sach, | Ø A Fext Fext Fmin Ø

15.  Depending on sentence stress, there is the alternative iâ/iâne (Paul/Wiehl/Grosse 1989 §23, 5), which results in the following scansions: A: |iâ ge| (Fext), B/C: |iân[e] ge| (Fext) with denuclearisation (cf. example 8).

Jespersen’s Cycle and the issue of prosodic ‘weakness’ 

B vñ ovch in ein div frovwe, di | er noh | nie ge|sach, | Ø A Fext Fext Fmin Ø C vñ ovch in ein div frowe, die | er noch | niene ge|sach,” | Ø *4 moras *Option I A Fext Fmin Fmin Fmin Ø vñ ovch in ein div frowe, die | er noch | nie|ne ge|sach,” | Ø * 5 metra *Option II A Fext Fmin Fmin Fmin Ø vñ ovch in ein div frowe, die er noch | nie|ne ge|sach,” | Ø ? t risyllabic anacrusis ?Option III A Fmin Fmin Fmin Ø vñ ovch in ein div frowe, die | er noch | niene ge|sach,” | Ø denucleari Option IV A Fext Fext Fmin Ø sation ‘And also the very lady, whom he had never seen, [esteemed] him’

However, the solution of denuclearisation is not always possible, because the integration of n depends on the consonantal strength of the final phoneme of the preceding syllable. For instance, a form ichn ‘I not’ does not offer a solution to the problem, because, just like ich ne, it consists of two syllables. The data from the Nibelungenlied with its various strategies of getting ne out of strong metrical positions demonstrate that there was a certain unease in Middle High German of stressing the negation particle. Typical resolution constellations with ne as a head were not realised. Instead prosodic or syntactic reorganisation took place. Reduced syllables other than the negation particle quite naturally form the heads of feet in the Nibelungenlied (cf. 9, manuscript B):

(9) Rhythmic stress on reduction syllables (A: 2232.4, B: 2292.4, C: 2354.4):

A ezn dorft chvnich so ivnger | nimmer | chvener | sin ge|wesen. Fext Fext Fext Fmin B ezen dorfte kvenech so iunger | nimmer | chue|ner ge|wesen. Fext Fmin Fext Fmin C vor dem ivngen kvnige; |niemen | dorfte | chvner | wesen. Fext Fext Fext Fmin ‘no young king could ever have been braver’

In this light, the rhythmic treatment of the negation particle is a special case. If reduced syllables can be rhythmically prominent, why is there an avoidance of stressing the negation particle? Part of the answer to the puzzle dates back to Early Old High German.

3.4  Reduction as a causing factor for Jespersen’s Cycle? The Old High German form ni is already the result of a reduction of PIE *ne. The vowel i indicates the pretonic reduced syllable. It can be compared to the Old High German

 Patrizia Noel Aziz Hanna

prefixes gi- and fir- instead of ga- und fur- (Braune/Eggers 1987: § 70). Thus ni proves in connection with the Proto-Indo-European form that the negation particle had been reduced prior to the oldest German texts. The OHG i in ni attests an intermediate state between the full vowel and the deletion of the vowel.16 The reduction of the particle is a gradual process (10): (10) The gradual reduction of the negation particle: PIE *ne > OHG ni > MHG ne/en > NHG Ø

The reduction of the negation particle preceded the initiation of Jespersen’s Cycle by centuries. Early Old High German had mononegation. Although the already weak negation particle was not strengthened, mononegation was stable. It follows that the phonetic reduction of the negation particle which resulted from its preverbal placement did not set Jespersen’s Cycle in motion. There were additional factors which led to a further weakening of the negation particle.

3.5  C  onspiracy: the weakening of the negation particle and the strengthening of V2 The ongoing reduction of the negation particle is the result of a conspiracy, a concurrence of unfavourable factors. It is caused by yet another interaction between syntax and phonology. There are 156 cases of ni in Otfrid’s Evangelienbuch at the absolute beginning of a sentence (cf. 11): (11) Old High German sentences with initial ni Ni bist es ío giloubo, sélbo thu ịz ni scówo; (Otfrid, Evb1.18.7) ‘You will never believe it as long as you have not seen it [the paradise] yourself.’ Níst si so gisúngan, mit régulu bithuúngan (Otfrid, Evb1.1.35) ‘It has not been sung [in Franconian] this way; the language does not yet obey the rule.’

In Middle High German, however, the negation particle ne very rarely starts the sentence, i.e., only in V1 sentences. In the Nibelungenlied, there is not a single sentence in versions A, B, and C with ne at the absolute beginning of a sentence. In a language with initial stress, the first syllable of a sentence is likely to be prominent. However, the negation particle ne is not accepted as the pre-field of a V2 sentence and thus does not occur in this rhythmically prominent position. 16.  The realisation of PIE e as OHG i is a reduction, because i is both higher and shorter than e (intrinsic duration). Cf. also Hitchcock/Greenberg (2001) who state that “Vowel reduction phenomena […] may merely represent a conflation of stress accent, vowel height and duration.”

Jespersen’s Cycle and the issue of prosodic ‘weakness’ 

The reason for the positional change of the negation particle from Old High German to Middle High German is the stabilisation of V2 in the declarative sentence.17 As a consequence, Middle High German ne does not occur in a position in which the Old High German negation particle could be stressed. Therefore, the strengthening of V2 adds to the reduction of ne. In comparison to the Proto-Indo-European negation particle, the Middle High German particle has three disadvantages: 1) Since it occurs preverbally, it is usually followed by a stressed syllable. 2) It cannot fill the pre-field on its own. 3) It is a monosyllabic function word which does not usually carry rhythmic stress. In a system of accent-based quantity, these factors lead to the deletion of the syllable.

4.  Summary The paper started with three questions: 1. What does ‘weakness of the German negation particle’ mean in terms of stress categories (sentence stress, main stress, secondary stress)? 2. Why is the negation particle weakened? 3. What are the specific criteria that caused the increasing reduction of the negation particle? The avoidance of secondary stress placement was a gradual development. I suggest the following answers to these questions: 1. The negation particle is weakened because it does not attract secondary stress. Neither sentence stress restrictions nor main stress restrictions play a role in this development. 2. The phonetic reduction alone is not the catalyst for Jespersen’s Cycle, as the Early Old High German mononegation shows. The negation particle is weakened because of an interplay of rhythmic, segmental and syntactic influencing factors. 3. The reasons for the avoidance of rhythmic stress on the negation particle are: a. Rhythmic well-formedness on the level of the sentence: The negation particle does not form a foot on its own and rarely resolves with the right-adjacent syllable because of its preverbal position. b. Vowel height: There is a tendency to avoid secondary stress on open syllables ending in a high vowel. This avoidance includes OHG ni. 17.  V1 remained as the pattern of imperative sentences and interrogative sentences.

 Patrizia Noel Aziz Hanna

c. The strengthening of V2: The syntactic process of restructuring is directly associated with the further reduction of the negation particle, because the negation particle cannot fill the Middle High German pre-field. Jespersen’s Cycle is the result of an interaction between syntax and phonology. The outcome of this interaction was that the chances for the negation particle to attract rhythmic stress decreased in the history of German. As a consequence, the inherited negation particle became ineffective.

Appendix: Glosses

(3) Examples of stressed ní in Otfrid’s Evangelienbuch

“Ní dúemes”, quádun se, “lés! not do:opt.ps.1pl say:3pl.pret they:nom.pl interj.

(Otfrid, Evb4.28.11)

wértisal thes wérkes; damage:acc.sg this:gen.sg work:gen.sg ‘They said: “Oh, let us not destroy this cloth here”’ ní sie sih ginérien not they:nom.pl themselves nourish:3pl.opt.ps

(Otfrid, Evb2.22. 12)

joh scóno giwerien. and beautifully dress:3pl.opt.ps ‘[The birds do not lack anything], neither food nor clothes.’ mit mánagfalten éhtin with many:dat.pl.f treasures:dat.pl

(Otfrid, Evb1.1.68)

níst iz bi unsen fréhtin. not.is:3sg.ps it:nom.sg by our:dat.1pl.f merit:dat.pl ‘[A country] with many treasures. It has not been our merit.’

(4) A strophe from the Nibelungenlied (A: 361):

Si sprach: “vil lieber pruoder, She:nom.sg.f said:3sg.prt most dear:nom.sg.m brother.nom.sg ir mohtet noch bestân you:2.pl.nom can:2pl.conj18.prt still stay.inf ‘She spake: “O dearest brother, / still here thou mightest stay

18.  The umlaut is not indicated (Bavarian manuscript).

Jespersen’s Cycle and the issue of prosodic ‘weakness’ 

vnd wurbet ander frovwen and woo:2pl.conj.prt other:gen.sg.f mistress:gen.sg daz hiez ich wol getân that:acc.sg.n call:1sg.conj.prt I:nom.sg well done:ptcp.prt ‘And woo another woman— / that were the better way—’ vnde dâ iv niht en-stvnde if there you:dat.pl not not-stand:3.sg.conj.prt enwâge sô der lîp. at.risk so the:nom.sg.m life:nom.sg ‘Where so sore endangered / stood not thus thy life.’ ir muget hie nâhen vinden you:2.pl.nom can:2pl.prs here near find.inf ein als hôch geborn wîp.” a:acc.sg.n equally high born:ptcp.prt woman:acc.sg ‘Here nearer canst thou find thee / equally a high-born wife.”’

(5) Emphatically stressed ne (A: 1680.2, B: 1739.2, C: 1782.2)

A daz ich der Niblunge that.conjunction I:nom.sg the:gen.pl Nibelung:gen.pl hortes nie gepflach. hoard:gen.sg never care:1.sg.prt B daz ich hort der Nibelvnge that.conjunction I:nom.sg hoard:acc.sg the:gen.pl Nibelung:gen.pl nie nie gepflach. never never care:1.sg.prt C deich hort der Nibelunge that.conjunction.I:nom.sg hoard:acc.sg the:gen.pl Nibelung:gen.pl niene gepflach. never.not care:1.sg.prt ‘that I did not [never ever] take care of the hoard of the Nibelungs anymore’

(6) ne as part of the preceding foot (A: 1691.4, B: 1750.4, C: 1794.4)

A ich lâze ivch daz beschowen, I:nom.sg let:1sg.prs you:acc.pl that:acc.sg.n see:inf daz ich gelogen niene hân.” that.conjunction I:nom.sg lie:ptcp.prt never.not have:1sg.prs

 Patrizia Noel Aziz Hanna

B ich lâze ivch daz wol schowen, I:nom.sg let:1sg.prs you:acc.pl that:acc.sg.n certainly see:inf daz ich gelogen nine han.” that.conjunction I:nom.sg lie:ptcp.prt never.not have:1sg.prs C ich lâze ivch daz wol schowen, I:nom.sg let:1sg.prs you:acc.pl that:acc.sg.n certainly see:inf daz ich gelogen niene han.” that.conjunction I:nom.sg lie:ptcp.prt never.not have:1sg.prs ‘I will show you that I did not lie’

(7) Deletion of ne (A: 809.2, B: 863.2, C: 874.2)

A ia gediende Sîfrit nie alsolhen haz, yes deserve:3sg.prt Sîfrit never such:acc.sg.m hatred:acc.sg B iane gediente Sîfrit nie alsolhen haz, yes.not deserve:3sg.prt Sîfrit never such:acc.sg.m hatred:acc.sg C iane gediente Sîvrit nie alsolhen haz,” yes.not deserve:3sg.prt Sîvrit never such:acc.sg.m hatred:acc.sg ‘Yes, Siegfried has never deserved such hatred’

(8) Denuclearisation: loss of the reduced vowel (A: 131.3, B: 130.3, C: 133.3)

A vnd ovch in ein diu frovwe, and also him:acc.sg.m very the:nom.sg.f lady:nom.sg die er noch nie gesach, whom:acc.sg he:nom.sg yet never see:3sg.prt B vñ ovch in ein div frovwe, and also him:acc.sg.m very the:nom.sg.f lady:nom.sg di er noh nie gesach, whom:acc.sg he:nom.sg yet never see:3sg.prt C vñ ovch in ein div frowe, and also him:acc.sg.m very the:nom.sg.f lady:nom.sg die er noch niene gesach,” whom:acc.sg he:nom.sg yet never.not see:3sg.prt ‘And also the very lady, whom he had never seen, [esteemed] him’

(9) Rhythmic stress on reduction syllables (A: 2232.4, B: 2292.4, C: 2354.4):

A ezn dorft chvnich so ivnger it:3sg.n.not can:3sg.conj.prt king:nom.sg so young:nom.sg.m nimmer chvener sin gewesen. never brave:nom.sg.comp be:inf be: ptcp.prt

Jespersen’s Cycle and the issue of prosodic ‘weakness’ 

B ezen dorfte kvenech so iunger it:3sg.n.not can:3sg.conj.prt king:nom.sg so young:nom.sg.m nimmer chuener gewesen. never brave:comp.sg be: ptcp.prt C vor dem ivngen kvnige; beyond the:dat.sg young:dat.sg king:dat.sg niemen dorfte chvner wesen. nobody can:3sg.prt brave:nom.sg.comp be:inf ‘no young king could ever have been braver’ (10) Old High German sentences with initial ni Ni bist es ío giloubo, not be:2sg.prs it:gen.sg.n ever believing:nom.sg.m

(Otfrid, Evb1.18.7)

sélbo thu ịz ni scówo; self:nom.sg.m you:nom.sg it:acc.sg.n not beholder:nom.sg ‘You will never believe it as long as you have not seen it [the paradise] yourself.’ Níst si so gisúngan, not-is:3sg.prs she:nom.sg.f so sung:ptcp.prt

(Otfrid, Evb1.1.35)

mit régulu bithuúngan with rule:dat.sg master:ptcp.prt ‘It has not been sung [in Franconian] this way; the language does not yet obey the rule.’

References Becker, P.J. 1977. Handschriften und Frühdrucke mittelhochdeutscher Epen. Wiesbaden: Reichert. Behaghel, O. 1932. Deutsche Syntax. Vol. IV. Heidelberg: Carl Winters Universitätsbuchhandlung. Bolinger, D. 1972. Accent is predictable (if you’re a mind reader). Language 48: 633–644. Braune, W. & Eggers, H. 1987. Althochdeutsche Grammatik, 14th Edn. Tübingen: Niemeyer. Detges, U. 2003. La grammaticalisation des constructions de négation dans une perspective onomasiologique, ou: La déconstruction d’une illusion d’optique. In Kognitive romanische Onomasiologie und Semasiologie, A.  Blank & P.  Koch (Eds), 213–233. Tübingen: Niemeyer. Donhauser, K. 1996. Negationssyntax in der deutschen Sprachgeschichte: Grammatikalisierung oder Degrammatikalisierung? In Deutsch – typologisch, E. Lang & G. Zifonun (Eds), 201–217. Berlin: Walter de Gruyter. Dresher, B.E. & Lahiri, A. 1991. The Germanic foot: Metrical coherence in Old English. Linguistic Inquiry 22: 251–286. Dufter, A. 2003. Typen sprachrhythmischer Konturbildung [Linguistische Arbeiten 475]. Tübingen: Niemeyer.

 Patrizia Noel Aziz Hanna Eisenberg, P. 2006. Das Wort. Grundriss der deutschen Grammatik. Stuttgart: Metzler. Ernst, G. 1985. Gesprochenes Französisch zu Beginn des 17. Jahrhundert. Direkte Rede in Jean Héroards “Histoire particulière de Louis XIII”. Tübingen: Niemeyer. Firbas, J. 1989. Degrees of communicative dynamism and degrees of prosodic prominence (weight). Brno Studies in English 18: 21–66. Giegerich, H.J. 1985. Metrical Phonology and Phonological Structure:  German and English. Cambridge: CUP. Hitchcock, L. & Greenberg, S. 2001. Vowel height is intimately associated with stress accent in spontaneous American English discourse. Proceedings of the 7th European Conference on Speech Communication and Technology (Eurospeech-2001) 1, P. Dalsgaard, B. Lindberg, H. Benner & Z. Tan (Eds), 79–82. Aalborg. Hogg, R.M. 2000. On the (non-)existence of High Vowel Deletion. In Analogy, levelling, markedness, A. Lahiri (Ed.), 353–376. Berlin: Mouton de Gruyter. Hyman, L. 1977. On the nature of linguistic stress. In Studies in Stress and Accent [Southern California Occasional Papers in Linguistics 4], L. Hyman (Ed.), 37–82. Los Angeles CA: Department of Linguistics, University of Southern California. Jespersen, O. 1917. Negation in English and other languages [Historisk-filologiske Meddelelser I,5]. Kopenhagen: Høst. Kelle, J. 1963. Otfrid von Weißenburg: Evangelienbuch. Vol.  1. Regensburg:  Manz. (First published 1856). Kenstowicz, M. 1994. Sonority-driven stress. Ms., MIT. Pp. 28. ROA-33. Rutgers Optimality Archive, http://roa.rutgers.edu/. Kuryłowicz, J. 1945. Le changement accentuel dans la langue française du XVIe siècle. Bulletin Linguistique 13: 39–45. Lehiste, I. 1970. Suprasegmentals. Cambridge MA: The MIT Press. Miller, C.W.E. 1902. The relation of the rhythm of poetry to that of the spoken language with special reference to ancient Greek. In Studies in Honor of Basil L. Gildersleeve. Baltimore MD: Johns Hopkins Press. Das Nibelungenlied. Paralleldruck der Handschriften A, B und C nebst Lesarten der übrigen Handschriften, edited by M.S. Batts 1971. Tübingen: Niemeyer The Nibelungenlied 1904. Translated into rhymed English verse in the metre of the original by G.H. Needler. New York NY: Holt. Noel Aziz Hanna, P. 2003. Sprachrhythmus in Metrik und Alltagssprache. Untersuchungen zur Funktion des neuhochdeutschen Nebenakzents [Studien zur Theoretischen Linguistik 15]. Paderborn: Wilhelm Fink Verlag. Noel Aziz Hanna, P. 2006. Integrating quantitative meter in non-quantitative metrical systems: The rise and fall of the German hexameter. Ars Metrica 1. Noel Aziz Hanna, P., Lindner, K. & Dufter, A. 2002. The meter of nursery rhymes:  Universal versus language-specific patterns. In Sounds and Systems. A Festschrift for Theo Vennemann, D. Restle & D. Zaefferer (Eds), 241–267. Berlin: Mouton de Gruyter. Otfrid von Weißenburg, Evangelienbuch. Edited by O.  Erdmann [Altdeutsche Textbibliothek 49]. 6th Edn. by D. Wolff 1973. Tübingen: Niemeyer. Paul, H. Wiehl, P. & Grosse, S. 1989. Mittelhochdeutsche Grammatik. Tübingen: Niemeyer. Saran, F. 1907. Deutsche Verslehre. München: Beck. Tatian. Lateinisch und altdeutsch mit ausführlichem Glossar. Edited by E.  Sievers. 2nd Edn. [1892]. Reprint Paderborn 1966: Schöningh.

Jespersen’s Cycle and the issue of prosodic ‘weakness’ 

Vennemann, T. 1974. Topics, subjects, and word order:  From SXV to SVX via TVX. In Historical Linguistics: Proceedings of the First International Congress of Historical Linguistics, Edinburgh, September 1973, J. Anderson & C. Jones (Eds), Vol. II., 339–376. Amsterdam: North Holland. Vennemann, T. 1986. Neuere Entwicklungen in der Phonologie. Berlin: Mouton de Gruyter. Vennemann, T. 1990. Syllable structure and simplex accent in Modern Standard German. In Papers from the Regional Meeting of the Chicago Linguistic Society 26, Vol. II:  The Parasession, M. Ziolkowski, M. Noske & K. Deaton (Eds), 399–412. Chicago IL: Chicago Linguistic Society. Vennemann, T. 1993. Language change as language improvement. In Historical Linguistics: Problems and Perspectives, C. Jones (Ed.), 319–344. London: Longman. (Reprint). Vennemann, T. 1994. Universelle Nuklearphonologie mit epiphänomenaler Silbenstruktur. In Universale phonologische Prozesse und Strukturen [Lingustische Arbeiten 310], K.H. Vater & H. Wode (Eds), 7–54. Tübingen: Niemeyer. Vennemann, T. 1995. Der Zusammenbruch der Metrik im Spätmittelalter und sein Einfluß auf die Metrik. In Quantitätsproblematik und Metrik. Greifswalder Symposion zur germanischen Grammatik [Amsterdamer Beiträge zur Älteren Germanistik 42], H.  Fix (Ed.), 185–223 Amsterdam: Rodopi.

Holmberg’s Generalization Blocking and push up Hans Broekhuis

Leiden University Center for Linguistics Holmberg’s (1999) formulation of Holmberg’s Generalization states that Scandinavian object shift cannot cross any phonologically realized VP-internal material. This correctly predicts that object shift may not apply in, e.g., embedded clauses in Danish: since in these languages V-to-I applies in main clauses only, the main verb occupies a VP-internal position in embedded clauses, and object shift would therefore violate HG. Generally, this is considered the end of the story, but it is not as HG can in principle be satisfied in two ways: either the verb blocks object shift, or object shift pushes the verb up into the I-position. A full explanation therefore requires an answer to the question of why the latter option is not chosen in Danish.

1.  The problem This article adopts as its point of departure the version of Holmberg’s Generalization (HG) in (1), according to which Scandinavian object shift (OS) cannot cross a VP-internal verb or VP-internal arguments, but it can cross their traces (as well as adverbial phrases).1

(1) Holmberg’s Generalization (Holmberg 1999: 15): Scandinavian OS cannot apply across a phonologically visible category asymmetrically c-commanding the object position except adjuncts (≈ cannot cross a VP-internal verb or argument).

Now consider the examples in (2), taken from Vikner (2006), which show that Danish has pronominal OS in main but not in embedded clauses. Since Danish has no V-to-I in embedded clauses, this is precisely what we expect on the basis of (1): OS is allowed in main clause (2a) since it only crosses the trace of the main verb, which has been

1.  In Swedish, but not the other Scandinavian languages, OS can also be blocked by verbal particles; cf., e.g., Holmberg (1999) and Vikner (2006: section 2.1.2). I will not go into this here and refer to Broekhuis (2008: section 3.2.4) for discussion and analysis.

 Hans Broekhuis

moved via the I-position into C, whereas OS is excluded in embedded clause (2b) since it crosses the main verb itself.2 (2) a. Hvorfor læste Peter 〈den〉 aldrig [VP tV 〈*den〉]? why read Peter   it never b. Jeg spurgte why Peter 〈*den〉 aldrig [VP læste 〈den〉]. I asked why Peter  it never read

Generally, this is considered the end of the story, but it is not since HG could in principle be satisfied in two ways: either the verb blocks OS, as in the example (2b), or OS pushes the verb up into the empty I-position. (3) a. Blocking strategy: a VP-internal main verb blocks OS. b. Push-up strategy: OS forces V-to-I of the main verb.

Option (3b) gives rise to a hypothetical language that behaves like Danish, but in which OS forces V-to-I to apply in embedded clauses. Given that Danish has no OS with lexical DPs, this would result in the asymmetry given in (4): V-to-I applies in embedded clauses when the object is a weak pronoun, but not when it is a lexical DP. (4) • Hypothetical language: “Danish” with push up of the main verb a. Jeg spurgte hvorfor Peter aldrig [VP læste den.her bog] I asked why Peter never read this book b. Jeg spurgte hvorfor Peter læste den aldrig [VP tV tDO]. I asked why Peter read it never

Since example (4b) satisfies HG, we must answer the question of why Danish prefers the blocking strategy in (2b) to the push-up strategy in (4b). A possible answer is that this reflects an inherent property of the language system. One option would be that the preference for blocking follows from the architecture of the language system. This comes close to the proposal by Holmberg (1999), who adopts the standard assumption that syntactic derivations are cyclic and adds to this the claim that any step in the derivation must satisfy HG in (1). Holmberg notes that these conditions cannot be simultaneously satisfied if OS is a syntactic operation: applying OS before V-to-I results in a violation of HG, and applying V-to-I before OS results in a counter-cyclic derivation. Holmberg therefore concludes that OS is a post-syntactic operation. From

2.  It is not categorically impossible to move an object across the main verb: wh-movement and Topicalization are possible, and the same holds for Neg/Q-movement (movement of negated and certain quantified DPs). These movements differ from OS in that they arguably involve A′-movement instead of A-movement. For more discussion on wh-movement and Topicalization, see Holmberg and Platzack (1995), and for Neg/Q-movement, see Svenonius (2000), Christensen (2005) and Broekhuis (2008: section 4.3.2).

Holmberg’s Generalization 

this, it may follow that OS is inherently incapable of forcing syntactic operations like V-to-I as it applies in a grammatical module that can only be accessed after the application of the rules of core syntax. Push up is consequently excluded for principled reasons, so that blocking is the only remaining option. However, there is good reason to doubt that OS is a post-syntactic operation: Chomsky (2001: 15) has pointed out that it is not expected for post-syntactic rules to be semantically restricted, so that the fact illustrated in (5) below that Icelandic OS can only apply to presuppositional material leads to the conclusion that OS is a syntactic operation after all. It also implies that we must drop Holmberg’s assumption that HG is a condition on derivations, and assume that it is rather a filter on output representations. A plausible alternative hypothesis of a more syntactic nature takes recourse to the notion of economy, and relies on the fact that the push-up strategy involves application of an otherwise unforced movement:  in Danish, V-to-I need not apply in embedded clauses and therefore its application in (4b) is blocked by economy. The discussion of the Icelandic examples below will show that this hypothesis cannot be maintained either. Of course, Icelandic does not involve push up of the finite verb given that V-to-I is obligatory in Icelandic for independent reasons (cf. section 4), but we can observe the push-up strategy in double object constructions. First, consider the examples in (5). The first two examples show that OS of lexical DPs is “optional” in Icelandic, the choice between them being dependent on the information structure of the clause: in (5a) the DP is part of the focus (‘new’ information) of the clause, whereas in (5b) it is part of the presupposition of the clause. Example (5c) shows that pronoun shift is obligatory. (5) • Icelandic (core data) a. Jón las ekki þessa bók. Jón read not this book

(þessa bók ⊂ focus)

b. Jón las þessa bók ekki tDO. Jón read this book not

(þessa bók ⊂ presupposition)

c. Jón las 〈hana〉 ekki 〈*hana〉 Jón read  it not

(hana = pronoun)

These examples lead to certain expectations for Icelandic double object constructions. The most interesting case is the one in (6) with an indirect object DP and a pronominal direct object. First, assume that the indirect object is part of the presupposition of the clause. The judgments are then exactly as expected on the basis of the observations in (5): both the indirect object and the pronominal object will undergo OS, as in (6c). (6) a. *Pétur sýndi oft Maríu hana. Pétur showed often Maríu it b. *Pétur sýndi Maríu oft tIO hana. c. Pétur sýndi Maríui hanaj oft tIO tDO. d. *Pétur sýndi hanaj oft Maríu tDO.

 Hans Broekhuis

We do not expect, however, that (6c) must also be used when the indirect object is part of the focus of the clause: the observations in (5) rather lead to the expectation that in this case the indirect object prefers to stay in its base position, whereas the direct object undergoes OS. Both preferences can be satisfied simultaneously by shifting the direct object across the indirect object, as in (6d), but this is excluded by the formulation of HG in (1). This leaves us with the two alternatives in (7): the blocking strategy wrongly selects example (6a) as the grammatical one, whereas the push-up strategy correctly selects (6c). This shows that, in this case, it is the push-up strategy that has the desired result.

(7) a. Blocking strategy: a VP-internal non-presuppositional indirect object blocks OS of a pronominal direct object. b. Push-up strategy: OS of a pronominal direct object forces OS of a VP-internal non-presuppositional indirect object.

The discussion above unequivocally shows that the hypothesis that push up is blocked by considerations of economy cannot be maintained. Blocking and push up are both available: failure to apply V-to-I to the main verb in Danish blocks OS, whereas OS of a pronominal direct object forces OS of a non-pronominal indirect object in Icelandic. This article will investigate what determines which of the two strategies is applied in which case. The proposed analysis contains three basic ingredients. First, HG in (1) will be derived by assuming that Scandinavian OS is subject to a number of shape conservation constraints (cf. Williams 2003; Müller 2000/2001, and Fox & Pesetsky 2005). Secondly, I will propose a maximally simple theory of V-to-I that accounts for some of the core facts in the Germanic languages. Thirdly, I will formulate a proposal concerning the question of how order preservation and V-to-I interact. The analysis is phrased in terms of the derivation-and-evaluation analysis of OS in Broekhuis (2000/2008), which will be reviewed in section 2.3

2.  The derivation-and-evaluation model This section briefly introduces some background assumptions adopted in the derivation-and-evaluation (D&E) framework; cf. Broekhuis (2008) for a more elaborate discussion. In much current linguistic theorizing, the focus of attention is onesidedly restricted either to the derivation (as in most minimalist studies) or to the evaluation of the output of the derivation (as in most optimality-theoretic work). 3.  Given the space limitations, it is neither possible to do justice to the vast literature on OS nor to discuss the relation between OS and Dutch/German Scrambling. I refer to Broekhuis (2008) for extensive discussion of these issues.

Holmberg’s Generalization 

The basic claim of the D&E framework is that in order to arrive at a descriptively and explanatorily adequate theory, it is needed to integrate both aspects into a single overarching framework. In order to obtain this, Broekhuis and Dekkers (2000) have argued in favor of the model in Figure 1, in which some version of the computational system CHL functions as a generator, which produces an output that is evaluated in an optimality-theoretic manner.



Output representations


Optimal output

Figure 1.  The derivation-and-evaluation model

Just as in most minimalist work, CHL consists of at least two operations, which are subject to inviolable conditions. The first operation is external merge that selects/copies some element from the lexicon and merges it with some other lexical element or with some syntactic object already formed.4 I will refer to this operation as Select, and follow Hornstein (2001) in assuming that it is subject to last resort, that is, triggered by some unsaturated θ-feature. The second operation is internal merge that selects/copies some element from the structure already formed and merges it to the root of that structure. I will refer to this operation as Move and follow the standard assumption that it is subject to last resort (triggered by some unvalued formal feature), locality conditions, the extension condition, etc. The main difference between D&E and the ‘standard’ versions of the minimalist program is that the former claims that CHL is a truly autonomous system in the sense that it is not sensitive to strength/epp-features that force or block the application of a certain operation; in fact, it is claimed that there are no such features. At any point in the derivation, then, CHL may randomly choose between applying or not applying the operation(s) that could in principle be performed (= satisfy the last resort condition), as in Figure 2. Consequently, the number of candidates in the candidate set is therefore at most 2n, where n is the number of operations that satisfy Last Resort. The actual number of candidates even radically decreases when we adopt phase theory: when we have a single cycle with 16 operations that satisfy last resort, this derives at most 65.536 (= 216) candidates; however, when we divide this cycle into 4 phases of 4 operations each, the number of candidates is at most 64 (= 4x24), provided, of course, that we have some form of cyclic optimization.

4.  See Broekhuis and Klooster (2007) and Broekhuis (2008) for empirical reasons against the claim that selection of lexical elements involves the construction of a numeration and in favor of the assumption that lexical items are taken from the lexicon directly.

 Hans Broekhuis

Operation 1 yes


Operation 2 yes



Operation 3 yes



no yes

no no



Figure 2.  The construction of the candidate set

The advantage of assuming that CHL restricts the number of candidates in the candidate set is that this enables us to drastically reduce the universal constraint set CON postulated by OT. Furthermore, since the candidates differ from each other in a small number of predictable ways (see Figure 2), we may also expect the number of constraint types to be rather small. Let us therefore assume that the OT-evaluator is actually a formalization of the interface conditions postulated in the minimalist program. If so, the syntactic constraints in CON must be related to the three components involved: the computational system CHL, which creates the relevant syntactic representations in the candidate set, and the articulatory-perceptual and the conceptual-intentional systems, which interpret these representations. Consequently, the syntactic constraints are of the two basic types in (8a&b). (8) The syntactic constraints in CON a. CHL constraints: (i) epp constraints: probe F attracts its goal (cf. (9)). (ii) Economy constraints: constraints on internal (or external) merge (cf. (10)). b. Interface constraints: (i) LF constraints: semantic constraints on Move (cf. (12)). (ii) PF constraints: constraints on e.g., linearization (cf. (15)) and deletion.

As is shown in (8a), the CHL constraints come in two subtypes. Constraints of the first subtype favor movement, and will be referred to as epp constraints since they all require that some unvalued formal feature attracts its goal. The epp constraints used below are given in (9). The constraints epp(φ) and epp(case) define two instances of OS: section 3 will briefly discuss OS triggered by the unvalued φ-features on V, and section 5 will discuss OS triggered by the unvalued case features on v. The constraint *stray feature will enter the analysis of verb movement in section 4. (9) epp constraints: a. epp(φ): unvalued φ-features attract their goal. b. epp(case): an unvalued case-feature attracts its goal. c. *stray feature: amalgamate formal features of the functional heads with the root they are associated with (e.g., unvalued verbal features on v, Asp and I attract their goal).

Holmberg’s Generalization 

The second subtype of CHL constraints are economy constraints that disfavor movement. The constraints used in the discussion below are taken from Grimshaw (1997) and given in (10). (10) Economy constraints: a. *move (stay): don’t apply internal merge. b. nolexm: don’t apply internal merge to lexical (θ-role assigning) verbs.

Word order variation between languages can now be accounted for by assuming that the epp and economy constraints interact in an OT-fashion. Ranking (11a), for example, expresses that the case feature on v (normally) does not trigger OS, because it is more important to satisfy the economy constraint *move than epp(case). This ranking will be called ‘weak’, since it is more or less equivalent to saying that the case feature is weak or has no epp-feature associated with it. Ranking (11b) expresses that the unvalued case feature (normally) triggers OS, because it is more important to satisfy epp(case) than *move. This ranking will be called ‘strong’, since it is more or less equivalent to saying that the case feature is strong or has an epp-feature associated with it. The rankings in (11a&b) thus distinguish languages like Icelandic, in which the case feature on v normally triggers OS, from languages in which this is not the case. An important advantage of this formalization of “feature strength” is that we do not categorically block or force movement: even under the weak ranking OS can be forced provided that there is some higher ranked constraint a that favors it (cf. (11a′)), and even under the strong ranking OS can be blocked provided there is some higher ranked constraint b that disfavors it (cf. (11b′)). (11) a. *move >> epp(case) (Weak ranking: case features normally do not trigger OS) a′. a >> *move >> epp(case)  (If a favors OS, ‘Procrastinate’ is overruled) b. epp(case) >> *move (Strong ranking: case features normally trigger OS) b′. b >> epp(case) >> *move (If b disfavors OS, ‘Strength’ is overruled)

The weak and strong rankings are typically overruled by the interface constraints in (8b):  the role of a and b in (11a′&b′) can be respectively performed by the LF-constraints in (12a&b). The constraint d-pronoun in (12a) essentially adapts Diesing’s (1997) claim that definite pronouns are variables that cannot occur in the domain of existential closure.5 The constraint alignfocus, taken from Costa (1998),

5.  I am passing over at least two important issues here. First, there are also proposals that attribute the tendency of pronouns to shift to their phonological weakness (cf. Vogel 2006), which would make d-pronoun a PF-constraint. It is not really important for the present discussion which proposal is the correct one, given that we would be dealing with an interface condition in both cases. Second, the pronoun of course moves into the local domain of v, so that one could dispute the claim that its landing site is vP-externally: what is intended here is that the pronoun is moved across the base position of the subject.

 Hans Broekhuis

formalizes the well-known observation that new information tend to occur in the right periphery of the clause. (12) LF constraints: a. d-pronoun: a definite pronoun must be vP-external: *[vP … pron[+def] … ]. b. alignfocus (af): the prosodically unmarked focus is the rightmost constituent in its clause.

First, consider the case in (11a′), which can be illustrated by means of the Danish examples in (13). Broekhuis (2000) argued that Danish OS in (13a) is blocked by the weak ranking *move >> epp(case). The fact that OS is nevertheless possible when the object is a definite pronoun is due to the fact that *move is outranked by d-pronoun, which requires that the pronominal object be vP-external. (13) Danish: d-pronoun >> *move >> epp(case) a. Hvorfor læste studenterne 〈*artiklen〉 ikke 〈artiklen〉? Why read the.students  the.article not b. Hvorfor læste studenterne 〈den〉 ikke 〈*den〉? why read the.students  it not

This shows that we can readily account for the fact that languages differ to the extent that they exhibit OS:  languages like Icelandic allow OS both with pronominal and lexical DPs due to the fact that the have a strong ranking of epp(case), languages like Danish have the ranking d-pronoun >> *move >> epp(case) and therefore allow OS of pronouns only, and languages like Finnish Swedish do not have any form of regular OS because *move outranks both epp(case) and d-pronoun. This gives rise to the following macro-parameterization.

*MOVE >> EPP(case) No full object shift

>> *MOVE Pronoun shift: Danish



>> *MOVE Full object shift: Icelandic

*MOVE >> D-PRONOUN No object shift: Finnish-Swedish

Figure 3.  Macro-parameterization of languages with respect to OS

Now, consider the case in (11b′), which can be illustrated by means of the Icelandic examples in (5a&b), repeated here in a slightly different form as (14), in which the angled brackets indicate alternative placements of the object. By assuming that alignfocus outranks epp(case), we account for the fact that OS is excluded when

Holmberg’s Generalization 

the object is part of the focus of the clause given that the high ranking of alignfocus requires it to be the rightmost constituent in its clause. (14) Icelandic: af >> epp(case) >> *move a. Jón las 〈*þessa bók〉 ekki 〈þessa bók〉. (þessa bók ⊂ focus) b. Jón las 〈þessa bók〉 ekki 〈*þessa bók〉. (þessa bók ⊂ presupposition)

We will see later that the PF-constraints in (15a&b) also block OS that would normally apply if this results in distorting the underlying word order: this will derive HG in (1). The constraint in (15c) may force verb movement that is otherwise not allowed: we will see that this constraint accounts for the asymmetry that some languages exhibit with respect to V-to-I in embedded and main clauses. (15) PF constraints: a. h-compl: a head precedes all terminals dominated by its complement. b. Relativized Minimality (relmin): If the foot of X-chain α c-commands the foot of X-chain β, the head of X-chain α c-commands the head of X-chain β (where X-chain = A-, A′-, or head chain). c. lexically fill top F (lftf): the highest head position in an extended projection must be lexically filled.

3.  VO- versus OV-languages and short OS The shape conservation constraints in (15a) and (15b) prohibit changing the underlying order of respectively heads and their complements, and of the arguments. These constraints imply some version of Kayne’s (1994) “universal base” hypothesis, and this, in turn, raises the question of how so-called VO- and OV-languages differ. This will be the topic of this section.

3.1  The universal base hypothesis For reasons discussed in Broekhuis (2008), I will follow Kayne (1994) in assuming that the universal order is specifier-head-complement. Furthermore, I will adopt Hale and Keyser’s (1993) proposal that verbs consist of a root and some verbalizing element: I will refer to the verbalizing element that combines with finite verbs as v, and to the verbalizing element that combines with participles as Asp. The base orders for present/past and perfect tense constructions are as given in (16a) and (16b), respectively. Finally, I follow Broekhuis (2000) in assuming that the root V/aux and the light verb v are associated with different nominal features: V/aux has unvalued φ-features, whereas v has unvalued case-features (cf. Chomsky 2005, who adopts a similar claim in the guise of feature inheritance).

 Hans Broekhuis

(16) • Universal base hypothesis a. [… I … [(S) … v[ucase] [… V[uφ] OBJ]]] b. [… I … [(S) … v[ucase] [… aux[uφ] [… Asp [… V[uφ] OBJ]]]]]

Since unvalued features act as probes, (16) implies that we must distinguish two types of OS. The first type is triggered by the case features on v, and since it is this type that is normally discussed in the literature on OS, I will refer to it as regular OS. The second type is triggered by the unvalued φ-features on the verbal roots V/aux; since the landing site of this type of OS is lower than that of regular OS, I will refer to this type as short OS. This section focuses on short OS, since it can readily be seen from (16) that the question of whether a certain language surfaces with a VO- or an OV-order depends on the application of this movement and V-to-v/Asp, that is, the ranking of the epp constraints epp(φ) and *stray feature in (9a&c) with respect to the economy constraint *move.

3.2  Germanic VO-languages (simple tense constructions) The constraints epp(φ), *stray feature and *move can in principle be ranked in six different ways, and five of these rankings will derive the VO-order. These five rankings are given in (17), where {a,b} expresses that the rankings a >> b and b >> a give rise to the same result. The rankings in (17) define three alternative ways of deriving the VOorder. The first option is to assume that the economy constraint *move outranks both epp(φ) and *stray feature: this predicts that V-to-v and short OS are both blocked. The second option is to assume that *move is outranked by *stray feature but not by epp(φ): this predicts V-to-v to be possible but short OS to be excluded. The third option is to assume that the two epp constraints both outrank *move: this predicts that V-to-v and short OS apply both. Finally, assume that VP-adverbs are adjoined to the projection of the verb root V, and that both V-to-v and short OS cross these adverbs (Chomsky 1995a: 329ff.). The ranking in (17a) then predicts that both the verb and the object will follow the VP-adverb, (17b) that the VP-adverb appears in between the verb and the object, and (17c) that the verb and the object both precede the VP-adverb, and (18) shows that (17c) makes the correct predictions for English.6

6.  I have nothing to say about the question of what determines whether a certain adverbial phrase is used as a VP- or as a sentential adverbial: for our present purpose it suffices that adverbs that are normally assumed to be ‘low’ (like manner, time and place adverbials) must follow both the verb and the object, whereas those that are assumed to be ‘high’ (like modal and many frequency adverbials) precede them. This means that the question of why every day rather behaves like a time than like a frequency adverb is far beyond the scope of this paper.

Holmberg’s Generalization 

(17) Germanic VO-languages: 5 potential rankings a. Option I: *move >> {epp(φ), *stray feature} ⇒ [… v adverb V OBJ] b. Option II: *stray feature >> *move >> epp(φ) ⇒ [… v+V adverb tV OBJ] c. Option III: {epp(φ), *stray feature} >> *move ⇒ [v+V OBJ adverb tV tOBJ] (18) a. *that John every day reads books. b. *that John reads every day tV books. c. that John reads books every day tV tDO.

The evaluation of the examples in (18) is given in Tableau 1. Since there are two operations that can in principle apply (in the relevant part of the derivation), there are four (= 22) candidates. The selection of (18c) as the optimal candidate does not depend on the ranking of the constraints epp(φ) and *stray feature, which is indicated in Tableau 1 by placing a dashed line between the two constraints; the angled brackets “>” and “


[v+V adv [tV O]]


[v [O adv [V tO]]] [v+V [O adv [tV tO]]]

* *!


* **

There are additional reasons to assume that option III, with short OS and V-to-v, is the correct one for English. For example, Johnson (1991) has argued on the basis of the placement of the object in particle verb constructions like John looked 〈the information〉 up 〈the information〉 that English has some sort of object movement that moves the object in front of the particle. Overt OS has also been proposed by Lasnik (1999a/1999b) on the basis of examples like The DA proved the defendants to be guilty

7.  The present proposal also correctly predicts that sentential objects must follow the postverbal (low) adverbs:  since sentential objects do not have φ-features, they are not suitable goals for the unvalued φ-features on the root V so that they must remain in their base position following the VP-adverbs; the strong ranking of *stray feature forces V-to-v and as a result the adverb appears in between the verb and the sentential object: that John said yesterday tV that he will come. The proposal does, however, not predict that PP-complements may either precede or follow the VP-adverbs in English.

 Hans Broekhuis

during each other’s trials: the subject of the infinitival clause must be shifted into the matrix clause in order to be able to bind the anaphor each other embedded in the adverbial phrase of the matrix clause. A cursory look at the position of the VP-adverbs in the Scandinavian languages suggests that they behave just like English. For this reason, I will adopt as a working hypothesis that all Germanic VO-languages are like English in having obligatory short OS, and hence that they all have the partial constraint ranking in (17c).

3.3  Germanic OV-languages (Dutch/German) Example (17) gives us five out of the six possible rankings of the constraints epp(φ), *stray feature and *move. This leaves the ranking in (19), and it is this ranking that gives rise to the Germanic OV-languages. The subranking epp(φ) >> *move expresses that the φ-features of V trigger short OS, whereas the subranking *move >> *stray feature expresses that V-to-v normally does not apply. The evaluation in Tableau 2 shows that this correctly predicts that the nominal object is obligatorily placed in a position preceding the clause-final verb.8 (19) Germanic OV-languages: epp(φ) >> *move >> *stray feature a. *dat Jan leest dat boek. that Jan reads that book b. dat Jan dat boek leest tDO. Tableau 2: OV-languages (simple tense) epp(φ) [v [V O]]


[v+V [tV O]]


[v [O [V tO]]]

[v+V [O tV [tV tO]]]


*stray feature *

* *



So far we have only considered examples with simple tenses. The Dutch examples in (20) show that in perfect tense examples the object must also precede the auxiliary.

8.  The present proposal also correctly predicts that sentential objects do not precede the verb; since sentential objects do not have φ-features, they are not suitable goals for the unvalued φ-features on the root V so that they must remain in their base position following the clause-final verb: dat Jan zei [S dat hij ziek was] ‘that Jan said that he was ill’. The proposal does, however, not predict that PP-complements may either precede or follow the clause-final verbs in Dutch.

Holmberg’s Generalization 

This follows under the assumption in (16b) that both the root of the main verb and the root of the auxiliary are endowed with φ-features: the strong ranking of epp(φ) does not only force movement of the object into the local domain of the verbal root of V but also into that of the auxiliary. Note that I have added the sentential adverb waarschijnlijk ‘probably’ to these examples in order to exclude the possibility that (20c) is derived by regular OS, which would target a position to the left of this adverb. The proposal that auxiliaries have φ-features is originally due to Broekhuis and Van Dijk (1995), who argue that auxiliaries can also assign case, in more current terminology, that auxiliaries are associated with a light verb v.9 (20) a. *dat Jan waarschijnlijk heeft gelezen dit boek. b. *dat Jan waarschijnlijk heeft dit boek gelezen tDO. c. dat Jan waarschijnlijk dit boek heeft t′DO gelezen tDO. that Jan probably this book has read ‘that Jan has probably read this book.’

The evaluation is given in the following tableau; in order to save space, this tableau does not include candidates with V-to-Asp or aux-to-v, which are all excluded due to the weak ranking of *stray feature, which is also omitted from this tableau. Tableau 3: OV-languages (complex tenses) candidates with V-movement ignored


… v [… aux [… Asp [… V O]]]]


… v [… aux [… Asp [O… V tO]]]]


… v [O … aux [… Asp+V [tO… V tO]]]]


* **

3.4  The Germanic VO-languages again: perfect tense examples The proposed analysis of the perfect tense constructions in Dutch creates a new problem for the analysis of the VO-languages. Since we have seen that these languages also have a strong ranking of epp(φ), we wrongly predict that in these languages the object must also be moved into the local domain of the auxiliary root, and should therefore cross the main verb. This means that short OS must be

9.  Of course, it is also possible to have the participle in the position preceding the auxiliary (which is, in fact, the only option in German). Since the space is simply lacking to discuss this issue properly, I have to refer the reader to Broekhuis (2008: section 5.3) for an analysis of this fact in terms of predicate movement.

 Hans Broekhuis

selectively blocked: the shift in (21b) must be forced, whereas the additional shift in (21b′), which results in a violation of HG in (1), must be blocked. This can be obtained by assuming that the shape conserving constraint h-compl in (15a) is ranked above epp(φ), as in (21). (21) Type A: h-compl >> {epp(φ), *stray feature} >> *move a. [… v+V [OBJ … tV tOBJ]] (simple tense) b. [… v+aux [… taux [… Asp+V [OBJ … tV tOBJ]]]] (perfect tense) b′. *[… v+aux [OBJ … taux [t′OBJ … Asp+V [tOBJ … tV tOBJ]]]]

The evaluations of the simple and perfect tense examples in Tableaux 4 and 5 show that the ranking in (21) allows short OS to apply as long as it does not invert the object and the main verb; this excludes the output representation in (21b′). In order to save space, Tableau 5 only contains the candidates with verb movement; the candidates without verb movement are all excluded due to the strong ranking of *stray feature, which is also omitted from this tableau. Tableau 4: VO-languages, hypothetical type A (simple tense) epp(φ)

*stray feature

[v adv [V O]]



[v+V adv [tV O]]



[v [O adv [V tO]]] [v+V [O adv [tV tO]]]





* **

Tableau 5: VO-languages, hypothetical type A (perfect tense) candidates without V-movement ignored


[v +aux [… taux [… Asp+V [… tV O]]]] [v +aux [… taux [… Asp+V [O… tV tO]]]] [v +aux [O… taux [… Asp+V [tO… tV tO]]]]

 *!






*** ****

Now consider the evaluations in Tableaux 6 and 7. These tableaux show that the introduction of the partial ranking h-compl >> epp(φ) >> *move makes it possible to derive the same result while postulating a weak ranking of *stray feature. Languages with this ranking exhibit the VO-order due to a push-up strategy: epp(φ) >> *move favors short OS; h-compl >> epp(φ) disfavors movement of the object across the verb, and epp(φ) >> *move >> *stray feature favors push up of the verbal root

Holmberg’s Generalization 

V over blocking of short OS. Below I will argue that this VO-type indeed exists and is instantiated by Danish.10 (22) Type B: h-compl >> epp(φ) >> *move >> *stray feature a. [… v+V [OBJ … tV tOBJ]] b. [… v+aux [… taux [… Asp+V [OBJ … tV tOBJ]]]] b′. *[… v+aux [OBJ … taux [t′OBJ … Asp+V [tOBJ … tV tOBJ]]]] Tableau 6: VO-languages, hypothetical type B (simple tense) simple tense


[v adv [V O]]


[v+V adv [tV O]]


[v [O adv [V tO]]]



*stray feature *






[v+V [O adv [tV tO]]] 

Tableau 7: VO-languages, hypothetical type B (perfect tense) h-compl epp(φ) *move *stray feature [v [… aux [… Asp [… V O]]]]


[v [… aux [… Asp+V [… tV O]]]]




[v+aux [… taux [… Asp [… V O]]]]




[v+aux [… taux [… Asp+V [… tV O]]]]














[v [… aux [… Asp [O … V tO]]]] [v [… aux [… Asp+V [O … tV tO]]]]

*! 

[v+aux [… taux [… Asp [O … V tO]]]]


[v+aux [… taux [… Asp+V [O … tV tO]]]]


[v [O … aux [… Asp [tO … V tO]]]]




[v [O … aux [… Asp+V [tO … tV tO]]]]




[v+aux [O … taux [… Asp [tO … V tO]]]]




[v+aux [O … taux [… Asp+V [tO … tV tO]]]]



10.  Type B languages should not have V-to-v when the object remains in situ. In order to account for the fact that V-to-v also applies with sentential complements, we can no longer adopt the assumption from fn.7/8 that these complements remain VP-internally. Broekhuis (2008) solves this problem by following Den Dikken (1995) and Kayne (2005:  ch.11) in assuming that “extraposed” clauses are actually “intraposed” and end up in postverbal position as the result of remnant VP-movement to the left. An analysis along the same line is given for the placement of PP-complements.

 Hans Broekhuis

3.5  Conclusion This section has argued that there are in principle two different types of Germanic VOlanguages. Type A in (21) requires V-to-v/Asp because of a strong ranking of *stray feature. type B in (22) has a weak ranking of *stray feature, but exhibits V-to-v/Asp as a result of push up due to a high ranking of h-compl. The OV-languages also have a weak ranking of *stray feature, but do not exhibit push up of the verbal root due to a low ranking of h-compl.11

4.  V-to-I This section argues that the two types of VO-languages distinguished in Section 3 do indeed occur in Germanic: Icelandic is a language of type A, while Danish is of type B. The reason for this claim is that the two language types can be distinguished by taking into consideration other verb movement operations like V-to-I. Since type A has a strong ranking of *stray feature it is expected that languages of this type (normally) have V-to-I, as is indeed the case in Icelandic. Type B in (22), on the other hand, has a weak ranking of *stray feature, so that it is expected that in languages of this type V-to-I is (normally) blocked; this is the case in Danish, in which V-to-I is more or less restricted to root clauses. Note that the OV-languages also have a weak ranking of *stray feature, and are therefore correctly expected to behave like Danish with respect to V-to-I.12

11.  One of the reviewers raises the interesting question of what the low ranking of h-compl in the Germanic OV-languages implies for categories other than the verb, given that we know that nominal and adjectival phrases are mostly head initial. Of course, the low ranking of h-compl does not entail that projections must be head-final; this will only be the case if movement of the complement across the head is needed for independent reasons. Given that nouns and adjectives normally neither agree with nor assign case to their complement, the latter need not be moved so that they will follow the head. In this connection it is important to note that the exceptional adjectives that do take a nominal complement do end up after it (cf., e.g., Van Riemsdijk 1983): Jan is dat gezeur zat ‘Jan is sick of that nagging’. This would follow if, contrary to what is normally assumed, these adjectives assign structural case to their nominal complement. The big challenge is the word order in PPs: prepositions do assign case to their complements but still may precede them in Dutch and German. At least in Dutch there is evidence that the word order is affected by semantic considerations: directional PPs are mostly postpositional, whereas non-directional PPs are prepositional. It will be clear that this opens up the possibility that at least one additional LF-constraint may be involved that may overrule the strong ranking of epp(case) in this case. 12.  The relative order of epp(φ) and *stray feature has not yet been established for Icelandic, but I will provisionally assume that it is the same as in Danish in order to make

Holmberg’s Generalization 

(23) a. Germanic VO-languages: (i) Icelandic: h-compl >> epp(φ) >> *stray feature >> *move ⇒ V-to-I. (ii) Danish: h-compl >> epp(φ) >> *move >> *stray feature ⇒ no V-to-I. b. Germanic OV-languages: epp(φ) >> *move >> {*stray feature, h-compl} ⇒ no V-to-I.

Since Icelandic has obligatory V-to-I in most contexts, nothing more needs be said about this language, but the fact that Danish and the OV-languages do have V-to-I in root contexts still need to be explained. I propose that this is due to the constraint lexically fill top F in (15c), which requires that the highest head in an extended projection be lexically filled (cf. also Zwart 2001). When this constraint outranks the economy constraint *move, V-second will be forced in main clauses. In embedded clauses, on the other hand, the highest head position in the extended verbal projection (= C) is lexically filled by the complementizer, so that V-to-I is blocked by the weak ranking of *stray feature.13 (24) a. Icelandic: *stray feature >> *move ⇒ V-to-I in all context. b. Danish: lftf >> *move >> *stray feature ⇒ V-to-I in main clauses only. c. Dutch/German: lftf >> *move >> *stray feature ⇒ V-to-I in main clauses only.

Putting aside exceptional cases of embedded V-second, the above proposal captures the basic distinctions concerning V-to-I. However, the proposal raises a new question: Given that Danish is a VO-language of type B, in which short OS pushes V into v, why doesn’t regular OS push up the v+V complex into I in Danish? In order to answer this question, we first have to look at the analysis of regular OS in the D&E framework.

5.  Push up and blocking with regular OS This section will discuss the push up and blocking strategy in OS constructions in Icelandic and Danish. Section 5.1 and 5.2 will start with discussing push up and blocking in double object constructions, and section 5.3 will conclude with discussing the question of why regular OS in Danish does not force V-to-I in embedded clauses.

the difference between the two rankings in (a) as small as possible. Given that *stray feature outranks *move, this seems innocuous. 13.  We must of course still account for the fact that insertion of a lexical complementizer is excluded in root contexts. A plausible assumption is that declarative/interrogative complementizers like that/if must be licensed/selected by some matrix verb. Also note that the notion ‘lexically filled’ in (15c) must be distinguished from the notion ‘phonetically realized’ since the constraint tel from Pesetsky (1998) may result in complementizers that are syntactically present but phonetically empty; cf. embedded wh-questions like I wonder [CP what Ø [IP Bill bought twhat]]].

 Hans Broekhuis

5.1  Icelandic regular OS in double object constructions In Icelandic double object constructions regular OS can derive the orders in (25a–c). The order in (25d) violates HG in (1) and is ungrammatical. We will see below that the actual choice of the word orders in (25) is related to the information structure of the clause. Since we have seen earlier that the relevant properties of Icelandic regular OS can be derived by postulating the subranking af >> epp(case) >> *move, let us see what this subranking predicts for (25).14 (25) a. Pétur sýndi oft Maríu bókina. Pétur showed often Maríu the.book b. Pétur sýndi Maríu oft tIO bókina. c. Pétur sýndi Maríu bókina oft tIO tDO. d. *Pétur sýndi bókina oft Maríu tDO.

When both the direct and the indirect object belong to the presupposition of the clause, the strong ranking of epp(case) predicts that both objects must move, so that (25c) is correctly selected as the optimal candidate. Since alignfocus overrules the strong ranking of epp(case), we predict that OS of the two objects is blocked when both are part of the focus of the clause, so that (25a) is correctly selected as the optimal candidate in this case. The ranking af >> epp(case) >> *move also correctly predicts that (25b) is acceptable: this order arises when the indirect object is part of the presupposition and the direct object is part the focus of the clause. A problem arises, however, when the direct object is part of the presupposition and the indirect object is part of the focus of the clause: the strong ranking of epp(case) now forces movement of the direct object, while alignfocus blocks OS of the indirect object, so that we wrongly predict (25d) to be grammatical. This problem can be solved by adopting the constraint relmin in (15b), which disfavors the output configuration of this example. When we assume that relmin outranks alignfocus, the evaluation is given as in Tableau 8, which correctly selects example (25a) as the optimal candidate. Note that alignfocus is a so-called gradient constraint, which means that a violation is added for each constituent that follows the focused phrases: in the b- and c-candidate the indirect object is followed by the adverbial phrase and the direct object, which results in two violations of alignfocus.

14.  The fact that the indirect object may undergo OS implies that dative is a structural case assigned by the verb. This seems uncontroversial as far as English and the Scandinavian languages are concerned, given that the indirect object may be promoted to subject in passive constructions (Holmberg & Platzack 1995:  215ff.), but in fact the same can be shown for Dutch and German, in which it is normally the direct object that is promoted to subject in passive constructions; see Broekhuis and Cornips (1994) for discussion.

Holmberg’s Generalization 

Tableau 8: Icelandic (IO in focus; DO not in focus) af

epp (case)

a. Subj I+V Adv tV IO DO 



b. Subj I+V IO Adv tV tIO DO



c. Subj I+V IO DO Adv tV tIO tDO


rel min

d. Subj I+V DO Adv tV IO tDO

*move (tO only) * **




Tableau 8 shows that violation of relmin is avoided by means of the blocking strategy, that is, by not applying OS of the direct object. When the direct object is a pronoun, on the other hand, it is clear that the alternative push-up strategy is used. This can be readily observed from (26): if pronoun shift were blocked by an indirect object that is part of the focus of the clause, we incorrectly predict (26a) to be acceptable next to (26c), which would then only arise in case the indirect object is part of the presupposition of the clause. (26) a. *Pétur sýndi oft Maríu hana. Pétur showed often Maríu it

◀ * blocking ▶

b. *Pétur sýndi Maríu oft tIO hana. c. Pétur sýndi Maríu hana oft tIO tDO. d. *Pétur sýndi hana oft Maríu tDO.

◀ √ push up ▶ ◀ Holmberg’s Generalization ▶

The fact that in the case of pronouns the push-up strategy is used can be readily accounted for by assuming that the constraint d-pronoun, which forces regular OS of weak pronouns, is ranked higher than alignfocus, which disfavors regular OS of an indirect object that is part of the focus of the clause. Tableaux 9 and 10 provide the evaluation of the examples in (26). Tableau 9 shows that when the indirect object is part of the presupposition of the clause, the strong ranking of epp(case) forces OS of the indirect object, so that pronoun shift can apply without causing a violation of relmin; (26c) is therefore selected as the optimal candidate. Tableau 9: Icelandic (DO pronoun; IO not in focus) relmin d-pronoun



a. Subj I+V Adv tV IO pron



b. Subj I+V IO Adv tV tIO pron



* **

c. Subj I+V IO pron Adv tV tIO tDO  d. Subj I+V pron Adv tV IO tDO

*move (tO only)




 Hans Broekhuis

Tableau 10 provides the evaluation for cases in which the indirect object is part of the focus of the clause: alignfocus disfavors movement of the indirect object and hence favors the blocking strategy, but this is overruled by the higher ranked constraint d-pronoun which favors the push-up strategy. Consequently, (26c) is again selected as the optimal candidate. Tableau 10: Icelandic (DO pronoun; IO in focus) relmin d-pronoun



a. Subj I+V Adv tV IO pron




b. Subj I+V IO Adv tV tIO pron





c. Subj I+V IO pron Adv tV tIO tDO  d. Subj I+V pron Adv tV IO tDO


*move (tO only) * **



This section has shown that, due to the subranking alignfocus >> epp(case), a non-presuppositional indirect object blocks regular OS of a presuppositional direct object (cf. Tableau 8), whereas due to the d-pronoun >> alignfocus a pronominal direct object forces regular OS of a non-presuppositional indirect object (cf. Tableau 10). This shows that both the blocking and the push-up strategy are attested.

5.2  Danish regular OS in double object constructions The previous subsection has shown that we can find both the blocking and the push up strategy in Icelandic double object constructions. This section will show that the same holds for Danish double object constructions. But let us first briefly discuss the less problematic cases: we will restrict our attention to examples that contain at least one pronoun given that OS of lexical DPs is normally blocked in Danish due to the weak ranking of epp(case). First consider the examples in (27), in which both the direct and the indirect object are a pronoun. The earlier established subranking d-pronoun >> *move >> epp(case) straightforwardly predicts that the object pronouns must shift, so that (27c) is the optimal candidate. This subranking also straightforwardly predicts that in (28), in which the indirect object is a pronoun and the direct object is a lexical DP, only the former can be shifted. (27) a. *Peter viste jo hende den. Peter showed indeed her it b. *Peter viste hende jo tIO den. c. Peter viste hende den jo tIO tDO. d. *Peter viste den jo hende tDO.

Holmberg’s Generalization 

(28) a. *Peter viste jo hende bogen. Peter showed indeed her the.book b. Peter viste hende jo tIO bogen. c. *Peter viste hende bogen jo tIO tDO. d. *Peter viste bogen jo hende tDO.

The situation is more complex, however, when the indirect object is a lexical DP and the direct object a pronoun, as in (29). The judgments assigned to these examples by Vikner (1989/1990) suggest that there is some uncertainty concerning the grammatical status of the examples in (29a&c). Furthermore, there seem to be disagreement on example (29a) as similar examples are judged fully acceptable by Christensen (2005: 155). Unfortunately, Christensen does not discuss examples like (29c), but since Vikner judges the two examples as equally acceptable, it seems safe to assume that both are grammatical (the same is reported for Swedish in Anagnostopoulou 2003: 128). (29) Danish (judgments from Vikner 1989) a. %Peter viste jo Marie den. Peter showed indeed Marie it b. *Peter viste Marie jo tIO den. c. %Peter viste Marie den jo tIO tDO d. *Peter viste den jo Marie tDO

The conclusion that (29c) is acceptable is remarkable, as this shows that Danish allows OS of lexical DPs in double object constructions, despite the fact that this is categorically blocked in monotransitive constructions.15 The evaluation in Tableau 11 shows that this follows immediately when we introduce the constraint relmin and assume that it outranks *move, which is independently needed to exclude example (29d). The evaluation in Tableau 11 also makes it clear that, just like in Icelandic, the constraint d-pronoun favors the push-up strategy. Tableau 11: Danish (DO-pronoun only) relmin


a. Subj I+V Adv tV IO pron


b. Subj I+V IO Adv tV tIO pron


epp(case) **




c. Subj I+V IO pron Adv tV tIO tDO  d. Subj I+V pron Adv tV IO tDO

*move (tO only)




15.  Tarald Taraldsen has brought to my attention that in Norwegian and some varieties of Swedish, OS of lexical indirect object DPs is possible without push up. This does, however, not hold for Danish (Holmberg & Platzack 1995: 172).

 Hans Broekhuis

What we have not accounted for yet is that example (29a) is also acceptable in Danish. Since this example involves the blocking strategy, and we have seen in the discussion of Icelandic that alignfocus favors this strategy, we may conclude that alignfocus outranks d-pronoun. Tableaux 12 and 13 further show that this will correctly predict the pattern in (29) provided that alignfocus in its turn is outranked by relmin: the blocking strategy in (29a) will be preferred over the push-up strategy when the indirect object is part of the focus of the clause, whereas the push-up strategy in (29c) is preferred when the indirect object is part of the presupposition of the clause. Tableau 12: Danish (DO-pronoun; IO not in focus) relmin



a. Subj I+V Adv tV IO pron


b. Subj I+V IO Adv tV tIO pron


** *



c. Subj I+V IO pron Adv tV tIO tDO  d. Subj I+V pron Adv tV IO tDO

*move epp(case) (tO only)




Tableau 13: Danish (DO-pronoun; IO in focus) relmin





b. Subj I+V IO Adv tV tIO pron



c. Subj I+V IO pron Adv tV tIO tDO


a. Subj I+V Adv tV IO pron

d. Subj I+V pron Adv tV IO tDO


*move epp(case) (tO only) ** *


** *


The conclusion we can draw from this section is that Danish double object constructions exhibit both the blocking and the push-up strategy. The results are summarized in (30). (30) Danish: relmin >> alignfocus >> d-pronoun >> *move >> epp(case) a. d-pronoun >> *move >> epp(case) ⇒ a pronominal direct object forces OS of a presuppositional indirect object DP b. alignfocus >> d-pronoun ⇒ a non-presuppositional indirect object DP blocks OS of a pronominal direct object.

Holmberg’s Generalization 

5.3  V-to-I in Danish embedded clauses Now that we have also discussed regular OS in Danish, we can finally tackle the question of why regular OS does not force V-to-I in (31), despite the fact that short OS does push up the verbal root V into v. (31) a. Jeg spurgte hvorfor Peter aldrig [VP læste den.her bog] I asked why Peter never read this book b. Jeg spurgte hvorfor Peter aldrig [VP læste den] I asked why Peter never read it c. *Jeg spurgte hvorfor Peter læste den aldrig [VP tV tDO]. I asked why Peter read it never

(Danish) ◀ √ blocking ▶ ◀ * push up ▶

Let us first determine the ranking of the constraints needed for our analysis. I want to propose the subranking in (32), which was in large part already independently established above; cf. (23), (24) and (30). The only addition is that epp(φ) outranks alignfocus, which has been independently motivated above by pointing out that short OS is not sensitive to the information structure of the clause. (32) Danish: {h-compl, lftf} >> epp(φ) >> alignfocus >> d-pronoun >> *move >> {*stray feature, epp(case)}

We have seen that the subranking h-compl >> epp(φ) >> *move >> *stray feature correctly predicts that short OS forces V-to-v: we can satisfy the highly ranked constraints h-compl and epp(φ) by applying both short OS and V-to-v at the expense of violating the lower ranked constraint *move. However, for the same reason we expect on the basis of the subranking h-compl >> d-pronoun >> *move >> *stray feature that regular OS of a pronominal object forces v-to-I: we can satisfy the highly ranked constraints epp(φ) and d-pronoun by applying both V-to-I and regular OS at the expense of violating the lower ranked constraint *move. However, this expectation is not borne out. We can solve this problem by assuming that the independently motivated constraint nolexm in (10b), which was introduced by Grimshaw (1997) in order to capture Pollock’s (1989) insight that in English V-to-I is blocked only when the finite verb is a θ-role assigner, is ranked in between epp(φ) and d-pronoun. The evaluation in Tableau 14 shows that the subranking epp(φ) >> nolexm ensures that short OS will push up the verbal root V in order to satisfy h-compl, whereas the subranking nolexm >> d-pronoun will block regular OS, so that the push-up strategy cannot be used to satisfy h-compl. Consequently, the candidate with short OS and V-to-v/Asp, but no regular OS and V-to-I is now correctly selected as the optimal candidate. Note that none of the violations of d-pronoun is fatal. This correctly predicts that when we replace the pronominal object by a lexical DP the same candidate will be selected as the optimal one.

 Hans Broekhuis

(33) a. a′. a′. b. b′. b′′. c. c′. c′′.

C … I … v … V pronoun C … I … v … pronoun V tO C … I … pronoun v … t′O V tO C … I … V+v … tV pronoun C … I … V+v … pronoun tV tO C … I … pronoun V+v … t′O tV tO C … V+v+I … tv tV pronoun C … V+v+I … tv pronoun tV tO C … V+v+I … pronoun tv … t′O tV tO

Tableau 14: V-to-I and regular object shift in Danish embedded clauses h-compl lftf epp(φ) nolexm d-pronoun *move *strayf (33a)






(33b) (33b′) (33b′′) (33c)

* *

*!  *!




















* *!

epp (case)













For completeness’ sake, note that in main clauses (33c′′) will be selected as the optimal candidate. The a- and b-candidates, in which V-to-I fails to apply, are then all excluded by h-compl and lexically fill top F. Since the three remaining candidates violate nolexm to the same extent, epp(φ) and d-pronoun get the final say: the optimal candidate is the one in which both short and regular OS of the pronominal object have applied, as in (34b). Note that when the object is a lexical DP, as in (34a), D-pronoun will not be violated, so that regular OS will be blocked by the weak ranking *move >> epp(case). (34) a. Hvorfor læste Peter 〈*den.her bog〉 aldrig 〈den her bog〉. Why read Peter this book never b. Hvorfor læste Peter 〈den〉 aldrig 〈*den〉. Why read Peter  it never

This concludes our discussion of the Danish verb movement puzzle. The puzzle is solved by ranking the independently motivated economy constraint nolexm in

Holmberg’s Generalization 

between epp(φ) and d-pronoun: the subranking epp(φ) >> nolexm accounts for the fact that short OS pushes up the verbal root V into v, whereas the subranking nolexm >> d-pronoun accounts for the fact that regular OS of pronouns is blocked when V-to-I does not apply for independent reasons.

6.  Conclusion This article has discussed Holmberg’s Generalization in (1), and has shown that there are in principle two ways to satisfy this generalization. Either OS is blocked when it has to cross the main verb or some co-argument, or the moved object forces movement of these elements as well. We have seen that both strategies exist. This paper studied the two strategies and gave an analysis that straightforwardly predicts which strategy applies in which case. The desired effects were derived by the interaction of the PF-constraints h-compl and relmin, which disfavor movement of a direct object across the verb or a co-argument, respectively. Given Kayne’s (1994) Linear Correspondence Axiom, one might expect that there is also a constraint spec-h that favors output representations in which heads follow the terminals dominated by their specifier. This was not investigated here, but if this is indeed the case, the three order preservation constraints spec-h, h-compl, and relmin may conspire such that, among other things, they derive the effects of Koeneman’s (2006) principle of Thematic Shape Conservation (TSC), which requires that within the thematic domain of the clause the underlying order of the thematic categories (= θ-role assigners and assignees) is maintained. As Koeneman points out, the TSC (and therefore also the present approach) immediately accounts for the fact that OS can be blocked by thematic categories, but not by adverbs. This voids the argument found in Holmberg (1999) and Bobaljik (2002) in favor of the claim that adverbs are PF-invisible, and for this reason do not block regular object shift.

References Anagnostopoulou, E. 2003. The Syntax of Ditransitives. Berlin: Mouton de Gruyter. Bobaljik, J.D. 2002. A-chains at the PF-interface: copies and ‘covert’ movement. Natural Language and Linguistic Theory 20: 197–267. Broekhuis, H. 2000. Against feature strength:  the case of Scandinavian object shift. Natural Language and Linguistic Theory 18: 673–721. Broekhuis, H. 2008. Derivations and Evaluations:  Object Shift in the Germanic Languages. Berlin: Mouton de Gruyter. Broekhuis, H. & Cornips, L. 1994. Undative constructions. Linguistics 32: 173–190. Broekhuis, H. & Dekkers, J. 2000. The minimalist program and optimality theory: Derivations and evaluations. In Optimality Theory: Phonology, Syntax and Acquisition, J. Dekkers et al. (Eds), 386–422. Oxford: OUP.

 Hans Broekhuis Broekhuis, H. & Klooster, W. 2007. Merge and Move as costly operations. Groninger Arbeiten zur germanistischen Linguistik 45:  17–37.  http://gagl.eldoc.ub.rug.nl/root/Volume45/ broekhuisklooster/. Broekhuis, H. & Van Dijk, K. 1995. The syntactic function of the auxiliary of time. In Linguistics in the Netherlands 1995, M. Den Dikken & K. Hengeveld (Eds), Amsterdam: John Benjamins. Chomsky, N. 1995a. The Minimalist Program. Cambridge MA: The MIT Press. Chomsky, N. 2001. Derivation by phase. In Ken Hale. A Life in Language, M. Kenstowicz (Ed.), 1–52. Cambridge MA: The MIT Press. Chomsky, N. 2005. On phases. Ms, MIT. Christensen, K.R. 2005. Interfaces. Negation  – Syntax  – Brain. Ph.D. dissertation, University of Aarhus. Costa, J. 1998. Word Order Variation. A Constraint-based Approach. Ph.D. dissertation, University of Leiden. den Dikken, M. 1995. Extraposition as intraposition, and the syntax of English tag questions. Ms, Free University, Amsterdam. Diesing, M. 1997. Yiddish VP order and the typology of object movement in Germanic. Natural Language and Linguistic Theory 17: 369–427. Fox, D. & Pesetsky, D. 2005. Cyclic linearization of syntactic structure. Theoretical Linguistics 31: 1–45. Grimshaw, J. 1997. Projection, heads and optimality. Linguistic Inquiry 28: 373–422. Hale, K. & Keyser, S. 1993. On argument structure and the lexical expression of syntactic relations. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, K. Hale & S. Keyser (Eds), Cambridge MA: The MIT Press. Holmberg, A. 1999. Remarks on Holmberg’s generalization. Studia Linguistica 53: 1–39. Holmberg, A. & Platzack, C. 1995. The role of inflection in Scandinavian syntax. Oxford: OUP. Hornstein, N. 2001. Move! A Minimalist Theory of Construal. Malden MA: Blackwell. Johnson, K. 1991. Object positions. Natural Language and Linguistic Theory 9: 577–636. Kayne, R.S. 1994. The Antisymmetry of Syntax. Cambridge MA: The MIT Press. Kayne, R.S. 2005. Movement and Silence. Oxford: OUP. Koeneman, O. 2006. Shape conservation, Holmberg’s generalization and predication. In Comparative Studies in Germanic Syntax, J.M. Hartmann & L. Molnárfi (Eds), Amsterdam: John Benjamins. Lasnik, H. 1999a. Minimalist Analysis. Oxford: Blackwell. Lasnik, H. 1999b. Chains of arguments. In Working minimalism, S.D. Epstein & N. Hornstein (Eds), Cambridge MA: The MIT Press. Müller, G. 2000. Shape conservation and remnant movement. In Proceedings of NELS 30, A. Hirotani et al. (Eds), 525–539. Amherst MA: GLSA. Müller, G. 2001. Order preservation, parallel movement, and the emergence of the unmarked. In Optimality-theoretic Syntax, G. Legendre et al. (Eds), 113–142. Cambridge MA: MIT Press/ MITWPL. Pesetsky, D. 1998. Some optimality principles of sentence pronunciation. In Is the Best Good Enough?, P. Barbosa et al. (Eds), 337–383. Cambridge MA: The MIT Press/MITWPL. Pollock, J.-Y. 1989. Verb movement, Universal Grammar and the structure of IP. Linguistic Inquiry 20: 365–424. Svenonius, P. 2000. Quantifier movement in Icelandic. In The derivation of VO and OV, P. Svenonius (Ed.), 255–292. Amsterdam: John Benjamins.

Holmberg’s Generalization 

van Riemsdijk, H. 1983. The case of German adjectives. In Linguistic Categories: Auxiliaries and Related Puzzles, Vol. I, F. Heny (Ed.), Dordrecht: Reidel. Vikner, S. 1989. Object shift and double objects in Danish. Working Papers in Scandinavian Syntax 44: 141–155. Vikner, S. 1990. Verb movement and licensing of NP positions in the Germanic languages. Ph.D. dissertation (draft version). Vikner, S. 2006. Object Shift. In The Blackwell companion to Syntax, M.  Everaert & H.  van Riemsdijk (Eds), Oxford: Blackwell. Vogel, R. 2006. Weak function word shift. Linguistics 44: 1059–1093. Williams, E. 2003. Representation Theory. Cambridge MA: The MIT Press. Zwart, J.W. 2001. Syntactic and phonological verb movement. Syntax 4: 34–62.

part iii

Thematic relations and NP realization

The No Case Generalization Halldór Ármann Sigurðsson Lund University

This paper argues that syntax has no case features, case instead being an interpretative feature or features operative in the PF morphology of individual languages, where it overtly distinguishes between arguments (or NPs). The paper also argues that the non-syntactic nature of case is to be expected, given Non-Isomorphism, that is, the fundamental non-isomorphic nature of the derivation. Nonetheless, the different PF case-marking strategies in different languages operate on the basis of common syntactic matching relations, including matching of Voice and marked v (v*, v**). The dependency of structural accusative upon structural nominative (the Sibling Correlation/Burzio’s Generalization) is accounted for in terms of double versus single Voice matching.

1.  Introduction* The question of how syntax relates to morphology and word order, that is to say to pf in general, is one of the truly big questions in linguistic research. GB theory (Chomsky 1981) basically suggested that the category of case was responsible for three central properties of language or at least of languages that are in some sense similar to English: – –

First, by the Case Filter, it should not be possible to spell out an NP that lacks case. Second, NP-movement was suggested to be driven by a ‘case-need’ – unless an NP moved under certain conditions, it would not get case, hence it would violate the Case Filter and be ruled out.

*This paper it is partly based on my presentation at CGSW 22 in Stuttgart, 8–9 June 2007, and it was also presented at the Mini-Case Workshop in Stuttgart, 9 October 2007. Thanks to two anonymous reviewers and to Joan Maling as well as to my audiences in Stuttgart for valuable comments and suggestions. Special thanks also to Terje Lohndal for much appreciated discussions and help. Last but not least, I am grateful to Noam Chomsky for an important and enlightening discussion about the nature of case in the fall 2003. The research for this work was supported by a grant from the Swedish Research Council, VR 421-2006-2086.

 Halldór Ármann Sigurðsson

Third, an NP had to be governed in order to get case. It followed that PRO had to be caseless, as it had to be ungoverned by the PRO Theorem. Thus, absence of case was made responsible for the silence problem posed by PRO, that is to say, the fact that PRO cannot be spelled out as a pronoun.

If this had been on the right track, case would have been a central feature of grammar, both of universal syntax and of the morphology of individual case languages. However, as it turns out a quarter of a century later, case cannot be assumed to have this central position in language. In fact, it has to be eliminated from syntactic theory. Thus, I will here argue that the No Case Ceneralization in (1) holds true (as suggested in Sigurðsson 2009a):

(1) Syntax has no case features

That is, case is an interpretative feature (or features) operated with in pf morphology, where it overtly distinguishes between arguments (or NPs) and enters into disambiguating agreement processes.1 Case-marking is indeed based on the syntactic (and the morphological) computation, but syntax does not operate with or on case, its assignment taking place after transfer to the expressive (externalizing) component, commonly referred to as pf. It follows that syntactic processes, in contrast to morphological processes, cannot be driven by case or operate with case features (for related ideas see e.g., Marantz 2000; McFadden 2004; Platzack 2006; Landau 2006). Legate (2008: 62ff) discusses non-finite contexts in ergative languages (Walpiri, Enga, Hindi) where transitive objects (O) can carry default absolutive case, in contrast to intransitive subjects (S). Legate takes this fact to show that the default absolutive represents abstract nominative on S but abstract accusative on O, and argues that these circumstances provide evidence for abstract Case in syntax. However, the facts she discusses only show that certain contexts are blocked from spelling out certain subjects as opposed to objects.2

1.  Syntactic Agree must be sharply distinguished from morphological agreement (Sigurðsson 2004b, 2006b). 2.  According to Legate (2008:  63, 65), some of the non-finite contexts where absolutive S is disallowed in Walpiri and Hindi allow dative (Walpiri) or genitive (Hindi) subjects. However, both the Walpiri and the Hindi contexts are nominalization constructions, and Legate does not offer any analysis of their strutural properties (i.e., it is not obvious whether or how the ungrammaticality of absolutive S in these constructions is different from the ungrammaticality of John in *John investigation as compared to John’s in John’s investigation). Legate’s presentation of the data is also rather incomplete. No examples showing the ungrammaticality of absolutive S in the Hindi construction are given (suggesting that it may be difficult to come up with relevant minimal pairs, with absolutive vs. genitive S). The Walpiri examples with an ungrammatical absolutive S are said to “become grammatical when the dative case suffix is added” (Legate 2008: 63), but this is not exemplified, that is, no minimal pair is given here either (whereas an example with a dative

The No Case Generalization 

Commonly, claims that case is syntactic are unspecific about what it means to be ‘syntactic’ and hence difficult to assess and discuss. If such a claim only means that morphological case is derivative of the syntactic and the morphological computation, then it is a very weak claim, if indeed it is a claim and not just a trivial factual statement. In the following, I will argue against two more specific and interesting standpoints: (1) that case features are syntactic primitives, input to (and driving) the syntactic derivation, or (2) that case features are discrete syntactic objects ‘produced’ or ‘activated’ in the course of the syntactic derivation, thus accessible to some syntactic processes (and not only to morphological processes). In the approach pursued here (and argued for in previous work), morphology is radically divorced from syntax, interpreting it rather than being part of it.3 More specifically, I argue that linguistic processes are non-isomorphic, and that there can thus be no one-to-one mappings between syntax and morphology (or any other levels or derivational stages in language). If so, syntax cannot operate with or on morphological features, such as ‘nominative case’, ‘1st person’ or ‘past tense’. Morphological categories are pf interpretations of abstract matching relations, such as: ‘an event participant that is identical with the speaker’ and ‘an event time prior to the time of speech’. In all such relations, at least two distinct elements, e.g., ‘event participant’ and ‘speaker’, ‘event time’ and ‘speech time’, ‘NP’ and ‘little v*’, are matched and valued in relation to each other, yielding a single output in morphology: ‘1st person’, ‘past tense’, ‘accusative’, etc. (Sigurðsson 2004a et seq.). In this paper, I will be focusing on case from this perspective, arguing that syntactic structures are not interpreted in terms of case features until in morphology.

transitive subject (A) is given). The facts mentioned by Legate are potentially interesting, but they do not warrant any conclusions about case in syntax. Nominative-accusative languages have certain inflectional paradigms with syncretic nom/acc forms, cf. the Icelandic name Jón.nom/acc as compared to Ólafur.nom, Ólaf.acc. The fact that the form Jón is grammatical as an object form in PRO infinitives but excluded from the subject position (of both infintives and nominalized verb phrases) has no bearing on the question of whether syntax has case features. It demonstrates a well-known and an easily observable fact, namely that the subject position in PRO infinitives cannot be lexicalized regardless of the form of the subject. 3.  I have occasionally referred to this approach as Radically Disentangled Morphology. It has some properties in common with mainstream Distributed Morphology, but it is also rather different from it, assuming ‘lexical interpretation’ of syntax rather than ‘lexical insertion’ and denying that there are any one-to-one mappings between syntax and morphology (the two operating with different elements). Following Sigurðsson 2004b, 2006a, I assume that the ‘initial lexicon’ accessible to and operated on by narrow syntax, i.e., the initial input to Agree and (external) Merge, does not contain any composite items (not even clusters of abstract features), all ‘physical clustering’ taking place in PF. Adopting a fairly standard view, I assume that some abstract feature clustering takes place in syntax, but exactly how grammar divides the ‘segmentation labor’ between syntax and PF is an intriguing and, in my view, a largely open question. I will not discuss it further here.

 Halldór Ármann Sigurðsson

2.  Case, NP-movement, PRO It is a well-known and a widely discussed fact that case has no general positional effects.4 This is especially well-established for Icelandic, due to its quirky subjects (e.g., Zaenen et al. 1985; Sigurðsson 1989, 1991, 2000, etc.; Marantz 2000; McFadden 2004, 2009; Thráinsson 2007). Thus, inherently case-marked NPs are subject to NPmovement in the same manner as structurally case-marked NPs: (2) a. *… að þá mundi hafa verið boðið okkur. … that then would have been invited us.dat


b. … að okkur mundi þá hafa verið boðið. … that us.dat would then have been invited ‘… that we would then have been invited.’ (3) a. *… að þá mundum hafa verið kosnir við. … that then would have been elected we.nom


b. … að við mundum þá hafa verið kosnir. … that we.nom would then have been elected ‘… that we would then have been elected.’

In contrast to definite subjects, indefinite, non-specific subjects often do not raise, regardless of their case:5 (4) a. … að þá mundi hafa verið boðið fjórum demókrötum. … that then would have been invited four Democrats.dat ‘… that there would then have been (some) four Democrats invited.’


b. … að þá mundu hafa verið kosnir fjórir demókratar. … that then would have been elected four Democrats.nom ‘… that there would then have been (some) four Democrats elected.’


As has also been widely discussed in the literature (Andrews 1976; Thráinsson 1979 inter alia, including many of my own works, e.g., Sigurðsson 1989, 2003), Icelandic quirky subjects behave like structurally case-marked subjects with respect to other phenomena that were taken to be driven by case in Government and Binding theory. This is illustrated (in part only) in (5)–(8): 4.  There are well-known positional differences between NPs, PPs and CPs that have commonly been taken to boil down to case in the generative literature (Stowell 1981 and many others, e.g., McClosky 1996). The argument is circular, losing force if one does not adopt the axiomatic assumption that CPs and PPs are caseless (cf. Thráinsson 1979; Sigurðsson 2006b: 29). 5.  The indefinites in (4) can raise, but then they get a specific reading. Bare indefinites, on the other hand, do not generally raise in Icelandic (in contrast to English), an issue that I will put aside here.

The No Case Generalization 

Subject Raising: (5) a. Þá virtist þeim [hafa verið boðið].  then seemed them.dat  have been invited ‘Then they seemed to have been invited.’


b. Þá virtist [hafa verið boðið of mörgum repúblíkönum/*þeim]. then seemed  have been invited too many Republicans.dat/*them. dat ‘Then there seemed to have been too many Democrats invited.’ (6) a. Þá virtust þeir [hafa verið kosnir].  then seemed they.nom  have been elected


b. Þá virtust [hafa verið kosnir of margir repúblikanar/*þeir]. then seemed  have been elected to many Republicans.nom/*they. nom

ECM: (7) a. Ég taldi [þeim hafa verið boðið].  I believed  them.dat have been invited ‘I believed them to have been invited.’


b. Ég taldi [hafa verið boðið of mörgum repúblíkönum/*þeim]. I believed  have been invited too many Republicans.dat/*them.dat ‘I believed there to have been too many Republicans invited.’ (8) a. Ég taldi [þá hafa verið kosna].  I believed  them.acc have been elected ‘I believed them to have been elected.’


b. Ég taldi [hafa verið kosna of marga repúblikana/*þá]. I believed  have been elected too many Republicans.acc/*them.acc ‘I believed there to have been too many Republicans elected.’

The central conclusions that can be drawn from these and related facts are stated in (9): (9) a. Case is irrelevant with respect to NP-movement b. Personal pronouns obligatorily undergo NP-movement

It thus seems that Person is the most important factor triggering NP-movement. This follows in the approach to the computation I have developed in previous work (e.g., Sigurðsson 2004a, 2009a, 2009b).6

6.  This is strongly supported by the quirky agreement facts, so briefly presented below (but thoroughly studied in previous work). Notice that while Person seems to be the central factor triggering NP-movement, it is evidently not the only factor that affects NP-placement. Thus, focus commonly affects it, to an extent, and so does definiteness, it would seem (but distinguishing between Person and definiteness is a nontrivial task, cf. Sigurðsson 2009b). Case, in contrast, is notably far from showing any correlation with NP-placement.

 Halldór Ármann Sigurðsson

Rather than taking facts of this sort at face value, numerous researchers (Belletti 1988; Jónsson 1996; Chomsky 2000, etc.) have assumed that Icelandic quirky case is exceptional and therefore does not really bear on the GB theory of abstract Case. Proponents of this idea, make the simple assumption that Icelandic quirky subjects are assigned invisible or abstract nominative ‘Case’, in addition to their inherent morphological case. This double case approach is not a priori implausible, nor is it particularly abstract or far-fetched. However, it is made suspicious by dat-nom constructions, where the nominative object controls number agreement of the verb: (10) Honum mundu ekki líka þeir. him.dat would.3pl not like they.nom ‘He would not like them.’

In contrast, Icelandic nominative objects never control (unambigouous) 1st and 2nd person agreement, as illustrated in (11):7 (11) a. *Honum mundum ekki líka við. him.dat would.1pl not like we.nom Intended: ‘He would not like us.’ b. *Honum munduð ekki líka þið. him.dat would.2pl not like you.nom.pl Intended: ‘He would not like you.’

The agreement contrast between (10) and (11) is understandable if Icelandic quirky subjects are as ‘subjecty’ as they are partly because they enter into a covert Agree relation with the clausal Person head (Pn), Person thus not being able to also overtly agree with the nominative object (in contrast to Number, Nr, see Boeckx 2000; Sigurðsson & Holmberg 2008 and the references cited there). This is illustrated in (10)′ and (11)′: (10)′ him. would.3 not like they.


Pn / Nr


(11)′ a. *him. would.1 not like we. b. *him. would.2 not like you..



↑__overt__↑ Pn / Nr





Thus, the extra abstract feature matched by raised subjects, quirky as well as nominative, is evidently Person, and not extra case. That, in turn, is not surprising if Person is a computational feature, matched by the subject (regardless of whether the subject

7.  The intended readings can be expressed by several alternative means, different for different predicates (thus the verb líka may take a PP, type/him.dat would like with us.acc/= ‘He would like us’).

The No Case Generalization 

subsequently triggers uninterpretable verb agreement in morphology), whereas case is not assigned until in morphology. See the theory of abstract Person as a computational feature developed in previous work (e.g., Sigurðsson 2004a, 2009b; cf. also Shlonsky 1989; Rizzi 2008). As I have also argued previously (Sigurðsson 1989, 1991, 2002, 2008), Icelandic offers pervasive evidence that PRO is case active (see also e.g., Landau 2006 and Bobaljik & Landau 2008). One piece of evidence showing this comes from case agreement of floating quantifiers, as in (12) (from Sigurðsson 2008); the case agreeing quantifiers are set in boldface, whereas the case agreement trigger is underlined:8 (12) a. Bræðrunum líkaði illa [að PRO vera ekki báðir kosnir]. brothers.the.D.m.pl liked ill  to N be not both.N.m.pl elected ‘The brothers disliked not being both elected.’ b. Bræðurnir æsktu þess [að PRO vera báðum boðið]. brothers.the.N.m.pl wished(for) it.G  to D be both.D.pl invited ‘The brothers wished to be both invited.’

Notice that the matrix subjects are case different from PRO in both examples, that is, the agreeing case of the quantifiers báðir (nom) and báðum (dat) is not transmitted from the matrix clause, instead being triggered by the case of PRO. Parallel facts are found in finite clauses, where the overt (local) subject is the agreement trigger, as shown in (13): (13) a. Bræðurnir voru ekki báðir kosnir í stjórnina. brothers.the.n.m.pl were not both.N.m.pl elected to board.the ‘The brothers were not both elected to the board.’ b. Bræðrunum var báðum boðið á fundinn. brothers.the.D.m.pl was both.D.pl invited.dft to meeting.the ‘The brothers were both invited to the meeting.’

In short, case has no bearing on either NP-movement or the licensing of PRO. That is not surprising if case is not a syntactic category.9 The question of why PRO cannot be spelled out as an overt pronoun is an intriguing and a very important issue that I will however not consider here. It is discussed in Sigurðsson (2008), where it is argued that PRO infinitives (of the canonical

8.  Other forms of the quantifiers, e.g., báðum in (12a) and báðir in (12b), are sharply ungrammatical. The abbreviations used in the examples are capital N, D, G for nominative, dative and genitive case, small capital m for masculine, and pl for plural. 9.  As pointed out in Sigurðsson 2008, the assumption that an NP has to have an ‘unsatisfied’ syntactic case feature in order to be syntactically active is redundant, hence vacuous (i.e., it is non-distinct from saying that the n-feature makes NPs syntactically active or from simply saying that NPs are syntactically active until they have been fully matched).

 Halldór Ármann Sigurðsson

English/Icelandic type) have a defective Person head that cannot be matched by an argument that is lexically φ-specified.

3.  More facts Agreement phenomena are commonly contingent on or triggered by case. If case is not a syntactic category, it follows that case-dependent agreement is not syntactic either. Important conceptual reasons as well as extensive empirical evidence suggests that this is precisely the correct conclusion (Sigurðsson 2004b, 2006a; McFadden 2004; Bobaljik 2006). Like case, however, agreement is based on the syntactic computation, but it does not follow that it takes place in syntax. It takes place in post-syntactic PF morphology. Formal agreement features, like number and person agreement on verbs, are contentless in the sense that they do not contribute anything to interpretation (Chomsky 1995, etc.). Thus, inasmuch as speakers accept non-agreement as in Them is there or The girls is there (see Henry 1995; cf. also Quinn 2005), the absence of verb agreement does not lead to any poorer or different semantics than in the standard They are here, The girls are here.10 Individual cases in individual languages are clearly not entirely divorced from content in this sense. However, it is also rather obvious that they could not be unitary features or primitives in syntax either. Thus, dative case is, for instance, used to mark the following kinds of NP relations in Icelandic: (14) a. b. c. d. e. f. g. h. i.

Agentive NPs in af- ‘by’ phrases in the passive Experiencer subjects of certain predicates Theme subjects of certain predicates Free benefactives Most benefactive indirect objects Numerous direct objects (with certain thematic and aspectual readings) Complements of many prepositions Complements of certain adjectives Certain adverbial NPs (instrumental, possessive, comparative)

The examples in (15) illustrate this; henni is the dative form of the third person singular feminine pronoun (nom, acc, gen = hún, hana, hennar, respectively): 10.  Rich agreement languages have certain instances of semantically related agreement, facilitating processing, much as rich case languages have semantically related cases. However, this only shows that morphology is based on syntax – it does not show that morphology takes place in syntax. For further discussion, see Sigurðsson 2006a, 2009a.

The No Case Generalization 

(15) a. Ég var studdur af henni. b. I was supported by her ‘I was supported by her.’

Henni líður vel. her feels good ‘She feels good.’

c. Henni fór fram d. her went forth ‘She made progress.’

Ég orti henni ljóð. I made her poem ‘I wrote her a poem.’

e. Ég sendi henni bréf. f. I sent her letter ’I sent her a letter.’

Ég bauð I invited ‘I invited her.’

g. Ég var með henni. h. I was with her ‘I was with her.’

Ég var henni góður. I was her good ‘I was nice/kind to her.’

henni. her

i. Ég var henni eldri. I was her older ‘I was older than she.’

It is also instructive to observe that prepositions with heterogeneous semantics obligatorily assign or require dative case, as illustrated for a few prepositions in (16): (16) a. c. e. g.

að mér ‘towards me’ b. frá mér ‘from me’ d. gengt mér ‘opposite to me’ f. út af mér ‘because of me’

af mér ‘off me’ gegn mér ‘against me’ hjá mér ‘at me, with me’ etc.


The arbitrariness of this becomes rather obvious when it is compared with certain other prepositions that require the genitive, as the ones in (17): (17) a. c. e.

auk mín ‘in addition to me’ b. milli okkar ‘between us’ d. vegna mín ‘because of me’

án mín ‘without me’ gen til mín ‘to me, towards me’ etc.

All the cases are used for multiple purposes (see Barðdal 2001; Jónsson 2005). Thus, nominative is used to mark the following relations: (18) a. Agentive subjects in finite clauses b. Numerous non-agentive subjects (of various kinds of predicates) c. Subjects of ECM-like infinitival and small clause complements of certain matrix verbs that take a dative subject d. Objects of certain verbs that take a dative subject e. Predicative NPs in finite clauses f. Predicative NPs in PRO infinitives g. Many left and right dislocated NPs, vocatives and other addressing expressions, most listed NPs, certain exclamative NPs

 Halldór Ármann Sigurðsson

The examples in (19) illustrate this; hún is the nominative form of the third person singular feminine pronoun: (19) a. Hún hefur skrifað margar bækur. she has written many books

+AG subj

b. Hún hvarf. / Hún var she disappeared / she was

–AG subj

kosin. elected

c. Mér hafði virst [hún vera þreytt]. me.dat had seemed  she be.inf tired. ‘It had seemed to me that she was tired.’ d. Mér hafði alltaf leiðst hún. me.dat had always bored she ‘I had always found her boring.’

ECM subj


e. Nú hefur þú verið hún þrisvar sinnum. now have you been she three times ‘You have now been her three times.’

Pred NP

f. Mig langar ekki [að verða hún í næsta lífi]. me.acc longs not [to become.inf she in next life ‘I don’t want to become her in my next life.’

Pred NP

g. Hún forseti! she president ‘Her for president!’

Excl NP

Saying that syntax operates with a (single) +nom feature amounts to saying that this feature is syntactically assigned not only to many subjects in Icelandic but also to some objects, certain ECM subjects, NP predicates, certain exclamative NPs, etc. It is unclear, to say the least, why that would or should be the case and what kind of syntactic case assignment or case decision mechanism would be required. Such an approach is not just an innocent or a neutral mainstream assumption: It actually claims that all the NP functions listed in (18) have some syntactic property in common, and it also makes the prediction that the +nom feature should have much the same distribution across nominative-accusative languages, contrary to fact (nominative being rare on objects and ECM subjects, predicative NPs commonly being non-nominative in many languages, etc.). Any approach claiming that +nom is a (single) syntactic feature has to come up with a syntactically unifying analysis of all the nominatives in (19) and it also has to offer some account of the fact that nominative case has different domains and functions in even closely related languages, like Icelandic, German and the Mainland Scandinavian languages (see further shortly). Many of the functions or relations listed in (14) and (18) are quite complex, a fact that suggests that the cases are not atomic features or primitives in language. Thus,

The No Case Generalization 

seeing to it that agentive subjects in finite clauses show up in the nominative requires a rule or a statement that takes, roughly, the following form: (20) ∀x: (x ∈ a finite clause & NP(x) & subject(x) & agent(x)) → nominative(x)

Similarly, the formula in (21), where the quantifier +∃̀ stands for ‘most’, would see to it that most benefactive indirect objects get assigned dative case:11 (21) +∃x: (NP(x) & indirect object(x) & benefactive(x)) → dative(x)

Notice that there is no way of linking only thematic content like benefactive or agent directly with the cases, there for instance being both nominative benefactives and dative agents (in af- ‘by’ phrases in passives). That is, the case-marking is essentially based on a combination of different kinds of information, as seen in the formulas.12 Moreover, formulas like these are just descriptive generalizations stated in terms of traditional notions that are themselves not syntactic primitives, such as ‘subject’ (cf. Chomsky 1981: 10; McCloskey 1997), ‘object’ and ‘finite clause’. I will return to the issue of case assignment in section 4, where I consider the question of how syntax ‘feeds’ morphological case-marking. As mentioned above, the cases differ from formal agreement features in typically relating to semantics (in different ways in different languages). It might seem to follow, and it is commonly assumed to follow, that at least the inherent cases are legible to the semantic interface (cf. Chomsky 2002: 113). If that was the case, however, we would expect cases to show up in more or less the same fashion across languages. Nothing could be further from the truth. Consider this for only a handful of Indo-European languages (see, e.g., Comrie 1990; Blake 2001):13 (22) a. b. c. d. e.

Pr-Indo-European: nom Lithuanian: nom Polish: nom Latin: nom Russian: nom

acc gen acc gen acc gen acc gen acc gen

dat voc abl inst dat voc inst dat voc inst dat voc abl dat inst

loc loc (Ill/Ade/All) loc (loc) loc

11.  There are more (and probably better) ways to get the same result, but the technical details are not important here. The point I’m making is that the cases typically represent complex relations, involving a number of factors. 12.  This holds true, even if thematic information is encoded by Voice and little v heads, see below. 13.  The case abbreviations are: Nom(inative), Acc(usative), Dat(ive), Gen(itive), Voc(ative), Abl(ative), Inst(rumental), Loc(ative), Ill(ative), Ade(ssive), and All(ative). As pointed out by a reviewer, comparing case systems in terms of traditional case labels is often misleading (‘genitive’ for instance having a larger domain in Modern Greek than in Ancient Greek or German), but this does not affect the simple point I am making, namely that this kind of variation is unexpected if the cases are syntactic objects.

 Halldór Ármann Sigurðsson

f. g. h. i. j. k.

Albanian: Ancient Greek: German/Icelandic: Modern Greek: Faroese (spoken): Rumanian:

nom acc gen dat abl nom acc gen dat voc nom acc gen dat nom acc gen voc nom acc dat n/a g/d (voc)

If syntax operates with an +ablative feature, for instance, it is unclear why it should, at some time point, stop doing so in some languages. In contrast, if the cases are morphological markers, their variability and historical instability can be analyzed in similar terms as other morphological variation – not a trivial task, but at least a conceivable one. On such a morphological approach to case variation, languages can be understood to have clusters of semantic and structural properties that are marked by different, language specific PF strategies, including for example ablative marking in Latin, dative or accusative marking in corresponding constructions in Icelandic, prepositional marking or no marking in English, etc. Related languages, with basically the same case systems, can show quite different distribution and function of their cases. This is even true to an extent of case-poor ‘sisters’ like Danish and Swedish, as illustrated in (23) (facts of this sort across the Germanic languages are discussed in Sigurðsson 2006b, see also Maling & Sprouse 1995): (23) a. Det er os/*vi. Danish b. it is us.acc ‘It is us.’

Det är it is ‘It is us.’

vi/*oss. we.nom


Thus, structures that are arguably exactly the same can get different representations in morphology (PF) in even closely related languages. Icelandic vs. German is another very clear example of this (see Maling 2001, 2002; Wunderlich 2003), an issue I will return to in the next section. In general, relations that are expressed with some particular case in one language are commonly expressed with different cases or by other means in other languages. Thus, Russian and the Germanic languages don’t have any special partitive case, while Finnish does. However, partitive and pseudopartitive relations are often marked with the genitive in Russian and commonly with prepositions in the Germanic languages but also sometimes with the genitive or even with no marking, as in the German and Swedish pseudopartitive construction zwei Flaschen Wein/två flaskor vin ‘two bottles of wine’ (see Neidle 1988; Vainikka & Mailing 1996; Blake 2001; Delsing 1993 and Sigurðsson 2003, for some discussion of these and related issues). It is not as if case languages don’t have any similarities in their case systems. On the contrary, such similarities are numerous. However, the point is that if individual cases were syntactic features, we would not expect any differences of this sort, given the

The No Case Generalization 

basic assumption that syntax operates with universal features only.14 It is of course not inconceivable that this basic assumption is on the wrong track, but it is unclear, to say the least, what alternative assumptions there could be. It might seem to be a way out here to assume that case variation boils down to parametric variation, but, to put it bluntly (perhaps), that could hardly seem to be a plausible alternative to anyone who has ever spent some time on studying morphological case in more than minimally complex case languages of the English or the Romance type. Thus, it is not clear, to say the least, how a parameter, given or implanted in Universal Grammar, would account for the fact that the preposition meaning ‘without’ selects acc in German (ohne), gen in Modern Icelandic (án) and acc, gen or, most commonly, dat in Old Norse (án), to mention only one of hundreds or thousands of tiny as well as more general differences of this sort (some of which will be discussed below).  – After all, it is hardly a coincidence that case parameters have not been successfully proposed in the generative literature. Also, languages apply various means, other than case or in addition to case, to mark the relation between an NP and its linguistic environment, including suprasegmental marking and some marking of a non-NP member of the relevant syntactic relation (see Sigurðsson 2003, for some discussion). Nichols (1992) studied dependency marking with respect to the typological notions of A(gent)-S(ubject)-P(atient). In her sample of 155 (relevant) languages, 148 or 95,5% had some such marking, and these in turn split into about equally large groups, with and without case-marking (see Nichols 1992:  90; cf. also Nichols & Bickel 2005:  98ff). This is presumably not very different from the result one might expect if different marking strategies (i.e., NP-marking (=case) vs. non-NP marking = agreement, etc.) are randomly spread across languages. Moreover, if one looks at constructions, rather than only at whole languages, the little evidence there is suggests that no marking (as in German/Swedish zwei Flaschen Wein/två flaskor vin, lit. ‘two bottles wine’) is a third, highly common alternative. As far as can be seen, the only general pattern in this is that NP relations are commonly marked in some manner. One could even phrase the putative ‘generalization’

14.  A reviewer points out that this argument should carry over to other categories, such as Tense. As briefly mentioned in the introduction, it does, morphological tense markers not being syntactic objects. The underlying relations are arguably syntactic and universal, but they always involve at least two temporal elements or ‘arguments’ (‘an event time X prior to the time of speech Y  ’, etc.). It is thus impossible to state the universal syntactic relations in terms of single features like +past, even though these relations are commonly expressed in morphology by overt markers that are traditionally characterized as ‘past’, ‘future’, etc. See further Sigurðsson & Maling (2009) and see also the analysis of abstract syntactic Person in other previous work (Sigurðsson 2004a etc.).

 Halldór Ármann Sigurðsson

such that NPs enter into syntactic relations, and that these relations are often marked or highlighted in one way or another, somewhere on the NP itself, or on some of its neighbors, or suprasegmentally, althought nothing of this is necessarily the case, and commonly is not the case  …

4.  On case assignment 4.1  Non-Isomorphism Reconsider the distribution of the Icelandic dative and nominative sketched in (14) and (18) above, restated in a simplified manner in (24) and (25): (24) The Icelandic dative marks: a) Agentive NPs in af- ‘by’ phrases; b) Certain experiencer subjects; c) Certain theme subjects; d) Free benefactives; e) Most benefactive indirect objects; f) Numerous direct objects; g) Complements of many prepositions; h) Complements of certain adjectives; i) Certain adverbial NPs (25) The Icelandic nominative marks: a) Agentive subjects; b) Numerous non-agentive subjects; c) Certain ECM subjects; d) Certain objects; e) Predicative NPs in finite clauses; f) Predicative NPs in infinitives; g) Many other NP types

Clearly, it does not make much sense to assume there to be syntactic dat and nom ‘features’ that would be the common denominators for all the relations in (24) and (25), respectively. In other words, case instructions do not take the simple form in (26), hence the stars: Narrow Syntax (26) a. * +nom b. * +dat

transfer → →

Morphology (PF) morphological nominative case morphological dative case, etc.

As a matter of fact, it seems to be a fundamental property of language that it never applies mappings of this sort between any levels or derivational stages. Thus, there are no one-to-one mappings from features in phonology onto soundwaves in phonetics, nor are there any such mappings from morphological features like +plural and +feminine onto phonological features such as [–high] and [+labial] (not even onto bundles of phonological features, as simply evidenced by allomorphy). I refer to this fundamental fact about language (and plausibly about any biological transformation process), as Non-Isomorphism, ni: (27) NI: Linguistic processes are non-isomorphic

4.2  Surface adjustments Given Non-Isomorphism, the question of how exactly individual cases are ‘produced’ in individual languages must be addressed. A part of the answer to this question is

The No Case Generalization 

trivial: Certain instances of case-marking are simple adjustment rules, taking, roughly, the form in (28): (28) Xα + NP → Xα + NPcase-α

This is illustrated in (29) for the above mentioned, most common prepositions meaning ‘without’ in German and Modern Icelandic: (29) a. ohne + NP → ohne + NPacc b. án + NP → án + NPgen

e.g., ohne mich/*meiner e.g., án mín/*mig

German Icelandic

This is not to say that the syntax of prepositions is simple, but it is to say that their case-marking properties are commonly trivial.15 Thus, any Icelandic preposition containing the string /um/, like um ‘about’, kringum ‘around’, umfram ‘in addition to’, etc., assigns accusative, and any (single-word) preposition or adverb containing /an/, innan ‘within’, sunnan ‘south of ’, etc., assigns genitive. Another type of surface case adjustments involves case agreement, as in adjectival, participial and NP predicates in Icelandic, and yet another instance of ‘mechanic’ case-marking strategy is genitive marking in ‘possessive’ NP/NP constructions (with an array of different semantic/syntactic properties). Again, the syntax of the constructions in question is everything but simple, while their case-marking properties are arguably trivial, at least in individual languages (notwithstanding the fact that, for instance, predicative case-marking shows curious variation across even closely related languages). As for the case-marking of subjects and objects, two simple relational rules can be discerned for nominative-accusative languages like Icelandic and German (Yip et al. 1987 and many since): (30) a. Inherent over Structural, I>S: Semantically related case (inherent case) takes precedence over non-semantically related case (so-called ‘structural’ case).16 b. Nominative over Accusative, N>A: Among the structural cases, nominative (case 1) takes precedence over accusative (case 2), that is, accusative cannot usually be assigned to an argument unless nominative is assigned to another argument in the same clause, whereas nominative is independent of the presence of an accusative argument (Burzio’s Generalization or the Sibling Correlation, see Sigurðsson 2003, 2006b).

15.  That does however not extend to prepositions that can assign either acc or dat, depending on the semantics of the P + NP relation. 16.  As discussed in Sigurðsson (2006b), ‘relational case’ is a more fortunate term than ‘structural case’ (there being a precedence relation between nominative and accusative), but I will be using the term ‘structural’ here, for expository convenience. Notice that ‘semantically related’ does not imply that the case itself is input to semantic interpretation.

 Halldór Ármann Sigurðsson

N>A is a typical elsewhere process in morphology, only taking place after I>S or when I>S does not take place.17 There are certain well-known apparent exceptions to N>A, but I will not discuss them here (many of them are described and discussed in Sigurðsson 2006b). Consider how this works for quirky vs. non-quirky case patterns, as in (31) and (32): (31) Mér áskotnuðust fjórir gullpeningar. me.dat acquired.3pl four gold-medals.nom ‘I got four gold medals (by some luck or coincidence).’


(32) a. Ég fékk fjóra gullpeninga. I.nom got.1sg four gold-medals.acc ‘I got four gold medals.’

nom/-ag acc

b. Ég tók fjóra gullpeninga. I.nom took.1sg four gold-medals.acc

nom/+ag acc


Evidently, some feature or features in (31) require that the subject of áskotnast ‘acquire, gain, get’ be dative, whereas the object is not subject to any such inherent case precedence or priority. Thus, by I>S, the subject cannot get structurally case-marked, and, by N>A, the object gets assigned nominative, rather than accusative. In (32), on the other hand, the subject (agentive or not) is assigned nominative by N>A, the object being assigned accusative as case 2.

4.3  Voice and subject case A number of complex issues arise. The fact that all agentive subjects in finite clauses are nominative, whereas numerous non-agentive subjects are quirky, suggests that subject case variation is in part a reflection of different Voice type heads (cf. Svenonius 2006). In the approach developed in Alexiadou et al. (2006) and Schäfer (2008), agentive subjects of finite clauses are introduced in the specifier of Voice, where Voice is marked

17.  Accusative structures (nom-acc) are more complex than plain nominative structures, thus taking priority in a sense (as pointed out by a reviewer). However, in the approach in Sigurðsson (2006c) ‘nominatives to be’ are actually merged lower than other arguments, nom thus taking structural priority (in a derivational approach). That approach is compatible with my present analysis, although I’m not pursuing it here. In an approach where accusative case is assigned by v* (as in Chomsky 2001; cf. section 4.5 below), acc can be understood a being assigned only when the structure contains v* (as opposed to plain v), but something more is needed to account for the fact that acc is normally only assigned when nom is also activated, later in the derivation. An even simpler approach than both Chomsky’s approach and the present one is analyzing nom as a non-assigned ‘no case’, showing up whenever an NP does not get case by (morphological) assignment, an ‘Anti Case Filter’ approach, as it were. That is arguably the most coherent analysis, but for expository ease, I assume the more traditional N>A here.

The No Case Generalization 

[+ag(entive)]. I will instead assume that the subject is generated vP-internally but enters a matching relation with an ag(entive) Voice feature, Voice/ag, as well as with the Fin(iteness) head.18 Evidently, the subject also has to match the clausal Person head, Pn (see Sigurðsson & Holmberg 2008 and the references cited there, see also the discussion around examples (10)–(11) above).19 However, quirky subjects also match Fin and Pn.20 Thus, as indicated with the connecting arrow in (33), only matching of Voice/ag seems to matter for PF case assignment to agentive subjects: (33) [CP … Fin [IP Pn … Voice/ …

NP …

NP/ in PF morphology


A different Voice head is involved in the Fate Accusative construction (type /us.acc drove to land/ = ‘We drifted ashore’, etc.), call it Voice/fate.21 As discussed in Sigurðsson (2006b), this fate Voice feature and Voice/ag are mutually exclusive (as one would expect). Thus, the matching relations in (34) commonly give rise to accusative marking in standard Icelandic (fate accusatives are like other quirky subjects in matching Pn and Fin, cf. Sigurðsson 2006b, 2009b): (34) [CP … Fin [IP Pn … Voice/ … V NP … ↑_______↑__↑

Voice features in dative subject constructions include experience, Voice/exp, and success/failure, or, more generally, (non-agentive) gain/loss:  Voice/gain, Voice/loss.22 Thus, the matching correlations in (35) (where I abstract away from little v type heads), often yield dative case in Icelandic morphology: (35) [CP … Fin [IP Pn … Voice/ … V NP1 … (NP2) … ↑________↑__↑

18.  In the sense of Fin in Sigurðsson 2004a, 2009b (inspired, in turn, by Rizzi 1997). See also Sigurðsson 2004b, 2006a on the nature of feature matching and syntactic Agree (as distinct from morphological agreement). 19.  I am abstracting away from other phi-features. 20.  Conversely, nominative PRO does not match Fin but is nonetheless assigned case. 21.  Voice/x is just a convenient notation, where x is the active feature. It should thus read something like: ‘x, an active Voice feature, excluding other Voice features’. The term ‘Voice’ itself is a cover term, much like ‘Aspect’ in the approach of Cinque (1999). 22.  For a more detailed classification, see Barðdal (2001). See also the lists in Jónsson (2003, 2005; but note that Jónsson’s lists do not reflect the idiomatic nature of many quirky constructions). Notice that I have no intention of giving a complete taxonomy of ‘quirky relations’ here. Notice also that many quirky subjects are derived by unaccusative/passive ‘promotion’ (cf. Sigurðsson 1989), their case thus being decided in relation to vP internal categories as well as in relation to Voice type heads (see section 4.5).

 Halldór Ármann Sigurðsson

Notice that the verb is involved in the NP matching relations in (34) and (35), as opposed to (33). This is a necessary feature of the analysis, in view of the fact that quirky case is commonly licensed in part by lexical properties of individual predicates (as has been the received understanding since at least Zaenen et al. 1985).23 Many quirky constructions are partly or entirely idiomatic. Thus, there does not seem to be any general reason why áskotnast ‘acquire, gain, get’ takes a dative subject, as in (31), whereas fá, the most central verb meaning ‘get’, takes a nominative subject, as in (32a).24 Rather, fá is arguably embedded under Voice/gain, without the matching relation between the two leading to dative marking in morphology (see further below on object case marking). Similarly, Voice/fate in (34) does not necessarily lead to or trigger accusative subject case, that is, some ‘fate predicates’ regularly take a nominative rather than an accusative subject.25 In addition, many speakers actually use nominative rather than accusative in the Fate ‘Accusative’ construction (Eythórsson 2000a, 2000b), without any concomitant semantic effects, it seems. Again, complex issues arise. Thus, some general approach to the inventory of Voice heads and the relation between their properties and argument theta- and case-properties across languages (including ergative languages) needs to be developed. However, the wonders of case-markings in individual languages are not central to my present purposes (but for some recent discussion, see, e.g., Jónsson 2003, 2005; McFadden 2004, 2009; McIntyre 2006; Svenonius 2005, 2006; Thráinsson 2007). What matters here is the following:

23.  However, ‘lexical items’ represent a larger structure than just the lexical (or conceptual) root. It is thus not self-evident what should count as a ‘lexical property’ and as an ‘individual predicate’. I will not pursue these (extensively discussed) issues here. 24.  The same applies to a large number of similar minimal pairs (see Barðdal 2001 & Jónsson 2003, 2005 for examples and Sigurðsson 2003 for some general discussion). In this particular case, there might be some connection with the ‘middle’ -st-suffix in áskotnast, cf. also pairs like the nom-acc verbs læra ‘learn’ and skilja ‘understand’ vs. dat-nom taking lærast ‘learn’ (without purposeful or conscious effort) and skiljast ‘understand’ (also without trying to, but merely by experience or circumstances), but the putative connection could not be a straightforward one, as most -st-verbs do in fact take nominative subjects (cf. Zaenen & Maling 1984 and many since, e.g., Svenonius 2006). One such verb is actually öðlast, meaning ‘acquire, gain, get, obtain’, much as áskotnast, but differing from it in taking a nominative subject and an accusative abstract object (‘strength’, ‘courage’, etc.). 25.  Thus, hrekja ‘drive’ can take a fate accusative subject (type /us.acc drove there because of the weather/ = ‘We were driven there by the weather’), whereas hrekjast with the fate reading ‘be driven’ takes a nominative subject (type /we.nom drove-st there because of the weather/ = ‘We were driven there by the weather’).

The No Case Generalization 

First, while Voice/ag in (33) certainly precludes quirky case-marking of subjects (presumably by precluding ‘quirky Voice features’), there is no mention of a +nom feature in the syntactic derivation (and there is also no mention of +acc or +dat in (34) and (35)). Second, the other instances of nominative case-marking, listed for Icelandic in (18)/(25), are not interpretations or ‘translations’ of the Voice/ag matching relation in (33), instead interpreting several different syntactic correlations.

Consider the second point for only nominative objects, as in (31) = (36): (36) Mér áskotnuðust fjórir gullpeningar. me.dat acquired.3pl four gold-medals.nom ‘I got four gold medals (by some luck or coincidence).’

As illustrated in (35), the dative subject enters into a matching ‘chain’ with V and Voice, in addition to Pn and Fin (the nominative object, in turn, matching Number, cf. (10) above). Thus, matching Voice does not ‘produce’ nominative case. One could argue that specifically matching Voice/ag triggers nominative marking, but different kinds of matching relations yield nominative as well, as in (36) (see also the discussion around (19) above). The most coherent understanding is thus that Voice/ag (e.g., in (32b) above), is not a ‘nominative assigner’, instead precluding Voice features that are active in quirky constructions, thereby allowing the PF strategy of Nominative over Accusative, N>A, in (30b) to apply to the subject, in the absence of I>S.

4.4  On dative direct objects Discussing all the case facts listed in (24)–(25) requires much more space than available here. However, consider at least the fact, stated in (24f), that Icelandic has numerous direct objects that are assigned dative case. Typically, the corresponding verbs assign accusative in German. Compare the Icelandic a- and the German b-examples in (37)–(42): (37) a. Hún kastaði steininum/*steininn.  she threw stone.the.dat/*acc ‘She threw the stone.’

Icelandic dat

b. Sie hat den Stein/*dem Stein geworfen.  she has the stone.acc/*dat thrown

German acc

(38) a. Hún hellti víninu/*vínið niður.  she poured wine.the.dat/*acc down ‘She spilled the wine.’

Icelandic dat

b. Sie hat den Wein/*dem Wein verschüttet.  she has the wine.acc/*dat spilled

German acc

 Halldór Ármann Sigurðsson

(39) a. Hún stýrði skipinu/*skipið. she steered ship.the.dat/*acc ‘She steered the ship.’

Icelandic dat

b. Sie hat das Schiff/*dem Schiff gesteuert. she has the ship.acc/*dat steered

German acc

(40) a. Hún gleymdi mér/*mig.  she forgot me.dat/*acc ‘She forgot me.’

Icelandic dat

b. Sie hat mich/*mir vergessen.  she has me.acc/*dat forgotten

German acc

(41) a. Hún heilsaði mér/*mig ekki.  she greeted me.dat/*acc not ‘She did not greet me.’

Icelandic dat

b. Sie hat mich/*mir nicht gegrüßt.  she has me.acc/*dat not greeted

German acc

(42) a. Hún bauð mér/*mig ekki.  she invited me.dat/*acc not ‘She did not invite me.’

Icelandic dat

b. Sie hat mich/*mir nicht eingeladen.  she has me.acc/*dat not invited

German acc

Minimal pair differences of this sort between these two closely related languages, with the same inventory of cases, are strikingly numerous. I quote Maling (2002: 31): Maling (1996) [an unpublished work] contains a list of more than 750 [Icelandic] verbs which in at least one sense occur with a dative object … The corresponding number of verbs for German is approximately 140, and for Russian fewer than 60  …

Dative direct objects in Icelandic primarily have four thematic interpretations (for further discussion, see e.g., Barðdal 2001; Maling 2002; Svenonius 2002; Jónsson 2005; Thráinsson 2007: 208ff): (43) a. The object (as a whole) is put into movement: ausa ‘scoop’ bylta ‘overturn’ fleygja ‘throw away’ fleyta ‘float’ henda ‘throw, throw away’ ýta ‘push, shift’

dreifa ‘spread’ hella ‘pour’ etc.

b. The object (as a whole) is under external control: beina ‘direct’ fljúga ‘fly’ (e.g., an aeroplain) ráða ‘decide’ ríða ‘ride’ (e.g., a horse) róa ‘row’ sigla ‘sail’ snúa ‘turn’ stjórna ‘control, govern’ etc.

The No Case Generalization 

c. the object is benefactive: bjarga ‘rescue’ borga ‘pay’ hjúkra ‘nurse’ hlífa ‘protect, spare’ þjóna ‘serve’ þóknast ‘please’

hjálpa ‘help’ launa ‘pay, reward’ etc.

d. the action described by the verb is potentially reciprocal:26 andmæla ‘contradict’ blandast ‘get mixed with’ fagna ‘welcome’ giftast ‘marry’ heilsa ‘greet’ misþyrma ‘torture’ skrifa ‘write to’ svara ‘answer’ etc.

The first two classes, in (43a) and (43b), seem to have a common aspect of wholeness (the opposite to partitive). Thus, some verbs can either take a dative object that is moved or controlled as a whole or an accusative object that is effected or affected. This is illustrated in (44):



(44) a. b. c. d. e. f.

hlaða steinum ‘pile bricks’ moka sandi ‘shovel sand’ nudda kremi ‘rub cream (onto/into)’ skjóta kúlunni ‘shoot the bullet’ snúa lyklinum ‘turn the key’ sópa ryki ‘sweep up dust’

hlaða hús ‘build a house (of bricks)’ moka skurð ‘dig a ditch’ nudda augun ‘rub one’s eyes’ skjóta dýrið ‘shoot the animal’ snúa fótinn ‘twist one’s foot/ankle’ sópa gólfið ‘sweep the floor’

Similarly, a few verbs make a distinction between dative benefactive objects (commonly animate) and accusative affected objects (commonly inanimate). This is illustrated in (45):



(45) a. b. c. d.

greiða barninu ‘comb the child’ strjúkja henni ‘stroke her’ þurrka sér ‘dry oneself ’ þvo sér ‘wash oneself ’

greiða hárið ‘comb the hair’ strjúkja enni hennar ‘stroke her forehead’ þurrka heyið ‘dry the hay’ þvo bílinn ‘wash the car’

Assume that the dative direct object in Icelandic matches one of a limited number of a little v-type heads, vwholeness, vgain, and perhaps a few other, call them simply v** in general (on a par with v* in Chomsky’s work, see below).27 This would yield the following 26.  Thus, both the subject and the object of these verbs are commonly (but not exclusively) +human. To be ’potentially reciprocal’ is not a very specific characterization, but I have not been able to come up with a better one. 27.  Similar ideas have been proposed for other kinds of arguments (with different labels of the heads involved and more elaborated structural proposals), for instance indirect objects and free datives. See McFadden (2004) and Schäfer (2008) and the references cited in these works. It is of course an important task of linguistics to study the inventory and nature of little

 Halldór Ármann Sigurðsson

matching relations (where higher matching relations are not indicated, but see shortly): (46) … [IP …

v**… V



Again, there is no mention of a case feature, like +dat, in the syntactic derivation. Importantly, also, there are no discernable semantic correlates with the IcelandicGerman case differences exemplified in (37)–(42). There is thus no reason to assume that German, or other languages for that matter, lack v**-type heads or features. Rather, the relevant difference here between the languages is morphological: Icelandic morphology commonly interprets the presence of a syntactic v** feature in terms of dative object case, whereas German morphology does so much less frequently.28 There are exceptions from this central pattern in both languages, that is, German has some dative direct objects and Icelandic has a number of accusative objects with thematic properties that are otherwise typical of dative direct objects in the language. This is illustrated by pairs like the following ones: dative: (47) a. b. c. d. e. f. g. h.

bifa ‘(slightly) move’ ýta ‘push, shift’ fylgja ‘follow, accompany’ ljúka ‘finish’ hjálpa ‘help’ þjóna ‘serve (e.g., at the table)’ bjarga ‘rescue’ hjúkra ‘nurse’

accusative: hreyfa ‘move’ flytja (til) ‘shift, move’ elta ‘follow, pursue’ klára ‘finish’ aðstoða ‘assist’ uppvarta ‘serve (at the table)’ lífga (við) ‘revive’ lækna ‘cure’

The natural interpretation of this fact is that not only v** but also the verb itself (or some verb related feature) matches the object, as shown in (46), and therefore can ‘switch off ’ or block the otherwise prevailing case interpretation of v**. In short, the presence of a v** feature is commonly signaled on direct objects by dative case in Icelandic morphology as opposed to German morphology, but there are

v-type heads, much as the inventory and nature of the Voice heads discussed above, but this is not one of the goals of the present study. Svenonius (2006) refers to the relevant head type as Vdat, but using the dat ‘index’ implies that syntax has a unitary object or feature that could be referred to as ‘dative’, contrary to fact (a more positive interpretation is to understand the dat ‘index’ as just a convenient way of saying that whatever semantic/syntactic properties the head and its matching relations may have, they will be represented by dative case in Icelandic morphology). 28.  I am abstracting away from free (bare, non-prepositional) datives, highly common in German (see, e.g., Schäfer 2007) but relatively rare in Icelandic (for some Icelandic examples, see Thráinsson 2007: 218f).

The No Case Generalization 

numerous exceptions from this generalization in both languages. Such idiosyncratic exceptions, as well as sporadic changes of the case-marking of individual items, are expected if case-marking takes place in post-syntactic morphology, but they would be truly troublesome if case-marking took place already in syntax. Assuming that case-marking is syntactic apparently forces one of two rather unfortunate options. The first one is that syntax operates with arbitrary features, which would preclude coherent semantic interpretation of syntax (thus being incompatible with the central goal of linguistics, to develop an understanding of the content-form relationship in language). The second one is to assume arbitrary case-deletion (cf. Chomsky’s deletion approach to agreement features in 2000, 2001, etc.). Thus, one could say that uppvarta ‘serve (at the table)’ in (47f) takes a ‘deep’ dative, just like þjóna ‘serve (e.g., at the table)’, and that its deep dative is subsequently deleted prior to or under transfer to PF (leading to accusative object case in PF morphology, as case 2, in accordance with N>A in (30b)). Presumably, one would furthermore have to say that dative deletion is also involved in the derivation of the German examples in (37b)–(42b) above. There is a conceivable alternative here, though, namely that the semantic factors involved in inherent case-making of subjects and objects are too subtle to be detectable. As for dative vs. accusative direct objects one could for instance hypothesize that events can be seen from either the subject’s or the object’s point of view (regardless of animacy), and that only the object’s point of view activates v**, resulting in dative object case in PF. If so, one would have to conclude that Icelandic for some reason expresses object point of view more frequently than German. I leave it to the reader to judge the viability of these potentially conceivable alternatives.

4.5  Case ‘preservation’, Voice, and little v The passive and certain other NP-movement constructions ‘preserve’ or respect the matching relations in (46), thereby also ‘preserving’ the inherent case interpretation of these relations, as has been widely discussed (Zaenen & Maling 1984; Zaenen et al. 1985; Sigurðsson 1989; Jónsson 1996; Svenonius 2006; Thráinsson 2007, among many). In addition, however, case ‘preservation’ is dependent on Voice. This is, for instance, suggested by the fact that anticausative (‘middle’) -st-verbs differ from passives in not ‘preserving’ inherent case on themes, as illustrated in (48): (48) a. Við lokuðum glugganum. we closed window.the.dat b. Glugganum var window.the.dat was

lokað. closed (by sby)

c. Glugginn lokaðist. window.the.nom closed-st ‘The window closed.’

Active nom-dati Passive dati Anticausative nomi

 Halldór Ármann Sigurðsson

The reason why the anticausative cannot ‘preserve’ dative case (on themes) is arguably that it represents or expresses a different Voice structure than the passive (Svenonius 2006; see also Zaenen & Maling 1984). Suppose that passives have Voicepass/ag whereas anticausatives have an expletive Voice feature, Voice/expl (closely following Alexiadou et al. 2006 and Schäfer 2008). If so, dative themes are not only licensed by a v** feature but also by active Voice/ag and Voicepass/ag in contrast to Voice/expl. The relevant matching relations for both the nominative subject and the dative object in (48a) are thus as illustrated in (49):29 (49)

[CP … Fin [IP Pn … Voice/ …

NP1 … v** …V



→ NP1/ in PF


→ NP2/ in PF

The relevant matching relations for the quirky passive in (48b) are sketched in (50): (50)

[CP … Fin [IP Pn … VoicePASS/ …

v** … V



NP/ in PF

Plain little v, in contrast, does not affect case-marking, as sketched in (51) for regular, non-quirky passives: (51)

[CP … Fin [IP Pn … VoicePASS/ …

v… V



NP/ in PF

Notice, however, that there is no acc-to-nom ‘conversion’. What is going on here is simply that the single argument in the clause is assigned nominative, in accordance with Nominative over Accusative, N>A in (30b). In this respect, passives are no different from regular unaccusatives, as illustrated in (52): (52) a. Þá var byggð ný kirkja í þorpinu. then was built new church.nom in village.the ‘Then a new church was built in the village.’ b. Það var þá horfin mynd úr safninu. there was then disappeared painting.nom from gallery.the ‘Then a painting had disappeared from the gallery.’



29.  In the approach in Sigurðsson (2006c), a nominative subject is merged lower or sooner than its object, later being shifted around the object for independent reasons, having to do with object Pn and Nr matching. As previously mentioned, the present approach is compatible with this low analysis of ‘nominatives to be’, but it is also independent of it (i.e., assuming that NP1 in (49) has been raised prior to Voice matching does not alter the present analysis). As I am not considering object Pn and Nr here, I put this aside (but see McFadden 2009 for a discussion).

The No Case Generalization 

As for anticausatives like lokast ‘close’ in (48c), it is evident that their Voice/expl ‘deactivates’ v**, the result being nominative case assignment in PF by N>A, as in non-quirky passives and unaccusatives.30 Regular nom-acc constructions, in contrast, can be analyzed as in (53), where I adopt v* from Chomsky (2001, etc.): (53) [CP … Fin [IP Pn … Voice …

NP1 … v* … V



NP1/ in PF


NP2/ in PF

Voice in transitive structures may or may not be ag(entive). It is not ag in (32a) above, for instance. The hypothesis that accusative objects not only match v* but also Voice captures Nominative over Accusative (the Sibling Correlation), namely, that assignment of ‘structural’ accusative is dependent on nominative being activated in the clause, whereas nominative is independent of accusative. This analysis presupposes that a head may in certain cases probe more than one goal. Other facts, for instance multiple case agreement facts, suggest that this is needed in any event. Notice also that the approach forces one of two conclusions: Either vP is not a full phase, or objects generally shift (even when there is no visible Object Shift) to the left edge of vP, where they can be probed by Voice without inducing a violation of PIC, the Phase-Impenetrability Condition (see Chomsky 2000: 108, 2001: 13f, 2008). I will not pursue these issues here, though. On the present approach, the syntactic message sent to the morphology/PF interface under transfer does not contain any morphological information, that is, the only information transferred is abstract structural information of the sort sketched in (49)–(51) and (53) above. PF is evidently a complex, layered system, with roughly the following ordered sub-interfaces in oral languages (see Sigurðsson 2006a: 204):31

(54) [NS

transfer to PF


Sign formation → Morphology → Phonology → Phonetics

After transfer and sign formation, morphology interprets syntactic matching relations like the ones in (49)–(51) and (53) in terms of abstract morphological features, such

30.  The same applies to nominalizations and adjectival ‘passives’ (The door is unlocked, etc.). I will not consider the mechanism of v**-deactivation here (but see Svenonius 2006 for discussion and a suggestion). 31.  How much of ‘PF’ (Perceptible Form) is common to oral languages and sign languages is an intriguing and an important question (see e.g., MacNeilage 2008: 273ff).

 Halldór Ármann Sigurðsson

as +nom, etc. The morphological features are subsequently transferred to phonology, which in turn interprets them in terms of phonological features.32 Syntax does not operate with or even ‘produce’ case features. Morphology does.

4.6  A note on A′-movement and case Movement of arguments, be it A′-movement or A-movement, does not affect casemarking (case being “divorced from movement”, Chomsky 2001: 17). This was illustrated for various A-movement contexts in (2)–(8) and it is seen again in (52) above. Nominative and accusative case-marking of subjects and objects is plainly decided by single versus double Voice matching in syntax and Nominative over Accusative in PF morphology.33 In the absence of inherent case-marking, double Voice matching (plus v* matching) is interpreted as nom-acc in morphology, whereas single Voice matching is interpreted as nom. Whether the argument in question undergoes A- or A′-movement is irrelevant. The question of why A-movement is CP clause-bounded, in contrast to A′-movement, is an interesting but a different question. A-movement is largely driven by IP-internal matching relations, most importantly Person matching, whereas A′-movement is triggered by more distant, IP-external relations. Even so, A′-movement differs from A-movement in showing that morphological case-marking sometimes ‘survives’ across CP-boundaries, as in Whom do you believe [Mary has invited whom]?34 This is unproblematic (and uninteresting) on an approach where whom is assigned a case feature in syntax, carrying it along under movement. On the present approach, in contrast, it is an intriguing issue. It could be taken to suggest that A′-movement applies in PF (rendering the reconstruction observations in Chomsky 1993:  37ff unaccounted for) or that A′-moved constituents ‘absorb’ or inherit the morphological properties of their silent copies by some kind of a late PF probing, in violation of PIC. Alternatively, and perhaps more plausibly, it could be taken to suggest that the standard phase theory (Chomsky 2007, 2008) must be revised somehow, presumably relativized with respect to features and domains. Other long distance relations, including nominal reference, can in fact cross arbitrarily many phase

32.  Thus, to give just two examples, when combined with the Icelandic root /rós/ ‘rose’ the feature complex nom.fem.pl gets the phonological value /ir/, whereas it gets /ar/ when combined with the root /nál/ ‘needle’, in the forms rósir and skálar ([-Ir] and [-ar] in subsequent phonetics). 33.  Alternatively, nom is decided in morphology by the ‘Anti Case Filter’ mentioned in fn. 17 above. 34.  I am indebted to Terje Lohndal and a reviewer for a helpful and knowledgeable discussion of this issue (and to Noam Chomsky, p.c., for having raised it in relation to the research questions pursued here).

The No Case Generalization 

boundaries, suggesting that a relativized phase theory is needed in any case.35 I leave it at that. A convincing account of ‘case preservation’ under A′-movement remains to be developed.

5.  Conclusion The present approach to the syntax underlying argument PF case (albeit not ‘producing’ it) is close in spirit to the ideas pursued by Chomsky in his minimalist research (see, in particular, Chomsky 2000, 2001).36 The following differences are important, though: –

First, it is Voice matching, and not Tense matching, that is commonly interpreted or reflected by morphological case.37 Other things being equal, it follows that PRO infinitives should get a parallel case interpretation as finite clauses, and, as briefly mentioned in section 2, that is indeed the case. Second, direct objects match not only marked little v (v* or v**) but also Voice; this captures the Sibling Correlation (Burzio’s Generalization).

Crucially, however, syntax contains no case features, such as +nom and +dat (or any other more general or abstract case features like ± oblique; all such features are morphological, inasmuch as they are ‘real’). Thus, syntactic processes, such as NPmovement, could not be driven by a ‘case-need’, such a need being nonexistent in syntax. Syntax computes relations between various kinds of elements. Such relations include matching relations between Voice heads, little v-heads and arguments (= argument structure), and these relations are commonly interpreted in terms of nominative and accusative case in PF. However, as we have seen, other NP relations may also be interpreted or expressed by either nominative or accusative case. This is not in principle any different from other relations between syntax and morphology. Thus, past tense verb forms in e.g., Romance and Germanic not only express shifted

35.  For a more specific structural approach to phase extension (in terms of head movement), see den Dikken 2007. 36.  But it is different in spirit from Sigurðsson (2006b, 2006c), where I did not try to pin down the structural correlates of PF case. The present results could not have been achieved without the ‘case in tiers’ insights of Yip et al. (1987) and the introduction of Voice into the minimalist discussion of Icelandic case by Svenonius (2005, 2006). It seems to me that the development of this subfield is thus reassuringly or at least hopefully becoming convergent. 37.  This first difference is however not radically distinct from Chomsky’s approach if T in his work is understood to be a cover term for a T-feature domain, including not only T itself but also clausal Pn, Nr and Voice features.

 Halldór Ármann Sigurðsson

(true) past time readings (in relation to speech time) but also unshifted (simultaneous) or future subjunctive readings (in which case the past morphology is copied by PF tense agreement across clause boundaries, in apparent violation of PIC), suggesting that syntax does not contain and operate with a unitary +past feature. Similarly, to mention just one additional well-known example (see Thráinsson 2007:  465ff), Icelandic reflexives not only express locally bound anaphors but also long distance and even (overtly) unbound logophors, commonly expressed by plain pronouns in related languages, including English. In sum:  Syntax contains abstract relations (such as NP-matching of Voice and little v), and the morphology of individual languages interprets or expresses these relations with the optimal (commonly the least ambiguous) means available to them in their language-specific PFs. Thus, there are no one-to-one mappings from syntax onto morphology or PF in general, the derivation instead being fundamentally non-isomorphic.38 Case features operate in morphology. Hence, they cannot be syntactic as well.

References Alexiadou, A., Anagnostopoulou E. & Schäfer, F. 2006. The properties of anticausatives crosslinguistically. In Phases of Interpretation, M.  Frascarelli (Ed.), 187–211. Berlin:  Mouton de Gruyter. Andrews, A. 1976. The VP complement analysis in Modern Icelandic. NELS 6: 1–21. Barðdal, J. 2001. Case in Icelandic:  A Synchronc, Diachronic and Comparative Approach. Lund: Studentlitteratur. Belletti, A. 1988. The case of unaccusatives. Linguistic Inquiry 19: 1–34. Blake, B.J. 2001. Case. 2nd Edn. Cambridge: CUP. Bobaljik, J. 2006. Where’s φ? Agreement as a postsyntactic operation. To appear in PhiTheory: Phi features across interfaces and modules, D.  Harbour, D.  Adger, & S.  Béjar (Eds), Oxford: OUP. Bobaljik, J. & Landau, I. 2008. Fact and fiction in Icelandic control. To appear in Linguistic Inquiry. Boeckx, C. 2000. Quirky agreement. Studia Linguistica 54: 354–380. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris.

38.  See also previous work (e.g., Sigurðsson 2006a, 2009a, 2009b). As mentioned in note 3 above, this is different from Distributed Morphology (where morphology is basically an extension of syntax). It seems possible that morphology has some universal features, but I do not know of any Universal Morphology, or Universal PF in general (cf. MacNeilage 2008). Putative Universal PF should, for instance, offer a coherent account of how oral languages relate to sign languages, and also of how both oral languages and sign languages relate to extinct written languages, such as Sumerian. No such account has been developed. In Otto Jespersen’s words (1992: 52, cited in Chomsky 1995: 3, and also in a preliminary version of Chomsky 2007): “no one ever dreamed of a universal morphology.”

The No Case Generalization 

Chomsky, N. 1993. A minimalist program for linguistic theory. In The View from Building 20, K. Hale & S.J. Keyser (Eds), 1–52. Cambridge MA: The MIT Press. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, R. Martin, D. Michaels & J. Uriagareka (Eds), 89–155. Cambridge MA: The MIT Press. Chomsky, N. 2001. Derivation by phase. In Ken Hale: A Life in Language, M. Kenstowicz (Ed.), 1–52. Cambridge MA: The MIT Press. Chomsky, N. 2002. On Nature and Language. Ed. by A. Belletti & L. Rizzi. Cambridge: CUP. Chomsky, N. 2007. Approaching UG from Below. In Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax-Semantics, H.M. Gärtner & U. Sauerland (Eds), 1–30. Berlin: Mouton de Gruyter. Chomsky, N. 2008. On phases. In Foundational Issues in Linguistic Theory. Essays in Honor of Jean-Roger Vergnaud, R. Freidin, C.P. Otero & M.L. Zubizarreta (Eds), Cambridge MA: The MIT Press. Cinque, G. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: OUP. Comrie, B. (Ed.), 1990. The World’s Major Languages. Oxford: OUP. Delsing, L.-O. 1993. The Internal Structure of Noun Phrases in the Scandinavian Languages. Lund. den Dikken, M. 2007. Phase extension: Contours of a theory of the role of head movement in phrasal extraction. Theoretical Linguistics 33: 1–41. Eythórsson, T. 2000a. Dative vs nominative:  Changes in quirky subjects in Icelandic. Leeds Working Papers in Linguistics 8: 27–44. Eythórsson, T. 2000b. Fall á fallanda fæti? [Case in retreat?] Íslenskt mál og almenn málfræði 22: 185–204. Henry, A. 1995. Belfast English and Standard English: Dialect Variation and Parameter Setting. Oxford: OUP. Jespersen, O. 1992. The Philosophy of Grammar. With a new introduction and index by J.D. McCawley. Chicago IL:  The University of Chicago Press (first published 1924 in London by George Allen and Unwin). Jónsson, J.G. 1996. Clausal Architecture and Case in Icelandic. Ph.D. dissertation, University of Massachuessets, Amherst. Jónsson, J.G. 2003. Not so quirky:  On subject case in Icelandic. In New Perspectives on Case Theory, E. Brandner & H. Zinsmeister (Eds), 127–163. Stanford CA: CSLI. Jónsson, J.G. 2005. Merkingarhlutverk, rökliðir og fallmörkun (Thematic roles, arguments and case-marking). In Íslensk tunga III:  Setningar, H. Thráinsson (Ed.), 350–409. Reykjavík: Almenna bókafélagið. Landau, I. 2006. Severing the distribution of PRO from Case. Syntax 9: 32–66. Legate, J.A. 2008. Morphological and abstract Case. Linguistic Inquiry 39: 55–101. MacNeilage, P.F. 2008. The Origin of Speech. Oxford: OUP. Maling, J. 2001. Dative:  The heterogeneity of the mapping among morphological case, grammatical functions, and thematic roles. Lingua 111: 419–464. Maling, J. 2002. Það rignir þágufalli á Íslandi. Verbs with dative objects in Icelandic. Íslenskt mál og almenn málfræði 24: 31–105. Maling, J. & Sprouse, R. 1995. Structural case, specifier-head relations, and the case of predicate NPs. In Studies in Comparative Germanic Syntax, H. Haider, S. Olsen & S. Vikner (Eds), 167–186. Dordrecht: Kluwer.

 Halldór Ármann Sigurðsson Marantz, A. 2000. Case and licensing. In Arguments and Case: Explaining Burzio’s Generalization, E. Reuland (Ed.), 11–30. Amsterdam: John Benjamins. McCloskey, J. 1996. Subjects and subject positions in Irish. In The Syntax of the Celtic Languages: A Comparative Perspective, R. Borsley & I. Roberts (Eds), 241–283. Cambridge: CUP. McCloskey, J. 1997. Subjecthood and subject positions. In Elements of Grammar: Handbook in Generative Syntax, L. Haegeman (Ed.), 196–235. Dordrecht: Kluwer. McFadden, T. 2004. The Position of Morphological Case in the Derivation:  A Study on the Syntax-morphology Interface. Ph.D. dissertation, University of Pennsylvania. McFadden, T. 2009. Structural case, locality, and cyclicity. In Explorations of Phase Theory: Features and Arguments [Interface Explorations], K. Grohman (Ed.), 107–130,. Berlin: Mouton de Gruyter. McIntyre, A. 2006. The interpretation of German datives and English have. In Dative and Other Cases, D. Hole, A. Meinunger & W. Abraham (Eds), 185–211. Amsterdam: John Benjamins. Neidle, C. 1988. The Role of Case in Russian Syntax. Dordrecht: Kluwer. Nichols, J. 1992. Linguistic Diversity in Space and Time. Chicago IL: The University of Chicago Press. Nichols, J. & Bickel, B. 2005. Locus of marking in the clause; Locus of marking in possessive noun phrases; Locus of marking:  whole-language typology (3 chapters). In World Atlas of Language Structures, M.  Haspelmath, M.  Dryer, D.  Gil & B.  Comrie (Eds), 98–109. Oxford: OUP. Platzack, C. 2006. Case as Agree marker. Working Papers in Scandinavian Syntax 77: 71–99. Quinn, H. 2005. The Distribution of Pronoun Case Forms in English. Amsterdam: John Benjamins. Rizzi, L. 1997. The fine structure of the left periphery. In Elements of Grammar. Handbook in Generative Syntax, L. Haegeman (Ed.), 281–337. Dordrecht: Kluwer. Rizzi, L. 2008. On delimiting movement. Paper presented at the 31st GLOW Colloquiem, Newcastle upon Tyne. Schäfer, F. 2008. The Syntax of (Anti-)Causatives: External Arguments in Change-of-state Contexts. Amsterdam & Philadelphia: John Benjamins. Shlonsky, U. 1989. The hierarchical representation of subject verb agreement. Ms., Haifa University. Sigurðsson, H.Á. 1989. Verbal Syntax and Case in Icelandic. Lund (republished 1992 in Reykjavík: Institute of Linguistics). Sigurðsson, H.Á. 1991. Icelandic Case-marked PRO and the licensing of lexical arguments. Natural Language and Linguistic Theory 9: 327–363. Sigurðsson, H.Á. 2000. The locus of case and agreement. Working Papers in Scandinavian Syntax 65: 65–108. Sigurðsson, H.Á. 2002. To be an oblique subject: Russian vs. Icelandic. Natural Language and Linguistic Theory 20: 691–724. Sigurðsson, H.Á. 2003. Case: abstract vs morphological. In New Perspectives on Case Theory, E. Brandner & H. Zinsmeister (Eds), 223–268. Stanford CA: CSLI. Sigurðsson, H.Á. 2004a. The syntax of person, tense, and speech features. Italian Journal of Linguistics 16: 219–251 (a special issue, ed. by V. Bianchi & K. Safir). Sigurðsson, H.Á. 2004b. Agree and agreement: Evidence from Germanic. In Focus on Germanic Typology, W. Abraham (Ed.), 61–103. Berlin: Akademie Verlag. Sigurðsson, H.Á. 2006a. Agree in syntax, agreement in signs. In Agreement Systems, C. Boeckx (Ed.), 201–237. Amsterdam: John Benjamins.

The No Case Generalization 

Sigurðsson, H.Á. 2006b. The Nom/Acc alternation in Germanic. In Issues in Comparative Germanic Syntax, J. Hartmann & L. Molnarfi (Eds), 13–50. Amsterdam: John Benjamins. Sigurðsson, H.Á. 2006c. The nominative puzzle and the low nominative hypothesis. Linguistic Inquiry 37: 289–308. Sigurðsson, H.Á. 2008. The case of PRO. In Natural Language and Linguistic Theory 26: 403–450. Sigurðsson, H.Á. 2009a. Remarks on features. In Explorations of Phase Theory: Features and Arguments [Interface Explorations], K. Grohman (Ed.), 21–52. Berlin: Mouton de Gruyter. Sigurðsson, H.Á. 2009b (in print). On EPP effects. Studia Linguistica 64. Sigurðsson, H.Á. & A. Holmberg. 2008. Icelandic dative intervention. In Agreement Restrictions, R. D’Alessandro, S. Fischer & G. Hrafnbjargarson, 251–279. Berlin: Mouton de Gruyter. Sigurðsson, H.Á. & Maling, J. 2009 (in print). Functional Projections: A Festschrift for Guglielmo Cinque, ed. by L. Bruge et al. Oxford: Oxford University Press. Stowell, T. 1981. Origin of Phrase Structure. Ph.D. dissertation, MIT. Svenonius, P. 2002. Icelandic case and the structure of events. Journal of Comparative Germanic Linguistics 6: 197–225. Svenonius, P. 2005. The nanosyntax of the Icelandic passive. Paper presented at the Lund Grammar Colloquium, May 26, 2005. Svenonius, P. 2006. Case alternations in the Icelandic passive and middle. To appear in Passives and Impersonals in European Languages, S. Manninen, K. Hiietam, E. Kaiser & V. Vihman (Eds), (See lingBuzz). Thráinsson, H. 1979. On Complementation in Icelandic. New York NY: Garland. Thráinsson, H. 2007. The Syntax of Icelandic. Cambridge: CUP. Vainikka, A. & Maling, J. 1996. Is partitive case inherent or structural? In Partitives: Studies in the Syntax and Semantics of Partitive and Related Constructions, J. Hoeksema (Ed.), 179–208. Berlin: Mouton de Gruyter. Wunderlich, D. 2003. Optimal case patterns: German and Icelandic compared. In New Perspectives on Case Theory, E. Brandner & H. Zinsmeister (Eds), 331–367. Stanford CA: CSLI. Yip, M., Maling, J. & Jackendoff, R. 1987. Case in tiers. Language 63: 217–250. Zaenen, A. & Maling, J. 1984. Unaccusative, passive and quirky case. In Proceedings of the Third West Coast Conference on Formal Linguistics, M. Cobler, S. MacKaye & M. Wescoat (Eds), 317–329. Stanford CA: CSLA. Zaenen, A., Maling, J. & Thráinsson, Höskuldur. 1985. Case and grammatical functions: The Icelandic passive. Natural Language and Linguistic Theory 3: 441–483.

The new impersonal as a true passive* Jóhannes Gísli Jónsson University of Iceland

This paper discusses a new impersonal construction in Icelandic. This construction has passive morphology but differs from canonical passives of transitive verbs in that the DP complement of the passive participle stays in situ and displays object properties. Contra Maling & Sigurjónsdóttir (2002), I argue that this is a true passive, not an active construction with a thematic null subject. As illustrated in the paper, there are some clear similarities between canonical passives and new impersonals that support a passive analysis of the latter construction but no clear differences to justify an active analysis.

1.  Introduction The topic of this paper is an innovative syntactic construction in Icelandic that has been subject to some controversy in recent years, a construction that I will refer to as the new impersonal. I will present arguments that the new impersonal is a true passive with an understood agent, as the passive morphology of this construction suggests.1 In doing so, I will argue against the view of Maling & Sigurjónsdóttir (2002) that the new impersonal is an active construction with a thematic null argument in subject position. In addition to arguments in favor of a passive analysis of the new impersonal, it will be shown that there are no clear differences between new impersonals and passives in Icelandic to support an active analysis. For concreteness, the analysis advocated here will be referred to as the Passive Analysis, in contrast to the Active Analysis of Maling & Sigurjónsdóttir (2002).

*I wish to thank two anonymous reviewers for their comments and Thórhallur Eythórsson and Joan Maling for lively discussions about the new impersonal over the years. Earlier versions of this paper were presented in Solv (ScandiaSyn meeting), Stuttgart (CGSW 22) and London (LAGB). I thank the audience of all these conferences for their comments. This study is part of a larger study of syntactic variation in Icelandic, supported by a grant from the Icelandic Science Fund (Rannís), principal investigator Höskuldur Thráinsson. 1.  Since the understood argument of Icelandic passives is usually an agent, I will use the term agent to refer to this argument even if it may bear other theta-roles.

 Jóhannes Gísli Jónsson

Although the Passive Analysis and the Active Analysis are two contrasting analyses of new impersonals, it is still quite possible that each analysis is correct for a certain class of speakers; the new impersonal may be an active construction for some speakers but a passive construction for others. In this paper, I will argue that the Passive Analysis holds for all speakers using the new impersonal as I take this to be the strongest hypothesis consistent with the data. The paper is organized as follows. Section 2 is a brief introduction to Icelandic passives and the new impersonal construction. This is followed by a discussion of some theoretical issues concerning the new impersonal in section two, in particular the properties of the understood agent and checking of accusative case. After a brief review of comparative data from the -no/to construction in Polish and Ukranian, a critical evaluation of the Active Analysis of Maling & Sigurjónsdóttir (2002) for Icelandic is presented in section 4. Some further arguments against the Active Analysis are presented in section 5, based on a new study of the new impersonal. Finally, the main points of the paper are summarized in section 6.

2.  Icelandic passives and the new impersonal Icelandic has a personal passive with DP-movement, as shown in (2). In such passives, structural accusative case disappears, as seen in (1a) vs. (2a), whereas lexical (dative or genitive) case is preserved, as shown in (1b) vs. (2b).2 (1) a. Einhver barði mig someone hit me.acc b. Einhver hjálpaði stelpunum someone helped the.girls.dat (2) a. Ég var barinn I.nom was hit.m.nom.sg. b. Stelpunum var hjálpað the.girls.dat was.3.sg helped.def ‘The girls were helped’

Nominative subjects of passives trigger agreement with the finite verb and the passive participle but default features appear in the presence of a dative subject, i.e., third

2.  The following abbreviations are used in the glosses: 3 = third person, acc = accusative case, dat = dative case, def = default agreement (nominative, neuter, singular), f = feminine, imp = impersonal, inst = instrumental case, m = masculine, n = neuter, nom = nominative case, pl = plural, pass = passive, prt = particle, refl = reflexive, and sg = singular.

The new impersonal as a true passive 

person singular on the finite verb and neuter singular nominative on the passive participle. I will use the term personal passive to refer to canonical passives of transitive verbs, irrespective of whether the finite verb agrees with the structural subject or not. DP-movement is not obligatory in personal passives if the DP complement of the passive participle is indefinite. In such passives, expletive það is inserted, as in (3a), unless the clause-initial position is occupied, e.g., by a null operator, as in (3b). As shown in (4), definite DPs must undergo DP-movement: (3) a. Það voru keyptir stólar there were bought.m.nom.pl chairs.m.nom.pl ‘Some chairs were bought’ b. Var kastað tómötum í söngvarann? was thrown.def tomatoes.dat at the.singer ‘Were tomatoes thrown at the singer?’ (4) a. *Það voru keyptir stólarnir there were bought.m.nom.pl the.chairs.m.nom.pl ‘The chairs were bought’ b. *Var kastað tómötunum í söngvarann? was thrown.def the.tomatoes.dat at the.singer ‘Were the tomatoes thrown at the singer?’

Like most other Germanic languages, Icelandic has impersonal passives of unergative verbs. This is illustrated below with the verbs dansa ‘dance’ and horfa ‘look’: (5) a. Það var dansað í allt kvöld there was danced.def in all evening ‘There was dancing all evening’ b. Í kvöld verður horft á leikinn tonight will.be looked.def on the.match ‘Tonight, we will watch the match’

In addition to personal and impersonal passives, Icelandic has an impersonal construction where the DP complement of the passive verb stays in situ and behaves like an object. Thus, accusative and definite DPs are possible in this construction, as shown in (6). There is no agreement in these examples since only nominative DPs trigger agreement. (6) a. Það var barið mig there was hit.def me.acc ‘I was hit’ b. Þess vegna var hjálpað stelpunum therefore was helped.def the.girls.dat ‘Therefore the girls were helped’

 Jóhannes Gísli Jónsson

This construction uses the auxiliaries vera ‘be’ or verða ‘become’ and a passive participle just like a regular passive. Examples like (6a) e.g., are ungrammatical without passive morphology:3 (7) *Það barði mig there hit me.acc ‘I was hit’

In view of its affinity with canonical passives, this construction is often referred to as the new passive. However, I will continue to use the more theory-neutral term new impersonal (construction) in this paper in order not to prejudge the contentious issue of how this construction should be analysed. The new impersonal seems to be a relatively recent phenomenon in Icelandic as the oldest attested examples are from the middle of the 20th century.4 It is a substandard construction and mostly used by children and adolescents. By contrast, the personal passive is accepted by all speakers, including those who use the new impersonal (Maling & Sigurjónsdóttir 2002). As far as I know, impersonal passives of unergative verbs are also accepted by all speakers of Icelandic. Examples like (6a) display two characteristics of the postverbal DP in new impersonals: accusative case and definiteness. By contrast, examples with dative DPs like (6b) are new impersonals only by virtue of the definiteness of the DP. Examples with a postverbal indefinite dative like (3b) are therefore ambiguous between a personal passive without DP-movement and the new impersonal. No such ambiguity arises with indefinite accusative DPs; for instance, (8) below is clearly a new impersonal: (8) Það var keypt stóla there was bought.def chairs.acc ‘(Some) chairs were bought’

The DP complement in the new impersonal is an object. This is shown not only by the possibility of accusative case but also by control facts. As shown in (9c), the DP complement cannot control PRO in infinitival adjuncts in contrast to true subjects, as in (9a). An unmoved DP complement in personal passives can also control PRO, as in (9b), although such examples are less acceptable than examples with DP-movement like (9a).

3.  This example is grammatical in the irrelevant reading where það is interpreted as an argument. 4.  A few examples with a definite dative DP like (6b) are actually attested in Old Icelandic. However, since Old Icelandic had a relatively free word order compared to Modern Icelandic, these examples should probably be analysed as regular passives (see Eythórsson 2008 for further discussion).

(9) a.

The new impersonal as a true passive 

Tveir menn voru ráðnir án þess að two.nom men.nom were hired.m.nom.pl without it to

hafa næga menntun have enough education ‘Two men were hired without having enough education’ b. ?Þá voru ráðnir tveir menn án þess að then were hired.m.nom.pl two.nom men.nom without it to hafa have

næga menntun enough education

‘Then, two men were hired without having enough education’ c. *Þá var ráðið tvo menn án þess að then was hired.def two.acc men.acc without it to hafa næga menntun have enough education ‘Then, two men were hired without having enough education’

Another object property is that an accusative argument in new impersonals cannot undergo DP-movement (Maling & Sigurjónsdóttir 2002). Moreover, the postverbal argument is exempt from the definiteness restriction, which only affects subjects below the canonical subject position in Icelandic, as in (10):5 (10) *Það hefur komið Ólafur of seint í skólann í marga daga there has come Olaf too late in the.school in many days ‘Olaf has come to school too late for many days’

This example was tested by Maling & Sigurjónsdóttir (2002) and rejected by nearly all the participants. Thus, it seems that the definiteness restriction holds for all speakers, including those who use the new impersonal. If the definiteness restriction in Icelandic is indeed restricted to subjects, we do not expect to find any speakers that systematically obey a definiteness restriction in new impersonals by accepting only indefinite (accusative) DPs. As far as I know, this is correct, although it is probably true that indefinite accusative DPs are generally more acceptable in new impersonals than definite accusative DPs. In this paper, I will focus on the syntax of new impersonals and leave aside the pragmatics of the new impersonal. The pragmatics of the new impersonal have never been investigated but it seems to me that new impersonals are primarily used in “bare happenings”, a situation type characterized by low topicality of both agent and patient. Sansò (2006) claims that this is also the main function of the Polish -no/to construction 5.  As Thráinsson (2007:277) points out, this particular example is not ideal for testing the definiteness restriction since various indefinite DPs are not acceptable either in the postverbal position here.

 Jóhannes Gísli Jónsson

discussed in 4.1 below. As Sansò (2006:242) explains, this situation type involves “events in which the agent is conceptualised as sufficiently unimportant to be backgrounded even though the patients in these clauses are not particulary topical.” Personal passives in Icelandic have a wider range of uses since they cannot only be used in bare happenings; they also serve the function of highlighting the patient argument by moving it to subject position.

3.  Some theoretical issues 3.1  Maling & Sigurjónsdóttir (2002) In the most detailed study of the new impersonal to date, Maling & Sigurjónsdóttir (2002) argue that it is an active impersonal construction with a thematic null subject as shown in (11): (11) Það var [IP pro [VP barið mig ]] there was hit me.acc

This structure makes use of two important facts about Icelandic syntax: (i) Icelandic is a V2 language, and (ii) the expletive það must precede the finite verb. In my view, there are various problems associated with treating expletive það as a left-peripheral element, as in (11) above, rather than a structural subject, but this need not concern us here (see Jónsson 1996:46–50 and references cited there). On the Active Analysis of Maling & Sigurjónsdóttir (2002), the new impersonal contrasts with true passives in that the former construction has a thematic null subject but passives only have an understood agent.6 On the Passive Analysis, both constructions have an understood agent but no thematic subject. The presence of an understood or “implicit” agent in passives in Icelandic and many other languages is shown by the fact that the agent can license rationale clauses, as in (12a). By contrast, unaccusative verbs are incompatible with such clauses since they do not have any understood agent, as shown in (12b): (12) a.

Launin voru hækkuð til að halda starfsfólkinu ánægðu the.salaries.nom were raised to keep the.staff happy

b. *Launin hækkuðu til að halda starfsfólkinu ánægðu the.salaries.nom increased to keep the.staff happy

6.  On the possibility of expressing the agent of both constructions in a by-phrase, see 4.3.1 and 5.1 below.

The new impersonal as a true passive 

The understood agent of passives can be analyzed in at least two different ways: (i) as part of the lexical-semantic representation of the passive verb but bound in the argument structure (Grimshaw 1990), or (ii) respresented by the passives morpheme itself (Jaeggli 1986 and Baker, Johnson & Roberts 1989). I will leave the issue open here since nothing crucial hinges on the choice between these two approaches. Maling & Sigurjónsdóttir (2002) present various arguments in favor of their analysis and these will be reviewed in section 4.3 below. In the following two subsections some theoretical issues concerning the understood agent and accusative case checking in new impersonals will be discussed. The main point of this discussion is to show that the properties of the understood agent raise various questions for the Active Analysis whereas the preservation of accusative case presents a challenge to the Passive Analysis.

3.2  The agent An important property of the new impersonal construction is that the agent cannot be overtly expressed in subject position: (13) *Einhver var barið mig someone was hit me.acc ‘Someone hit me’

In this respect, new impersonals are like regular passives in Icelandic. For example, the agent of impersonal passives cannot be overtly realized in subject position: (14) *Einhver var dansað someone was danced ‘Someone danced’

The overt agent in (13) and (14) is the weak quantifier einhver ‘someone’, which is semantically suitable when new impersonals and impersonal passives denote a particular event with an unknown agent. Still, this word is impossible in the subject position of (13) and (14) just like any other lexical item. The agent in both canonical passives and new impersonals in Icelandic must be animate.7 Thus, neither (15a) nor (15b) below can be understood such that natural forces, e.g., the sun or the rain, saved the crop: (15) a.

Uppskerunni var bjargað the.crop.dat was saved

b. Það var bjargað uppskerunni there was saved the.crop.dat

7.  As discussed by Maling (2006) and Sigurðsson & Egerland (to appear), the animate agent of Icelandic passives is usually human as the use of non-human animate agents is heavily restricted.

 Jóhannes Gísli Jónsson

The parallels between new impersonals and canonical passives illustrated in (13) – (15) above look like strong arguments for the Passive Analysis of the new impersonal. However, since the active -no/to construction in Polish is restricted in the same way (Kibort 2004:252–253), these facts should be compatible with the Active Analysis of new impersonals. For arguments that the Polish -no/to construction is an active construction despite its passive origins, see section 4.1 below. Nevertheless, the Active Analysis raises the question why the null argument of new impersonals cannot be overtly expressed in subject position and why it can only be animate. Moreover, the licensing of the null argument must be accounted for under the Active Analysis, especially since Icelandic generally disallows referential null subjects (see Sigurðsson 1989:123–196).8 Maling & Sigurjónsdóttir (2002) do not address these important issues and in that respect their analysis is clearly incomplete.

3.3  Accusative case Under the Active Analysis of Maling & Sigurjónsdóttir (2002), accusative case is preserved in the new impersonal due to the presence of a null external argument bearing nominative case. Thus, new impersonals are just like active clauses with overt nominative subjects in this respect. The preservation of accusative case in new impersonals is potentially problematic for the Passive Analysis which does not postulate any nominative argument in subject position. This runs counter to Burzio’s Generalization which rules out structural accusative case checking unless the verb has an external argument. However, as many scholars have observed, this link between an external argument and structural accusative case is theoretically problematic since these two phenomena should be independent of one another. I will assume, therefore, that Burzio’s Generalization should be replaced by a requirement that nominative case take priority over structural accusative case (see Yip, Maling & Jackendoff 1987; Haider 2000 and Woolford 2003, 2007 among many others). For convenience, this will be referred to as the Nominative First Requirement (NFR).

8.  For discussion of the null subject in the autonomous construction in Irish, see Stenson (1989) and McCloskey (2007). Note that Irish allows referential null subjects quite freely and this may have facilitated the reanalysis of the autonomous construction from a passive to an active construction.

The new impersonal as a true passive 

The NFR entails that the “absorption” of accusative case in canonical passives is due to the absence of a nominative DP. It could be argued under this view that new impersonals differ from canonical passives in having a (null or overt) nominative expletive in subject position, thereby making accusative case checking possible. However, this predicts incorrectly that accusative case on the complement of unaccusative verbs should be possible for those who accept the new impersonal. As shown in (16b) below, this is excluded for all speakers. (16) a. Það höfðu komið gestir í heimsókn there had come guests.nom in visit ‘Guests had come for a visit’ b. *Það höfðu komið gesti í heimsókn there had come guests.acc in visit

This suggests that the crucial difference between new impersonals and canonical passives concerns passive participles rather than the subject position. To capture this difference, we can modify the NFR by assuming that it is actually about case checking by functional heads. More specifically, the functional head υ (taking VP as its complement) cannot check accusative case unless T checks nominative case. This is a very natural approach if we assume that these two heads are always present and form the core of the functional architecture of finite clauses. The new impersonal can now be analysed as a construction where accusative case on the DP object is checked by some functional head other than υ, e.g., a head associated with participial morphology. Another possibility is to assume that accusative case in new impersonals is checked by the passive verb itself without the involvement of any functional head. The plausibility of this option is strengthened by the fact that this would not be the only example of “lexical” accusative case on objects in Icelandic (see Yip, Maling & Jackendoff (1987) for relevant discussion). There are other analyses of how accusative case can be checked in passives (see Baker, Johnson and Roberts (1989), Bowers (2002) and Lavine (2005)), and these analyses have been proposed to account for the fact that preservation of accusative case is compatible with passivization in some languages. As illustrated in 4.2 below, accusative case is preserved in the passive -no/to construction in Ukranian and the same is true of passives in Kannada and Nepali (Goodall 1993). Finnish may be yet another example (Manninen & Nelson 2004), although the accusative there is restricted to human pronouns. Accusative case also occurs in double object passives in Norwegian, Swedish and some dialects of English (Woolford 1993). In view of this, it is fair to conclude that the Passive Analysis of new impersonals cannot be rejected on the grounds that accusative case is excluded in true passives.

 Jóhannes Gísli Jónsson

4.  The Active Analysis of Maling & Sigurjónsdóttir (2002) On the basis of comparative data from Polish and Ukranian, Maling & Sigurjónsdóttir (2002) identify the following syntactic properties of impersonal constructions with a null thematic subject: (17) a. b. c. d.

No agentive by-phrase is possible. Binding of anaphors (reflexive and reciprocal) is possible. Control of subject-oriented adjuncts is possible. Nonagentive (“unaccusative”) verbs can occur in the construction.

These properties are supposed to distinguish active impersonal constructions from true passives, which do not have these properties. The Active Analysis predicts that new impersonals will show all these properties, at least to the extent that new impersonals have been reanalyzed as actives. As discussed in more detail in 4.3 below, none of the properties listed in (17) distinguish new impersonals from personal passives in Icelandic and thus they fail to provide evidence for the Active Analysis. The same conclusion is also reached by Eythórsson (2008); see also Barðdal & Molnar (2003) for a critique of the Active Analysis. The properties listed in (17) are based on evidence from Polish and Ukranian, especially the so called -no/to construction, which is historically a passive construction. Maling & Sigurjónsdóttir (2002) illustrate quite clearly that there is a contrast between the two languages in that Polish -no/to has become an active construction, whereas Ukranian -no/to is still a passive. This will be shown in 4.1 and 4.2 below where all the Polish and Ukranian examples as well as the glosses are taken from Maling & Sigurjónsdóttir (2002).

4.1  Polish Example (18a) below illustrates the personal passive in Polish whereas (18b) exemplifies the impersonal construction formed by the suffix -no/to: (18) a. Świątynia była zbudowana w 1640 roku church.nom was built.f.sg in 1640 year b. Świątynię zbudowano w 1640 roku church.acc built.imp in 1640 year

The personal passive in Polish is a true passive with respect to all the tests in (17). The passive allows a by-phrase, as in (19a), and disallows binding of anaphors, as in (19b). Moreover, the subject-oriented adjunct in (19c) must predicate over the structural subject. It cannot refer to the implicit agent, i.e., (19c) cannot mean that the robbers

The new impersonal as a true passive 

were drunk. Finally, (19e) shows that the Polish personal passive is incompatible with unaccusative verbs:9 (19) a. Jan był obrabowany przez nich John.nom was robbed.3.m by them b. *Swoja refl

własna ojczyzna była chawalona own fatherland.nom was praised.f.sg

c. Jan był obrabowany po pijanemu John.nom was robbed.3.m.sg while drunk ‘John was robbed while drunk’ [John was drunk] d. *Wazędzie było tańczone everywhere was.n.sg danced.n.sg Intended: There was dancing everywhere e. *Dawniej było umeriane młodo before was.n.sg died.n.sg young Intended: In the old days, people died young

The -no/to construction in Polish contrasts with the personal passive as it behaves like an active impersonal with respect to all the tests in (17). By-phrases are excluded, the understood agent can bind anaphors and control a subject-oriented adjunct and unaccusative verbs are possible: (20) a. Jana obrabowano (*przez nich) John.acc robbed.imp (*by them) ‘They robbed John (*by them)’ b. Zamknięto się w fabryce locked.imp refl in factory ‘They locked themselves in the factory’ c. Chawalono swoją własną ojczyznę praised.imp refl own fatherland.f.acc.sg ‘They praised their own fatherland’ d. Jana obrabowano po pijanemu John.acc robbed.imp while drunk ‘They robbed John while (they were) drunk’ e. Tańczono wazędzie danced.imp everywhere ‘There was dancing everywhere’

9.  Passives of unergative verbs are also excluded as shown by the ungrammaticality of (19d).

 Jóhannes Gísli Jónsson

f. Dawniej umeriano młodo before died.imp young ‘In the old days, people died at a young age’

The status of Polish -no/to as an active construction is further supported by the fact that it can combine with the regular passive in Polish (see Kibort 2004:260–261 and references cited there). According to Frajzyngier (1982:273–4), there are two additional differences between passives in Polish and the -no/to construction:  (i) passives are incompatible with agent-oriented adverbs like ‘unwillingly’ in contrast to the -no/to construction, and (ii) passives are compatible with non-human causers whereas -no/to is not. While these differences do not necessarily show that -no/to is an active construction, they illustrate that there is a very clear distinction between canonical passives and -no/to in Polish.

4.2  Ukranian As Maling & Sigurjónsdóttir (2002) observe, the -no/to construction in Ukranian behaves like a passive construction (see also Sobin 1985 and Lavine 2005).10 As exemplified below, the agent can be expressed in an instrumental phrase, corresponding to an English by-phrase, the understood agent cannot bind anaphors or control a subject-oriented adjunct and unaccusative verbs are excluded in this construction: (21) a. Mojim mylym mene zradženo my.inst beloved.inst me.acc betrayed.pass ‘I was betrayed by my beloved’ b. *Svoju žinku bulo obmaneno self ’s wife.acc was deceived.pass Intended: Someone deceived his wife c. *Povernuvšys’ dodomu, hroši bulo znajdeno returning home money was found.pass Intended: Having returned home, the money was found. d. *Umerto/*Zaxvorito/*Prijixato died.pass/got.sick.pass/arrived.pass

The passive auxiliary bulo ‘was’ is possible in the Ukranian -no/to construction. This is not the case in Polish where the -no/to forms are finite. The morphology of -no/to thus provides further evidence that it is a passive construction in Ukranian but an active construction in Polish.

10.  Ukranian also has a regular periphrastic passive which patterns like the Polish periphrastic passive but this is not illustrated here.

The new impersonal as a true passive 

4.3  Icelandic The Active Analysis of Maling & Sigurjónsdóttir (2002) entails that the new impersonal in Icelandic is quite similar to the -no/to construction in Polish, despite the clear morphological differences between the two constructions. Maling & Sigurjónsdóttir (2002) base their claims on a written survey that was carried out in 1999–2000. The participants in this survey were 1731 tenth graders (age 15–16) in 65 schools in Iceland and 205 adult controls across the country. Results from 1695 students (845 males and 850 females) and 200 adults were used. The tenth graders were divided into two classes: those who live in “Inner Reykjavík” and those who live outside of that area, i.e., in the suburbs of Reykjavík and outside of Reykjavík. For convenience, the two groups of tenth graders will be referred to as the IR-group (Inner Reykjavík) and the E-group (elsewhere group). The justification for this division is that students in the latter group were much more likely to accept the new impersonal. This is exemplified in (22), where the percentages show the acceptability scores for these sentences: Triadic verbs with acc or dat indirect object

Elsewhere Inner Rvík Adults

(22) a. Það var beðið mig að vaska upp 74% it was asked me.acc to wash up ‘I was asked to do the dishes’



b. Það var sagt mér að taka til it was told me.dat to clean up ‘I was told to clean up’




These numbers show a clear difference between the E-group and the IR-group and also between the these two groups and the adults. In the following subsections, results from all these groups will be shown although the E-group is clearly the most important group in the study of Maling & Sigurjónsdóttir (2002).

4.3.1  By-phrases The Active Analysis predicts that by-phrases should be impossible in the new impersonal since by-phrases are the hallmarks of passive constructions. However, this is not borne out by the data. Let us first consider by-phrases in personal passives: Agentive by-phrase in grammatical control sentences

Elsewhere Inner Rvík Adults

(23) a. Honum var sagt upp af forstjóranum 87% he.dat was fired prt by the.director



b. Það var samþykkt af öllum í bekknum 95% it was agreed by all in the.class



að fara í keilu to go bowling

 Jóhannes Gísli Jónsson

The acceptability rate is quite high here in all the groups and only slightly lower than one would expect in personal passives without a by-phrase. A rather different picture emerges with by-phrases in the new impersonal: Agentive by-phrase in the new impersonal (24) a. Það var skoðað bílinn it was inspected the.car.acc af by b. Það it

Elsewhere Inner Rvík Adults 33%






bifvélavirkjanum the.car.mechanic var sagt honum upp was fired him.dat prt

af forstjóranum

by the.director

The results for the E-group are boldfaced to emphasize the fact that by-phrases in the new impersonal are much more acceptable than the Active Analysis predicts. Since agentive by-phrases are only possible in passives in Icelandic, we can conclude e.g., that at least 33% of speakers in the E-group analyse the new impersonal in (24a) as a passive. The average number for new impersonals with inanimate accusatives given by Maling & Sigurjónsdóttir (2002:113) suggest that (24a) without a by-phrase would have scored about 55% in the E-group. Thus, the presence of a by-phrase seems to reduce the acceptability rate of (24a) only by a half or even less in that group. It is worth emphasizing that agentive by-phrases are restricted to passives in Icelandic. They cannot e.g., be used to refer to the understood agents of causative complements or the impersonal modal construction: (25) a.

Ég lét I let

gera við tölvuna (*af Jóni ) repair the.computer  by John

b. Það þarf að þvo gólfið (*af einhverjum) there must to clean the.floor  by someone

Returning to the results in (24), the numbers for the IR-group and the adults are very low and this could be taken as an argument for the Active Analysis. In my view, these numbers only show that the presence of a by-phrase sharply reduces the acceptability of the new impersonal among speakers who generally reject this construction. Presumably, this drop in acceptability is due to the fact that by-phrases in Icelandic passives are usually bad if there is no DP-movement. For instance, by-phrases are excluded in impersonal passives like (26): (26) *Það var sungið af tveimur kórum there was sung by two choirs

The new impersonal as a true passive 

This suggests that there are two problems with examples like (24) for many speakers: (a) they are new impersonals, and (b) they have a by-phrase but no DP-movement. For those who accept the new impersonal it is only the latter problem that leads to a decline in acceptability.11 Hence, the acceptability rate for by-phrases in new impersonals is what we would expect in a passive construction that does not have DP-movement. Further support for this conclusion can be seen in the results of a new study of the new impersonal discussed in section 5 below.

4.3.2  Binding of anaphors Another prediction of the Active Analysis is that binding of anaphors by the null subject should be quite free in the new impersonal. However, the results reported by Maling & Sigurjónsdóttir (2002) show that this is quite restricted. This is exemplified in (27): Binding of possessive reflexives (27) a. Á kvöldin var skoðað in the.evenings was checked

Elsewhere Inner Rvík Adults 32%









d. Það var oft kaffært bróður 5% it was often dunked brother.acc



tölvupóstinn sinn the.e-mail.acc SELF’s b. Í morgun var hrint systur this morning was pushed sister.dat sinni af hjólinu SELF’s off the.bike c. Það var klippt hárið á it was cut the.hair.acc on dúkkunni sinni the.doll SELF’s

sinn í sundlauginni SELF’s in the.swimming.pool

I leave out of consideration all the examples of simple reflexives tested by Maling & Sigurjónsdóttir (2002) since these reflexives may not be true anaphors. I also ignore examples of anaphor binding in impersonal passives because they are not directly relevant to the issue at hand (but see Eythórsson 2008 for further discussion).

11.  We cannot test whether DP-movement improves by-phrases in new impersonals with monotransitive verbs since DP-movement is excluded in such cases. For discussion of DPmovement in new impersonals with ditransitive verbs, see 5.3 below.

 Jóhannes Gísli Jónsson

All the numbers in (27) are very low, except for (27a), but this is because the examples in (27b–d) are pragmatically odd.12 Still, the example in (27a) is fine and it shows that a reflexive possessive reduces the acceptability of new impersonals considerably, suggesting that they are passives rather than actives. Another problem for the Active Analysis is that anaphor binding is not a very reliable test for the presence of a null thematic subject because binding of anaphors by the implicit agent of personal passives is possible: (28) a. Á kvöldin var skoðaður tölvupóstur frá börnunum sínum in the.evening was checked e-mail.nom from the.children SELF’s b. Sumt er bara gert fyrir sjálfan sig some is just done for oneself ‘Some things you only do for yourself ’

Examples like (28) were not tested by Maling & Sigurjónsdóttir (2002) but in my judgment they are acceptable. Similar examples from other languages can also be found in the literature, e.g., Norwegian (Lødrup 2007) and English: (29) Such privileges should be kept to oneself (Baker, Johnson & Roberts 1989)

Apparently, the best examples of this kind in English involve generic sentences with the anaphor oneself like (29) (Ken Safir p.c.). Icelandic is slightly more liberal than English here by allowing reflexive binding both in examples with an habitual reading like (28a) and generic sentences like (28b).

4.3.3  Subject-oriented adjuncts Implicit agents of canonical passives in Icelandic license various agent-oriented adjuncts, e.g., adverbs like viljandi ‘deliberately’ and rationale clauses: (30) a. Stóllinn var eyðilagður viljandi the.chair was destroyed deliberately b. Málverkið verður selt til að afla fjár the.painting will.be sold to raise money

12.  The problem with the examples in (27b–d) seems to be that they are incompatible with the agent defocusing that is characteristic of new impersonals as well as passives. Thus, the presence of the reflexive possessive necessarily calls the hearer’s attention to the understood agent, the antecedent of the possessive, as it is difficult to make sense of these examples without knowing the identity of the agent. No such problem arises in (27a) where the possessive is essentially made redundant by real-world knowledge (i.e., people check their own e-mails rather than other people’s e-mails).

The new impersonal as a true passive 

Maling & Sigurjónsdóttir (2002) maintain that certain adjuncts require a syntactic (subject) controller. As an argument they provide an example of a personal passive and a participial adjunct which they give two question marks: (31) Valsinn var dansaður skellihlæjandi the.waltz was danced laughing.uproariously

This example is claimed to be bad because there is no syntactic controller here, only an understood agent. The numbers from the survey of Maling & Sigurjónsdóttir (2002) also seem to support their claim that the new impersonal contains a syntactic subject capable of licensing an adjunct: Participial adjuncts (32) Það var lesið minningargreinina grátandi it was read the.obituary.acc crying

Elsewhere Inner Rvík Adults 62%



The acceptability rate for (32) is quite high in the E-group and the IR-group, suggesting that participial adjuncts are perfectly fine in the new impersonal. However, the problem is that there is no clear contrast with personal passives. Examples like (31) above are acceptable in the right context, and the same is true of (33) below, which is the passive equivalent of (32): (33) Minningargreinin var lesin grátandi the.obituary.nom was read crying

Maling & Sigurjónsdóttir (2002) did not test examples like (33) and this is rather surprising since similar examples are acceptable for many speakers in other languages, e.g., English: (34) a. At the commune, breakfast is usually eaten nude (Collins 2005:101) b. This song must not be sung drunk (Baker 1988:318)

All the adjuncts tested by Maling & Sigurjónsdóttir (2002) were uninflected adjuncts with the present participle suffix -andi. The Active Analysis is further undermined by the fact that depictives cannot be used in passives or the new impersonal in Icelandic if they are predicated of the understood agent. This is illustrated in (35) below:13 (35) a. *Morgunmatur er alltaf borðaður nakinn breakfast.nom is always eaten naked b. *Það er alltaf borðað nakinn there is always eaten naked c. *Það er alltaf borðað morgunmat nakinn there is always eaten breakfast.acc naked

13.  Although (35c) has not been tested in a survey, I am fairly sure it is impossible for most speakers.

 Jóhannes Gísli Jónsson

The adjective here is masculine singular nominative, just like adjectives predicated of arbitrary PRO in infinitival clauses, but this makes no difference here as other inflectional features on the adjective would make examples like (35) even worse. Under the Active Analysis, new impersonals like (35c) should be fully acceptable, contrary to fact. In this way, new impersonals contrast with the active -no/to construction in Polish where inflected adjectives can be predicated of the unexpressed agent. Interestingly, such adjectives in Polish must be virile (plural) rather than masculine singular as in control infinitivals (Kibort 2004:254–255). To salvage the Active Analysis, one could argue that depictives are incompatible with the understood agent of new impersonals because the agent does not have any gender or number features to control agreement on the depictive.14 However, this would not explain the contrast between the new impersonal and the Polish -no/to construction. A null subject lacking number and gender features would also be very different from overt DPs in Icelandic and attributing unique properties to the null subject of new impersonals would simply undermine the hypothesized presence of a such a subject.

4.3.4  Unaccusative verbs One prediction of the Active Analysis is that the new impersonal should be possible with all kinds of verbs, including unaccusative verbs, since an active construction should not be subject to any lexical semantic restrictions on the main verb. Maling & Sigurjónsdóttir (2002) tested this prediction with the verbs detta ‘fall’, hverfa ‘disappear’, svitna ‘sweat’ and deyja ‘die’.15 The results are shown in (36) below: Unaccusative verbs (36) a. Það var dottið í hálkunni it was fallen on the.ice

Elsewhere Inner Rvík Adults 55%





fyrir framan blokkina in front of the.appartment.block ‘People fell on the ice in front of the block of flats’ b. Það var horfið sporlaust it was disappeared traceless


í stjörnustríðinu in the.star.war

‘People disappeared without a trace in the star war’

14.  McCloskey (2007) makes a similar point regarding the fact that the understood agent of the autonomous construction in Irish cannot bind reflexives. 15.  They also tested the verb koma ‘come’ which was accepted by 58% of the adults, presumably since koma is fully acceptable in impersonal passives in Icelandic.

The new impersonal as a true passive 

c. Í nótt var ekkert svitnað in night was nothing sweated


in the.sleeping.bag







‘Last night, nobody sweated in the sleeping bag’ d. Það var dáið í bílslysinu it was died in the.car.accident ‘People died in the car accident’


The examples above look exactly like impersonal passives but since impersonal passives are generally impossible with unaccusative verbs, one could argue that these examples are acceptable to the extent that they can be analyzed as new impersonals. Hence, the prediction is that speakers of the E-group accept these examples more readily than speakers of the other two groups. It is certainly true that the E-group has the highest acceptability rate with all these unaccusative verbs except for svitna ‘sweat’ where there is a tie between the E-group and the adults.16 According to Maling & Sigurjónsdóttir (2002:127), these results support the Active Analysis as they indicate that the new impersonal is “beginning to extend its usage to nonagentive verbs which do not form passives in the standard language”. In my view, this conclusion is not warranted because the difference between the three groups of speakers is much smaller than in examples of the new impersonal, especially the difference between the two groups of adolescents. There are clear differences between individual examples in (36). For instance, (36d) has the lowest acceptability rate in all the groups. A possible explanation is that the event denoted by the verb deyja ‘die’ involves the greatest degree of affectedness of the understood argument, a drastic change of state that is nearly always irreversible. This makes deyja ‘die’ quite different from the agentive verbs that work best in impersonal passives. On the other hand, detta ‘fall’ has the highest acceptability rate among the adolescents. The reason may be that this verb usually entails an agentive activity (e.g., walking or running) prior to the actual event. Another potential factor is that (36a) is well-suited for agent defocusing as the implicit argument of (36a) is naturally understood as referring to an unspecified group of people rather than particular individuals.

16.  The example with svitna differs from the other examples here by the absence of the expletive það, which is rarely used in formal registers. Therefore, this example sounds more formal than the other examples and this may have increased the acceptability rate for svitna among the adults.

 Jóhannes Gísli Jónsson

5.  A new study of the new impersonal This section presents results from a recent survey of the new impersonal. This survey included 808 speakers in 26 locations across Iceland and four age groups (14–15 years, 20–25 years, 40–45 years and 65–70 years).17 The results of the survey corroborate the basic findings of Maling & Sigurjónsdóttir (2002) that the new impersonal is mostly used by young speakers whereas the personal passive is accepted by virtually everyone. A written questionnaire with 125 test sentences was used in the survey, including 30 examples of new impersonals and 12 examples of canonical passives. The other sentences, as well as 24 sentence pairs, were designed to test various other syntactic phenomena in Icelandic. To offset potential ordering effects, half of the participants answered one version of the questionnaire and the other half answered another verison that had the opposite order of the test sentences. The participants in the survey were instructed to judge the examples according to their own intuitions and they were given three choices, i.e., they could judge each example as acceptable, dubious or impossible. Each test sentence was preceded by an introductory sentence to provide a natural context for the test sentence. An example of this is shown in (37) below: (37) a. Anna hrækti á Anna spat at

markmanninn the.goal.keeper

(introductory sentence)

b. Það var strax sent hana út af (test sentence) there was immediately sent her.acc off

There are two methodological differences between this survey and that of Maling & Sigurjónsdóttir (2002) that should be noted. First, the participants in the survey of Maling & Sigurjónsdóttir (2002) had only two options, i.e., they had to either accept or reject the test sentences. To facilitate comparison between the two surveys, the number for the option “dubious” in the new survey will be evenly divided between “acceptable” and “impossible”. Using this method, the acceptability rate for (37b) above was 56% among the 9th graders (14–15 years) and 13% among the three adult groups combined. Treating all the adults as one group makes the numbers for this survey comparable to the adult group in the survey of Maling & Sigurjónsdóttir

17.  The survey was administered to almost 1200 speakers but to keep a proper balance between the different age groups, results from many 9th graders were not calculated. A very small number of participants were excluded for other reasons, such as giving too many “wrong” answers to control questions.

The new impersonal as a true passive 

(2002).18 Moreover, since more than 90% of the 9th graders in the new survey live outside of central Reykjavík, the 9th graders are roughly comparable to the E-group in the old survey. The second difference is that all the test sentences in the survey of Maling & Sigurjónsdóttir (2002) were given without context and this may have affected the acceptability rate for the test sentences in some cases. In general, the acceptability scores for new impersonals were lower in the new survey among the 9th graders. For example, the average score for dative animate DPs was 74% in the survey of Maling & Sigurjónsdóttir (2002) but 57% in the new survey. It is not clear how to explain this contrast but one possible factor is an increased awareness of the substandard status of new impersonals in recent years. As a result, the adolescents may have been more reluctant to accept new impersonals in the new survey. Due to limitations of space, we can only discuss those results from the new survey that are directly relevant for the choice between the two competing analyses of the new impersonal. This is done in the following three subsections on by-phrases (5.1), non-agentive verbs (5.2) and ditransitive verbs (5.3).

5.1  By-phrases The new survey had four test sentences with by-phrases, three with canonical passives and one with a new impersonal. These examples are shown in (38) below. The upper number after each example is the acceptability score for the yongest group in the survey, the 9th graders, but the number in brackets is for the adults: (38) a. Hann var smíðaður af færeyska bátasmiðnum he.nom was built by the.Faroese boatbuilder ‘It was built by the Faroese boatbuilder’

79% (89%)

b. Áður var þó önnur þyrla skoðuð earlier was still another helicopter inspected

63% (70%)

af viðgerðarmanninum by the.repair.man c. Fyrst var þó skoðuð önnur flugvél first was yet inspected another plane

60% (52%)

af flugvirkjanum by the.air.mechanic ‘Still, another plane was first inspected by the air mechanic’

18.  However, this also obscures the fact that the new impersonal is more acceptable among the youngest group of adults (20–25 year olds) than the two older groups.

 Jóhannes Gísli Jónsson

d. Það var strax lagað bílinn there was immediately fixed the.car.acc

38% (6%)

af bifvélavirkjanum by the.mechanic ‘The car was immediately fixed by the car mechanic’

The crucial example here is the new impersonal in (38d) which shows a much higher acceptability rate for a by-phrase than one would expect under the Active Analysis. Note that new impersonals with a definite accusative DP but no by-phrase had an average acceptability score of 57% among 9th graders in the new survey. This suggest that the presence of a by-phrase reduces the acceptability rate of new impersonals by 1/3 in this group. This drop in acceptability is hardly surprising since by-phrases also seem to reduce the acceptability rate of regular passives, especially passives without DP-movement, as in (38c).

5.2  Non-agentive verbs As discussed in section 4.3.4, one clear prediction of the Active Analysis is that new impersonals should be spreading their use to non-agentive verbs that are excluded from canonical passives. Maling & Sigurjónsdóttir (2002) tested this prediction with unaccusative verbs but three non-agentive transitive verbs were tested in the new survey, eignast ‘acquire’, eiga ‘own’ and líka ‘like’. The results are shown in (39): (39) a. Um daginn var loksins eignast nýjan bíl on the.day was finally acquired new.acc car.acc ‘They got a new car the other day’

28% (5%)

b. Það var samt alltaf átt marga hesta there was still always owned many.acc horses.acc ‘People still kept owning many horses’

33% (6%)

c. Það er bara ekki líkað svona fólk there is just not liked such people.nom ‘Such people are just not liked’

16% (1%)

The corresponding personal passives, as in (40), are ungrammatical. Note that (40a,b) are passives without DP-movement but these examples would be equally bad with DP-movement. (40) a. *Um daginn var loksins eignastur nýr bíll on the.day was finally acquired new.nom car.nom ‘They got a new car the other day’ b. *Það voru samt alltaf áttir margir hestar there were still always owned many.nom horses.nom ‘People still kept owning many horses’

The new impersonal as a true passive 

c. *Svona fólk er bara ekki líkað such people.nom are just not liked ‘Such people are just not liked’

The results in (39) display a striking contrast between the 9th graders and the adults. Presumably, this is because adults strongly reject new impersonals if there is some further factor to reduce their acceptability, e.g., the presence of a by-phrase, as in (38d). The crucial issue here is whether the examples in (39) provide evidence for the Active Analysis of new impersonals. It is not clear to me that they do since these examples are significantly less acceptable than new impersonals with agentive verbs. This is especially clear in (39c) which features a verb taking a dative subject. To evaluate the acceptability scores in (39a–b), these examples can be compared to examples of the new impersonal with an indefinite accusative DP and an agentive monotransitive verb. There were seven such test sentences in the new survey and the acceptability rate for them varied between 46% – 72% among the adolescents and 11% – 33% among the adults. The average rate for these examples was 63% in the first group and 21% in the second group. Thus, the reduction in the acceptability rate for (39a,b) is approximately one half.

5.3  Ditransitives New impersonals are possible with ditransitive verbs, just as personal passives. New impersonals with ditransitive verbs come in two varieties; either both objects stay in situ following the passive verb, as in (41a), or the indirect object moves by DP-movement to the subject position, as in (41b).19 (41) a. Það var sýnt þeim bæklinga áður en þau fóru 52% there was shown them.dat brochures.acc before they left (12%) ‘They were shown brochures before they left’ b. Var þeim ekki einu sinni sýnt íbúðina fyrst? was them.dat not even shown the.apartment.acc first ‘Were they not even shown the apartment first?’

59% (19%)

The high acceptability rate for (41b) is very interesting because the presence of the indirect object in subject position rules out the possibility of a null thematic subject in the same position. Thus, examples like (41b) provide a very strong argument against the Active Analysis.20

20.  Recall that DP-movement of an accusative DP is excluded in new impersonals. 21.  It is quite clear that the dative DP in (41b) is in subject position, immediately following the finite verb.

 Jóhannes Gísli Jónsson

In view of examples like (41b), the Active Analysis of Maling & Sigurjónsdóttir (2002) could be modified by assuming that the null subject is in a lower subject position. Such a position is indeed available for indefinite subjects like einhver ‘somebody’ in active clauses in Icelandic, as shown in (42): (42) Það hafði einhver sýnt þeim there had somebody.nom shown them.dat

íbúðina the.apartment.acc

Still, this will not help for two reasons. First, DP-movement of the indirect object to the lower subject position is quite possible in new impersonals, as shown in (43), suggesting that this position is not occupied by a null subject: (43) Það var einhverjum sýnt íbúðina there was somebody.dat shown the.apartment.acc

The other problem is that movement of the indirect object across the lower subject position would violate well-known locality restrictions on movement, e.g., Shortest Move (Chomsky 1995) or Relativized Minimality (Rizzi 1990). The effect of these restrictions can be seen when an indirect object is moved across an overt subject, as in (44): (44) *Hafði þeim einhver sýnt íbúðina? had them.dat somebody.nom shown the.apartment.acc

The analysis of Collins (2005) offers a possible solution to the problem that examples like (41b) pose for the Active Analysis of new impersonals. The basic idea is that the complement of a passive verb can be “smuggled” across a null subject by moving it as part of a bigger phrase and then moving it independendly to the highest subject position. The problem is that this approach seems to require that the null subject is in a fairly low position where overt subjects are impossible. Moreover, smuggling is theoretically spurious since it it not clear what the driving force behind smuggling really is and how smuggling can be constrained so that it does not generate ill-formed structures of various kinds.

6.  Conclusion In this paper, I have discussed the new impersonal construction in Icelandic, a construction that displays passive morphology and hosts a DP complement that behaves like the object of a transitive verb. Contra Maling & Sigurjónsdóttir (2002), I have argued that the new impersonal is not an active construction with a thematic null subject. The new impersonal should rather be analysed as a true passive with an understood agent. Whereas some arguments concerning the status of the new impersonal are inconclusive and require further study, all the arguments that are reasonably clear suggest that the

The new impersonal as a true passive 

new impersonal is a passive construction. These arguments involve three important facts about the new impersonal: (i) the possibility of using an agentive by-phrase, (ii) the ban against depictives (inflected adjuncts), and (iii) the possibility of DP-movement of an indirect object. The first point can already be seen in the survey of Maling & Sigurjónsdóttir (2002) and is further corroborated by the results of a new survey of the new impersonal but the second and the third point are novel arguments.

References Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago IL: The University of Chicago Press. Baker, M, Johnson, K. & Roberts, I. 1989. Passive arguments raised. Linguistic Inquiry 20: 219–251. Barðdal, J. & Molnar, V. 2003. The passive in Icelandic – compared to mainland Scandinavian. In Structures of Focus and Grammatical Relations, J. Hetland & V. Molnar (Eds), 231–260. Tübingen: Niemeyer. Bowers, J. 2002. Transitivity. Linguistic Inquiry 33: 183–224. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Collins, C. 2005. A smuggling approach to the passive in English. Syntax 8(2): 81–120. Eythórsson, T. 2008. The new passive in Icelandic really is a passive. In Grammatical Change and Linguistic Theory: The Rosendal Papers, T. Eythórsson (Ed.), 173–219. Amsterdam: John Benjamins. Frajzyngier, Z. 1982. Indefinite agent, passive and impersonal passive: A functional study. Lingua 58: 267–290. Goodall, G. 1993. On case and the passive morpheme. Natural Language and Linguistic Theory 11: 31–44. Grimshaw, J. 1990. Argument Structure. Cambridge MA: The MIT Press. Haider, H. 2000. The license to license. Licensing of structural case plus economy yields Burzio’s Generalization. In Arguments and Case. Explaining Burzio’s Generalization, E. Reuland (Ed.), 31–55. Amsterdam: John Benjamins. Jaeggli, O. 1986. Passive. Linguistic Inquiry 17: 587–622. Jónsson, J.G. 1996. Clausal architecture and case in Icelandic. Ph.D. dissertation, University of Massachusetts, Amherst. Kibort, A. 2004. Passive and passive-like constructions in English and Polish. Ph.D. dissertation, University of Cambridge. Lavine, J.E. 2005. The morphosyntax of Polish and Ukranian –no/–to. Journal of Slavic Linguistics 13: 75–117. Lødrup, H. 2007. Norwegian anaphors without visible binders. Journal of Germanic Lingustics 19: 1–22. Maling, J. & Sigurjónsdóttir, S. 2002. The ‘New Impersonal’ construction in Icelandic. Journal of Comparative Germanic Linguistics 5: 97–142. Maling, J. 2006. From passive to active: Syntactic change in progress in Icelandic. In Demoting the agent: Passive, Middle and Other Voice Phenomena, B. Lyngfelt & T. Solstad (Eds), 197–223. Amsterdam: John Benjamins.

 Jóhannes Gísli Jónsson McCloskey, J. 2007. The grammar of autonomy in Irish. Natural Language and Linguistic Theory 25: 825–857. Manninen, S. & Nelson, D. 2004. What is a passive? The case of Finnish. Studia Linguistica 58: 212–251. Rizzi, L. 1990. Relativized Minimality. Cambridge MA: The MIT Press. Sansò, A. 2006. ‘Agent defocusing revisited’. Passive and impersonal constructions in some European languages. In Passivization and Typology: Form and function, W. Abraham & L. Leisiö (Eds), 232–273. Amsterdam: John Benjamins. Sigurðsson, H.Á. 1989. Verbal Syntax and Case in Icelandic. Ph.D. dissertation, University of Lund, Sweden. Sigurðsson, H.Á. & Egerland, V. To appear. Impersonal null-subjects in Icelandic and elsewhere. Studia Linguistica. Sobin, N. 1985. Case assignment in the Ukranian morphological passive construction. Linguistic Inquiry 16: 649–622. Stenson, N. 1989. Irish autonomous impersonals. Natural Language and Linguistic Theory  7: 379–406. Thráinsson, H. 2007. The syntax of Icelandic. Cambridge: CUP. Woolford, E. 1993. Symmetric and asymmetric passives. Natural Language and Linguistic Theory 11: 679–728. Woolford, E. 2003. Burzio’s Generalization, markedness, and constraints on nominative objects. In New Perspectives on Case Theory, E. Brandner & H. Zinsmeister (Eds), 301–329. Stanford CA: CSLI. Woolford, E. 2007. Case locality: Pure domains and object shift. Lingua 117: 1591–1616. Yip, M., Maling, J. & Jackendoff, R. 1987. Case in tiers. Language 63: 217–250.

Anaphoric distribution in the prepositional phrase Similarities between Norwegian and English Jenny Lederer The distribution of anaphoric pronouns in prepositional phrases has garnered much attention in the literature on antecedent binding since, contrary to fundamental binding principles, this syntactic environment appears to allow either reflexive or coreferential nonreflexive pronouns (c.f. Safir 2004; Reinhart & Reuland 1993; Pollard & Sag 1992). This paper takes a closer look at two prepositional phrase contexts in English and Norwegian, which seem to allow the reflexive pronoun when the PP superficially denotes directionality. With careful examination of the semantics of these constructions, it is shown that a vague notion of directionality evoked in the formal syntax notion of the functional projection PATH is insufficient to capture the data’s distribution pattern. The grammar must make reference to more detailed spatial configurations in order to model the real-world examples.

1.  Introduction The pursuit of a syntactic model for spatial terms has developed into a growing area in current formal syntax research. This research focuses on two issues: what types and classes of words should be included in the category P (Svenonius 2004a) as well as how to formally model the internal semantic structure of these PP heads (Svenonius 2004b). Within this model head Ps are allowed functional projections such as PATH or PLACE, which are meant to capture the categorical semantics of spatial adpositions and are used to model distributional regularities found among spatial systems. Functional projections have also served as a theoretical tool to explain certain interactions between spatial terms and other areas of the grammar. In this paper, we will specifically address one such area in which the semantics of prepositions in Germanic affects the distribution of anaphora in the prepositional phrase. We will conclude that a finite set of phrasal functional projections does not accurately model a real-world data set. Based on an interestingly similar distributional pattern in English and Norwegian, we will show that the semantics of prepositions need to be understood in the spatial

 Jenny Lederer

context in which they occur, and formal theories which insert only limited semantics into a syntactic apparatus fail to explain important idiosyncrasies in the data.

1.1  Preliminary data Wechsler (1997:15) noted that English directional prepositions pattern differently from locative prepositions with regard to anaphoric pronoun distribution. Sentence (1) shows a directional preposition occurring with the reflexive pronoun, and (2) shows a locative, nondirectional preposition occurring with the nonreflexive pronoun (data from the British National Corpus1):

(1) Hei could have stuck pins into himselfi and it would have taken ten seconds for his body to complain. (FSP 2109)

(2) Hei put the opened bottle down next to himi and smelled the top. (CA3 641)

A semantically similar phenomenon exists in Norwegian. Because Norwegian nonreflexive pronouns have an anti-subject orientation, the Norwegian equivalent of (2) above is not possible; however, in Norwegian, the use of a reflexive versus a nonreflexive possessive pronoun interacts with the semantics of motion conveyed in the verb phrase. Tungseth (2003:480) shows that the choice of pronoun dictates whether the motion in the sentence will be interpreted as directional or nondirectional: (3) Jeg kastet Peri i svømmebassenget sitti (directional). I threw Per in swimming pool-def refl ‘I threw Per into his swimming pool.’

(Tungseth 2003:480 #12a)

(4) Jeg kastet Peri i svømmebassenget hansi (locative or directional). I threw Per in swimming pool-def pron (Tungseth 2003:480 #12b) ‘I threw Per in his swimming pool.’

The sentence in (3) with the reflexive has only the directed motion reading, in which Per’s trajectory starts outside the swimming pool and ends inside the swimming pool, while (4) with the nonreflexive is ambiguous between the directed motion reading and the located motion reading, in which the throwing event occurs within the confines of the pool.

1.2  Paper outline In the first part of this paper, the English examples in (1) and (2) will be unified with the Norwegian examples in (3) and (4). In sections 2 and 3 we will present more

1.  Data cited herein has been extracted from the British National Corpus Online service, managed by Oxford University Computing Services on behalf of the BNC Consortium. All rights in the texts cited are reserved.

Anaphoric distribution in the prepositional phrase 

details about the English and Norwegian data. In section 4 we include a discussion of how the PP data relates to syntactic binding principles. Before detailing a semantic analysis in section 7, we will evaluate a syntactic analysis based on functional projections. Laid out in sections 5 and 6, the syntactic analysis can account for some of the different pronominal distributions found among directional and locative prepositional phrases but does not capture the complexity of real-world data. Supported by other constructions in English and Norwegian as well as reflexive patterns in Spanish, we will make the argument that the semantics of reflexivity incorporates a spatial notion of proximity, which is responsible for the appearance of the reflexive forms in the directional cases above.

2.  Directional PPs in English Based on an analysis of over 10,000 examples in the British National Corpus and a total count of 1667 examples involving coreference between antecedent and pronoun (specifically, ones in which the PP is part of the VP and the pronoun’s referent is the subject and a clausemate antecedent), we can confirm the occurrence of the reflexive and nonreflexive pronoun is not equally distributed across prepositions (χ2(16, N = 1667) = 722.58, p < .001) as indicated in Table 1: Table 1.  Counts of pronoun distribution according to head preposition Preposition next to beneath behind In front of before below above beyond on top of on toward(s) around out of into in inside within Total




5 24 593 130 58 5 16 1 4 73 129 176 0 0 97 36 18 1365

0 0 2 2 4 2 11 4 2 41 5 23 13 37 81 21 54 302

5 24 595 132 62 7 27 5 6 114 134 199 13 37 178 57 72 1667

 Jenny Lederer

Table 2.  Percentages of pronoun distribution according to head preposition Preposition next to beneath behind In front of before below above beyond on top of on toward(s) around out of into in inside within TOTAL




100% 100% 99.6% 98% 94% 71% 59% 20% 67% 64% 96% 88% 0% 0% 54% 63% 25% 83%

0% 0% 0.4% 2% 6% 29% 41% 80% 33% 36% 4% 12% 100% 100% 46% 37% 75% 17%

100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%

Table 2 characterizes these same pronoun counts as percentages. Thus, each cell represents the percentage of reflexive versus nonreflexive pronouns to occur within the individual preposition context. Directionality can be isolated as one semantic factor that contributes to the unequal distribution of reflexive versus nonreflexive pronoun in PPs; prepositions with directional senses, such as into, on, and around, have a higher rate of occurrence with the reflexive. Shown in Table 3, at the macroscopic level, directional PPs do pattern with a slightly higher rate of reflexive pronouns than nondirectional contexts (χ2(1, N = 1667) = 7.29, p A)  263f., 267, 271–274 non-isomorphism  249, 262 Northern Italian dialects  21 Norwegian  66, 95, 176 fn. 4, 239 fn. 15, 289, 296, 307ff., 311–316, 320–323, 359, 367, 377 null subject  23, 31, 33, 286, 288 fn. 8, 295, 298, 304 Number (Nr)  254 Numeration  62, 104–106, 223 fn. 4

  Index of subjects & languages O object affected  269 benefactive  269 indirect  66f., 221f., 236–240, 256, 259, 262, 269 fn. 17, 293, 303ff. nominative  107, 254, 267 position  70, 76f. shift  76 fn. 10, 78 ff., 219, 242f. 273 Occurrence (occ)  62 Old Irish  184 fn. 14 Old High German  171, 198f., 201–206, 209ff., 215 Old Italian  6 fn. 4 Old Norse  261, 377 Optimality Theory (OT)  224f. Otfrid  199, 201ff. OV-language  42, 52, 79, 227, 230f., 234f. P parallel interpretation  121 parallelism requirement  121, 123, 125 fn. 5, 127, 132f., 143 parameter  126, 226, 261, 317, 319, 368, 370 parasitic gap  17, 79 participle  281–284, 361, 364, 366f., 369ff., 375, 383f. passive  77, 111, 189, 256, 265 fn. 22, 271f., 281 passim, 341f. periphrastic do  377 Person (Pn)  253ff. PF-interface  43, 103 fn. 15, 113 fn. 26, 187, 224, 225 fn. 5, 227, 243, 249ff., 256, 260, 262, 265, 267, 271, 273, 273 fn. 31, 274ff. phase theory  119f., 127–131, 131 fn. 10, 133, 223, 274f. phase impenetrability condition (PIC)  124, 127ff., 273f. phrase structure  60ff., 122–127, 140–147 Polish  259, 282, 285, 288, 290ff., 298 PP-fronting  15f., 22 fn. 18, 34 fn. 30 pre-field  210ff.

prepositional phrase  307 passim preterit  360f., 364, 366ff., 370f., 375, 378, 383f. Principle of Minimal Compliance  151, 153, 157 PRO  250ff., 255, 257, 265 fn. 20, 275, 284, 298 proposition  73–76 prosodic weakness  197, 199 fn. 1, 202 Proto-Indo-European  197f., 203, 210f. pseudo-cleft  41, 46, 51f., 56, 373 fn. 21 pseudopartitive construction  260 push-up  220ff., 232, 237–241 Q quasi-verbs  14 R radically disentangled morphology  251 raising  327 passim Generalised Raising Hypothesis (GRH)  327f. reconstruction  18, 20, 31, 166ff., 274 reduction  42f., 48, 69, 197–204, 209–212, 214 relative pronoun  70f. resumptive preposing  3, 4 fn. 1, 10, 16f., 26, 30 Rhaeto-Romance  91 fn. 5 right dislocation  27 fn. 24, 36 right periphery  27 fn. 24, 226 Romance  3, 275, 363, 382 root context  3f., 36, 235, 235 fn. 13 Rumanian  260 Russian  259f., 268 S Sanskrit  192 fn. 23 Scandinavian  14, 42 fn. 2, 87, 91, 95, 108 fn. 21, 115, 219, 222, 230 236 fn. 14, 258, 357, 361f., 364, 366, 375, 383, 386

scope  20, 36, 63, 66, 97, 108, 110, 111fn. 23, 121, 124ff., 138 fn. 20, 146f., 156f., 161, 165ff., 168 fn. 9, 341f., 345 fn. 13 Sibling Correlation  249, 263, 273, 275 sign language  273 fn. 31, 276 fn. 38 sisterhood  60, 71, 80 sisters  59 Spanish  21, 309, 317ff., 345 stress  197–212, 332f., 333 fn. 3 clash  201 fn. 4 main  198ff., 211 rhythmic  198–201, 209, 211f. 214 secondary  198–201, 201 fn. 4, 204, 206, 211 sentence  198ff., 203, 208 fn. 15, 211 strict cycle  102, 106, 350 strong pronoun  4, 31 structural parallelism  131f. subject criterion (see also EPP)  5, 22 fn. 18, 23f., 26, 29f. position  21, 70, 75f. quirky  252ff., 253 fn. 6, 264–267, 272, 334f. subject-initial main clause  44, 73, 76, 80 Swedish  219 fn. 1, 239, 239 fn. 15, 260f., 289, 362 fn. 5, 385 syntactic conspiracy  121 syntactic position  59–62, 79, 119, 121, 134f., 178, 313 T temporal anaphor  171–191, 362, 378 pronoun  362, 378 tense  75f., 178, 185f., 251, 357 passim absolute  360 relative  360 thematic roles  327 passim thematic shape conservation  243 Topic  3 passim, 63, 65, 70, 72, 74, 89, 91f., 109, 111–114, 174, 176f., 182–185

Index of subjects & languages    aboutness-shift  5ff., 13 familiar  5, 7, 13, 36 topicalization  3–5, 11–17, 21f., 26–31, 36, 59, 73f., 87f., 220 fn. 2 Transfer  250, 262, 271, 273ff. transitivity failure  60, 62, 64, 69 U Ukranian  282, 289f., 292 unaccusative  22 fn. 18, 76f., 111, 265 fn. 22, 272f., 286, 289–292, 298f., 302 understood agent  281f., 286f., 291f., 294, 296 fn. 12, 297f., 298 fn. 14, 304 unergative  32, 283f., 291 fn. 9 Universal Base Hypothesis  227f.

Universal Grammar (UG)  28, 37, 261, 385 V v (little)  76f., 224f., 227f., 231, 249, 251, 259 fn. 12, 264 fn. 17, 269–273, 275f., 372 Venetian  65, 74, 92 fn. 7 verb second (V2)  60, 73, 73 fn. 9, 85–99, 107–115, 119, 122, 171, 173ff., 187–190, 210ff., 357, 364f., 368ff., 375, 383 generalized  365, 368, 370 obligatory  365 requirement  122, 139ff. residual V2 context  29, 365, 375, 377 fn. 26 V3-construction  136 verb projection raising (VPR)  124, 126, 126 fn. 7, 141ff.

Voice  249, 259 fn. 12, 264–267, 270 fn. 27, 271–276 VO-language  42, 228–235 VP shell  99, 106, 108 fn. 22 W Walpiri  250, 250 fn. 2 weak pronoun  5, 30 fn. 17, 21, 23, 31, 33, 88, 95, 220, 237 West Germanic  41, 43–49, 52–57, 60, 72–76, 79, 119 passim, 129 fn. 9 WH-element/phrase  17, 21, 29 fn. 26, 34f., 56, 64f., 70, 72f., 74f., 88, 110, 111, 111 fn. 23, 123 fn. 2, 131, 137, 157f., 173, 175, 175 fn. 5 Y Yiddish  87

Linguistik Aktuell/Linguistics Today A complete list of titles in this series can be found on the publishers’ website, www.benjamins.com 145 Westergaard, Marit: The Acquisition of Word Order. Micro-cues, information structure, and economy. xii, 242 pp. + index. Expected June 2009 144 Putnam, Michael T. (ed.): Towards a Derivational Syntax. Survive-minimalism. x, 264 pp. + index. Expected July 2009 143 Rothmayr, Antonia: The Structure of Stative Verbs. xv, 212 pp. + index. Expected June 2009 142 Nunes, Jairo (ed.): Minimalist Essays on Brazilian Portuguese Syntax. 2009. vi, 243 pp. 141 Alexiadou, Artemis, Jorge Hankamer, Thomas McFadden, Justin Nuger and Florian Schäfer (eds.): Advances in Comparative Germanic Syntax. 2009. xv, 395 pp. 140 Roehrs, Dorian: Demonstratives and Definite Articles as Nominal Auxiliaries. 2009. xii, 196 pp. 139 Hicks, Glyn: The Derivation of Anaphoric Relations. 2009. xii, 309 pp. 138 Siddiqi, Daniel: Syntax within the Word. Economy, allomorphy, and argument selection in Distributed Morphology. 2009. xii, 138 pp. 137 Pfau, Roland: Grammar as Processor. A Distributed Morphology account of spontaneous speech errors. 2009. xiii, 372 pp. 136 Kandybowicz, Jason: The Grammar of Repetition. Nupe grammar at the syntax–phonology interface. 2008. xiii, 168 pp. 135 Lewis, William D., Simin Karimi, Heidi Harley and Scott O. Farrar (eds.): Time and Again. Theoretical perspectives on formal linguistics. In honor of D. Terence Langendoen. 2009. xiv, 265 pp. 134 Armon-Lotem, Sharon, Gabi Danon and Susan Rothstein (eds.): Current Issues in Generative Hebrew Linguistics. 2008. vii, 393 pp. 133 MacDonald, Jonathan E.: The Syntactic Nature of Inner Aspect. A minimalist perspective. 2008. xv, 241 pp. 132 Biberauer, Theresa (ed.): The Limits of Syntactic Variation. 2008. vii, 521 pp. 131 De Cat, Cécile and Katherine Demuth (eds.): The Bantu–Romance Connection. A comparative investigation of verbal agreement, DPs, and information structure. 2008. xix, 355 pp. 130 Kallulli, Dalina and Liliane Tasmowski (eds.): Clitic Doubling in the Balkan Languages. 2008. ix, 442 pp. 129 Sturgeon, Anne: The Left Periphery. The interaction of syntax, pragmatics and prosody in Czech. 2008. xi, 143 pp. 128 Taleghani, Azita H.: Modality, Aspect and Negation in Persian. 2008. ix, 183 pp. 127 Durrleman-Tame, Stephanie: The Syntax of Jamaican Creole. A cartographic perspective. 2008. xii, 190 pp. 126 Schäfer, Florian: The Syntax of (Anti-)Causatives. External arguments in change-of-state contexts. 2008. xi, 324 pp. 125 Rothstein, Björn: The Perfect Time Span. On the present perfect in German, Swedish and English. 2008. xi, 171 pp. 124 Ihsane, Tabea: The Layered DP. Form and meaning of French indefinites. 2008. ix, 260 pp. 123 Stoyanova, Marina: Unique Focus. Languages without multiple wh-questions. 2008. xi, 184 pp. 122 Oosterhof, Albert: The Semantics of Generics in Dutch and Related Languages. 2008. xviii, 286 pp. 121 Tungseth, Mai Ellin: Verbal Prepositions and Argument Structure. Path, place and possession in Norwegian. 2008. ix, 187 pp. 120 Asbury, Anna, Jakub Dotlačil, Berit Gehrke and Rick Nouwen (eds.): Syntax and Semantics of Spatial P. 2008. vi, 416 pp. 119 Fortuny, Jordi: The Emergence of Order in Syntax. 2008. viii, 211 pp. 118 Jäger, Agnes: History of German Negation. 2008. ix, 350 pp. 117 Haugen, Jason D.: Morphology at the Interfaces. Reduplication and Noun Incorporation in Uto-Aztecan. 2008. xv, 257 pp. 116 Endo, Yoshio: Locality and Information Structure. A cartographic approach to Japanese. 2007. x, 235 pp. 115 Putnam, Michael T.: Scrambling and the Survive Principle. 2007. x, 216 pp. 114 Lee-Schoenfeld, Vera: Beyond Coherence. The syntax of opacity in German. 2007. viii, 206 pp.

113 Eythórsson, Thórhallur (ed.): Grammatical Change and Linguistic Theory. The Rosendal papers. 2008. vi, 441 pp. 112 Axel, Katrin: Studies on Old High German Syntax. Left sentence periphery, verb placement and verbsecond. 2007. xii, 364 pp. 111 Eguren, Luis and Olga Fernández Soriano (eds.): Coreference, Modality, and Focus. Studies on the syntax–semantics interface. 2007. xii, 239 pp. 110 Rothstein, Susan (ed.): Theoretical and Crosslinguistic Approaches to the Semantics of Aspect. 2008. viii, 453 pp. 109 Chocano, Gema: Narrow Syntax and Phonological Form. Scrambling in the Germanic languages. 2007. x, 333 pp. 108 Reuland, Eric, Tanmoy Bhattacharya and Giorgos Spathas (eds.): Argument Structure. 2007. xviii, 243 pp. 107 Corver, Norbert and Jairo Nunes (eds.): The Copy Theory of Movement. 2007. vi, 388 pp. 106 Dehé, Nicole and Yordanka Kavalova (eds.): Parentheticals. 2007. xii, 314 pp. 105 Haumann, Dagmar: Adverb Licensing and Clause Structure in English. 2007. ix, 438 pp. 104 Jeong, Youngmi: Applicatives. Structure and interpretation from a minimalist perspective. 2007. vii, 144 pp. 103 Wurff, Wim van der (ed.): Imperative Clauses in Generative Grammar. Studies in honour of Frits Beukema. 2007. viii, 352 pp. 102 Bayer, Josef, Tanmoy Bhattacharya and M.T. Hany Babu (eds.): Linguistic Theory and South Asian Languages. Essays in honour of K. A. Jayaseelan. 2007. x, 282 pp. 101 Karimi, Simin, Vida Samiian and Wendy K. Wilkins (eds.): Phrasal and Clausal Architecture. Syntactic derivation and interpretation. In honor of Joseph E. Emonds. 2007. vi, 424 pp. 100 Schwabe, Kerstin and Susanne Winkler (eds.): On Information Structure, Meaning and Form. Generalizations across languages. 2007. vii, 570 pp. 99 Martínez-Gil, Fernando and Sonia Colina (eds.): Optimality-Theoretic Studies in Spanish Phonology. 2007. viii, 564 pp. 98 Pires, Acrisio: The Minimalist Syntax of Defective Domains. Gerunds and infinitives. 2006. xiv, 188 pp. 97 Hartmann, Jutta M. and László Molnárfi (eds.): Comparative Studies in Germanic Syntax. From Afrikaans to Zurich German. 2006. vi, 332 pp. 96 Lyngfelt, Benjamin and Torgrim Solstad (eds.): Demoting the Agent. Passive, middle and other voice phenomena. 2006. x, 333 pp. 95 Vogeleer, Svetlana and Liliane Tasmowski (eds.): Non-definiteness and Plurality. 2006. vi, 358 pp. 94 Arche, María J.: Individuals in Time. Tense, aspect and the individual/stage distinction. 2006. xiv, 281 pp. 93 Progovac, Ljiljana, Kate Paesani, Eugenia Casielles and Ellen Barton (eds.): The Syntax of Nonsententials. Multidisciplinary perspectives. 2006. x, 372 pp. 92 Boeckx, Cedric (ed.): Agreement Systems. 2006. ix, 346 pp. 91 Boeckx, Cedric (ed.): Minimalist Essays. 2006. xvi, 399 pp. 90 Dalmi, Gréte: The Role of Agreement in Non-Finite Predication. 2005. xvi, 222 pp. 89 Velde, John R. te: Deriving Coordinate Symmetries. A phase-based approach integrating Select, Merge, Copy and Match. 2006. x, 385 pp. 88 Mohr, Sabine: Clausal Architecture and Subject Positions. Impersonal constructions in the Germanic languages. 2005. viii, 207 pp. 87 Julien, Marit: Nominal Phrases from a Scandinavian Perspective. 2005. xvi, 348 pp. 86 Costa, João and Maria Cristina Figueiredo Silva (eds.): Studies on Agreement. 2006. vi, 285 pp. 85 Mikkelsen, Line: Copular Clauses. Specification, predication and equation. 2005. viii, 210 pp. 84 Pafel, Jürgen: Quantifier Scope in German. 2006. xvi, 312 pp. 83 Schweikert, Walter: The Order of Prepositional Phrases in the Structure of the Clause. 2005. xii, 338 pp. 82 Quinn, Heidi: The Distribution of Pronoun Case Forms in English. 2005. xii, 409 pp. 81 FuSS, Eric: The Rise of Agreement. A formal approach to the syntax and grammaticalization of verbal inflection. 2005. xii, 336 pp. 80 Burkhardt, Petra: The Syntax–Discourse Interface. Representing and interpreting dependency. 2005. xii, 259 pp.

79 Schmid, Tanja: Infinitival Syntax. Infinitivus Pro Participio as a repair strategy. 2005. xiv, 251 pp. 78 Dikken, Marcel den and Christina Tortora (eds.): The Function of Function Words and Functional Categories. 2005. vii, 292 pp. 77 Öztürk, Balkız: Case, Referentiality and Phrase Structure. 2005. x, 268 pp. 76 Stavrou, Melita and Arhonto Terzi (eds.): Advances in Greek Generative Syntax. In honor of Dimitra Theophanopoulou-Kontou. 2005. viii, 366 pp. 75 Di Sciullo, Anna Maria (ed.): UG and External Systems. Language, brain and computation. 2005. xviii, 398 pp. 74 Heggie, Lorie and Francisco Ordóñez (eds.): Clitic and Affix Combinations. Theoretical perspectives. 2005. viii, 390 pp. 73 Carnie, Andrew, Heidi Harley and Sheila Ann Dooley (eds.): Verb First. On the syntax of verbinitial languages. 2005. xiv, 434 pp. 72 FuSS, Eric and Carola Trips (eds.): Diachronic Clues to Synchronic Grammar. 2004. viii, 228 pp. 71 Gelderen, Elly van: Grammaticalization as Economy. 2004. xvi, 320 pp. 70 Austin, Jennifer R., Stefan Engelberg and Gisa Rauh (eds.): Adverbials. The interplay between meaning, context, and syntactic structure. 2004. x, 346 pp. 69 Kiss, Katalin É. and Henk van Riemsdijk (eds.): Verb Clusters. A study of Hungarian, German and Dutch. 2004. vi, 514 pp. 68 Breul, Carsten: Focus Structure in Generative Grammar. An integrated syntactic, semantic and intonational approach. 2004. x, 432 pp. 67 Mišeska Tomić, Olga (ed.): Balkan Syntax and Semantics. 2004. xvi, 499 pp. 66 Grohmann, Kleanthes K.: Prolific Domains. On the Anti-Locality of movement dependencies. 2003. xvi, 372 pp. 65 Manninen, Satu Helena: Small Phrase Layers. A study of Finnish Manner Adverbials. 2003. xii, 275 pp. 64 Boeckx, Cedric and Kleanthes K. Grohmann (eds.): Multiple Wh-Fronting. 2003. x, 292 pp. 63 Boeckx, Cedric: Islands and Chains. Resumption as stranding. 2003. xii, 224 pp. 62 Carnie, Andrew, Heidi Harley and MaryAnn Willie (eds.): Formal Approaches to Function in Grammar. In honor of Eloise Jelinek. 2003. xii, 378 pp. 61 Schwabe, Kerstin and Susanne Winkler (eds.): The Interfaces. Deriving and interpreting omitted structures. 2003. vi, 403 pp. 60 Trips, Carola: From OV to VO in Early Middle English. 2002. xiv, 359 pp. 59 Dehé, Nicole: Particle Verbs in English. Syntax, information structure and intonation. 2002. xii, 305 pp. 58 Di Sciullo, Anna Maria (ed.): Asymmetry in Grammar. Volume 2: Morphology, phonology, acquisition. 2003. vi, 309 pp. 57 Di Sciullo, Anna Maria (ed.): Asymmetry in Grammar. Volume 1: Syntax and semantics. 2003. vi, 405 pp. 56 Coene, Martine and Yves D’hulst (eds.): From NP to DP. Volume 2: The expression of possession in noun phrases. 2003. x, 295 pp. 55 Coene, Martine and Yves D’hulst (eds.): From NP to DP. Volume 1: The syntax and semantics of noun phrases. 2003. vi, 362 pp. 54 Baptista, Marlyse: The Syntax of Cape Verdean Creole. The Sotavento varieties. 2003. xxii, 294 pp.  (incl. CD-rom). 53 Zwart, C. Jan-Wouter and Werner Abraham (eds.): Studies in Comparative Germanic Syntax. Proceedings from the 15th Workshop on Comparative Germanic Syntax (Groningen, May 26–27, 2000). 2002. xiv, 407 pp. 52 Simon, Horst J. and Heike Wiese (eds.): Pronouns – Grammar and Representation. 2002. xii, 294 pp. 51 Gerlach, Birgit: Clitics between Syntax and Lexicon. 2002. xii, 282 pp. 50 Steinbach, Markus: Middle Voice. A comparative study in the syntax-semantics interface of German. 2002. xii, 340 pp. 49 Alexiadou, Artemis (ed.): Theoretical Approaches to Universals. 2002. viii, 319 pp. 48 Alexiadou, Artemis, Elena Anagnostopoulou, Sjef Barbiers and Hans-Martin Gärtner (eds.): Dimensions of Movement. From features to remnants. 2002. vi, 345 pp. 47 Barbiers, Sjef, Frits Beukema and Wim van der Wurff (eds.): Modality and its Interaction with the Verbal System. 2002. x, 290 pp.

46 Panagiotidis, E. Phoevos: Pronouns, Clitics and Empty Nouns. ‘Pronominality’ and licensing in syntax. 2002. x, 214 pp. 45 Abraham, Werner and C. Jan-Wouter Zwart (eds.): Issues in Formal German(ic) Typology. 2002. xviii, 336 pp. 44 Taylan, Eser Erguvanlı (ed.): The Verb in Turkish. 2002. xviii, 267 pp. 43 Featherston, Sam: Empty Categories in Sentence Processing. 2001. xvi, 279 pp. 42 Alexiadou, Artemis: Functional Structure in Nominals. Nominalization and ergativity. 2001. x, 233 pp. 41 Zeller, Jochen: Particle Verbs and Local Domains. 2001. xii, 325 pp. 40 Hoeksema, Jack, Hotze Rullmann, Víctor Sánchez-Valencia and Ton van der Wouden (eds.): Perspectives on Negation and Polarity Items. 2001. xii, 368 pp. 39 Gelderen, Elly van: A History of English Reflexive Pronouns. Person, Self, and Interpretability. 2000. xiv, 279 pp. 38 Meinunger, André: Syntactic Aspects of Topic and Comment. 2000. xii, 247 pp. 37 Lutz, Uli, Gereon Müller and Arnim von Stechow (eds.): Wh-Scope Marking. 2000. vi, 483 pp. 36 Gerlach, Birgit and Janet Grijzenhout (eds.): Clitics in Phonology, Morphology and Syntax. 2001. xii, 441 pp. 35 Hróarsdóttir, Thorbjörg: Word Order Change in Icelandic. From OV to VO. 2001. xiv, 385 pp. 34 Reuland, Eric (ed.): Arguments and Case. Explaining Burzio’s Generalization. 2000. xii, 255 pp. 33 Puskás, Genoveva: Word Order in Hungarian. The syntax of Ā-positions. 2000. xvi, 398 pp. 32 Alexiadou, Artemis, Paul Law, André Meinunger and Chris Wilder (eds.): The Syntax of Relative Clauses. 2000. vi, 397 pp. 31 Svenonius, Peter (ed.): The Derivation of VO and OV. 2000. vi, 372 pp. 30 Beukema, Frits and Marcel den Dikken (eds.): Clitic Phenomena in European Languages. 2000. x, 324 pp. 29 Miyamoto, Tadao: The Light Verb Construction in Japanese. The role of the verbal noun. 2000. xiv, 232 pp. 28 Hermans, Ben and Marc van Oostendorp (eds.): The Derivational Residue in Phonological Optimality Theory. 2000. viii, 322 pp. 27 Růžička, Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study. 1999. x, 206 pp. 26 Ackema, Peter: Issues in Morphosyntax. 1999. viii, 310 pp. 25 Felser, Claudia: Verbal Complement Clauses. A minimalist study of direct perception constructions. 1999. xiv, 278 pp. 24 Rebuschi, Georges and Laurice Tuller (eds.): The Grammar of Focus. 1999. vi, 366 pp. 23 Giannakidou, Anastasia: Polarity Sensitivity as (Non)Veridical Dependency. 1998. xvi, 282 pp. 22 Alexiadou, Artemis and Chris Wilder (eds.): Possessors, Predicates and Movement in the Determiner Phrase. 1998. vi, 388 pp. 21 Klein, Henny: Adverbs of Degree in Dutch and Related Languages. 1998. x, 232 pp. 20 Laenzlinger, Christopher: Comparative Studies in Word Order Variation. Adverbs, pronouns, and clause structure in Romance and Germanic. 1998. x, 371 pp. 19 Josefsson, Gunlög: Minimal Words in a Minimal Syntax. Word formation in Swedish. 1998. ix, 199 pp. 18 Alexiadou, Artemis: Adverb Placement. A case study in antisymmetric syntax. 1997. x, 256 pp. 17 Beermann, Dorothee, David LeBlanc and Henk van Riemsdijk (eds.): Rightward Movement. 1997. vi, 410 pp. 16 Liu, Feng-hsi: Scope and Specificity. 1997. viii, 187 pp. 15 Rohrbacher, Bernhard Wolfgang: Morphology-Driven Syntax. A theory of V to I raising and prodrop. 1999. viii, 296 pp. 14 Anagnostopoulou, Elena, Henk van Riemsdijk and Frans Zwarts (eds.): Materials on Left Dislocation. 1997. viii, 349 pp. 13 Alexiadou, Artemis and T. Alan Hall (eds.): Studies on Universal Grammar and Typological Variation. 1997. viii, 252 pp. 12 Abraham, Werner, Samuel David Epstein, Höskuldur Thráinsson and C. Jan-Wouter Zwart (eds.): Minimal Ideas. Syntactic studies in the minimalist framework. 1996. xii, 364 pp. 11 Lutz, Uli and Jürgen Pafel (eds.): On Extraction and Extraposition in German. 1996. xii, 315 pp. 10 Cinque, Guglielmo and Giuliana Giusti (eds.): Advances in Roumanian Linguistics. 1995. xi, 172 pp.

9 8 7 6 5 4 3 2 1

Gelderen, Elly van: The Rise of Functional Categories. 1993. x, 224 pp. Fanselow, Gisbert (ed.): The Parametrization of Universal Grammar. 1993. xvii, 232 pp. Åfarlí, Tor A.: The Syntax of Norwegian Passive Constructions. 1992. xii, 177 pp. Bhatt, Christa, Elisabeth Löbel and Claudia Maria Schmidt (eds.): Syntactic Phrase Structure Phenomena in Noun Phrases and Sentences. 1989. ix, 187 pp. Grewendorf, Günther and Wolfgang Sternefeld (eds.): Scrambling and Barriers. 1990. vi, 442 pp. Abraham, Werner and Sjaak De Meij (eds.): Topic, Focus and Configurationality. Papers from the 6th Groningen Grammar Talks, Groningen, 1984. 1986. v, 349 pp. Abraham, Werner (ed.): On the Formal Syntax of the Westgermania. Papers from the 3rd Groningen Grammar Talks (3e Groninger Grammatikgespräche), Groningen, January 1981. 1983. vi, 242 pp. Ehlich, Konrad and Jürgen Rehbein: Augenkommunikation. Methodenreflexion und Beispielanalyse. 1982. viii, 150 pp. With many photographic ills. Klappenbach, Ruth (1911–1977): Studien zur Modernen Deutschen Lexikographie. Auswahl aus den Lexikographischen Arbeiten von Ruth Klappenbach, erweitert um drei Beiträge von Helene MaligeKlappenbach. (Written in German). 1980. xxiii, 313 pp.

View more...


Copyright © 2017 ZILADOC Inc.