Language and Logos: Studies in theoretical and computational linguistics 9783050062365, 9783050049311

Citation preview

Language and Logos Studies in Theoretical and Computational Linguistics T h o m a s Hanneforth, Gisbert Fanselow (Eds.)

studia grammatica Herausgegeben von Manfred Bierwisch unter Mitwirkung von Hubert Haider, Stuttgart Paul Kiparsky, Stanford Angelika Kratzer, Amherst Jürgen Kunze, Berlin David Pesetsky, Cambridge (Massachusetts) Dieter Wunderlich, Düsseldorf

studia grammatica 72

G"a a n n sëi e o f w rth '

(Eds)

Language and Logos Studies in Theoretical and Computational Linguistics

Akademie Verlag

Bibliografische Information der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über http://dnb.d-nb.de abrufbar.

ISBN 978-3-05-004931-1 ISSN 0081-6469

© Akademie Verlag GmbH, Berlin 2010 Das eingesetzte Papier ist alterungsbeständig nach DIN/ISO 9706. Alle Rechte, insbesondere die der Übersetzung in andere Sprachen, vorbehalten. Kein Teil des Buches darf ohne Genehmigung des Verlages in irgendeiner Form - durch Fotokopie, Mikroverfilmung oder irgendein anderes Verfahren - reproduziert oder in eine von Maschinen, insbesondere von Datenverarbeitungsmaschinen, verwendbare Sprache übertragen oder übersetzt werden. All rights reserved (including those of translation into other languages). No part of this book may be reproduced in any form - by photoprinting, microfilm, or any other means - nor transmitted or translated into a machine language without written permission from the publishers. Druck: Books on Demand, Norderstedt Printed in the Federal Republic of Germany

Festschrift for Peter Staudacher on his 70th Birthday

CONTENTS

Preface "Ich muss keinen mehr beeindrucken"

9

I Syntax

15

Joanna Biaszczak A Spurious Genitive Puzzle in Polish

17

Gisbert Fanselow Semantic Type Effects on Crossing Movement in German

48

Sascha W. Felix Me and Chomsky

64

Günther Grewendorf On the Typology of Verb Second

72

Gereon Müller Movement from Verb-Second Clauses Revisited

97

Andreas Pankau, Craig Thiersch, Kay-Michael Würzner Spurious Ambiguities and the Parentheticals Debate

II Semantics

129

147

Christian Ebert, Cornelia Ebert On Squeamishness of the Royal Kind

149

Caroline Féry Information Structure of schon

160

6

Contents

Stefan Hinterwimmer When-Clauses, Factive Verbs and Correlates

176

Gerhard Jäger The Proof Theory of Partial Variables

190

Marcus Kracht Brentano's Apple

202

Manfred Krifka How to Interpret "Expletive" Negation under bevor in German

214

Wolfgang Sternefeld Wide Scope in situ

237

Thomas Ede Zimmermann What it Takes to be Missing

255

III

Automata Theory

267

Jörg Didakowski Robust Parsing as a Constraint Optimization Problem within a Finite-state Approach

269

Thomas Hanneforth, Colin de la Higuera ε-Removal by Loop Reduction for Finite-state Automata

297

Bryan Jurish Efficient Online k-Best Lookup in Weighted Finite-State Cascades

313

Daniel Quernheim Tomita's Algorithm Revisited

328

IV

Mathematical Linguistics

337

Hans-Martin Gärtner, Jens Michaelis On the Treatment of Multiple-Wh-Interrogatives in MGs

339

Uwe Mönnich Some Remarks on Mildly Context-Sensitive Copying

367

Contents

V

Computational Linguistics

1

391

Damir Cavar On Statistical Metrics for Selection and Phrasality

393

Christian Wartena Testing the Distributional Hypothesis for Collaborative Tagging Systems

407

VI

417

Classical Studies

Herbert E. Brekle Herstellungstechniken von Inschriften auf römischen Wasserleitungsrohren aus Blei

419

Preface "Ich muss keinen mehr beeindrucken "

1. Hot summer days Roughly a generation ago, one of the two editors of this book (Gisbert) sat in a class taught by Peter Staudacher. It was a hot summer afternoon, with many teachers sitting with their students on the meadows of Regensburg's university campus, holding their seminars there. The co-editor suggested to Peter Staudacher that his class should do the same, which was objected to on the ground that teaching in the course required a blackboard. The co-editor pointed out that one could very well carry the blackboard from the classroom to the meadows, and Peter replied "Ich bin schon verheiratet. Ich muss keinen mehr beeindrucken" (I am already married. I do not have to impress anyone anymore). The co-editor does not know why he recalls this incident so vividly. Perhaps, the answer simply is that human memory in fact records everything, and the only issue always is if and how the stored information can be retrieved. But, in any event, we think that the sentence "I do not have to impress anyone" perfectly characterizes Peter Staudacher scientific and personal approach. He is the perfect example of unpretentiousness combined with wisdom and scientific rigor. Therefore, while Peter never attempted to impress anyone, he in fact has impressed so many, some of which contribute to this volume in order to honor this outstanding scientist, and to say "thank you" for all they got from him. Many more could have contributed but do not appear here, simply because of limitations of space (and deadlines).

2. Edutainment When a major German university recently tried to fill a linguistics professorship and when it came to praising the top candidate in the official search committee's report, the candidate's innovative concept of academic teaching was highlighted. The committee was enthusiastic about the candidate's edutainment approach. Edutainment is a kind of crime in academic teaching that Peter certainly cannot be accused of. His approach to teaching is a different one, he managed to get young people, including the two editors, fascinated by the study of language without small movies,

10

Ich muss keinen mehr

beeindrucken

quizzes, jokes, and other such niceties. Rather, Peter's seminars involved blackboards filled with formulae as part of some logical proof, and when the chalk was replaced (or complemented) by modern technology, Peter added innumerable overhead slides. In the early nineties, Peter established the record of going through roughly 80 slides in 45 minutes, and the slides were densely filled by formula (in small font) and unenriched by the graphics, pictures, and cartoons that others would put there. At the end of the eighties, he switched from slides to endless slide rolls - the latest media technology available in Regensburg at that time. He filled page by page of this endless tape, almost like the read-write head of a Turing machine - and sometimes he was even talking about Turing machines - constantly winding up the tape. From time to time, it happens that a student asked a question. One co-editor (Thomas) remembers well the "squeak-squeak" noises the device generated as Peter was winding down the tape for half a minute or so, looking for the topic the student was asking about. We will not mention the dense, 400-pages PowerPoint presentations he later used in Potsdam. Thomas still uses them in his automata theory introduction which he inherited from Peter - after some refurbishments to meet the needs of the video game generation. But, note, everyone in the audiences in Regensburg and Potsdam felt they had learnt a lot from these heavyish presentations. What is the secret, then, of Peter's successful teaching? Perfect and detailed preparation is certainly one aspect. When you sit in his class, you notice that each step taken was well-planned, but a person with knowledge of Peter Staudacher can combine this with true flexibility in response to questions and ideas formulated by the students. But, what is much more: Peter has always practiced teaching as an academic dialogue between partners that are, in principle, peers in their quest for truth. His classes were a true reincarnation of Platonic dialogues, with students not being outsmarted and lured into something by fancy didactic methods, but being treated as equals that are capable of thinking, learning, and understanding. For us, when we had entered the university decades ago, this confrontation with a brilliant mind taking us serious was an experience we had not had before and that we did not have again later, it opened us the door to the beautiful world of scientific reasoning. We took all classes that Peter offered, and when, after two or three semesters, Peter repeated a class, both of us tried to take this class a second time (but were not admitted). Peter's Platonic dialogs with his students could last for hours, they continued after the official termination of the class. We have not met anyone else willing to spend nearly half a day going through the details of some scientific problem with a student (and being capable of doing so). Often, the only time constraint was the departure of the last train of the day - the student's last train, of course. By this method, Peter got us, and others many of which have submitted papers to this book, into linguistics. In the beginning, the circles were small - only very few people in Eastern Bavaria opted for linguistics as their major subject when they enrolled at the University of Regensburg where Peter Staudacher was teaching from the very early seventies until 1993. We learnt the import of the rule tres faciunt collegium in these days, but at the same time, it was frequently broken. You could end up, once in a while, as the only student in the class, usually taking place between 6 and 9 in the evening. After eight, most of the university was in darkness,

Preface

11

and the night-watchmen were often not really sure what was going when they peeped into the classroom. Later, when Peter Staudacher moved to Potsdam in 1994, he entered a university that attracted more students to linguistics, and as this book shows, he paved the way into science for a considerable list scholars.

3. Scientific Mass Production When a German university hires a professor these days, the qualification for edutainment is not the most important criterion. We select people on the basis the number of papers placed into the top peer-reviewed journals, and we measure their academic achievements by a Hirsch-index. Noam Chomsky would thus have a hard time being shortlisted. This development is certainly remarkable, given an insight communicated in the front line personal management seminars sponsored by the university administrations, viz. the insight that a good and efficient team does not consist of one type of personality only: the visionary guy with the splendid plans for the next collaborative research center must be in close contact with the critical mind questioning the sustainability of the high-flying ideas, the effective producer of journal papers concerned with mid-size problems needs a complementation by the reflective colleague who has a much broader picture in mind. It is not such a long time ago that universities could hire their scientific staff on the basis of such considerations. Peter Staudacher cannot (and would not!) boast of a long list of publications, but his impact on the field of linguistics is considerable. We have already formulated a few thoughts on his teaching shaping the minds of many young scholars and setting the standards (rarely met) for academic instruction. But Peter Staudacher is also the clandestine co-author of numerous books and articles that people in his academic vicinity have published, ranging from speech act theory (as one of his students was surprised to find out when he read the acknowledgements of the second linguistics book he bought) to computational linguistics, from syntax to semantics - and from the to be submitted journal article to the proposal for a collaborative research centre. When you asked for a critical evaluation, Peter Staudacher would never refuse the request, and he would respond by a thorough, deep, convincing critique of the draft. So many books and papers have become so much better in the interaction with Peter Staudacher, and so many proposals for large research grants got accepted because the devastating questions were raised by Peter before the draft came under the eyes of the reviewers. He would apply the same frank (calling things "sheep shit" when necessary) and well-founded critical attitude in the lectures and research seminars in the departments he worked at, so that he figured as the driving force of scientific interaction there. He could be stubborn when something did not convince him (so a visiting professor once sighed in despair: "That is really tiresome") but all open-minded partners in this scientific discourse confessed at some point that Peter had put his finger into the sore point of model to be defended. Thomas remembers a situation in a colloquium in Regensburg where a scholar from some other institute presented his "formalization" of a linguistic topic. Peter just friendly asked: "What does this arrow

12

Ich muss keinen mehr

beeindrucken

mean?" and, while insisting on that question, caused the referent (who had never thought about the meaning of the arrow) to lose his poise. The uncompromising and frank way in which Peter formulates his opinion was appreciated by all once they realized that Peter applied the same rigorous standards to himself. Such a person who you can lean to and whose feedback you can rely on is missing now in the department in Potsdam, and in the scientific circles in computational linguistics, syntax, and semantics that Peter belonged to, it is a real gap which we know can't be closed because of the current way of recruiting personnel at German universities. We miss Peter's outstanding knowledge not only in the fields just mentioned. Peter is a true polymath who brought in perspectives that we younger scientists have barely heard of. You can have profound conversations with him about music, the Austrian-Hungarian empire, philosophy, the current political situation, and much more. Peter's research journey posed a considerable challenge for the editors of this book. Their expertise is confined to that aspect of his linguistic knowledge that he highlighted when they attended his classes. Peter's research has now returned out to where it began. Peter was trained as a classical philologist, he wrote a philosophical dissertation on Plato's dialogue Parmenides, and found his way into linguistics more or less by chance: Peter Staudacher and Herbert Brekle knew each other because of the friendship between their wives, and when Herbert Brekle took over one of the very first German professorships for theoretical linguistics, he thought that linguistics would profit from a person with a profound mastery of formal logic, so he offered a position to Peter Staudacher. The first theoretical model in linguistics that really fascinated Peter Staudacher was Montague Grammar, appealing because of its formal rigor. Montague Grammar remained Peter Staudacher's favourite approach in semantics, and his love for semantics continued even when he turned to other linguistic approaches. At some point, Peter realized how unconvincing Montague grammar was in terms of syntax, Peter started a detailed interaction with syntax (Government-and-Binding-style) but later, he became dissatisfied with the rather lax standards concerning the formalization of the theoretical claims made. For Peter Staudacher, this dissatisfaction meant that he began formalizing the axioms of GBtheory himself, and this was the beginning of his career as a computational linguist which brought him to Potsdam, where he became the professor of theoretical computational linguistics in 1993, a position he kept until his retirement in 2005. In the early days, Peter designed one of the very first GB-parsers, written in Prolog, and he contributed to the theory of formal complexity of languages and grammars. Although holding a computational linguistics position in Potsdam, Peter Staudacher wasn't a computational linguist in the nowadays understanding of the words. As he often explained to first year students, Computerlinguistik (the German translation of computational linguistics which he found quite bad) has nothing to do with computers. Instead, he focused on the notion of computation, when he said that linguistics is basically computational linguistics. Understanding the formal properties of modern syntax theories was one of his major concerns, together with the problems of compositional semantics, in his Potsdam years. After retirement, Peter stayed in Potsdam as long as he was a PI in the collaborative research center on information structure. His moving back to Regensburg was also a

Preface

13

return to philosophy and classical philology - he is now working on the critical edition of Plato's Kratylos. Or, more precisely, it was not a return to these academic fields but rather a shift of focus - even when he was actively involved in linguistics, he would write articles on Plato, and work on critical editions. Above, we made a few critical remarks on recent trends in German academic life, but the import of placing one's work at certain clearly visible locations is of course not negative. Three decades ago, the world was different, with unpublished papers being more valued than what you could find in the journals. In retrospect, this was a highly undemocratic situation (you needed to be a guy to who such unpublished work was sent, or you needed to know such a guy (at least, xerography had already been invented)), so that the growing importance of journal publications also means a much fairer access to the "means of production" in research (at least for those working at a university rich enough for subscribing to the journal). Be this as it may, it is certainly deplorable that Peter Staudacher decided to not follow the journal publication strategy. This meant that important contributions that he made found their way into the international discussion more slowly than necessary. We just want to pick one example. The semantic analysis of so-called donkey sentences (every farmer who beats a donkey owns it) is a long standing problem of formal semantics. Peter Staudacher developed a solution in terms of dynamic predicate logic, in parallel with Groenendijk and Stokhof. It led to a number of correct predictions (such as the availability of a bound reading for the pronoun in either no burglar was here or he already left the flat) and is now considered one of the standard approaches to the problem. Peter Staudacher published this work not in a leading journal, but in the Festschrift for Herbert Brekle, which did, of course, not focus on formal semantics, and this implied that the article initially did not get the attention it deserved. At least nowadays, it is duly acknowledged. We hope that one or two papers of comparable importance are included in the present Festschrift as well! Peter Staudacher often decided to communicate and promote his research insights in the form of research projects funded by the Deutsche Forschungsgemeinschaft, both inside and outside larger research structures such as the Graduate School Economy and Complexity of Language (Peter was the director of its Potsdam branch) and the Collaborative Research Center Information Structure. We again only want to mention one particularly important and influential aspect. Peter was worried about the computational complexity of current approaches to syntax but at the same time very skeptical about attempts of formalize language in terms of context free grammars - because of their empirical shortcomings. In a highly regarded paper published in the ACL proceedings series, Peter developed a new class of formal languages/grammars, distributed-index languages with very promising computational properties. This line of research was continued in a number of research projects which led the international discussion of the proper formal treatment of minimalist grammars.

14

Ich muss keinen mehr beeindrucken

4. Conclusion People contribute to Festschrifts because they want to express their gratitude for and their admiration of outstanding scientific achievements. Very often - and this particularly holds for this book dedicated to Peter Staudacher - their gratitude and admiration also extends to the personal domain. Given this is a personal matter, little can and should be said at a public place like this. We only want to say: when you needed him, he was present.

Acknowledgements A number of people contributed to this book. We would like to express our gratitude to Wolfgang Sternefeld for the IATgX-stylcfile and Kay-Michael Würzner for adapting it to our needs. Kay also answered numerous questions concerning typesetting matters. We also thank the Potsdam students Maria Lesinski and Anne Mucha for their thorough proof-reading of the manuscripts and Katja Richter and Veit Friemert at the AkademieVerlag for their support. We are indebted to Manfred Bierwisch for his offer to publish the Festschrift in the studia grammatica series. Special thanks go to Ursel Staudacher (Peter's wife implicitly mentioned in the beginning) for her conspiracy. Potsdam, April 2010 Thomas Hanneforth Gisbert Fanselow

Parti

Syntax

A Spurious Genitive Puzzle in Polish Joanna

Blaszczak

1. Introducing the Puzzle In Polish, like in many other languages, direct objects of transitive verbs normally receive the ACC(usative). However, when the verb is negated, the case of the direct object obligatorily changes to GEN (hence the name: "Genitive of Negation" (GoN)); cf. ( 1 ). ( 1 ) a. Ewa czyta / g a z e t y / * gazet. Eve reads / newspapersACC / * newspapers0EN 'Eve reads/is reading newspapers.' b. Ewa nie czyta / g a z e t / * gazety. Eve NEG reads /newspapers GEN / * newspapersACC 'Eve does not read/is not reading newspapers.'

/ A C C / * GEN

NEG / G E N / * ACC

GoN in Polish, unlike its Russian 'kin' (see section 2.1), is a purely syntactic phenomenon, i.e., it is syntactically obligatory and does not have any additional semantic effects. In other words, any direct object of a negated transitive verb will be marked for GEN in Polish, irrespective of its semantic properties or discourse status. Thus, both human/animate objects and inanimate objects, both countable and mass noun objects, both singular and plural objects are likewise marked for GEN under negation; cf. (2) for illustration. Also a distinction along the lines of definiteness/indefiniteness, which often correlates with discourse properties ("old" versus "new" information; see also section 2.3 and 3.2.2.2), does not have any effect on the GoN in Polish; cf. (3) and (4). (2) Nie mogg znalezc Jana I sera / ksiqzki I ksiqzek. NEG can i , s g . P R E S findINF John GEN / cheese GEN / book GEN / books GEN Ί cannot find John / cheese / a book / books.'

NEG GEN

(3) Nie czytalam niczego / "Wojny ipokoju". NEG read|. SG . p nothing GEN / "War and Peace" Ί didn't read anything.' / Ί didn't read "War and Peace".'

NEG GEN

F

(4)

AST

a. Co ζ tymi ksiqikami? ('What about these books?') Tych ksiqzek Jan znowu nie przeczytal. [these books] GEN John again NEG read3.sc M.PAST 'John hasn't read these books again.'

18

Joanna Blaszczak

Argument/verb type 1. external arguments of a) transitive verbs b) unergative verbs 2. internal arguments a) (underlying) direct objects of

transitive verbs unaccusative verbs passive verbs b) other internal arguments indirect objects prepositional objects oblique (e.g., INSTR) objects

GoN

otherwise marked

*

Nominative Nominative

*

*

ok

Accusative

*

Nominative Nominative

*

* * *

Table 1: Distribution ofGoNin

Dative 'Prepositional Case' Oblique (e.g., INSTR) Case Polish

b. Czego Jan nie przeczytatl ('What has John not read?') {Jan nie przeczytal) jakiejs waznej ksiqzki. (John NEG read3.S0.M.PAST) [some important book]CEN 'John hasn't read some important book. (I don't remember the title).' Moreover, GoN in Polish, unlike its Russian 'kin' (see section 2.1), is a very restricted phenomenon: its occurrence is confined to just one configuration, namely to the direct (ACC) object position of transitive verbs; cf. Table 1 above. In other words, only direct objects of transitive verbs that are assigned the structural ACC case in the syntax obligatorily change to GEN under negation in Polish. More importantly, even subjects of unaccusative verbs, which are - according to the standard assumptions - base-generated in the direct object position (cf. (5)), are excluded from the GoN rule.1 Note that even the default, non-agreeing form of the verb does not improve the acceptability of the GEN marked NP in such cases; cf. 5b'. 1

That verbs like umrzec 'to die' are indeed unaccusative in Polish can be proved by using the unaccusativity/unergativity diagnostics employed by Cetnarowska (2000) for Polish. Firstly, one can form a resultative adjective terminating in -ty from umrzec 'to die', which is a deep unaccusativity diagnostic; cf. (i). Secondly, umrzec is infelicitous in the impersonal -nol-to construction, which is an unergativity diagnostic in Polish; cf. (ii). Thirdly, umrzec allows for a distributive /?o-subject (cf. (iii)), which is one of the unaccusativity diagnostics proposed by Pesetsky (1982) for Russian and used by Cetnarowska (2000) as a diagnostic for surface unaccusativity in Polish (Examples taken from Cetnarowska, 2000, p. 37, 39, 41):

(i)

umarty 'dead' (from umrzec 'to die')

(ii)

* Umarto ζ glodu. to-diedpERK from hunger 'They died of hunger.'

(iii) [U]marlo po dziecku ζ kazdego przedszkola. died3S0.N.P1.:RF po childuic from each kindergarten Ά child died from each kindergarten.'

A Spurious Genitive Puzzle in Polish (5)

19

a. (W tym szpitalu) umarl wczoraj pacjent. NOM (in this hospital) died3. SG M yesterday patient N 0 M.M.sG '(In this hospital) a patient died yesterday.' b. (W tym szpitalu) f[zaden pacjent] /* [zadnego pacjenta] nie umarl. (in this hospital) / [ n o patient]N0M.M.sG /* [no patient] G E N M S G NEG died 3 . SC M '(In this hospital) no patient died.' NOM b tym szpitalu) [zadnego pacjenta] nie umarlo. *GEN (in this hospital) [no patient] 0EN . M . SG NEG died3.SG.N

Now, given the facts presented above, it is surprising to find examples like (6) in Polish. Such examples are not expected - in light of what was said about the GoN in Polish - since first, the GEN marking seems here not to be obligatory, second, the GEN seems to have here a semantic / interpretational effect, and third, the GEN seems to alternate here with the NOM and not with the usual ACC. This latter fact is particularly baffling: the GEN marked NP, 2 in (6a) is most likely the internal argument of the unaccusative verb BE, 3 but, as was pointed out above, internal arguments of unaccusative verbs ("unaccusative subjects") cannot be marked for GEN under negation in Polish. (6) a. We wsi nie bylo (zadnego) lekarza. in village NEG was 3 . SG N (no GEN SG M ) doctor GEN . SOM 'There was no doctor in the village.' b. Lekarz nie byl we wsi.

GEN

NOM

doctor N O M.sG.M

NEG w a s 3 . S G . M in village 'The doctor was not in the village.'

The aim of the paper is to show that in fact, there is nothing puzzling about the examples in (6). The puzzle quickly turns out to be a spurious one, once it is recognized that (6a) and (6b) are derived from two different "base" structures. More importantly, the GEN marking in (6a) - instead of being a puzzle - will be shown to be compatible with the general rule of GoN in Polish. Before doing so, it will first be argued that all other potential solutions to the puzzling facts in (6) do not work or are not really convincing. In section 2.1 we will see that the facts in (6) cannot be reduced to some Russian sort of GoN. Also the claim that (6a) and (6b) are derived from a uniform "underlying" structure is anything but convincing (section 2.2). Neither can the problem at hand - this is the topic of sections 2.3 and 2.4 - be explained away by referring to different scopes of negation or different information or perspective structures. Section 3 presents the solution to the puzzle posed by the examples in (6) and discusses some of its welcome consequences (see Blaszczak, 2007 for an extensive discussion).

2

Throughout this paper I will use the general term 'nominal phrase' (NP) to refer to any nominal phrase to avoid a discussion about the NP- vs. DP-distinction, which is especially problematic in languages such as Polish which do not have the category article.

3

Usually BE is taken to be an unaccusative verb par excellence (cf. Babyonyshev, 1996; Brown, 1996; see also Moro, 1997). This view will be modified in section 3.2.

20

Joanna Blaszczak

2. Looking for a Solution to the Puzzle 2.1 GEN in (6a) as an Instance of a GoN of the Russian Sort While the facts presented in (6) are unexpected from a strictly Polish point of view, they seem at first to perfectly suit the Russian facts. The Polish GoN and its Russian 'kin' differ in two important aspects, which seem to be relevant to the problem at hand. Firstly, the GoN in Russian shows a broader distribution. As indicated in Table 2 below, unlike what is the case in Polish, in Russian subjects of unaccusative, passive and so-called existential predicates might be marked for GEN under negation. Secondly, the GoN in Russian, unlike the GoN in Polish, is not a purely syntactic phenomenon. In fact, it is syntactically not obligatory, i.e., it may alternate with the ACC or the NOM. Note that the GoN in Russian has a clear semantic effect in that "[... ] the GEN marked arguments tend to receive an existential interpretation, while those marked either NOM or ACC receive either an existential interpretation or a presuppositional/generic reading" (Brown (1996, p. 48)); cf. (7) and (8) (from Babyonyshev (2003, p. 50-51)). 4 (7) a. Ja ne polucala zurnalov. I NEG received magazineGENPL Ί received no magazines.' b. Ja ne polucala zurnaly. I

GEN

ACC

N E G received magazineACC.PL

Ί didn't receive the magazines.' / '*? I received no magazines.' (8) a. V klasse ne

pojavilos '

studentov.

in class N E G appeareds0.N student GENPL 'No students appeared in class.' b. V klasse ne pojavilis studenty. in class N E G appearedPL studentN0M.PL

GEN

NOM

'The students did not appear in class.' / 'No students appeared in class.' At first glance it might seem that the GEN puzzle posed by sentences like those in (6) above could be solved by simply claiming that this GEN is in some sense like the GoN in Russian. A closer inspection of the facts reveals, however, that such a conclusion would be too rash or even problematic. As pointed out above, the GoN rule in Russian, unlike that in Polish, does not apply exclusively to direct objects of transitive verbs which are otherwise marked for ACC. 4

(i)

Some scholars, like, for example, Babyonyshev (2003) (cf. also Pereltsvaig, 1999), even claim that nomináis that are unambiguously definite cannot occur in GEN in Russian; cf. (i) (cited from Babyonyshev, 2003, p. 51-52; but see Harves, 2002, p. 45f and Richardson, 2003, p. 134 for critical discussion and apparent counterexamples). Vanja ne procitai "Vojnu i mir". / * "Vojny i mira". Vanya NEG read3.S0.M.pAsT [War and Peace] AC c / * [War and Peace]0KN 'Vanya didn't read "War and Peace".'

21

A Spurious Genitive Puzzle in Polish Argument/verb type 1. external arguments of a) transitive verbs b) unergative verbs 2. internal arguments a) (underlying) direct objects of transitive verbs unaccusative verbs passive verbs b) other internal arguments indirect objects prepositional objects oblique (e.g., INSTR) objects

GoN

*

otherwise marked Nominative Nominative

*

ok ok ok

Accusative Nominative Nominative

*

Dative 'Prepositional Case' Oblique (e.g., INSTR) Case

* *

Table 2: Distribution of GoN in Russian

This opens the possibility of using the GoN rule of the Russian sort to cover both the canonical cases of GoN in Polish (i.e., direct objects of transitive verbs) and the "exceptional" case of the GEN case marking of the "subject" nominal phrase in negated existential BE-sentences in Polish (recall (6a)). However, by making such an assumption we would run into the following problem. The GoN in Russian might also apply to subjects of negated unaccusative and passive verbs as well as to subjects of negated existentials. Now, on the assumption that what these examples have in common is the fact that the GEN marked nominal argument originates in the "underlying" internal argument ("direct object") position, it is not clear why the Polish version of this GEN should be confined to negated existential BE sentences. The question is thus, why subjects of negated unaccusative verbs (recall ex. (5)) cannot be marked for GEN, or even more surprisingly, why it is so that no other verb - apart from BE - which might be used to indicate the existence of some object at some location, allows its "subject" to be marked for GEN under negation; cf. the contrast between (9) and (10).5 (9) a. Na stole lezala ksiqzka. on table lie3.S0.F.PAST bookNOM.sG.F 'There was a book (lying) on the table.' b. Na stole nie lezala (zadna) ksiqzka. on table NEG lie3.SG.F.PAST (noNOM.sG.F) bookNOM.sG.F There was no book (lying) on the table.'

5

(i)

/NOM

/ NOM

It should be noted that (10b') is actually acceptable on a contrastive/narrow scope reading of negation; cf. (i). Na stole nie byla ksiqzka, tylko gazeta. on table NEG was3.sc> bookNoM.sG.i· but newspaperNoM.sG.K 'There was not a book but a newspaper on the table.'

22

Joanna Blaszczak b' *Na stole nie leíala Heialo ksiqzki. on table NEG lie3.SG.F.PAST / lie3.SG.N.PAST book G E N S G F

* GEN

(10) a. Na stole byla ksiqzka. on table be3.SG.RPAST b o o k N O M . s G . F 'There was a book on the table.' b. Na stole nie bylo (zadnej) ksiqzki. on table NEG b e 3 . S G . N . P A S T (noGEN SG.F) book OEN . SGF 'There was no book on the table.' b *Na stole nie byla ksiqzka. on table NEG b e 3 . S G . F . P A S T b o o k N O M . s G . F

/NOM

/ GEN

* NOM

Thus, we arrive in some sense at a paradoxical situation. While the distribution of regular GoN in Polish is too narrow to account for the puzzling GEN in negated existential BE sentences, the distribution of the GoN of the Russian sort is too broad since it would allow for GEN marked "subjects" not only with existential BE but also with unaccusastive predicates in general, a situation which is not attested in Polish. The problem is even more serious if one takes into account that there are cases where what appears to be a subject of an unergative verb can be subject to the GoN rule in Russian; cf. (11) (due to Babby, 2000). (11) Na zabrosennom zavode upal i razbilsja Sasa. S tex por tam (bol'se) at abandoned factory fell and hurt-self Sasha. From that time there (anymore) ne igraet nikakix detej. NEG play3.SG [NEG-kind children] GENPL 'Sasha fell and hurt himself at the abandoned factory. Since that time, children don't play there (anymore).' To account for cases like those in (11) above, among others, Babby (2000) proposes an analysis according to which the GoN is assigned to available NPs in the scope of VP negation without regard to their function and, therefore, the domain of the GoN can be represented by the schema in (12), which includes, but is not limited to the direct object. Note that according to Babby (2000, p. 18f.), in the example in (11) we are dealing with an existential sentence in which the inverted subject NP is not in the direct object position but is adjoined to the matrix VP. That is, the VP in existential sentences has the structure given in (13), which conforms to the GoN configuration in (12) (see Babby, 2000 for details). (12)

[ V ( P ) N P ]VP

-»•

[ne [V(P) N P

G E N

NP ]

V P

]NEGP

(13) [[V]vp NPJvp In a similar vein, Partee & Borschev (2004) (cf. also Borschev & Partee, 1998) assume that verbs can be semantically "bleached" to just an "existential meaning", which means that the following equivalence holds in the given context of utterance: V(THING, LOC) BE(THING, LOC). In other words, if a verb that has its normal lexical meaning, which in most cases is not simply 'exist' or 'be', is used in a GoN construction, the hearer

A Spurious Genitive Puzzle in Polish

23

uses contextual information to support an accommodation of the presupposition, perhaps shifting the verb meaning to make it "less agentive"; cf. (14). (14)

a. N E S (Negated Existential Sentence):

Ne

belelo

parusov

na gorizonte.

N E G shone.white S G N sails G E N . P L M on horizon 'No sails were shining white on the horizon.' b. Presupposed Equivalence: Ά sail shone white on the horizon.' 'There was a sail on the horizon.' c. 'Dictionary axiom' (part of lexical semantics): to shine-white to be white (in the field of vision) d. Dictionary or encyclopaedic axiom; 'common knowledge': 'Sails as a rule are white.' With this in mind, let us notice that if we were to assume that the G E N marking in negated existential BE-sentences in Polish is an instance of a GoN of the Russian sort, the question would be why similar shifting operations are not possible in Polish. Why can't Polish verbs be "bleached" just to the "existential meaning", or - assuming that they might be "bleached" in a given context - why is it so that such a "bleached" verb allows for a GoN in Russian and not in Polish? 6 Thus, appearances to the contrary, it seems that we cannot solve the problem of the puzzling G E N in (6a) by trying to attribute it to some special property of "existential" constructions in Polish. Though the problematic G E N is connected with existential sentences, it appears to be an exclusive property of negated existential B E sentences and not a property of existential constructions in general. Hence, in the next sections we will narrow down the discussion to B E sentences.

2.2

A Simple Uniform Analysis - Too Simple

The simplest way of analyzing BE-sentences such as those in (6) would be to assume that they are derived from the same underlying structure in (15), following, among others, Hoekstra & Mulder (1990), Freeze (1992), den Dikken (1995, 1 9 9 7 , 2 0 0 6 ) , Moro (1997). 7 6

Note that examples of the kind discussed by Babby (2000) and Borschev & Partee (1998) / Partee & Borschev (2004) are drastically ungrammatical in Polish; cf. (i) and (ii):

(i)

a.

b. (ii)

a.

b.

7

Od tego czasu nie bawiq sig tam wigcej zadne dzieci. since that time NEG play j PL P R l . s self there more no N n M PL children NnM PL 'Since that time, children don't play there (anymore).' * Od tego czasu nie bawi sig lam wigcej zadnych dzieci. since that time NEG playj.SG.pR>i.s self there more ΠΟΟΚΝ.ΡΙ. childrenGEN.pL Na horyzoncie nie bielity sig zadne zagle. on horizon NEG sh0ne.white3.PL self noNOM.pL sails N()M .p L 'No sails were shining white on the horizon.' * Na horyzoncie nie bielilo sig zadnych ¿agli. on horizon NEG shone.white3.so.Ν self nocKN.PL sailsotN.PL

(cf. (11))

(cf. (14a))

For reasons of space, I cannot discuss these analyses in due detail. See Blaszczak (2007, 2008b) for a detailed discussion.

24 (15)

Joanna BE [ s c N P t h e m e

RR

L0CATI0N

Blaszczak

J

The different types of construction would then arise as the result of moving either the N P T H E M E (locative sentences; cf. 16a)) or the P P L O C A T I O N (existential sentences; cf. 16b)) into a sentence-initial position (mostly understood as [Spec,IP]): (16)

a. b.

BE BE

NPTHEME PPLOCATION

[T [ NP

NP

locative existential

PPLOCATION]

T H E

ME

UP

]

Simple and attractive as it might be, such a "uniform" analysis is problematic at least for two reasons. First of all, if locative and existential sentences have the same underlying structure, why is there a difference in case marking of the Theme NP argument in negated variants of the respective sentences (as in (6))? Not only that this analysis does not provide any insightful explanation of this fact, even worse, it makes wrong predictions. Given that negation can obviously influence the case marking of the Theme argument and given furthermore that "properties of the probe/selector a must be exhausted before new elements of the lexical subarray are accessed to derive further operations" (Chomsky, 2000, p. 132), the question arises as to how the Theme-NP is ever able to escape the GEN marking (by negation); cf. (17). (17)

a.

B E

b. NEG+BE I

[

S C

[SC

NPTHEME

PPLOCATION

]

N P T H E M E PPLOCATION ]

GEN OK/expected

/

NQM?/unexpected

t

Another problematic issue is that it is not clear what accounts for different interpretations if locatives and existentials have underlyingly the same argument structure. What decides which element ( N P E M E or P P L O C A T I O N ) has to move? Freeze (1992) solves this problem by assuming that the question of which argument (the THEME or the LOCATION) moves to [Spec,IP] is governed by the [ ± definite] feature of the THEME; cf. (18). T H

(18)

a-

B E

[sc

NPTHEME

PPLOCATION]

BE b. BE

[doctor in village] [doctor in village] [+definite] b'. locative sentence (NP > BE > PP) Lekarz byl we wsi. doctorNOM SG.M was3 SG.M in village 'The doctor was in the village.' c. BE [doctor in village] [-definite] c . existential sentence (PP > BE > NP) We wsi byl lekarz in village WAS3.sc.M doctor NOM .SG.M 'There was a doctor in the village.'

A Spurious Genitive Puzzle in Polish

25

Even if it works in (18), this account is too simple. For example, in negated existential sentences the THEME argument can very well be definite/topical and move to some sentence initial position; cf. (19b/b') and (20). (19) a. existential sentence (PP > BE > NPGEN) We wsi nie bylo lekarza. in village NEG was3.SG.N doctorOEN.SG.M 'There was no doctor in the village.' b. BUT: existential sentence (NP o e n > BE > PP) Lekarza nie bylo we wsi. doctorGEN . M NEG w a s 3 . s o . N in village ~ 'The doctor was not in the village.' b . (Context: John had a small accident and went to see his doctor. Unfortunately) Lekarza nie bylo dzisiaj we wsi. doctorGEN.SG.M NEG was3.SG.N today in village 'The doctor was not in the village today.' (He had some important meeting in the town.) S 0

(20)

existential sentence (NPOEN > BE > PP) (Context: John's teacher talking to John's mother: "I must tell you something about your son.") Jana nie bylo dzisiaj znowu na lekcji. JohnOEN NEG was3.SG N today again on lecture 'John was not at the lecture today again.'

To conclude, a simple uniform analysis of existential and locative sentences sketched above leaves us without any clear answer to the question about the case marking of nominal arguments and about possible word orders in such sentences. Is there another way of explaining the problematic facts?

2.3 An Analysis in Terms of Scope of Negation and Information Structure - Not Good Either According to Babby (1980), there are two factors that are relevant for the GEN marking of the "subject" in negated existential sentences, namely: (i) scope of negation and (ii) information structure (understood here in terms of Theme-Rheme structure). He distinguishes between existential sentences (ES), on the one hand, and declarative sentences (DS) (to which our locative sentences would belong), on the other hand. While in the former negation/assertion has scope over the entire sentence, in the latter the scope of negation/affirmation is restricted to the verbal predicate, i.e., the subject is not included in the scope of negation/assertion, but rather the entity denoted by the subject NP is presupposed to exist independently; cf. (21).

26

Joanna Btaszczak AFFIRMATIVE

(21)

NEGATED 9

EXISTENTIAL

a. [scope of a VP NP]

%

DECLARATIVE

c. NP [Scope of a VP]

=>

b. [NEG V P N P G E N ] d . N P N O M [neg V P ]

a. We wsi byl lekarz in village w a s 3 . S o . M d o c t o r N O M . s G . M 'There was a doctor in the village.' b. We wsi nie bylo (zadnego) lekarza. in village NEG w a s 3 . S G . N (noGEN SG M) doctor GENSG . M 'There was no doctor in the village.' c. Lekarz byl we wsi.

GEN

doctorNOM.sG.M was3.sG m in village

'The doctor was in the village.' d. Lekarz nie byl we wsi. doctorNOM.SG.M NEG was3.SG.m in village 'The doctor was not in the village.'

NOM

Next, Babby (1980) proposes identifying the scope of negation/assertion with the rheme part of a sentence. Accordingly, affirmative and negated existential sentences are analyzed as "rheme-only" sentences (possibly with an optional locative theme). The GEN is assigned to the "subject" NP in negated existential sentences (NES) in accordance with the following rule in (22): (22) Rule of Genitive marking in NES: [rheme V N P ] "4 s [ n e g [ V N P o e n ] Conditions: (a) NP is indefinite, (b) V is semantically empty 8 However, this analysis of the GEN marking of the nominal argument in negated existential sentences is problematic for reasons we have already discussed. First, we saw in (20) that the GEN marking is not an exclusive property of indefinite NPs. Second, as was illustrated in (19) and (20), a GEN marking is possible even if a given NP is not part of the rheme (contrary to what the condition in (22) would let us expect). Moreover, the assumption that the NOM/GEN case alternation in negated locative/ existential sentences can be reduced to the scope of negation is problematic as well. In Babby's analysis an NP has to occur in the scope of negation (which in turn is identified with the rheme part of a sentence) in order to be marked for GEN. At first glance it might appear that the scope of negation is indeed a decisive factor determining the case marking of the nominal argument in negated existential/locative sentences. Thus, the intuitive difference between (23a) and (23b) would be that in the former case the NP is outside the scope of negation while in the latter case it is still in the scope of negation.

8

Recall the discussion in section 2.1, where it was pointed out that a predicate (in Russian) can be semantically "bleached" to just an "existential meaning".

27

A Spurious Genitive Puzzle in Polish (23)

a. Lekarz

nie

byl

we wsi.

doctorNOM.sG.M NEG was3.S0.M in village 'The doctor was not in the village.' b. Lekarza nie bylo we wsi. doctorGEN.sG m NEG was3.so.N in village ? T h e r e was no doctor in the village.' 'The doctor was not in the village.'

This assumption is, however, problematic for at least two reasons (see Borschev & Partee, 2001; Partee & Borschev, 2002; Partee, 2000 for similar observations and extensive discussion of Russian facts). Firstly, notice that Polish has no articles, hence the definite/indefinite interpretation of nominal phrases must be rendered in some other way, mainly by word order or by some special (in)definiteness markers such as demonstrative or indefinite pronouns. Now, on the assumption that a GEN marked NP has to occur within the scope of negation (in the rhematic part of a sentence), we would actually expect that a bare nominal phrase like the NP 'doctor' in the example above (i.e., an NP that is not accompanied by any overt marker of definiteness like a demonstrative pronoun) should be interpreted as indefinite. Thus, we would expect that the example (23b) above has (mainly) the interpretation: 'There was no doctor in the village.' This interpretation is in fact quite difficult to get here. The most obvious reading (under neutral intonation), which is in accordance with the basic rule about the interplay of word order and the definite versus indefinite interpretation of noun phrases in Polish, 9 is the reading under which the NP 'doctor' is interpreted as definite (and as a matter of fact as belonging to the thematic part of the sentence). But note that - if the above observation is correct - this would actually mean that the GEN NP occurs outside the syntactic scope of negation. 10 Secondly, notice that in negated existential sentences and in negated locative sentences both arguments, Location and Theme, can (and must) be in the scope of negation if they are themselves negative pronouns, so-called 'η-words'; cf. (24) and (25) (see Borschev & Partee, 2001, p. 40f. for similar Russian examples). This is so because negative pronouns behave like negative polarity items in that - in order to be properly interpreted, i.e., to be grammatical - they must be licensed by (be in the scope of) an appropriate licenser 9

Cf. S z w e d e k (1974, p.208): " [ . . . ] N o u n s with indefinite interpretation appear in sentence final position only (unless explicitly m a r k e d indefinite in s o m e other way). [ . . . ] N o u n s with definite interpretation appear in non-final positions (again, unless explicitly marked otherwise)". See also Weiss (1982, p. 235) and Topolinska (1981, p. 64), w h o point out that a nominal phrase in the initial preverbal position (i.e., in the so-called thematic part of the sentence, under normal intonation characteristics of the initial position) must be accompanied by an indefinite pronoun when it functions as an unspecified (indefinite) argument. Otherwise it would automatically be interpreted as specified (definite).

10

Note that - as pointed out by Borschev & Partee (2001, p. 39f.) - in the case of inherently definite N P s like proper n a m e s the situation is different. It does not matter whether a proper n a m e is outside or within the scope of negation "since its structural scope will not be reflected in any semantic difference in the result". But in the case of a bare noun, being or not being in the scope of negation is expected to have a semantic effect (definite vs. indefinite interpretation), unless w e assume that the preverbal G E N N P is a presuppositional definite description, hence scope-insensitive just like a proper name.

28

Joanna

Btaszczak

(here: negation; see Blaszczak, 2001 as well as the references cited there for a detailed discussion). (24) a. Nikt

tarn nie

byl.

nobodyNOM.sG.M there N E G was 3 . SG .M

'nobody was there.' b. Jan (nigdzie) nie byl (nigdzie). JohnNOM.sG.M (nowhere) NEG was3.SG.M (nowhere) 'John was nowhere.' (25) a. (Nikogo) tarn nie bylo (nikogo). (nobody G E N S G M ) there NEG was3.SG.N (nobody GEN . SGM ) 'nobody was there.' / 'There was nobody there.' b. Nigdzie (Jana) nie bylo (Jana). Nowhere (JohnGEN SG M ) NEG was3.SG N (John GEN . SGM ) Lit.: 'Nowhere was John.' To sum up the discussion so far, it seems that the scope of negation alone cannot explain the difference in the case marking of the nominal argument in negated existential/locative sentences. Likewise, information structure alone, though relevant, is not the decisive factor either. If neither scope of negation nor information structure alone, what else could explain the problematic facts at hand?

2.4 An Analysis in Terms of Different Perspective Structure - A More Promising Approach Given the existence of examples like (19b) or (20), Borschev & Partee (2001,2002) argue that the decisive contrast between negated locative (NDS) and negated existential sentences (NES) is to be defined not in terms of the Theme-Rheme structure (and the scope of negation), but rather in terms of "Perspective Structure". They assume that existential and locative sentences have a common structure, as defined in (26) (cf. Borschev & Partee, 2001, p. 18). (26) The common structure of "existence/location situations" and their descriptions:

BE (THING, LOC) Now, the relevant assumption is that an "existence/location situation" may be structured either from the perspective of the THING or from the perspective of the LOCation. 11 The participant chosen as the point of departure for structuring the situation is called the Perspectival Center by Borschev and Partee (ibid.). 12 If THING is chosen as the Perspectival Center, the emerging structure of the interpretation is that of a locative ("Declarative") 11

Borschev & Partee (2001, p. 36f.) notice that their notion of Perspective is in some sense similar to the notion of Observer used by Paduceva (1992, 1997). The latter notion is also found in Klebanowska (1974).

12

Borschev & Partee (2001, p. 18) point out that "[their] Perspectival Center will play the role that 'Theme' played for Babby (1980)".

A Spurious Genitive Puzzle in Polish

29

sentence; cf. (27). In contrast, if LOC is chosen as the Perspectival Center the resulting structure of the interpretation is that of an existential sentence; cf. (28) (see Borschev & Partee, 2001, p. 18f.,28f. for a detailed discussion and more examples). Importantly, an NES denies the existence of the thing(s) described by the subject NP in the Perspectival Center LOCation (ibid., p. 19). (27) Perspective Structure: BE (THING, LOC) [THING is the Perspectival Center] a. Ivan byl na lekcii. Ivan N0M was3.so.M at lecture 'Ivan was at the lecture.' b. Ivan ne byl na lekcii. IvanNOM NEG was3.SG M at lecture 'Ivan was not at the lecture.'

DS

NDS

(28) Perspective Structure: BE (THING, LOC) [LOC is the Perspectival Center] a. Na lekcii byl Ivan. at lecture was3.sG.M IvanNOM Lit.: 'At the lecture there was Ivan.' / 'Ivan was at the lecture.' b. Na lekcii ne bylo Ivana. / Ivana ne bylo na lekcii. at lecture NEG was3.sc.ν IvanGEN Lit.: 'At the lecture there wasn't Ivan.' / 'Ivan wasn't at the lecture.'

ES

NES

Another important assumption is that "any Perspectival Center must be normally presupposed to exist" (Borschev & Partee, 2001, p. 19). Given this assumption, we expect that NDS (our negated locative sentences) and NES (our negated existential sentences) differ not only in terms of case marking of their "subjects", but also in terms of presupposition. More precisely, while in an NDS the existence of the THING (i.e., the NOM subject) is always presupposed, in an NES it is the existence of the LOC that is presupposed. Partee & Borschev (2002, p. 192) demonstrate this claim on the basis of the following Russian examples (here quoted from Harves (2002, p. 238-9)). Thus in (29), in a context that there was no concert, i.e., the concert failed to take place, only (29a) can be felicitously used. This is so because in (29a) due to its status as an NDS, the existence of LOC, understood here as a location involving an event (concert), is not presupposed. Unlike (29a), (29b) is an NES, meaning that it is precisely the LOC whose existence is presupposed. Because of this, (29b) is not appropriate in the context at hand. (29) Context 1 : 'There was no concert.' a.

OK

Ni odin student na koncerte ne byl. [not one s t u d e n t ] N O M . s G . M at concert NEG was 3 SG M 'Not a single student was at the concert.' b. # Ni odnogo studenta na koncerte ne bylo. [not one student] GENSG . M at concert NEG was3.SG N 'There was not a single student at the concert.'

NDS

NES

Joanna Blaszczak

30

In contrast, in a context that there are no students in the town, only (30b) can be felicitously used. This is so because due to its status as an NES, in (30b) it is the existence of LOC and not that of THING which is presupposed. Accordingly, (30a), which is an NDS, meaning that it is the existence of THING 13 that is presupposed in this case, is not appropriate in the given context. (30) Context 2: 'There are no students in our town.' a.

#

Ni odin student na koncerte ne byl. [not one s t u d e n t ] N 0 M . s G . M at concert NEGwas3. SG . M 'Not a single student was at the concert.' b. OK Ni odnogo studenta na koncerte ne bylo. [not one student]GEN.SG.M at concert NEG was3.sc.ν 'There was not a single student at the concert.'

NDS

NES

An analysis in terms of Perspective Structure as such does not explain why the nominal argument in negated existential sentences is marked for GEN (i.e., what the factors are that allow for a GEN marking of what would appear to be a "subject" NP), but it seems to be more promising than the other analyses discussed in the previous sections in that it at least makes the right predictions, namely it restricts the GEN marking of a nominal argument to one particular type of sentence, i.e., to negated existential sentences in which the LOCATION is chosen as the Perspectival Center, meaning that it is the LOC and not the THING whose existence is presupposed. Since the Perspective Structure is a different kind of distinction than a Theme-Rheme structure (see below), it is no longer necessary for an NP to be the rheme (or part of the rheme) of the sentence in order to be marked for GEN. Thus, an NP which is the theme (or part of the theme) in a given sentence can still be marked for GEN as long as such an NP is the THING in a negated existential sentence (i.e., in a sentence where the LOC is chosen as the Perspectival Center). In the following additional arguments will be given to strengthen the view that in order to account for the NOM/GEN case alternation in negated locative/existential sentences more is needed than information structural distinctions. In particular, it is basically claimed that an analysis based on Perspective Structure, as proposed by Partee and Borschev in their numerous works, is in principle correct. However, the claim needs to be stronger: the difference in the interpretation of locative (Babby's 'declarative') and existential (Babby's 'existential') sentences, which Partee and Borschev attribute to the question of what (THING or LOC) is chosen as the Perspectival Center, is in fact just a reflex of fundamentally different structures of the respective sentences. That is, the difference between locative and existential sentences is above all structurally encoded and all other properties of the respective sentences follow from this. The precise structural analysis cannot be given here for reasons of space (see Blaszczak, 2007 for an extensive discussion) and only the relevant structures will be sketched to make the point mentioned above clear. 13

In the c a s e of a quantificational N P it is not the referent of the N P itself that is presupposed to exist, "but the domain over w h i c h it quantifies" (Partee & Borschev, 2 0 0 2 , p. 191).

A Spurious Genitive Puzzle in Polish

31

3. Solving the Puzzle 3.1 GEN/NOM Alternation is more than Information or Perspective Structure That something like Perspective Structure is more than just the issue of "taking a perspective", but must be structurally encoded (i.e., it has a structural correlate), is corroborated by the formal distinctions between negated locative and existential sentences: among others, different case markings of the nominal argument (NOM vs. GEN) and different agreement properties (agreement vs. lack of agreement, i.e., 'default' agreement). 14 Thus, the question is how the Perspective Structure can be conceptualized. With what other already existing linguistic notions/concepts can it be compared? Partee & Borschev (2004) point out that Perspective Structure "is basically a structuring at the model-theoretic level, like the telic/atelic distinction, or the distinction between Agents and Experiencers. These properties reflect cognitive structuring of the domains that we use language to talk about, and are not simply "given" by the nature of the external world. Correspondingly, all of them are properties with respect to which we find differences from language to language" (ibid., p. 8). They further notice that "the choice of Perspectival Center, as so described, has much in common with the choice of Theme (Topic) 15 on the one hand, and with the choice of grammatical Subject on the other: all three notions involve structuring something (a situation, a proposition, or a sentence) so that one part is picked out and the rest is in effect predicated of it." Partee and Borschev stress that Perspective Structure is not the same as information structure, although they admit that it is in some sense similar to information structure in that the element (LOC or THEME) chosen as the Perspectival Center is also the most likely candidate for the topic. According to Partee and Borschev, their Perspective Structure is also "not directly syntax, although it may be reflected in the syntax" (ibid.). Rather, they take the Perspective Structure "primarily to be a choice of what structure we want to impose on some piece of reality that we want to describe. And in this it has something in common with deciding whether to describe a buying or a selling. It is similar in some ways to figure-ground choices, as in choosing whether to say that A is above Β or that Β is below A" (ibid.).

14

Other formal distinctions between locative and existential sentences are discussed in detail in Blaszczak (2007, 2008a).

15

Partee and Borschev do not seem to make a distinction between 'theme' and 'topic'. See de Swart & de Hoop (2000) for a recent discussion as to how many distinctions (different concepts) with respect to information structure are really needed. For example, do we need both a Theme-Rheme distinction and a Topic-Comment distinction, or could these two distinctions coincide/be reduced to just one distinction? 'Theme' and 'topic' would then be understood as "what we are talking about", 'rheme' and 'comment' as "what we are saying about the theme" (ibid., p. 116).

32

Joanna

Blaszczak

3.2 Existential and Locative Sentences have Different Underlying Structures Unlike Partee and Borschev, I would like to propose that Perspective Structure is not only reflected in the syntax, it is directly encoded in the syntax. 16 Thus, the proposal is that in examples like (6a) and (6b) above, we have two different verbs BE with two different argument structures. In the former case, the nominal argument is projected as the external argument, and in the latter case it is the locative phrase that is the external argument; cf. (31). 17 In other words, what Partee and Borschev describe as taking Perspective on some piece of reality is in fact choosing between two verbs BE: a locative one or an existential one, each of which has its own argument structure. Given that in the case of the existential verb BE the (quasi) external argument is the Location (which might be understood a Possessor) while the nominal argument is the internal argument, the resulting structure is what Partee & Borschev (2004, p. 6) describe as "in an existential sentence, it is as if the predication is somehow "turned around", to assert of the LOCation that it has the THING in it." While for Partee and Borschev the answer to the question of "in what way and at what 'level' of structure the predication is 'turned around'" (ibid.) is at the level of Perspective Structure, the answer proposed here is that this is a matter of having two different verbs BE whose arguments are differently projected in the syntax; cf. (31). 18 16

Partee & Borschev (2004, p. 8, the page number here refers to the manuscript of this paper) are a little bit more precise about "where in the grammar the choice of Perspective Structure is registered," namely "it is a 'diathesis choice', a choice among two alternative argument structures for verbs that can take both a ' T H I N G ' and a ' L O C ' argument, analogous to the argument structure choices for verbs like spray, load or verbs like give, send. An alternative that might be preferred in some frameworks is to permit alternative surface syntactic choices f r o m a single underlying structure, as in the 'small clause' analyses of Chvany (1975), M o r o (1997) [...]."

17

M o r e correctly, one has to differentiate two types of locative sentences. For space reasons, I will ignore (ii).

(i)

[„p

(ii)

[vp ν [ V p

18

N P A G K N T [V·

V

NPTHEMR

[VP [V·

ν V

locative (agentive

PPLOCUI PPLOCUI

reading)

locative ("simpleposition"

reading)

Note that only the N P in locative sentences can have an agentive interpretation (cf. (i), due to Dziwirek, 1994, p. 173—4) and shows usual subject properties (in terms of binding; cf. (ii)). This is expected under the analysis proposed here since only locative sentences (on their agentive reading) have an external N P argument ("Agent") whereas the nominal argument in existential sentences is a T h e m e argument, generated in the underlying 'direct object' position.

(i)

a.

b.

(ii)

a.

b.

Celowo nie bylem na przyjgciu on purpose N E G w a s i . S G M at party Ί wasn't at Eve's party on purpose.' * Celowo nie bylo mnie na on purpose N E G was3 SG N meGKN at

u Ewy. at Eve przyjgciu u Ewy. party at Eve

Jani (nie) byl w swoimi / * jegOi pokoju. John N O M ( N E G ) waS3. S G M in R E F L / * his room 'John was (not) in his room.' Janai nie bylo w jego, / *? swoimi pokoju. JohnoEN N E G was3. SG N in his 'John wasn't in his room.'

/ *? R E F L

room

locative

existential

locative

existential

33

A Spurious Genitive Puzzle in Polish (31)

a.

[VP PPLOC [Ν ν [ V p V NP T H E ME]]]

existential

b. [vp NP [Ν ν [vp V PPLOC]]]

locative (agentive

reading)

Notice that by proposing that the difference between locative sentences and existential sentences is directly encoded in the syntax, the case marking of the nominal argument (NOM vs. GEN) becomes a matter of syntax. In other words, whether the nominal argument is going to be marked for NOM or GEN is a question of what syntactic structure a given sentence has. By doing this, we free the case marking in negated existential and locative sentences from the obligation to reflect the information structure differences or differences in terms of the scope of negation (recall the analysis by Babby, 1980). 3.2.1 Case Properties By assuming that locative and existential sentences have different underlying structures the case marking properties cease to be puzzling. It was pointed out in section 1 that GoN in Polish is restricted to just one configuration: direct object position of a negated transitive verb. Now, the analysis of existential sentences along the lines proposed in (31a) provides an immediate solution to the GEN dilemma in this case. The GEN marking of the nominal argument in negated existential sentences is no longer surprising, no more than the NOM marking of the nominal argument in negated locative sentences. On the contrary, the GEN marking is actually expected - given the structure in (31a) - and it is completely compatible with the usual GoN facts in Polish. Let us look at the structures provided in (32). (32)

a. NEG

[vp XP [Ν· Ν [yp V NPTHEME]]]

canonical

transitive

î

GEN b. N E G

[ v p PPLOC/POSSESSOR Iv Ν [ V p B E NP T H E ME]]]

existential

î GEN c. NEG

[,ρ N P a g e n t [„• V [VP BE PPLOC]]]

locative

('agentive')

î no GEN In order to be marked for GEN under negation, the nominal argument must be in the right configuration: the nominal argument has to be generated as a direct internal argument of the negated transitive. Thus, to put it in descriptive terms, it seems that for an internal argument to be GEN marked under negation in Polish, there must be another (external) argument present in the structure; cf. (32a). Now notice that on the analysis advocated in this paper only existential sentences offer the right configuration for GoN assignment; cf. (32b). The locative argument (location), which is generated in some sense "externally" to the verb (in the specifier position of a light verb), might be understood (in some abstract sense) as a possessor: 19 The situation that some entity exists/does not exist at

19

Interestingly, note that in regular possessive sentences with an N P possessor as an external argument, the internal argument ('the p o s s e s s e e ' ) is also marked for G E N under negation in Polish:

34

Joanna Blaszczak

some location can be understood in such a way that the location contains/does not contain some entity. 20 In contrast, locative ("agentive") sentences do not provide the right configuration: in this case (cf. (32b)), the NP is itself the external argument.

3.2.2 Information Structure / Discourse Properties 3.2.2.1 'Inner predication ' and 'outer predication ' Before we discuss the information structure properties in locative and existential sentences, let us make a distinction between 'inner predication' (i.e., the predication within the v-V-domain: the thematic-aspectual domain, inner domain) and 'outer predication' (i.e., the predication within the C-T-domain: the discourse-informational domain, outer domain). Note that these two predication domains determine what syntactically relevant phases are. 21 In the first phase of the derivation (the 'inner' phase) the arguments of a given verb are syntactically projected, resulting in a thematic-aspectual structure. For a "verbal phrase" to be thematically complete, all argument positions must be discharged, which means that also the eventuality argument (the referential argument) must be discharged, or to put it in other terms, existentially bound. Assuming that the existential binding of the eventuality argument takes place in the Polarity Phrase (PolP), PolP in this sense closes up the v-Vdomain, that is, PolP belongs to the first phase of derivation, the 'inner phase'. 2 2 , 2 3 The inner phase is also the domain within which "inner predication" takes place. Following Chierchia (2004:26), this predication relation is taken to be a relation which consists in predicating a property of an individual (importantly, an individual of any sort), the result of which is a proposition: "If r is a property and u an individual (of any sort) and ~ is the predication relation, then "r(u) is the proposition that u has property r " (ibid.). Syntactically, the predicator ( " ) can be regarded as being associated with the functional head v, VP functioning then as a property. In other words, there is a predication relation (i)

a.

b.

Jan ma samochód / * samochodu John has car AC c / * car0HN 'John has a car.' Jan nie ma samochodu / * samochód. John NEG has carGEN / * car A cc 'John has no car.'

20

Cf. Zamparelli (1995) for a similar analysis with respect to existential sentences in English and Italian; cf. also Hazout (2004).

21

See also den Dikken (2007) for the notion of an inherent phase which is also based on predication (subjectpredicate structure). However, den Dikken's analysis is very different from that proposed here.

22

Notice that it is also at this level (or to put it more precisely: at the first phase of derivation in a phasal model of Chomsky (2000) et seq.) that the fate of the case marking of the nominal argument in negated existential and locative sentences is decided.

23

Depending on the value of the head of the PolP, affirmative or negative, the meaning w e will get will be: there is an eventuality e such t h a t . . . ( 3 e ) or there is no eventuality e such t h a t . . . (~>3e), respectively.

A Spurious Genitive Puzzle in Polish

35

which is mediated by v°, and which consists in predicating a property, realized syntactically as a VP, of an individual occupying the Spec,vP position (the external argument). 24 In the second phase of the derivation (the 'outer' phase) the temporal properties (and also temporally related aspectual properties) and modal properties (including force/clause typing etc.) are determined. And, more importantly, also discourse-informational properties are settled here. The latter comprise, among others, determining what the sentence is about, that is, choosing the sentence topic. Normally, one of the verb's arguments, i.e., one of the elements of the preceding phase, is chosen to be the topic of a sentence. Given that the eventuality argument is the referential argument of the verb, it can be chosen to be the topic of a sentence as well. This is what happens in thetic sentences. 25 In this respect, word order and discourse-pragmatic properties, are decided on at the second phase (CP/TP), although they are in some sense derivatives of the decisions taken with respect to the first Spell-Out domain. With this background, let us have a closer look at some concrete examples. 3.2.2.2 Subjects of 'inner' and

'outerpredication'

In examples like (27) and (28), it is the nominal argument (the 'thing') in the former case and the Location argument in the latter case which is the "subject" of 'inner' predication. The "subject" of the 'inner' predication usually (by default) becomes a "subject" of 'outer' predication, understood here in terms of a Topic-Comment structure. This corresponds to the cross-linguistic observation that the subject of a sentence is by default also the topic of the sentence unless the subject is somehow explicitly marked as not being the topic (such marking strategies may include the use of special focus particles, the placement of the subject in some postverbal position, or the use of special intonation) (see Lambrecht, 2006 for a recent discussion). This assumption would also be in agreement with the view of Kiss (2002) concerning the interpretation of the standard EPP (Extended Projection Principle), namely the idea that it could be reinterpreted as the requirement that the sentence contain a subject of predication (see also Rothstein, 2001). 26 Notice that by making the assumption that the "subject" of the 'inner' predication becomes (by default) the Topic of the sentence, we also derive Borschev and Partee's observation concerning the presuppositional nature of their Perspectival Center. Given that what Borschev and Partee refer to as 'Perspectival Center', i.e., the Location argument and the nominal argument ('Thing'), is the "subject" of the 'inner' predication in existential and locative sentences respectively, and given furthermore that the "subject" of inner predication is by default also the Topic of the sentence, we derive the fact that "any Perspectival Center must be normally presupposed to exist" (Partee & Borschev, 2002, p. 188). This is 24

In the case of verbs lacking an external argument it can be assumed - following Chierchia (2004) - that they have an "expletive subject" via the "Expletivization Rule". Expletivization applies to a proposition and turns it into a property that is "predicated of an arbitrarily chosen funny object" (ibid., p. 32).

25

Cf. Krifka (2006), w h o observes that - though thetic judgments lack a topic constituent - they have a topic denotation, namely the situation itself. See below.

26

This is actually only one of the two functions of the standard EPP, referred to by Kiss (2002, p. 116) as EPP1: "A sentence expressing predication must contain a topic." The second function, called by Kiss (2002, p. 119) EPP2, is the following: "Of the arguments of a predicate, one must be marked as a subject". These two requirements usually, but not necessarily, coincide.

36

Joanna Blaszczak

so because we (normally) presuppose the existence of things (the topic) we are talking about. Or to put it more generally, the subject of predication is normally presupposed to exist.27 Thus, in the default case, the Location seems to become the topic in an existential sentence and the nominal argument ( ' T h i n g ' ) becomes the topic in a locative sentence. The remaining part of the sentence ( V plus the nominal argument in the former case, and V plus the Location argument in the latter case) represents the comment, and thus the new information in the unmarked case. This gives us the unmarked word orders of the respective sentences, i.e., the unmarked PP-(neg)V-NP order of existential sentences and the unmarked NP-(neg)V-PP order of locative sentences. But - as was illustrated in section 2.2 (cf. also ex. (28a) vs. (28b)) - negated existential sentences show more flexibility as far as the ordering of their constituents is concerned. This is so because - unlike affirmative existential sentences, which have a presentational function, i.e., are usually used to introduce new participants into a discourse - the negated existential clauses are in most cases not used discourse-initially, hence they contain discourse-linked material (old information, theme). Now, observe that the main principle governing the actual arrangement of lexical items in a sentence in Polish (as in other Slavic languages) is not grammatical, but communicative. Thus, as argued in Grzegorek (1984, p. 92), "lexical items occur in the order of their relative communicative value, i.e., according to the increasing degree of C. D. ( = Communicative Dynamism 28 ) or according to the topicality in Givon's terminology. [ . . . ] In languages such as Polish or Russian preverbal position usually marks the part of the sentence which is old information, whereas the verb functions either as transition or as part of the focal phrase". Given this principle, it is actually expected that constituents representing old information will be placed in a preverbal position. And this is exactly what happens in the Polish examples (33). (33)

a. [Preceding context: I was told that the doctor should be in the village at this time, but:] Lekarza we wsi nie bylo. doctorCEN.sG.M in

village N E G

was3.SG.N

'The doctor was not in the village.' b. [Preceding context: I went extra to Cracow to meet John, but:] Jana tarn nie bylo. JohnGEN

SG.M

there N E G was3. SGN

'John was not there.' 27

That the topic constituent comes together with an existential presupposition is presumably even more understandable if w e take sentences with an (overt) topic to correspond to categorical statements. Categorical judgments are 'double judgments', i.e., they consist o f two steps: ( i ) "naming an entity" and ( i i ) "making a statement about it" (see Sasse, 1987 and the references cited there for discussion). T h e first step can be taken to be equivalent with asserting the existence o f an entity about which - in the second step - a statement will be made.

28

"Communicative dynamism is the deep word order rendered by the left-right order o f the nodes o f the tectogrammatical representation" (de Swart & de Hoop (2000, p. 113)). This concept played an important role in the Prague School. In accordance with communicative dynamism, the 'less dynamic' topical material precedes the 'more dynamic' focal material. In the tectogrammatical representation there is a clear borderline between topic and focus: "There is an item A such that every item which is less dynamic than A belongs to the topic, whereas every item which is more dynamic belongs to the f o c u s " (de Swart & de H o o p (2000, p. 113)).

A Spurious Genitive Puzzle in Polish

37

Note that in the examples above both the GEN marked nominal argument as well as the Location argument represent old information, i.e., they belong to the background. Is there an explicit (distinguished) topic in such examples? Are we talking about 'the doctor' / 'John' or rather about 'the village' / 'Cracow'? It seems to me that it is certainly possible here to take 'the doctor' or 'John' to be the topic of the respective sentence ('the village' / 'Cracow' simply being the background or thematic/given information). Note that it is also possible that only the GEN marked NP appears in preverbal position whereas the locative phrase appears postverbally; cf. (34) (see also (19b) and (20)). (34)

[Preceding context: I was looking everywhere for the doctor, but:] Lekarza nie byto we wsi. d o c t o r G E N . s G . M NEG was3.sG.N in village 'The doctor was not in the village.'

Again, in this example both 'the doctor' and 'in the village' belong to the background (if 'in the village' is contextually understood as 'everywhere in the village'), but only the NP appears preverbally, hence only the NP seems to be a good candidate for the topic. This would actually correspond to the view advocated by Vallduvi (1993) that instead of a binary distinction like Background-Focus, a trinary distinction is needed since not every element belonging to the background is automatically the topic. The topic in fact usually corresponds to only one part of the background information, which he calls 'link'. The remaining part of the background (i.e., the background without the link) is called 'tail'; cf. (35) from Vallduvi (1993, p. 8).29 S

FOCUS GROUND

LINK (35) a. What about John? What does he do? b. What about John? What does he drink?

TAIL [ G [ L John]] [ F drinks BEER] [Q [L John] drinks ] [F BEER]

Assuming such an analysis, the GEN NP in (34) would be the link, and the locative PP would represent the tail; the focus of the sentence would be NEG + V. 29

That binary distinctions like Topic-Comment or Theme-Rheme are not always adequate to correctly describe the information structure of a given sentence has also been pointed out by Fibras (1965). He observes that often it is the case that, e.g., a verb in a given sentence does not belong to focus (new information, rheme), nor does it represent old information (theme). He proposes therefore that verbs play a transitional role between theme and rheme and postulates a trinary structure: Theme-Transition-Rheme. King (1995, p. 77) expands this idea: "If this three-way division is further expanded to allow elements other than the verb to appear in the transition, then this division closely resembles the division into topic, discourse-neutral material, and focus: the theme corresponds to the topic, the transition to discourse-neutral material, and the rheme [ . . . ] to the focus."

38

Joanna

Blaszczak

Finally, let us look at (36). Here, both the Location argument and the nominal argument appear postverbally. In such examples it could be assumed that the actual topic is the invisible event variable. However, it might be assumed that in such cases there is a situationally or contextually determined invisible restriction on the event variable: 'at that time' as in (36a) or more generally 'there and then' as in (36b). The postverbal material in (36) represents new (focal) information or discourse-neutral information. (36) a. (W owym czasie) nie bylo wewsi lekarza. (at that time) NEG was3.SG.N in village doctor GEN '(At that time) there was no doctor in the village.' b. Byl duzy ruch na glównej ulicy. was 3 . SG M [big traffic] N O M.sG.N on main street 'There was heavy traffic on the main street.'

SG M

The proposed analysis would in fact closely resemble the analysis of Kiss (2002). Kiss (2002, pp. 114-5), following the proposals put forward by Kratzer (1995) and ErteschikShir (1997) on the basis of Davidson (1967), that eventive verbs, hence eventive sentences, have an event argument serving for spatio-temporal specification, typically represented by a variable bound by an invisible existential. In eventive sentences the optional temporal and locative expressions are licensed by the invisible event variable; they function as restrictions on it and tend to appear in the topic position. Further Kiss assumes that "[... ] eventive sentences with no visible constituent in SpecTopP predicate about the situational or contextual restriction on their event variable". The subject of predication is then in such cases the situationally or contextually determined invisible restriction on the event variable (meaning 'here and now', or 'there and then')" (ibid., p. 115).30 To sum up the discussion so far, it has been argued that the difference between existential and locative sentences is syntactically encoded in terms of what (LOC or THING) is the subject of inner predication (at the vP/VP level). Also the case marking of the nominal argument has been claimed to be a matter of syntax. The actual word order in existential and locative sentences is pragmatically determined. For example, we have seen that in a (negated) existential sentence the nominal argument may appear postverbally or preverbally, depending on its discourse status. Accordingly, also a locative phrase may be placed postverbally if it is itself 'less dynamic' than the nominal argument, as in (34), or if it itself represents new information, as in (36). But importantly, in all these examples, independent of the actual position of the locative phrase in the sentence, it is the Location argument that is the subject of inner predication and the nominal argument ('Thing') that 30

(i)

That locative (or temporal) expressions might function as restrictions on the event variable appears to be intuitively clear. Kiss (2002, p. 115) notes that in Hungarian such temporal and locative expressions tend to appear in topic position; cf. (i) (ibid.). The truth value of a given sentence is evaluated with respect to the referent of such a temporal or locative element. Thus, the truth value of (i) is evaluated with respect to the referent of tegnap 'yesterday'. [TOPP Tegnap [vp meghívta Marit Jónos vacsorára]] yesterday invited Mary A cc John for.dinner 'Yesterday John invited Mary for dinner.'

A Spurious Genitive Puzzle in Polish

39

is the object. This shows us that the case marking of the nominal argument cannot be primarily a matter of information structure. More precisely, the nominal argument does not have to represent new information (to be the Rheme in the sentence) in order to be marked for GEN (recall the discussion of Babby's (1980) analysis in section 2.3). Note that this is exactly what happens in other sentences in Polish. So, for example, in all the sentences given in (37) below, 'John' is the subject of the inner predication (at the vP level). By default, the subject of inner predication becomes the subject of outer predication (the topic of the sentences; see the discussion above); cf. (37a). However, as was pointed above, the actual arrangement of constituents in a sentence in Polish is pragmatically determined, i.e., depending on their communicative value (discourse status), the subject and the object may appear in different positions in the clause; cf. (37b/c) (other orders are possible as well; see Witkos (1993, p. 29If.)). Note that the - admittedly - marked verb-initial word order in (37d) could be analyzed along the line proposed by Zybatov & Junghanns (1998) for verb-initial sentences in Russian. According to Zybatow and Junghanns, such sentences are best analyzed as thetic sentences which have an abstract Topic ("situation / eventuality argument topic") (cf. also the analysis by Kiss, 2002 mentioned above). More precisely, the topic feature is assigned to no overt constituent, but to the functional head Tense, and it must be realized or marked as such by the verb overtly raised to Τ (see Blaszczak, 2007 for a different implementation of this view). (37) a. Jan czyta ksiqzkç. JohnNOM read3.SG.pRES bookACC 'John is reading a book.' b. Ksiqzkg czyta Jan. bookACC read3.sG.pREs JohnNOM c. Ksiqzkç Jan czyta. bookACC JohnNOM read3.s0.pREs d. Czyta Jan ksiqzkç read3.SG.pRES JohnNOM bookACC

3.2.3 Scope of Negation: Licensing of n-Words The proposed analysis (i.e., the assumption that existential and locative sentences have different base structures) has yet other nice pay-off. Once we have freed the NOM/GEN alternation in negated locative/existential sentences from the scope of negation, the otherwise problematic facts concerning the licensing of η-words in such constructions (recall ex. (24) and (25), which are repeated below) receives a natural explanation. (24) a. Nikt

tarn nie byl. nobodynom.sg.m there NEG w a s 3 . S G . M 'nobody was there.' b. Jan (nigdzie) nie byl (nigdzie). John N O M .sG.M (nowhere) NEG was3.SG.M (nowhere) 'John was nowhere.'

40

Joanna Blaszczak

(25) a. (Nikogo) tarn nie bylo (nikogo). (nobodyOEN.SC M) there NEG was3.so.N (nobody GENSGM ) 'nobody was there.' / 'There was nobody there.' b. Nigdzie (Jana) nie bylo (Jana). Nowhere (John GEN . SGM ) NEG was3.SG.N (John GENS0 . M ) Lit.: 'Nowhere was John.' The licensing is done at the level of the first phase of a derivation, negated vP/VP (see Blaszczak, 2001). Note that the negation does not differentiate between locative and existential sentences here, as shown in (38). This emphasizes once again the argument that the case marking of the nominal argument in the examples at hand cannot be a matter of being in the scope of negation. (38)

a. [ N E G P N E G [ VP PP [VP B E NP NOM ]]]

a

b.

· [NEGP

NEG [vp nowhere [vp BE noboáy GEN ]]] I Î t 'licensing'

[NEGP N E G

[ V P N P N O M [VP B E

PP]]]

b . [NEGP NEG [vp nobodyNOM [ V P BE nowhere]]] I î î 'licensing'

Note that exactly the same happens in any other negated sentence in Polish. The negative pronouns are licensed in the scope of negation at the first phase level; cf. (39). If a negative pronoun is chosen to be the subject of outer predication, it will be interpreted partitively in the sense of 'no X of some contextually determined/presupposed set' (cf. also footnote 13). This is true for (39), where 'nobody' can mean, e.g., 'none of the students', and unsurprisingly - also for, e.g., (24a) and (25b). Thus (24a) might mean, depending on a given context, e.g., 'none of us', 'none of the professors', etc., and (25b) can mean 'at none of the places I was looking for John/him'. Hence, we see that the scope of negation in negated existential/locative sentences is determined in exactly the same way as in other negated sentences (in Polish). No special assumptions are needed here. (39) a. Nikt

nie przeczytal artykutu. n o b o d y N O M . s G . M NEG r e a d 3 . S G . M . P A S T . P E R F articleGEN.SG.M 'Nobody has read the article.' b. [NEGP [ NEG [ vP nobody [ V p read paperGEN ]]] I î 'licensing'

3.2.4 Additional Supporting Evidence: "Correction" and Ellipsis Another argument for the claim that the difference between existential and locative sentences is primarily syntactically encoded and is only secondarily reflected in terms of

A Spurious Genitive Puzzle in Polish

41

information structure is based on the observation that only in negated locative sentences with a NOM NP can a contrastive/correction phrase 'but ' be used; see Saloni & Swidzinski (1985, p. 143); cf. (40).31 (40) a. NEG/NOM / Chlopiec nie byl w Lublinie, tylko w Warszawie. boyNOM.sc.M NEG was3.sG.M in Lublin but in Warsaw 'The boy was not in Lublin but in Warsaw.' b. NEG/GEN * Chlopca nie bylo w Lublinie, tylko w Warszawie. boyGEN SG.M NEG was3.SGN in Lublin but in Warsaw Note further that examples like (40b) are bad irrespective of the word order. Thus, a counterpart of (40b) with a sentence-initial locative PP and a postverbal NP is also bad; cf. (41). Assuming that different word orders normally reflect different information structures in Polish, the lack of contrast between (40b) and (41) seems to suggest that the reason for the ungrammaticality of the examples at hand is rather to be looked for in some deeper structural properties of the respective sentences. Cf. also (42). (41) * W Lublinie nie bylo chlopca, tylko w Warszawie. in Lublin NEG was3.S0.N boy GENSGM but in Warsaw (42) a. * Jana nie bylo w Lublinie, tylko Piotra. JohnGEN NEG was3.SG N in Lublin but Peter 0EN (intended: 'Not John was not in Lublin, but Peter.' / 'It is not John, who was not in Lublin, but Peter.' ) b. ?* W Lublinie nie bylo Jana, tylko Piotra. in Lublin NEG was3 SG N JohnOEN but PeterGEN Before we explain the ungrammaticality of (42), let us go back to the contrast illustrated in (40) above. I would like to suggest that the reason why (40a) is fine and (40b) is bad is first of all structurally determined. This provides an indirect argument for the claim that the difference between locative and existential sentences is primarily structurally encoded in terms of what (Location or the nominal argument ('Thing')) is the subject of the inner predication and only secondarily reflected in the information structure of the corresponding sentences. So let us consider what exactly happens when a contrastive/corrective phrase 'but is used. It seems to me that the use of such a phrase makes the "structurally sentential" negation a contrastive negation, which means that the sentence is no longer negative, but "affirmative". This in turn requires that such a "previously negative" sentence has a syntactically identical affirmative counterpart. The reason for this is that the contrastive/corrective phrase is based on ellipsis (which in turn requires structural identity of the elided material). Now observe that only (43a), which corresponds to (40a), has a cor31

See Borschev & Partee (2001, p. 51 f.) for a relevant discussion in Serbian/Croatian and Russian.

42

Joanna Blaszczak

responding, i.e., structurally identical, affirmative counterpart, cf. (43b) and (43c), hence the ellipsis works fine in this case, as shown in (43d/e). (43) a. Jan nie byl w Lublinie. JohnNOM NEG w a s 3 . S G . M in Lublin 'John was not in Lublin.' —» "affirmative counterpart" b. Jan byl w Lublinie. JohnNOM was3.sG.M in Lublin 'John was in Lublin.' —• "underlying structure" (identical for both (43a) and (43b)) c. [ N E G P nie [vp Jan [ V p byl w Lublinie ]]] —>· "contrastive phrase + ellipsis"(bold-face = spell-out; struck through = elided material; italics = copies) d. Jan byl nie w Lublinie, tylko w Warszawie. JohnNOM NEG was3.SG.M in Lublin but in Warsaw 'John was not in Lublin, but in Warsaw.' e. Jan [NEGP nie byt [ vP Jen [yp byt w Lublinie ]]], tylko [ vP Ja» [yp byt w Warszawie ]].

[- (40a)]

In contrast with (43a), (44a), which corresponds to (40b), does not have a "structurally identical" affirmative counterpart, cf. (44b/c): the locative sentence in (44b) has a completely different underlying structure; cf. (44b'). The existential sentence in (44c) is also not identical: a different form of the verb is used and the nominal argument is differently case marked; cf. (44a') and (44c'). Given this, no ellipsis is allowed here; cf. (44d) and (44e/e'). (44) a. Jana nie bylo w Lublinie. JohnGEN NEG was3.SG.N in Lublin 'John was not in Lublin.' —> "underlying structure" · ' [ N E G P πίε [VP w Lublinie [vp bylo JanaoEN ]]] —> "affirmative counterparts" b. Jan byl w Lublinie. JohnNOM was3.sG.M in Lublin 'John was in Lublin.' c. W Lublinie byl Jan. in Lublin was3.SG.M JohnNOM 'John was in Lublin.' —)· "underlying structures" b . [vp Jan [yp byl w Lublinie ]] c.' [vp w Lublinie [yp byl Jan NOM ]] —> "contrastive phrase, NO ellipsis"

[=(40b)]

A

locative

existential

A Spurious Genitive Puzzle in Polish

43

d. * Jana nie bylo w Lublinie, tylko w Warszawie. John CEN NEG was3.SG.N in Lublin but in Warsaw (intended: 'John was not in Lublin, but in Warsaw.') e. Jana [NEGP nie byto [„P W Lublinie [ V p bylo JanaGEN ]]], tylko [ vP Jan [vp byl w Warszawie ]]. or: e.' Jana [NEGP nie byto [ vP w Lublinie [ V P bylo JanacEN ]]], tylko [1>P w Warszwawie [ V P byl JanNoM ]]· Note that the same kind of explanation could be used to account for the ungrammaticality of (42). Assuming again that the use of a contrastive/corrective phrase ' b u t . . . ' means that we contrast a negative sentence with its affirmative counterpart and subsequently elide the "identical parts", in the case of (42) no ellipsis is possible since the negative and affirmative counterparts are structurally/morphologically not identical. This is illustrated in (45) below. Note that by virtue of being the nominal argument in an affirmative existential (or locative) sentence the NP 'Peter' should be marked for NOM, and not for GEN. 32 (45)

Jana nie JohnGEN NEG *tylko Piotra but PeterGEN

byto w Lublinie. was3. SGN in Lublin byl w Lublinie. was3.SG m in Lublin

3.3 Conclusion To sum up the discussion, it has been argued that the difference between existential and locative sentences is primarily structurally encoded at the vP/VP level (at the first phase of a derivation). The crucial question is which argument of the verb BE (the Location or the nominal argument (the 'Thing')) is projected as the "external argument", i.e., which argument is the subject of inner predication. In the case of existential sentences it is the Location argument which is the subject of inner predication, and in the case of locative sentences it is the nominal argument. The subject of inner predication becomes by default also the subject of outer predication, i.e., the topic of the sentence. Hence, in the case of locative sentences the nominal argument, the 'Thing', is the subject of outer predication, i.e., the topic of the sentence, and in the case of existential sentences it is the Location which becomes the topic (or, as has been pointed out above, the actual topic (the subject of outer predication) might be the situational/event variable, and the Location functions as a restriction on it). However, the actual arrangement of constituents in the sentences under discussion, as in any other Polish sentence, is determined by the pragmatic/communicative principles. 33 Given this, it is reasonable to think that 32

Interestingly, both Location (PP) and Thing (NP) can in fact be contrasted in negated existential sentences as long as a second negation is used. This is so because by using two negations, a contrastive (constituent) negation and a sentential negation, the sentential negation does not need to function (to be interpreted) as a contrastive negation, as was the case in the examples discussed above. See Btaszczak (2007) for details.

33

This conclusion is, in fact, in some sense similar to the conclusion reached by Sasse (1987, p. 517) with respect to the thetic/categorical distinction, namely: "The thetic/categorical theory ( . . . ) is based on the

44

Joanna Blaszczak

the NOM/GEN case alternation in negated existential/locative sentences is primarily a matter of syntax, and not one of information structure or scope of negation. By making this assumption, the initial "GEN puzzle" in sentences like (6) in Polish turns out to be a spurious one.

References Babby, L. H. (1980). Existential Sentences and Negation in Russian. Ann Arbor, MI: Karoma. Babby, L. H. (2000). The Genitive of Negation and Unaccusativity. Ms., Princeton University. Babyonyshev, M. (1996). Structural Connections in Syntax and Processing: Studies in Russian and Japanese. Ph.D. thesis. MIT, Cambridge, MA. Babyonyshev, M. (2003). The Extended Projection Principle and the Genitive of Negation Construction. In: S. Brown & A. Przepiórkowski (eds.) Negation in Slavic. Slavica Publishers, Bloomington, IN. 31-69. Blaszczak, J. (2001). Investigation into the Interaction between the Indefinites and Negation. (Studia grammatica, vol. 51). Berlin: Akademie-Verlag. Blaszczak, J. (2007). Phase Syntax: The Polish Genitive of Negation. Habilitation thesis. University of Potsdam. Blaszczak, J. (2008a). Differential Subject Marking in Polish: The Case of 'X was not at Y' Constructions. In: H. de Hoop & P. de Swart (eds.) Differential Subject Marking. Dordrecht: Kluwer. 1-38. Blaszczak, J. (2008b). What HAS to BE used? Existential, Locative, and Possessive Sentences in Polish. In: A. Antonenko, J. Bailyn & C. Bethin (eds.) Formal Approaches to Slavic Linguistics. (The Stony Brook 2006 Meeting, vol. 16). Ann Arbor: Michigan Slavic Publications. 3 1 ^ 7 . Borschev, V. & B. Partee (1998). Formal and Lexical Semantics and the Genitive of Negated Existential Sentences in Russian. In: Z. Boskovic, S. Franks & W. Snyder (eds.) Formal Approaches to Slavic Linguistics: The Connecticut Meeting 1997. (vol. FASL 6). Ann Arbor, MI: Michigan Slavic Publications. 75-96. Borschev, V. & B. Partee (2001). The Russian Genitive of Negation in Existential Sentences: The Role of Theme-Rheme Structure Reconsidered. In: E. Hajicova, P. Sgall, H. J. & T. Hoskoves (eds.) Travaux de Cercle Linguistique de Prague (novelle série). (vol. 4). Amsterdam: John Benjamins. 185-250. Manuscript 2001, published 2002. Borschev, V. & B. Partee (2002). Existential Sentences, BE, and the Genitive of Negation in Russian. Manuscript. [Paper presented at the Conference on Existence: Semantics and Syntax. Nancy, September, 26-28, 2002], assumption of two fundamentally different types of statement, which operate independently of criteria of information structure (given/new)." In his view (ibid., p. 511), the thetic/categorical distinction is "not a matter of information structure", but it has something to do with the notion of 'predicativity' : "categorical sentences contain a predication base about which some state of affairs is predicated, while thetic sentences are simple nonpredicative assertations of states of affairs."

A Spurious Genitive Puzzle in Polish

45

Brown, S. (1996). The Syntax of Negation in Russian. Ph.D. thesis. Indiana University. Cetnarowska, B. (2000). The Unergative/Unaccusative Distinction in Polish. In: P. Banski & A. Przepiórkowski (eds.) GLiP-1: Proceedings of Generative Linguistics in Poland. Warsaw: Institute of Computer Science, Polish Academy of Sciences. 3 5 ^ 6 . Chierchia, G. (2004). A Semantics for Unaccusatives and its Syntactic Consequences. In: A. Alexiadou, E. Anagnostopoulou & M. Everaert (eds.) The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface. Oxford and New York: Oxford University Press. 22-59. Chomsky, N. (2000). Minimalist Inquiries: The Framework. In: R. Martin, D. Michaels & J. Uriagereka (eds.) Step by Step. Cambridge, MA: MIT Press. 89-155. Chvany, C. (1975). On the Syntax of BE-Sentences in Russian. Cambridge: Slavica. Davidson, D. (1967). The Logical Form of Action Sentences. In: N. Rescher (ed.) The Logic of Decision and Action. Pittsburgh: University of Pittsburgh Press. 81-95. den Dikken, M. (1995). Copulas. Manuscript, Vrije Universiteit Amsterdam/HIL. den Dikken, M. (1997). The Syntax of Possession and the Verb 'Have'. Lingua 101/3-4. 129-150. den Dikken, M. (2006). Relators and Linkers. The Syntax of Predication, Predicate Inversion, and Copulas. Cambridge, MA: The MIT Press. den Dikken, M. (2007). Phase Extension. Contours of a Theory of the Role of Head Movement in Phrasal Extraction. Theoretical Linguistics 33/1. 1-41. Dziwirek, K. (1994). Polish Subjects. New York, London: Garland Publishing. Erteschik-Shir, N. (1997). The Dynamics of Focus Structure. Cambridge: Cambridge University Press. Fibras, J. (1965). A Note on Transition Proper in functional Sentence Analysis. Philologica Pragensia 8. 170-176. Freeze, R. (1992). Existentials and Other Locatives. Language 68. 553-595. Grzegorek, M. (1984). Thematization in English and Polish. A Study in Word Order. (Seria Filologia Angielska). Poznan: University of Poznan. Harves, S. (2002). Unaccusative Syntax in Russian. Ph.D. thesis. Princeton University. Hazout, I. (2004). The Syntax of Existential Constructions. Linguistic Inquiry 35/3. 393^30. Hoekstra, T. & R. Mulder (1990). Unergatives as Copular Verbs: Locational and Existential Predication. Linguistic Review 7. 1-79. King, T. H. (1995). Configuring Topic and Focus in Russian. Stanford, CA: CSLI Publications. Kiss, K. E. (2002). The EPP in a Topic-Prominent Language. In: P. Svenonius (ed.) Subjects, Expletives, and the EPP. Oxford: Oxford University Press. 107-124. Klebanowska, B. (1974). 'Nie ma', 'nie bylo', 'nie bçdzie'. ['There is not', 'There was not', 'There will not be']. Ρ race Filologiczne [Philological Papers] XXV. 155-160. Kratzer, A. (1995). Stage-Level and Individual-Level Predicates. In: G. N. Carlson & F. J. Pelletier (eds.) The Generic Book. Chicago: Chicago University Press. 125-175. Krifka, M. (2006). Basic Notions of Information Structure. In: Working Papers of the SFB632, Interdisciplinary Studies on Information Structure (ISIS), (vol. 6). Potsdam: Universitätsverlag Potsdam. 13-56.

46

Joanna Btaszczak

Lambrecht, Κ. (2006). Grammar and Information. Paper presented at Potsdam University, June 13, 2006. Moro, A. (1997). The Raising of Predicates. Cambridge: Cambridge University Press. Paduceva, E. V. (1992). O semanticeskom podxode k sintaksisu i genitivnom sub"ekte glagola BYT' [On the Semantic Approach to Syntax and the Genitive Subject of the Verb BYT' 'BE']. Russian Linguistics 16. 53-63. Paduceva, E. V. (1997). Roditel'nyj sub"ekta ν otricatel'nom predlozenii: sintaksis ili semantika? [Genitive of Subject in a Negated Sentence: Syntax or Semantics?]. Voprosy Jazykoznanija 2. 101-116. Partee, Β. (2000). Topics under Negation: "But the answer never came.". In: J. Dölling & T. Pechmann (eds.) Prosodie — Struktur - Interpretation. (- Linguistische Arbeitsberichte, vol. 74). Institut für Linguistik, Universität Leipzig. 43-57. Partee, Β. & V. Borschev (2002). Genitive of Negation and Scope of Negation in Russian Existential Sentences. In: J. Toman (ed.) Formal Approaches to Slavic Linguistics: The Ann Arbor Meeting. Ann Arbor, MI: Michigan Slavic Publications. 181-200. Partee, Β. Η. & V. Borschev (2004). The Semantics of Russian Genitive of Negation: The Nature and Role of Perspectival Structure. In: K. Watanabe & R. B. Young (eds.) Proceedings of SALT XIV. Ithaca, NY: CLC Publications. 212-234. Pereltsvaig, A. (1999). The Genitive of Negation and Aspect in Russian. In: Y. Rose & J. Steele (eds.) McGill working papers in linguistics, (vol. 14). Montreal: McGill University. 111-140. Pesetsky, D. (1982). Path and Categories. Ph.D. thesis. MIT, Cambridge, MA. Richardson, K. (2003). The Case for Meaningful Case: The Interaction of Case, Aspect, and Case in Russian. Ph.D. thesis. Harvard University. Rothstein, S. (2001). Predicates and their Subjects. Dordrecht: Kluwer. Saloni, Ζ. & M. Swidzinski (1985). Skladnia wspólczesnego jçzyka polskiego. [The Syntax of Contemporary Polish]. Warszawa: Pánstwowe Wydawnictwo Naukowe. Sasse, H.-J. (1987). The Thetic/Categorical Distinction Revisited. Linguistics 25. 511580. de Swart, H. & H. de Hoop (2000). Topic and Focus. In: L. Cheng & S. R. (eds.) The First Glot International State-of-the-Article Book: The Latest in Linguistics. Berlin and New York: Walter de Gruyter. 105-130. Szwedek, A. (1974). Some Aspects of Definitness and Indefinitness of Noun in Polish. Papers and Studies in Contrastive Linguistics 2. 203-211. Topoliñska, Ζ. (1981). Remarks on the Slavic Noun Phrase. (Prace Instytutu Jçzyka Polskiego, vol. 37). Zaklad Narodowy im. Ossoliñskich. Wydawnictwo Polskiej Akademii Nauk. Vallduví, E. (1993). Information Packaging: A Survey. Edinburgh: HCRC Publications, University of Edinburgh. Weiss, D. (1982). Indefinite, definite und generische Referenz in artikellosen Sprachen. In: H. J. Mehlig (ed.) Slavistische Linguistik. München: Sagner. 229-261. Witkos, J. ( 1993). Some Aspects of Phrasal Movement in English and Polish. Ph.D. thesis. University of Poznan.

A Spurious Genitive Puzzle in Polish

47

Zamparelli, R. (1995). Layers in the Determiner Phrase. Ph.D. thesis. University of Rochester, NY. Zybatov, G. & U. Junghanns (1998). Topics im Russischen. (Sprache und Pragmatik 47). Lund: Germanistisches Institut der Universität Lund.

Semantic Type Effects on Crossing Movement in German Gisbert

Fanselow

1. Introduction In English, a so-called Superiority effect (Chomsky, 1973) arises when a wh-phrase moving to Spec,CP crosses a c-commanding wh-phrase. The presence of a structurally higher wh-phrase blocks the movement of lower wh-phrases (lb). Such crossing effects are, however, confined to wh-pronouns: subject which-phrases can be crossed by wh-objects, as shown in (2). (1) a. Who saw what? b.*What did who see? (2)

Which bird did which man see?

Recently, the status of (lb) as a violation of a grammatical constraint has been questioned (e.g., Hofmeister et al. (2007)). The idea is that crossing movement as in (lb) is just more difficult to process than its non-crossing counterpart (la), and that the perception of this processing difficulty is responsible for the (dramatically) reduced acceptability of (lb). Proponents of a grammatical account of the contrast in (1) need not deny a processing difficulty difference in ( 1 ) - they might rather claim that this processing difficulty difference has been grammaticalized in English. Some evidence for the correctness of a grammaticalization analysis of (1) comes from a comparison of English with Czech, German, and Russian. On the one hand, there is experimental evidence that genuine superiority effects are absent in Czech (Meyer, 2004), German (Fanselow et al., accepted) and Russian (Fedorenko & Gibson, submitted). The two sentences in German (3) do not differ in acceptability. Therefore, there is no constraint at work in German syntax that blocks crossing movement. (3) a. wer sah was? who saw what b. was sah wer? 0

The research presented here was supported by DFG grants FOR 375 (Project A3) and FA 255/6-1. I would like to thank Lyn Frazier, Stefan Frisch, Matthias Schlesewsky, Ralf Vogel, and Thomas Weskott for their help in various respects.

Semantic Type Effects on Crossing Movement in German

49

On the other hand, crossing movement nevertheless induces a loss of acceptability for certain combinations of wh-phrases in German, too, and it is likely that at least some of these structures come with a reduced acceptability because of a processing difficulty. E.g., effects of crossing movement on acceptability show up whenever the two wh-phrases involved are too similar to each other. Haider (1997) was the first to observe this. He pointed out that there is a contrast in acceptability between (4a) and (4b): an object of an embedded infinitival may be moved across a matrix wh-object, but only if the two whobjects differ in overt case morphology (4a). A lower dative crossing a higher dative (4b) considerably reduces acceptability (#). (4) a. wen hat er wem geraten über Malta t zu befragen? whoACC has he whoDAT advised about Malta to interview 'who did he advise to interview who about Malta' b.#wem hat er wem geraten t nach Malta zu folgen? whoDAT has he whoDAT advised to Malta to follow 'who did he advise to follow who to Malta' Likewise, when a subject crosses an object, the animacy of the two wh-phrases must differ. When an animate object wh-pronoun moves to the left of an animate subject whpronoun, as in (5b), the structure is normally judged as unacceptable. See Meyer (2004) for Czech, and Fanselow et al. (accepted) for German. (5) a. wer hat wen gesehen? whoNOM has whoACC seen 'who has seen who?' b.#wen hat wer gesehen? That processing difficulty contributes to this effect as well is made plausible by the following observation: An animacy effect also shows up when we consider the interaction of wh-phrases and quantifiers, as in (6). Fanselow et al. (accepted) found a (small) acceptability difference between (6a) and (6b), suggesting that there is an acceptability penalty for a wh-phrase crossing a quantifier with which it interacts in terms of scope and which has the same animacy status (6b) - compared to sentences in which the quantifier and the wh-phrase differ in animacy (6a). It is highly unlikely that (6b) could be ruled out by a grammatical constraint that only tolerates (6a). (6) a. was hat jeder gesehen? what has everybody seen 'what has everybody seen' b. wen hat jeder gesehen? whoACC has everybody seen 'who has everybody seen' However, the size of the effect in (6) is much smaller than the one in (5) (cf. Fanselow et al. (accepted)), which suggests that there must be an additional ban against the appear-

50

Gisbert Fanselow

ance of wh-pronouns in subject positions (see Arnon et al. (2005), Fedorenko & Gibson (submitted), Haider (2004)). The present paper reports the results of a number of acceptability rating experiments which were carried out in order to shed more light on the factors reducing acceptability of crossing movement in German. Section 1 will demonstrate that an animacy effect also shows up in the interaction of subject wh-pronouns with discourse linked which-phrases. Section 2 investigates "how many"-questions. This type of structure is interesting because English "how many"-questions pattern with "who-" rather than "which"-questions with respect to Superiority. Furthermore, a "how many"-phrase does not ask for persons/objects but for cardinalities. Semantically, they are thus maximally distinct from whpronouns such as wer 'who' or was 'what'. Formally, however, "how many"-questions are animate or inanimate, depending on the pertinent specification of the noun. We found no animacy related superiority effect in the interaction with wer in our experiments. This supports the view that the animacy effect in (3) and (5) is a consequence of a semantic rather than formal syntactic problem. Finally, in section 3, we compare crossing effects in scrambling and wh-movement contexts. Our experiments reveal that the reordering of wh-phrases with equal animacy specification reduces acceptability, but also that there is an additional, construction specific penalty for the crossing of a wh-phrase by wh-movement.

2. Wh-Pronouns and Discourse Linked Wh-Phrases As mentioned above, there is a constrast in acceptability between two types of multiple questions involving crossing movement: when the animate wh-pronoun wer 'who' in subject position is crossed by a further animate wh-pronoun (5b), this leads to a considerable decrease in acceptability that is absent when wer is crossed by the inanimate wh-pronoun was 'what' (3b). The reverse holds for the inanimate wh-pronoun was in subject position: without a loss in acceptability, it can be crossed by animate wh-phrases only (Fanselow et al., accepted). Crossing movement must involve wh-phrases differing in animacy when an in situ subject wh-pronoun is involved. What could be responsible for this difference? Note that the existence of an acceptability decrease in crossing movement structures in German is rather surprising: German is a V2 language in which, from a grammatical point of view, any category can move to the left periphery of a main clause, to the specifier position of CP. In particular, this implies that any category can, in principle, cross a higher constituent on its way to the left. Thus, non-wh objects are always able to move across subjects, as illustrated in (7). (7) Den Peter mag jeder. theACC Peter likes everyone 'Everyone likes Peter.'

Semantic Type Effects on Crossing Movement in German

51

This general freedom of movement is in line with the predictions of Chomsky (2005). Chomsky argues that movement to Spec,CP is not subject to any type of Minimal Link Condition' that would rule out crossing movement. The source for an effect of crossing movement on acceptability must thus be sought in the extrasyntactic domain. Syntactically unrestricted leftward movement is, e.g., subject to constraints related to information structure, either encoded directly in terms of mapping rules linking syntactic structures and discourse templates (as proposed, e.g., by Neeleman & van de Koot (2008)), or resulting from principles linking prosody and syntax (see Fanselow & Lenertovà (accepted)). Fanselow et al. (accepted) attempt an explanation of the contrast between (3b) and (5b) along these lines. Suppose that the choice of a more complex structure with accusative > nominative order is warranted only if it does not have the same interpretation as the simpler structure, as predicted by interface economy models such as Reinhart (1995). Then, the crossing constellation is acceptable only if the phrase moved to Spec,CP bears some function that it cannot have if it is left in situ. E.g., it could be topical in the sense of figuring as the "sorting key" (Comorovski, 1996) for the answers to the multiple question. The sorting key idea can be illustrated with (8). The information requests are shaped in different ways in these sentences. (8a) asks to provide, for each member χ of a contextually given set of girls, the CDs that χ bought. (8b) asks to provide, for each member a; of a contextually given set of CDs, the girls that bought x. The movement of the object (rather than the subject) in (8b) is thus licensed because it leads to a different sorting key. The two sentences in (8) are, therefore, equally acceptable. (8) a. which girl bought which CD? b. which CD did which girl buy? An object initial multiple question must thus come with a sorting key different from the one of its subject initial counterpart. Crossing movement as in (5b) is, therefore, problematic, since both subject and object are wh-pronouns quantifying over the whole set of (contextually accessible) humans. The sorting keys of (5a) and (5b) are therefore identical, so that crossing movement is not warranted. Matters are different in (3): (3a-b) come with different sorting keys: (3a) sorts the answers by persons while (3b) sorts them by objects or events. The choice of different word orders is correlated with different pragmatic properties, and is therefore compatible with interface economy. This reasoning predicts that there should be a negative effect of crossing movement even for pairs of which-phrases when the two phrases are identical in meaning as in (9), so that the choice of a different word order does not lead to a different sorting key. We have no experimental evidence on this issue. (9) a. welcher Student erkannte welchen Studenten? which N0M student recognized whichACC student b. welchen Studenten erkannte welcher Student? 'which student recognized which student' ' The Minimal Link Condition requires that a cannot move to c if there is a 6 that could also move to c and that is closer to c than a.

52

Gisbert Fanselow

Furthermore, (10b) should be as acceptable as (10a), because (10b) sorts the answers by professors, which (10a) does not do. We tested sentences such as (10), along with those in (11), in the second half of the first experiment reported in Fanselow et al. (accepted). The experiment is described there in more detail, so we will just sum up basic information here: 20 participants rated sentences such as (10) and (11) in a written 2 acceptability test (with 4 items/condition and 72 distractor sentences). (10) a. wer erkannte welchen Professori whoNoM recognized whichACC professor b. welchen Professor erkannte wer? whichACC professor recognized whoNOM 'who recognized which professor?' (11) a. welcher Professor erkannte wen? whichNoM professor recognized whoACC b. wen erkannte welcher Professor 'which professor recognized who?'

5.09 3.09

4.28 4.11

On a 7-point scale of acceptability, with 7 best and 1 worst, the subject initial version (10a) of the question with a subject wh-pronoun turned out to be more acceptable than its object initial counterpart (10b) (5.09 vs. 3.09), while the subject- and object initial versions of sentences with subject which-phrases (11) were equally acceptable (4.28 vs. 4.11). Crossing movement thus reduces acceptability when a wh-pronoun is crossed, even when the crossing element is a which-phrase. It is quite remarkable that the 'size' of the effect in (10) (5.09 vs. 3.09) is comparable to the one Fanselow et al. (accepted) found for (5) (5.39 vs. 2.44). A comparison of the "raw" means of acceptability between experiments is, of course, not really warranted, but note that the two sub-experiments at least used the same set of distractors, which should calibrate absolute acceptability ratings to a certain extent. From a crosslinguistic perspective, the findings concerning (10) partially do not come as a surprise. Arnon et al. (2005) and Fedorenko & Gibson (submitted) investigated the effect of the type of the wh-phrase (which-phrase vs. wh-pronoun) on the acceptability of multiple questions in English, and found that the type of the wh-phrase in Spec,CP is irrelevant for acceptability when the subject is a wh-pronoun. Crossing movement is excluded in this case. The type of the wh-object matters, however, for subject which-phrases: sentences are more acceptable when the object wh-phrase in Spec, CP is a which-phrase, too. The results for (10) therefore appear to imply that there is a separate ban on in situ wh-pronoun subjects. It is absolute in English, and relative in German, where it rules out crossing movement of wh-phrases with the same animacy value. The latter effect is not easy to capture in terms of the sorting key idea sketched above: the content of the 2

German wh-pronouns wer, wen, was. etc. can also be used as indefinite pronouns (unless they appear in Spec,CP), in which case they are deaccented. When appearing on a written questionnaire, ( 10b) is therefore ambiguous between a multiple question interpretation, and a single question interpretation "which professor was recognized by anybody?". We are convinced that we can nevertheless rely on the questionnaire method in our experiments, because the multiple question interpretation is the highly preferred one, cf. Fanselow et al. (2008) for experimental evidence.

Semantic Type Effects on Crossing Movement in German

53

wh-subject differs from the content of the wh-object in (10), so that different word orders should yield different sorting keys. The sorting key pragmatics of question-answer pairs thus does not suffice to account for the decreased acceptability of an in situ wh-subject, unless we assume that a rearrangement of potential sorting keys is not warranted when one sorting key is denotationally included in the other (but one may wonder why that should hold!). We still have to show that, just as in the case of (3), a difference in animacy between the wh-pronoun and the which-phrase overrides the ban against in situ wh-subjects. This is the result of a further acceptability rating study (EXP 1) with 23 participants who rated 16 experimental items (4 items/condition) and 98 filler items appearing on a written questionnaire. Experimental items could be subject- (12a,c) and object- (12b,d) initial, and either the subject (12a,b) or the object (12c,d) could be a wh-pronoun. In contrast to the element related to (10), the two wh-phrases differed in animacy. The results of the experiment as expressed on a 7-point scale are given behind the corresponding sentence type in (12). (12) a. Wer hat welchen Aufsatz fiir das heutige Treffen lesen sollen! whoNOM has whichACC paper for the today meeting read should? b. Welchen Aufsatz hat wer fiir das heutige Treffen lesen sollen ? 'Who was supposed to read which paper for today's meeting?' c. Welcher Mitarbeiter hat was für das heutige Treffen lesen sollen? which assistant has what for the today meeting read should d. Was hat welcher Mitarbeiter für das heutige Treffen lesen sollen? 'Which assistant was supposed to read what for today's meeting?'

5.09 4.65 5.03 5.14

Sentences of the type (12b) with an inanimate object crossing a wh-pronoun subject were only slightly worse (5.09 vs. 4.65) than their non-crossing counterparts (12a). This difference was only marginally significant, and in the subject analysis only ( F 1 ( l , 22) = 3.79, ρ = .06). No other effects were significant. The results of EXP1 are in line with our expectations. There is no pronounced difference in acceptability between crossing and non-crossing movement when a wh-pronoun is crossed over by a which-phrase with a differing animacy value. The small, very marginal and not significant effect of word order may just reflect the greater processing ease of subject initial sentences in German, if it is interpretable at all.3 It should be added that the results in EXP1 are not in line with the findings of Featherston (2005). Featherston (2005) has found a superiority effect for wh-pronoun/ whichphrase pairs of the kind used in EXP1. This difference between the two experiments is, apparently, due to the fact that our experiment used transitive verbs, while Featherston's experiments were constructed with ditransitive verbs, cf. again Fanselow et al. (accepted) for supporting evidence. We still have no clear answer to the question of why 3

If an experiment is considered in isolation, the absence of significant effect is of course difficult to interpret. Nevertheless, we feel justified in drawing conclusions from such null results here, because they appear in a large series of experiments (Fanselow et al., 2008, Fanselow et al., accepted) that have been constructed in identical or very similar fashion, and because they show up under the same conditions again and again.

54

Gisbert Fanselow

this difference in the material should have such a systematic effect on the outcome of the experiment, but a few tentative suggestions can be found in Fanselow et al.

3. "How many"-Questions Complex wh-phrases formed with "how many" rather than "which" may shed a further interesting light on the nature of the difficulties with crossing movement in German. Formally, "how many"-questions inherit the gender specification from their nouns, and it is therefore tempting to assume that the animacy distinction inherent to their nouns also percolates up to the "how many"-phrase, just as it does for which-phrases. On the other hand, "how many"-questions do not ask for individuals, rather, they ask for numbers, quite a different sort of objects that is neither animate nor inanimate. If the ban on wh-pronouns sitting in situ in the subject position is semantic in nature, and if our semantic characterization of "how many"-questions is correct, the crossing of wer "who" by a "how many"-phrase should not imply a loss in acceptability. If, however, "how-many"-phrases can be animate, and if the constraint on the crossing of wer is syntactically driven, "how many"-phrases should behave like which-questions in terms of the licensing of crossing movement. Our second experiment (EXP2) was designed to answer such questions. All experimental items began with a "how many"-phrase, that could be the object (13a,b) or the subject (13c,d). The second wh-phrase was a wh-pronoun. The "how many"-phrase either contained an animate (13b,d) or an inanimate (13a,c) noun. There were thus four experimental conditions. Each of the 48 participants saw 4 items per condition, in addition to 68 unrelated distractor items and 16 related distractors which differed from the 16 experimental items in terms of an agreement violation (verbs appearing in singular rather than plural form, and vice versa, cf. (13e)). (13) a. wie viele Waffen vernichtet jeweils wer? how many weapons destroys in each case w h o N O M 'who destroys how many weapons?' b. wie viele Feinde vernichtet jeweils wer? how many enemies destroys in each case w h o N O M 'who destroys how many enemies?' c. wie viele Waffen vernichten jeweils wen? how many weapons destroy in each case whoACC 'how many weapons destroy who?' d. wie viele Feinde vernichten jeweils wen? how many enemies destroy in each case whoACC 'how many enemies destroy who?' e. *wie viele Waffen vernichten jeweils wer?

4.08

4.18

4.33

4.44

The means of acceptability for each condition appearing after the pertinent example in (13) already indicates that no contrasts involving animacy can be observed with "howmany"-questions. Mean acceptability for subject initial questions was at 4.38, and at 4.18

Semantic Type Effects on Crossing Movement in German

55

for object initial sentences. The difference was not significant (Fi (1,47) = 1.80, ρ > .10, F 2 ( l , 15) = 2.01,ρ > .10). There was no main effect of animacy (Fi < 1, F 2 < 1) and there was no interaction between animacy and the grammatical function of the how many phrase ( F j < 1 ,F2 < 1). Subject wh-pronouns can be crossed by "how many"-phrases without a loss in acceptability. It plays no role whether the noun embedded in the crossing wh-phrase is animate or inanimate. The critical difference to which-phrases is that which-phrases ask for individuals, while "how many"-phrases do not. This suggests that the crossing constraint involves a semantic distinction: a crossing affect arises only if the wh-phrase quantifies over the same type of entities as the in situ wh-pronoun does. The following three experiments were meant to test whether a crossing effect arises when the in situ subject is a "how many"-phrase. In English, we observe a penalty for crossing movement whenever the in situ wh-subject is not discourse-linked. "How many"phrases show a superiority effect similar to who (see, e.g., Hornstein & Weinberg (1990)), arguably because they are not discourse-linked. This allows the expectation that in situ "how many"-subjects should behave like in situ wer in German. EXP 3 had the four conditions exemplified in (14): the subject was an animate "how many"-phrase, and the object was either wen 'who A C C ' (14a,b) or was 'what' (14c,d). In these multiple questions, either the subject (14a,c) or the object (14b,d) was fronted. 32 participants rated the 16 experimental items (4 per condition), in addition to 76 distractor items. Given the results of EXP2, we did not expect to find a difference in acceptability as a function of the animacy of the wh-pronoun, and this expectation was borne out: (14)

a. Wieviel Abgeordnete haben in der Fraktionssitzung how many representatives have in the club meeting

wen kritisiert? who A cc criticized 4.73 b. Wen haben in der Fraktionssitzung wie viel Abgeordnete kritisiert? 4.36 'how many representatives criticized who in the club meeting?' c. Wie viel Abgeordnete haben in der Sitzung was kritisiert? 5.37 how many representatives have in the club meeting what criticized? d. Was haben in der Sitzung wie viel Abgeordnete kritisiert? 4.66 'how many MPs criticized what in the club meeting?'

An advantage of subject-initial questions was, however, visible both for wen 'who' (4.73 vs. 4.36) and for was 'what' (5.37 vs. 4.66) objects in EXP3. A main effect of the animacy of the object was significant ( F i ( l , 31) = 14.872, ρ < .001, F 2 ( 1,15) = 7.5167, ρ < .05), just as the main effect of word order ( F i ( l , 3 1 ) = 6.8408, ρ < .05, F 2 ( l , 15) = 21.517, ρ < .001), but the two factors did not interact ( F i ( l , 3 1 ) = 2.4664, ρ > 0.1, F 2 ( 1 , 1 5 ) = 1.0223, ρ > 0.3). The main effect of animacy of the object is difficult to interpret. Probably, it either reflects a general preference for canonical transitive constellations with animate subjects and inanimate objects (but one wonders why this preference fails to have an effect in most other experiments in our series), or an unintended bias in the choice of verbs (overrepresentation of those preferring an inanimate object) in the experimental material.

56

Gisbert Fanselow

As in the preceding experiments, we observe an unspecific and small advantage for subject initial questions that we may attribute to the greater processing ease of sentences with canonical word order (cf. Krems (1984) and subsequent work). Crucially, the size of the effect is much smaller than the crossing effect for in situ wer 'who' in (5b) and (10b). The absence of an interaction between animacy and word order suggests that the acceptability of wh-movement crossing in situ "how many"-subjects constitutes a further difference between English and German: in situ "how many"-phrases pattern with whichphrases rather than wh-pronouns in the domain of triggering crossing effects. This conclusion can be drawn, however, only if there is no reduction of acceptability when one "how many"-phrase crosses the other. EXP2 has shown that the semantic type of "how many"-phrases is different from both animate and inanimate concrete entities, and if crossing movement is forbidden just in case the two wh-phrases belong to the same semantic category, then only "how many"-phrases themselves should be able to affect the extraction of "how many"-phrases in a negative way. With E X P 4 and E X P 5, we wanted to find out whether the crossing of one "how many"-phrase by another decreases acceptability. Both experiments involved multiple questions with two "how many" phrases and had two conditions each: the embedded multiple questions could be subject- (15-16a) or object- (15-16b) initial. Both experiments were embedded among 80 distractor items, and they were carried out with 32 (EXP4) and 36 (EXP5) participants, respectively. In both experiments, the subject was constructed with an animate noun. In EXP4, the object wh-phrase involved an inanimate noun, in EXP5, it was built up with an animate noun. ( 15) Zähle mir bitte auf : 'Please enumerate' : a. wie viele Studenten jeweils wie viele Leistungspunkte erworben How many students in each case how many credit points acquired haben. have b. wie viele Leistungspunkte jeweils wie viele Studenten erworben haben. 'how many BA students got how many credits points in each case' (16) In der Liste ist genau aufgeführt

6.25 5.23

'In the list, it is specified exactly'

a. wie viele Anwälte jeweils wie vielen korrupten Politikern helfen How many lawyers in each case how many corrupt politicians help mussten. mustpAST 5.51 'how many lawyers had to help how many corrupt politicians in each case' b. wie vielen korrupten Politikern jeweils wie viele Anwälte helfen mussten 5.33 The results of EXP4 involving pairs of one animate and one inanimate "how many"phrase resemble those of EXP3: We see a small but significant advantage of non-crossing sentences ( F i ( l , 3 1 ) = 24.351, ρ < .001, F2( 1,7) = 22.894, ρ < .01). The presence

Semantic Type Effects on Crossing Movement in German

57

of this effect is not surprising: object initiality is often penalized because it leads to a greater processing difficulty. However, the effect is much smaller than the decrease in acceptability that arises when a wh-pronoun is crossed in a subject position. This lends further support to the view that German "how many"-phrases do not pattern with whpronouns with respect to a superiority-like effect confined to wh-phrases of the same semantic type. The results of EXP5 are in line with this claim. The experimental items in EXP5 involved pairs of wh-phrases that agreed maximally: both wh-phrases were animate "how many"-phrases. The means of the two conditions of EXP5 did not differ from each other ( F j ( l , 3 5 ) = 2.6232, ρ > .1, F2 < 1), there was not even a small advantage of subject initial sentences. The experiments involving "how many"-phrases have revealed a number of interesting points. First, the sharp decrease in acceptability due to crossing wh-movement turns out to be confined to constellations in which the in situ phrase is a wh-pronoun. The discourselinking property plays no role: both discourse linked which-phrases and "how many"phrases can be crossed over easily. Second, the effect is confined to those constructions in which the crossing phrase asks for the same type of individual as the wh-pronoun. In this respect, wh-pronouns and which-phrases show the same behavior, and contrast with "how-many" phrases. In some experiments, a weak advantage of subject initial sentences could be detected that we attributed to the greater processing ease of sentences with canonical order. What determines whether such an effect shows up in an experiment or not is unclear to us.

4. Scrambling German wh-movement differs from English wh-movement not only in terms of the absence of a general Superiority effect. As was noted first by Haider (1981), one also finds apparent violations of the Weak Crossover Condition in German, and there are no thattrace-effects. Haider (1981) argued that these different patterns are due to a major parametrical contrast, viz. configurationality. While there are various ways in which free constituent order in German can be formally described, the flexibility of word order has far-reaching consequences for Superiority effects that are relatively independent of the model chosen. If the object can appear in an A-position in a sentence with OSV order, then the wh-movement of the object can in fact originate in the pre-subject position. Therefore, the derivation of an object initial multiple question need not involve any crossing wh-movement at all. The crucial A-barmovement step of was in the derivation (17) is the one linking t* and was, because the object crosses the subject as an instance of scrambling (linking t and t*), or because OSV order is already base-generated (17) a. vermutlich wer was gesagt hat"? Presumably who what said has b. vermutlich wasi wer ti gesagt hat

Gisbert Fanselow

58 c. hatj vermutlich wasi wer ti gesagt tj d. waSi hatj vermutlich ti* wer tt gesagt tj

If scrambling in German interacts with Α-movement properties (cf. Fanselow (2001), Frey (2004), Haider & Rosengren (2003), Neeleman & van de Koot (2008)), then we should see no genuine superiority effects for clausemates at all. Or, more precisely, the acceptability of a multiple question with inverted word order should mainly reflect properties of A-scrambling. The slight decrease in acceptability that we observed in crossing multiple questions in some of the experiments might find a straightforward explanation in this model: it could be due to the decrease in acceptability that scrambling often triggers. Furthermore, suppose that wen 'who ACC ' cannot be scrambled across wer 'who NOM ' without a significant drop in acceptability. The strong crossing effect we have observed in multiple questions formed with these two wh-pronouns would then reflect the interaction of the inacceptability of the scrambling step for wen preceding movement to Spec,CP, and the violation of whatever principle is responsible for superiority in those derivational attempts in which wen is wh-moved directly to Spec,CP across wer. We therefore need to investigate whether the crossing penalty for pairs of wh-pronouns with equal animacy status is due to a constraint on scrambling. This idea can be tested by comparing multiple questions with corresponding sentences involving scrambling. One needs to keep in mind, however, that 'normal' scrambling in German affects definite, thematic, given material rather than indefinite phrases that are in focus (because they are wh-words), so that one need not necessarily expect that scrambled structures and crossing multiple questions behave alike in all respect. EXP6 comparing scrambled sentences with multiple questions with inverted order of subject and object was constructed along the following lines. The overall structure of the experimental items was kept constant: it involved a matrix clause in which an indirect question was embedded. In the multiple question version, one wh-phrase was moved to the left periphery of the embedded clause, while the other remained in situ. The moved wh-phrase could either be a subject or an object wh-pronoun ( 18a,b). The scrambling version exploited the fact that German wh-pronouns come with an indefinite pronoun reading, too. The two occurences of wer and wen were embedded in an if-question ( 18c,d). This rules out the wh-interpretation of wer and wen. This version also appeared with subject initial order and with a scrambling-generated object initial order. (18) es ist egal

'it does not matter'

a. wer damals wen ständig angerufen hatte whoNOM then whoACC permanently called had b. wen damals wer ständig angerufen hatte 'who had permanently called whom then' c. ob damals wer wen ständig angerufen hatte whether then whoNOM whoACC permantently called had d. ob damals wen wer ständig angerufen hatte 'whether anybody had permanently called anybody else then'

6.16 3.70 3.91 2.97

Semantic Type Effects on Crossing Movement in German

59

32 participants rated 16 experimental items (4 per condition) plus 88 distractor items (among which there were 16 simple questions potentially involving a weak crossover violation). There was a main effect of the sentence type versions ( F i ( l , 3 1 ) = 61.27; ρ < .001; F 2 ( l , 15) = 69.43, ρ < .001), with multiple questions being more acceptable than sentences with an indefinite interpretation of the wh-pronouns in an indirect yes/no question (5.04 vs. 3.34). There was also a main effect of word order, with non-crossing sentences being better than crossing ones ( F i ( l , 3 1 ) = 60.91; ρ < .001; F2(l, 15) = 50.28, ρ < .001). Subject initial sentences were more acceptable than object initial ones (4.93 vs. 3.44). Furthermore, there was a significant interaction between the two factors of the experiment (Fi (1,31) = 27.65; ρ < .001; F 2 ( l , 1 5 ) = 16.09, ρ = .001). Indirect yes/no questions are worse than multiple questions. Perhaps, this difference is due to the appearance of two wh-pronouns with an indefinite interpretation in the former class of sentences - recall that the indefinite interpretation is the dispreferred one. Nothing hinges on the precise interpretation of this main effect. Furthermore, EXP6 replicates the findings of earlier experiments in that the huge acceptability difference between subject and object initial multiple questions with two animate wh-pronouns appeared again in a very clear way (6.2 vs. 3.7). Subject and object initial sentences differed significantly (¿i(l,31) = 3.35, ρ < .01; ¿2(1,15) = 3.35, ρ < .01) as well in the case of sentences with the same wh-pronouns but interpreted as indefinites (3.9 vs. 3.0). There was a significant ( F ( l , 31) = 5.68, ρ < .05) correlation between the word order effects in the multiple question and the indefinites condition ( R = .399,R2 = 16), which corroborates the view that the "soft" superiority-like effect observed in multiple question is partially related to the factor that influences the acceptability of scrambled structures. This correlation is not simply due to some participant-specific degree of "liberality" that would correlate all judgments of a participant. E.g., the word order effect in multiple questions turned out to not correlate with the similar factor (subject- or object question) in a weak crossover experiment appearing on the same questionnaire. However, the significant interaction between the word order factor and the sentence type/wh-interpretation factor also shows that the difference between object- and subjectinitial multiple questions with two animate wh-pronouns cannot exclusively be accounted for by the "scrambling factor". 4 There is something additional about the wh-pronoun wer crossed over in a wh-question by wen that makes the resulting object initial question even less acceptable than sentences with wen scrambled across wer in comparison to their subject initial counterpart. The strong crossing effect of multiple questions can therefore not be reduced to the constraint triggering the weak overall object initiality effect. It is due to an independent factor. The idea suggests itself to apply the design of EXP6 to wer-was-pairs of wh-pronouns, too. The correlation of the word order effect for the question- and the indefinite interpretation of the wh-pronouns should still hold (the weaker superiority-like effect is ubiquitous, because all crossing sentences are object-initial), but we should see no interaction between the interpretation type factor (question vs. indefinite) and the word order factor: the ad4

In principle, this interaction could also be due to a simple floor effect, but note that the acceptability value of ( 18d) is still comparatively high 1(2.91 on a 7-point scale) which makes a floor effect less likely.

60

Gisbert Fanselow

ditional ban against crossing wer in a multiple question applies in the case of animate wh-objects only. The makeup of the conditions of EXP7 illustrated in (19) differs only minimally from the one used in EXP6. 36 participants rated the 16 experimental items (4 per condition) and 72 distractors. (19) Ich frage mich schon lange "I have wondered for quite a while" a. wer damals was gesagt hat who then what said has b. was damals wer gesagt hat. 'who said what then' c. ob damals wer was gesagt hat. whether then who what said has d. ob damals was wer gesagt hat. Ί have been wondering for quite some time if anyone said anything then'

6.13 4.98 3.58 2.76

EXP 7 confirms all expectations. First, as in EXP6, sentences in which the wh-pronouns had to be interpreted as indefinites were less acceptable than the multiple questions (Fi(1,35) = 119.84, ρ < .001, F 2 ( l , 1 5 ) = 232.23, ρ < .001). Furthermore, the word order manipulation yielded a significant effect: subject initial sentences were better (6.13 for multiple questions and 3.58 for indefinites) than object initial sentences (4.98 and 2.76, respectively) ( F j ( l , 3 5 ) = 51.536, ρ < .001, F 2 ( l , 15) = 37.304, ρ < .001). Judgments for the two object initial conditions of the experiment (R 2 = 0.163, ρ < .05) were correlated, just as the judgments for the subject initial sentences (R 2 = 0.158, ρ < .05) (but correlations within interpretation types were higher: R2 = 0.547, ρ < .001 in the case of multiple questions, and R2 = 0.436, ρ < .001 for indefinites). The subject advantage in multiple wer-was-questions thus seems to be driven by the factor that is also responsible for the similar effect in scrambled structures. The two factors of the experiment did not interact ( F i ( l , 3 5 ) = 2.04, ρ > 0.1, F 2 (1, 15) = 1.5297, ρ > 0.2). Unlike what we found for wer-wen pairs, the acceptability of object-initial multiple questions formed with was is not additionally reduced by a further factor in the case of inanimate objects combined with an animate subject. This is in line with our expectations: the additional factor is operative only if the phrase crossing the animate wh-pronoun is animate itself. The composition of the questionnaire on which EXP7 appeared allows us to compute a set of further correlations. A different experiment on the questionnaire compared the acceptability of sentences with scrambled and topicalized objects with the judgments for their subject initial counterparts. Both arguments were definite noun phrases in this experiment, in which topicalization and scrambling indeed turned out to reduce acceptability. We compared the participants' behaviour in the superiority and the scrambling/topicalization experiment, and found that the acceptance of object-initial multiple questions is correlated with definite object scrambling (R2 = 0.223, ρ < .01), and less so with topicalization (R2 = 0.132, ρ < .05). The experiments reported here thus lend support to two observations. First, the weak decline in acceptability that we sometime observe in crossing multiple questions is cor-

Semantic Type Effects on Crossing Movement in German

61

related with the general decline in acceptability that we find in all kinds of structures (scrambling, topicalization) that begin with an object. Independent of what it is triggered by, it is not confined to multiple questions, and thus does not have to be explained in a theory of wh-movement. There is only one effect of crossing movement in multiple questions in a strict sense: the strong penalty for crossing wer and was by a wh-phrase that also ranges of animate / inanimate entities, respectively.

5. Concluding Remarks Often, it is not easy to disentangle the effects of processing difficulty and the violation of grammatical restrictions in the explanation of some pattern of acceptability. Processing difficulty may be the source of a grammaticalization process, and ungrammaticality will often lead to a processing difficulty at least because of attempts of the human parser to find a legal analysis for the ungrammatical string of words. Crosslinguistic comparisons turn out to be very helpful in this context. All other things being equal, we expect processing difficulty not to vary dramatically between languages. Longer movement paths are more difficult to parse than shorter movement paths, and effects of the length of a movement path on acceptability should be visible in all languages. Grammaticalization must be postulated whenever acceptability is good in one language because of the absence of a processing difficulty, but low in another: in this situation, it is not likely that there is a processing account for unacceptability in the latter language. In the area of superiority, we seem to be confronted with such a situation: German shows that the constellation in (20) can be hard to process. It has been known for long that the wh-subject must be a wh-pronoun for there to be a decrease in acceptability. Meyer (2004) and Fanselow et al. (accepted) added the observation that acceptability drops only if the two wh-phrases have an equal animacy specification. The present paper has clarified some of the details of this condition for a superiority effect, and it has shown that ZP must be CP for there to be an effect. A related effect arises when the subject position is filled by a universal quantifer taking scope over the wh-object: (20)

[zp wh-object... [χρ wh-subject...

We do not really know what the source of this peculiar effect is. But when the conditions just mentioned are not met, multiple questions with inverted subject-object order are acceptable in German, while they are not in English. Crossing movement is inacceptable in the former language even when, as suggested from German, there is no processing difficulty linked to crossing movement. In English, there is thus a grammatical ban against crossing movement. It is not only the case that we do not really know what makes the constellation (20) hard to parse, it is also difficult to relate the Superiority effect - even in the general form that we find in English - to a plausible general constraint of syntax. After all, (8b) shows that crossing wh-movement is not blocked in general, and attempts to reconcile (8b) with a general Minimal Link Condition very often become very complex, and the necessary assumptions are often not motivated independently. Therefore, the complexity

62

Gisbert Fanselow

of a grammar that explicitly rules out (21) (cf. Haider (2004) for a related proposal) rather than trying to derive its effect from an MLC with a complex system of exceptions may thus not be greater than that of the alternative "principle-based" model struggling with the fact that more structures are ruled out than necessary: (21) *[TP [DP +wh, +pron]

....

References Arnon, I., B. Estigarribia, P. Hofmeister, T. F. Jaeger, J. Pettibone, I. Sag & N. Snider (2005). Rethinking Superiority Effects: A Processing Model. Poster presented at the CUNY Sentence Processing Conference, University of Arizona. Chomsky, N. (1973). Conditions on Transformations. In: S. R. Anderson & P. Kiparsky (eds.) A Festschrift for Morris Halle. New York: Holt, Rinehart and Winston. 232-286. Chomsky, N. (2005). On Phases. Ms. MIT. Comorovski, I. (1996). Interrogative Phrases and the Syntax-Semantics Interface. Dordrecht: Kluwer Academic Publishers. Fanselow, G., & C. Féry (2008). Missing Superiority Effects: Long Movement in German (and other languages). In: J. Witkós & G. Fanselow (eds.) Elements of Slavic and Germanic Grammars: A Comparative View. Frankfurt: Lang. 67-87. Fanselow, G. (2001). Features, Theta-Roles, and Free Constituent Order. Linguistic Inquiry 32. 4 0 5 ^ 3 7 . Fanselow, G. & Lenertovà (accepted). Left Peripheral Focus: Mismatches between Syntax and Information Structure. Natural Language and Linguistic Theory Accepted. Fanselow, G., M. Schlesewsky & R. Vogel (accepted). Animacy Effects on Crossing Wh-Movement in German. Linguistics Accepted. Featherston, S. (2005). Universals and Grammaticality. Linguistics 43. 667-711. Fedorenko, E. & E. Gibson (submitted). Syntactic Parallelism as an Account of Superiority Effects: Empirical Investigations in English and Russian. Submitted. Frey, W. (2004). A Medial Topic Position for German. Linguistische Berichte 198. 154190. Haider, H. (1981). Empty Categories. On some differences between English and German. Wiener Linguistische Gazette 25. 13-36. Haider, H. (1997). Economy in Syntax is Projective Economy. In: C. Wilder, H.-M. Gärtner & M. Bierwisch (eds.) The Role of Economy Principles in Linguistic Theory. Berlin: Akademie Verlag. 205 - 226. Haider, H. (2004). The superiority conspiracy. In: A. Stepanov, G. Fanselow & R. Vogel (eds.) The Minimal Link Condition. Berlin: Mouton de Gruyter. 147-175. Haider, H. & I. Rosengren (2003). Scrambling: nontriggered chain formation in OV languages. Journal of Germanic Linguistics 15. 203-267. Hofmeister, P., T. F. Jaeger, I. Sag, I. Arnon & N. Snider (2007). Locality and Accessibility in Wh-Questions. In: S. Featherston & W. Sternefeld (eds.) Roots: Linguistics in Search of its Evidential Base. Berlin: de Gruyter.

Semantic Type Effects on Crossing Movement in German

63

Hornstein, Ν. & A. Weinberg ( 1990). The Necessity of LF. Linguistic Review 7.2. Krems, J. (1984). Erwartungsgeleitete Sprachverarbeitung. Frankfurt: Lang. Meyer, R. (2004). Syntax der Ergänzungsfrage. München: Sagner. Neeleman, A. & H. van de Koot (2008). Dutch Scrambling and the Nature of Discourse Templates. The Journal of Comparative Germanic Linguistics 11. 137-189. Reinhart, T. (1995). Interface Strategies. (Working papers in Linguistics). Utrecht: OTS.

Me and Chomsky Remarks from someone who quit Sascha W. Felix

Did you ever ask yourself why you became a linguist? Most probably, because you are somehow interested in language(s), but obviously that is far too trivial an answer. Chances are that you have been fascinated by languages ever since you entered high school. More or less by coincidence or possibly because the curriculum requested it you opted to learn a foreign language and you somehow found out that that is something you really liked. You realized that things can expressed equally well in languages other than English (something that your grandma never really believed) and you also became aware of the fact that saying things in different languages may also entail some amount of mental reorganization required by the specific structure of the language used. You found out that different languages do different things in different ways and yet you also discovered that no matter how different two languages may appear to be, there are a lot of common properties and regularities which seem to characterize any human language. Motivated by all these insights you decided that learning one foreign language just isn't enough, so in the following years you continued to look at all kinds of different languages. Some you learned to speak with more or less success, others you just studied for their specific properties. Anyway, while others got a kick out of collecting stamps, watching birds, learning to play piano, or whatever, you spent much of your spare time reading foreign books, or watching foreign movies, thus learning something about other cultures and other peoples' way of thinking and acting. When you entered college, you felt that linguistics is the field to go for, because it promised to give you the opportunity of continuously looking at interesting and exciting data from all kinds of languages. Since this was exactly what you enjoyed doing, the field of linguistics was obviously the right choice. As you advanced, you hit upon generative grammar and you felt that this framework offered you something meaningful that carried you beyond the point of merely collecting all kinds of language data and observations. It seemed that generative grammar provided you with something deeply insightful about the way languages are organized in general. Nevertheless, while at the theoretical level doing work in generative grammar proved to be a real intellectual challenge, this framework also gave you the opportunity of doing what you really liked, namely playing around with all kinds of linguistic data, analyzing their structure, sorting out regularities, searching for differences and commonalities. After all, linguistics - so you were told - was an empirical science, so the basis of whatever you

Me and Chomsky

65

did in terms of generative analyses appeared to be language data. So, again, this was obviously the right way for you to go. After a while, what you offered and published in terms of generative analyses of language data did not seem to be too unsuccessful; so as a consequence, one day you were offered a job at your university's linguistics department. The next morning you woke up realizing that you had become a professional linguist just like Chomsky and many others whose work you had previously read and studied. Well, maybe this is not exactly your life story, but it comes pretty close to being mine which, I suspect, I more or less share with innumerable others. So now both you and Chomsky are linguists, but what is the difference between you two? Trivially, there are a whole bunch of differences, but two of them may be particularly worth mentioning. First of all, Chomsky is a lot smarter than you are. In fact, he is so dazzling smart that some people in his vicinity could not bear the weight of his personality and argumentation any longer and thus decided to quit and go for their own ways. As you soon found out, joining the generative enterprise meant at the same time that you acknowledged Chomsky as the leading figure of the field and the undisputed authority on generative truths. He was the guy who said what was right or wrong and in not infrequent cases seriously challenging him was bound to lead to disaster, both in terms of your academic career and thus your family's food supply. Well, while all this is essentially an aspect of academic sociology and the way academia works, in itself it is not particularly enlightening. The crucial difference between you and Chomsky is something else. In contrast to you, Chomsky has never been interested in languages per se. He never learned any foreign language apart from Hebrew which was more or less a patrimony of his Jewish background and he never published anything in a language other than English. The fact that he started out analyzing language data was just a coincidence due to the fact that his dad happened to work in the field of linguistics and he was thus familiarized with the technical methods of analyzing language data. What Chomsky was actually interested in was cognitive psychology, that is, he wanted to find out something about the fundamental properties of the human mind conceived of as a blueprint for any kind of human behaviour. In other words, he wanted to know what distinguishes humans from other organisms and what it is that makes them human. Due to his educational background he believed that language and language data may provide a reasonable empirical basis to investigate the types of problems he was interested in. In the early days of transformational grammar Chomsky developed a system of formal machinery for the analysis of individual languages. While he applied this system almost exclusively to the description of English - the only language he really knew well - , it gave you the opportunity to extend the application to your mother tongue and any other language you were familiar with. So again, you could do what you really liked; namely play around with language data, writing huge chunks of rule systems, investigating their mutually dependencies, order of application and formal properties. The fact that Chomsky was really interested in something else, regarding linguistic analyses on a par with a chemist shaking his test tubes was not of too much concern to you. He did what he liked, namely searching for the structure of the human mind, and

66

Sascha W. Felix

you did what you liked, namely collecting and analyzing language data. This discrepancy in focus was something you just noticed in passing because it really did not do any harm to you and your personal interests. At the same time you were both fascinated and impressed by the depth and impact of Chomsky's philosophical, psychological, (and possibly also political) remarks. Not that these exercised any immediate influence on your daily linguistic work which remained largely as before, but the way Chomsky writings transcended the linguistic world and were received with great enthusiasm in a large domain of neighbouring fields was something that made you feel good and important, because it gave you the impression of being part of a deeply meaningful enterprise which might revolutionize the way in which not only linguistic facts, but also their impact on other domains were being viewed. This changed in detail, though not in principle with the advent of government-andbinding theory, or, if you prefer, principle-and-parameters theory. The perspective on linguistic data became more restrictive. While in the old days of transformational grammar you could legitimately look at any kind of language data that happened to be there, the new GB framework forced you to be more selective. Only those linguistic data were relevant that told you something about the universal properties of human languages in general. Any analysis you offered for a given set of linguistic data in a given language had to be reduced to a set of universal principles and had to be checked against a large set of data and analyses that had been worked out by others for a possibly completely different set of data from a large variety of different languages. Still worse, your analysis had to be compatible with insights from language acquisition, language processing, language loss, diachronic linguistics, and the like. In some sense, this more restrictive view placed and additional burden on, and provided an additional challenge to the way in which you proceeded in your daily work. But this made things also more interesting and rewarding. You had to look at more data, consider more facts, and provide more sophisticated solutions than in the old transformational days. But, crucially, it did not change anything with respect to the way in which you proceeded in principle. The empirical basis of all your work was still specific data from specific languages, be it English, Chinese, Warlpiri, Icelandic, or whatever. You could continue doing what you really liked, namely play around with language data, though admittedly on a higher level of abstraction and sophistication. For Chomsky the transition from transformational grammar to government-and-binding was, in some sense, the most natural move from what he called descriptive to explanatory adequacy. If, in principle, any natural language can be learned by any human being and if it is true that linguistic properties and regularities are fairly abstract, then it is clear that this human language learning ability must be based on some set of cognitive structures which in some way reflect universal properties of human language. That is, humans in contrast to other organisms must be endowed with some kind of innate knowledge about what is a possible human language. If you are not so much interested in playing around with language data, regularities and properties, but rather focus your concern on the structure of the human mind in some specific domain, then it becomes fairly obvious that the search for universal properties

Me and Chomsky

67

will be of primary, if no exclusive importance, because such universal properties can be hypothesized as reflecting innate cognitive structures of the human mind. But even if you are not that much interested in questions of cognitive psychology, the GB framework provided a powerful (perhaps too powerful) tool for the description and analysis of language data. At the same time it also formulated a particular intellectual change: describing some property X in some language Y is one thing, but finding universal properties and principles that hold for all languages - past and present - is quite another. You had to be, to some degree, both smart and ingenious to succeed in such as task. As is well known, the 80's were largely characterized by endeavours to determine such universal properties and regularities. As a consequence, a large number of fairly abstract principles were proposed from which certain properties of individual languages could be derived. The rationale behind these endeavours was as simple as it was compelling. If a reasonable set of such principles could be formulated, then these could explain why humans are able to learn any natural language under the well-known constraints of a stimulus-impoverished environment. What is more important in the present context is that both in transformational grammar and in government-and-binding - though these are miles apart in theoretical perspective - the empirical basis was essentially the same, namely specific language data. If you wanted to find out if a proposed transformational analysis of some English constructions was descriptively correct, you had to look at English data. Maybe you found out that for some additional constructions the original analysis did not hold. So you proposed a new analysis which could successfully describe both the original data as well as the new data which you yourself had discovered. In a similar vein, you might investigate whether or not the original formulation of subjacency did, in fact, hold universally. If your mother tongue happened to be Italian, you would obviously look at the corresponding Italian data. You might then discover that subjacency was either erroneous or somehow worked differently in Italian. To come up with the idea of parameters is, to be sure, an independent intellectual achievement which cannot be directly derived from specific Italian data. But that is fairly trivial, because it only shows that there are no discovery procedures which lead in a straightforward way from data to descriptions. Nevertheless, whatever problem arises, whatever question you try to answer, at least as a first step you look at some specific language data. If you wanted to find out whether languages are universally configurational, again you had to look at specific language data, this time properties of maybe Hungarian or Japanese. If it appeared that these languages appear to be non-configurational, but you still believed in the universality of configurational structure, then you had to come up with some clever idea of how to reconcile the apparently discrepant Hungarian and Japanese facts with the notion of configurationality. You might discover some "new" hitherto unnoticed Hungarian and Japanese constructions which seemed to prove the basic configurational character of these two languages leading you to propose analyses in which both the old and the new data could be accounted for. If you observed that some language overtly move wh-words and quantifiers while others do not, and at the same time display the same scope properties in either case, you

68

Sascha W. Felix

might come up with the notion of logical form showing that all languages move, though at different levels. Again, the empirical basis for such theoretical notions is constituted by specific linguistic facts and regularities. To give a final example, if you wanted to argue that the ECP can be reduced to subjacency or, conversely, that subjacency can be reduced to the ECP, then, again, you would do so by looking at specific data. In other words, both under a transformational analysis as well as within the GB framework, the starting point for any kind of analysis or proposal is always a specific set of language data or facts. If you, like Chomsky, are primarily interested in cognitive psychology, your specific perspective on the entire generative enterprise might be somewhat different from the one of someone who is just interested in language and language data. But, crucially, even if your primary perspectives and concerns may differ, both of you are happy with what the GB framework has to offer. If you enjoy linguistic analysis proper for the sake of looking at, and analyzing language data, you can happily proceed with your daily work. If your primary concern is with questions of cognitive psychology, you can equally well enjoy the machinery that GB has to offer. In other words, GB (and for that matter transformational analysis) makes everyone happy because it provides a powerful tool for the analysis of language data and at the same provides a framework for dealing with questions concerning the structure of the human mind. Which perspective you prefer is largely a matter of personal taste. All this changed radically, if not dramatically, when Chomsky introduced the minimalist program in the late 80's and early 90's. For Chomsky himself and those interested in the human mind the transition from GB to the minimalist program was again a most natural move from explanatory adequacy to problems and questions beyond explanatory adequacy. One of the major conceptual problems of the GB framework was the fact that, while universal principles constrained the possible properties of natural languages, the principles themselves were largely unconstrained. That is, any formulation of a principle was acceptable as long as it correctly captured the relevant facts. As a consequence, the number of principles proposed during the 80's successively increased in a dramatic way. It was therefore more than natural to go one step further and explore possibilities of constraining the principles themselves. As is well-known, Chomsky took an even more radical step by abandoning GB principles altogether, allowing at first only for principles that could be derived from mere "conceptual necessity". Ignoring for the moment questions concerning the distinction between rules and (representational) principles, GB essentially asked: why do natural languages have the properties they have? The general answer was: because they are constrained by a set of abstract principles which somehow reflect cognitive properties of the human mind. In contrast, the minimalist program asks: why are natural languages constrained by the principles (or rules) by which they are constrained. As far as I can tell, the answer is still pending, but it somehow seems to move around the notion of optimal design. More specifically, the minimalist program asks why natural languages have transformations, why they are constrained by islands, bounding restrictions, or why distinctions between logical form and phonetic form obtain. To put it differently, GB attempted to

Me and Chomsky

69

explain properties of language constructions and facts, while the minimalist program endeavours to explain principles (and rules) of language design. Crucially, this change in perspective as well as the ensuing move to a still higher level of abstraction has dramatic consequences for the community of linguists and their everyday work. To take just one example: the question of why natural languages have transformations cannot be answered by looking at language data. There isn't a single language fact, datum, or construction that could possibly provide an answer to this question. In a similar vein, the question of why natural languages display island constraints cannot be answered by analyzing whatever set of constructions in whatever language. And if you continue to ask whether or not some given language mechanism also obtains in cognitive domains other than language, then this question, again, cannot be answered by looking at language data as a matter of mere logic. Trivially, you have to look at other cognitive domains and their regularities. That is not to say, that the minimalist program is in any sense unempirical. But the empirical basis is not linguistic facts, but rather principles explaining linguistic facts. The minimalist program does, to be sure, provide technical machinery for the description and analysis of natural language(s). So, in principle, one could easily ignore the conceptual shift and proceed to simply reformulate known facts in a new terminological framework. Though, as far as I can tell, this has, in fact, been done, it certainly does not do justice to the spirit of the minimalist program. When transformational theory in the spirit of Aspects was replaced by GB in the early 80's, a legitimate question was how to handle, say, pronominalization under the new framework. Many who had been raised in the traditional transformational framework struggled for some time with familiarizing themselves with the new descriptive machinery. Though it took some time, most of us eventually succeeded. In contrast, the question of how to handle, say, German V2 within the minimalist framework is not only irrelevant, it does not even make sense. And this is simply because this framework is not concerned with specific language constructions, but rather with the principles governing language constructions. Take the basic notion of Merge, for example. As a matter of mere logic, you can only merge from outside or inside a given structure, there is nothing else. And this explains why languages have what they have, namely phrase structure and movement. The merit of Merge is not to replace X-bar and move a by some new descriptive device, but rather to explain why natural languages have both phrase structure (rules) and movement. As Chomsky has frequently emphasized, the existence of transformations had for a long time been regarded as a mysterious deficiency of natural languages. Under Merge this is no longer the case. And, again, the transition from X-bar/ move a to Merge is nothing you can arrive at by looking at linguistic data. The move from GB to the minimalist program is dramatic in another sense, namely that the generative enterprise lost part of its crew, namely all those whose primary concern is not so much with cognitive psychology, but rather with analyzing linguistic data. While GB made (almost) everybody happy, the minimalist program does not. The questions that minimalism asks are of a very general and abstract type and if you are simply interested in playing with language data, the minimalist program does not have too much to offer.

70

Sascha W. Felix

Personally, I find the consequences that followed from this transition quite intriguing, although, admittedly, these are much more about academic sociology than about linguistic science proper. First of all, there is, of course, the group of those who might be called the hard-core Chomskyans. They moved from GB to the minimalist program in much the same way they had previously moved from transformational grammar to GB. They take up the challenge posed by Chomsky himself and proceed to investigate the type of questions which constitute the core and spirit of the minimalist program. I suspect that at least some people in this group took the new framework as a mere change in descriptive machinery rephrasing their analyses with the new terminology not caring too much about what the minimalist program was all about. But obviously, this is mere speculation. Secondly there are those who largely ignore what has happened. They continue to offer analyses within the GB framework noting at best in a footnote that their analysis could be rendered in minimalist terminology in such and such a way. This is a somewhat curious fact, if we consider previous developments in generative grammar. The transition from transformational grammar to GB occurred virtually overnight. Obviously, this is a slight exaggeration, but nevertheless in the early 80's, shortly after Chomsky's Lectures had become generally available, almost everyone switched from transformational grammar to GB within a relatively short period of time. It is hard to imagine that anyone could have successfully placed a traditional transformational analysis, say, in 1983 in any of the high prestige journals. In contrast, even 20 years after the introduction of the minimalist program papers in the spirit of GB are far from rare. So in some sense GB analyses and minimalist analyses coexist side by side. To avoid being accused of being someone who is ignorant of recent developments, you might cautiously talk about subjacency effects or ECP phenomena pretending to view those as descriptive generalizations rather than as UG principles proper, but apart from these terminological tricks the basic spirit is still that of GB. Finally, there is the group of "traitors" and "heretics". They simply moved away from Chomskyan linguistics in the sense that they either proceeded to pursue their own interests largely uninfluenced by minimalism or they took refuge to some other framework, e.g. optimality theory, cognitive grammar, or the like which - they believed - served their zest for language data analyses better than minimalism ever could. In my personal estimation, I find this development to be highly unfortunate and regrettable. Obviously, those who had always been appalled by (the success of) generative grammar rejoice and simply conclude that generative grammar is dead. This, however, is nothing but wishful thinking, because many who work in a non-minimalist framework would still consider themselves as generative grammarians in much the same way as those pursuing lexical-functional grammar or generalized phrase structure grammar in the late 70's and early 80's considered their work as generative grammar even though they heavily opposed Chomsky's views at that time. Of course, one may be inclined to welcome the increasingly broader variety of perspectives and approaches that has developed in the post-GB period. One may also judge this development as enriching the field which for many decades had been under the pressure

Me and Chomsky

71

of Chomsky's judgments and verdicts. And I also know of many who welcome the return to what they call a more descriptive style in current linguistic work. My personal view, for what it is worth, is rather the opposite. It seems to me that the field of generative grammar has lost much of its homogeneity and uniformity which, I believe, was one of its crucial strengths eventually responsible for its overwhelming success both inside and outside the world of language. As far as I can see, the field is continuously and unsystematically dispersing into all kinds of directions and outside the group of hard-core Chomskyans (see above) it is frequently unclear what the whole business is about beyond the mere systematic description of language data. As early as the mid 90's I was somewhat surprised to be confronted with more and more papers that simply offered a description for some arbitrary set of data in some arbitrary language. There was no mention of why these data are interesting beyond the mere fact that they simply existed. There was no comment on what the analysis offered showed in any deeper theoretical sense. This was definitely different for work in the 80's and before. Unless you could show why the data you selected were interesting for questions of linguistic theory and why your analysis was superior to some competing analysis, you had no chance of getting your paper published in any serious journal. In some sense I feel that much (but obviously not all) of current linguistic work displays a relapse to the spirit prevailing in pre-Chomskyan times. Linguistics is about describing language data. Period. Beyond this there is no deeper epistemologica! goal. Of course, those who became linguists because they like to play around with language data could not care less, because they can pursue their interests under any development of the field, nowadays possibly with less pressure and stress. Personally I felt that much of what I was offered to read in recent years was intolerably boring and that the field of linguistics was becoming increasingly uninteresting and trivialized. So I decided to quit and turn to other fields. But, of course, that's a very personal decision.

On the Typology of Verb Second Günther Grewendorf

1. The Linear V2 Restriction If we try to answer the question of what a V2 language is with reference to a simple model of sentence structure, such as the theory of topological fields, the answer seems to be obvious: in declarative main clauses of a V2 language, the finite verb occupies the position of the left sentence bracket. Since the so-called prefield represents exactly one position that precedes the left bracket, we can derive the linear V2 restriction, according to which only one XP can precede the finite verb in main declarative clauses. However, this characterization of a V2 language is already faced with problems in view of constructions such as left dislocation in German, where the linear V2 restriction appears to be violated: (1)

Den neuen Roman von Händler, den habe ich noch nicht gelesen. the new novel by Händler it have I not yet read

Similar problems arise also with other languages, such as Old Italian, Old High German, Old English, Old French, which have traditionally been characterized as V2 languages, 1 Rhaetoromance, Mocheno, Sappadino, Cimbrian. As the examples in (2)-(4) show, some of these languages allow V3, V4, etc. orders: (2)

[Old High German] a. [Dhea uuehhun] [auur] [in heilegim quhidim] arfullant sibun iaar. the weeks however in sacred language fulfill seven years 'The weeks, however, take seven years in sacred language.' {Isidor, 457, Robinson, 1997; Fuß, 2008, p. 57) b. endi bidhiu iu chiuuisso quham christ and therefore already certainly came Christ 'And that's why Christ has certainly already come.' (Isidor 464; Axel, 2007, p. 221)

0

The present paper grew out of joint research with Cecilia Poletto. Part of it was presented in a talk delivered at the University of Massachusetts in Amherst in October 2009. I would like to thank Kyle Johnsen, Tom Roeper and Maggie Speas for useful comments and suggestions. If not indicated otherwise the data from Cimbrian (variety of Luserna) is the result of empirical research that I have carried out with Cecilia Poletto. Thanks very much to Fiorenzo Nicolussi for his help and patience.

1

Kaiser (2002) argues against the V2 nature of Old French.

On the Typology of Verb Second (3)

73

[Old Italian] [Ad ogni matto] [I .ναν;J paiono matti, [« come] [ai savi] [; to every fool the wise men seem fools such as to the wise men the matti] paiono veramente matti. fools seem really fools (Old Florentine: Novellino, p.40; Poletto, 2009)

(4)

[Old French] car ja cust cors qui ci gist ne sera remuez de son leu. for never this body which here lies neg will-be removed from its place 'for never will this body which lies here be removed from its place' (La Queste del Saint Graal 37,1 ; Vance, 1997, p. 62)

(5)

[Cimbrian] a. Gestarn dar pua iz furse gestânt untar in reng. yesterday the boy is perhaps stood under the rain. b. Pit fadige dar maurar hatt augemacht 'ζ haus. with difficulty the mason has built the house (Panieri, 2006, p. 307)

If these languages are in fact V2 languages, then our examples suggest that there are at least two different types of V2 languages depending on the position the verb occupies in the left periphery: (a) V2 languages that (with a few exceptions) obey the linear V2 restriction (German, Scandinavian). (b) V2 languages that allow at least V3 (Old English, Old High German, Cimbrian) but sometimes also V 4 , . . . (Old Italian) But if the linear V2 restriction does not provide us with a V2 criterion, how can we be sure that the languages under consideration are in fact V2 languages? Let us therefore turn to another property that has been considered a characteristic of V2 languages: the so-called "Germanic subject inversion". This property is, in fact, present in Old Italian, as can be seen from (6): (6)

[Old Italian] ... primieramente avea ella fatta a llui ingiuria. . . . for first had.3sg she done to him injury (BL, Rett., 116; Grewendorf & Poletto, 2009)

Cimbrian too shows subject inversion. However, unlike Old Italian, where this kind of inversion can be observed with DP-subjects in general, Cimbrian only has inversion with clitic subjects, cf. (7) vs. (8):

74 (7)

Günther Grewendorf [Cimbrian] a. Gestarn hatt-ar gisekk in has. yesterday has-he seen the hare (Panieri, 2006, p. 309) b. *Gestarn ar-hatt gisekk in has.

(8)

[Cimbrian] a..* Untar in reng iz a pua gestànt. under the rain is a boy stood b .*Gestarn hatt dar pua gesekk in has. yesterday has the boy seen the hare

According to Bidese (2009), the restriction that subject inversion is only possible with clitic subjects can already be observed in Old Cimbrian. On the other hand, Bidese points out that unlike modem Cimbrian, Old Cimbrian obeys the linear V2 restriction. The situation in Old Cimbrian is thus inverse to what we observe in Old Italian, which has DP-subject inversion but violates the linear V2 restriction. Interestingly, the ban against DP-subject inversion in modern Cimbrian does not extend to quantified subjects: (9) a. 'Zhatta

niamat

telefonaart.

it has-Part nobody called b. 'Z hatta eparummas telefonaart. it has-Part somebody called Let us assume that the existence of some kind of subject inversion in declarative main clauses is a characteristic of V2 languages, and that this characteristic can be analyzed as the result of verb movement to the left periphery of the clause. The question is whether the critical role of this property can receive further confirmation from independent V2 characteristics. Let us therefore consider another syntactic property that has been claimed to be typical of V2 languages. There is an asymmetry of main and embedded clauses that can typically be observed in V2 languages like German and the continental Scandinavian languages. This asymmetry concerns the position of the finite verb and is generally attributed to the complementary distribution of finite verb and complementizer: (10)

[German] a. Den neuen Roman von Händler the new novel by Händler b. Jutta glaubt, dass ich den neuen Jutta thinks that I the new

(11)

habe ich noch nicht gelesen. have I not yet read Roman von Händler noch nicht gelesen habe. novel by Händler not yet read have

[Norwegian] a. Jens skj0nte ikke dette sp0rsmâlet. Jens understood not this question b. *Jens ikke skj0nte dette sp0rsmâlet.

On the Typology of Verb Second

c. Vi vet

at

75

Jens ikke skjOnte

dette sp0rsmälet.

we know that Jens not understood this

question

(Taraldsen, 1986) (12)

[Swedish]

a. Den boken köpte

Erik i London.

the book bought Erik in London

b. Jag vet I

att han inte kommer.

know that he

not comes

(Platzack, 1986) If it is true that this kind of asymmetry is due to the complementary distribution of finite verb and complementizer, then we predict the absence of this asymmetry in V 2 languages where the complementizer occupies a position different from the left-peripheral target position of the finite verb. This is what we can in fact observe in Cimbrian. We will therefore take a closer look at the connection between the V 2 property and the complementizer system.

2.

V2 and the Complementizer System

Bhatt & Yoon (1991) distinguish between two different kinds of complementizers. Complementizers that are not associated with specific mood properties but only signal the subordinate status of a clause are called "pure subordinators". Pure subordinators such as Korean -ko, Japanese -to, Yiddish -az, Icelandic -ad, Hungarian -hogy, Kashimiri ki, Persian ke are compatible with a variety of mood markers. In the presence of a pure subordinator, mood marking can take place through the morphology of the finite verb or specific mood particles. In the Kashmiri examples ( 13), the pure subordinator ki cooccurs with the subjunctive marker and the interrogative marker of the verb, and in the Persian example (14), the pure subordinator ke cooccurs with the interrogative particle aya: (13)

[Kashmiri]

a. tem dop ki he

su kheyi-hee-na yi.

said SUB he eat-subj-Neg this

'He said that he would not eat it.'

b. Bill-an prutsh maajl ki

swa heky-aa az

Bill ERG asked mother SUB she can-Q

yith

today came

'Bill asked (his) mother if she can come today.' (Bhatt, 1999, p. 153) (14)

[Persian] (man) nemîdânam ke I

NEGknow

äyä

an

pesar zabänsenäsT mïxânad.

SUB PRT D E M boy

linguistics

Ί don't know if the boy studies linguistics.' (Ohl & Korn, 2006)

studies

Günther Grewendorf

76

The second type of complementizer, as found in German, English and Continental Scandinavian is represented by mood indicating complementizers, which conflate subordination and mood marking. As far as V2 languages are concerned, pure subordinators are typically found in symmetrical V2 languages while mood indicating complementizers occur in asymmetrical V2 languages (due to the complementary distribution of complementizer and verb second as illustrated in (15)). (15)

[German] a. Ich I b. Ich c. Ich

glaube believe glaube, glaube,

dass Maria den Studenten that Maria the student Maria hat den Studenten den Studenten hat Maria

geküsst hat. kissed has geküsst. geküsst.

Furthermore, if a language allows cooccurrence of complementizer and ννΛ-element in Wi-questions, we can observe that pure subordinators usually precede w/z-elements, as illustrated with the Kashmiri example (16): (16)

[Kashmiri] a. tse chay khabar ki kyaa kor tern. you aux know that what did he 'You know what he did.' b. *tse chay khabar kyaa ki kor tem. 'You know what he did.' (Bhatt, 1999, p. 1590

In contrast to pure subordinators, mood indicating complementizers follow wh-elements. This is the case with complementizers in Norwegian, Swedish and Bavarian, as the examples (17)-(19) illustrate. (17)

[Norwegian] Vi vet hvem som ikke skj0nte dette sp0rsmâlet. we know who that not understood this question (Taraldsen, 1986, p. 8)

(18)

[Swedish] Han undrade vem som inte hade öppnat dörren. he wondered who that not had opened the door (Platzack, 1986)

(19)

[Bavarian] I mecht wissen wea dass des gsogt hod. I want know who that this said has

The way in which the complementizer type interacts with the position of the finite verb can best be studied in languages which have both kinds of complementizers. Such a language is Cimbrian, which has two classes of complementizers (Panieri, 2006, p. 338-

77

On the Typology of Verb Second

342). "Ke-type" complementizers such as ke ('that'), benn ('when'), umbromm ('because'), bia ('how') act as pure subordinators, while "az-type" complementizers such as az ('that'/'if'), bal ('if'), intánto az ('while'), dopo az ('after'),fin az ('until'), bo (relative clause complementizer) show the typical behavior of mood indicating complementizers. The two classes of complementizers are associated with different verb order properties, as can be seen from a series of tests which concern the position of negation, the position of sentence particles, and the position of clitics (see Grewendorf & Poletto, 2009). While sentences introduced by complementizers of the former type show the behavior of main clauses in several respects, sentences with az-type complementizers behave differently. This can be seen from the relative position of finite verb and negation. The negative particle net ('not') follows the finite main verb in main clauses and ^-clauses (and precedes the past participle). This is shown for main clauses in (20) and for /»-clauses in (21): (20)

[Cimbrian] a. I gea net ka miss. I go not to mass b. */ net gea ka miss. (Panieri, 2006, p. 331)

(21)

[Cimbrian] a. I boas I know b.*I boas I know Panieri

ke du geast nèt that you go not ke du nèt geast that you not go (2006, p. 339)

ka to ka to

Tria. Trient Tria. Trient

The situation is different with az-clauses, where negation must precede the finite main verb: 2 (22) a. Dar he b. *Dar he

hat has hat has

geböllt wanted geböllt wanted

azz-e that-I azz-e that-I

nèt vortgea. not away-go vortgea nèt. away-go not

These differences in the relative position of verb and negation are reminiscent of similar contrasts between Islandic and Continental Scandinavian. Negation follows the verb in Icelandic and precedes it in Continental Scandinavian. This difference is usually attributed to different positions of the verb. The general assumption is that the verb raises to a position above negation in Icelandic, while it remains in the VP in Continental Scandinavian. Applying this account to Cimbrian, we can conclude from the contrast between (20)/(21 ) on the one hand and (22) on the other hand that in main clauses and ^-clauses, the finite verb occupies a higher position than in az-clauses. Note, however, that this 2

I have ignored the relative position o f negation and auxiliary/modal verbs since az-clauses optionally allow main clause order in this case.

78

Günther Grewendorf

contrast does not yet tell us anything about whether or not the finite verb in main and keclauses occupies a left-peripheral position. It only shows that the verb raises to a position higher than negation and thus satisfies a prerequisite for V2. A more revealing test is offered by the position of the particle da. This particle is homophonous but not identical with the locative element da ('there'), which is reflected in the fact that it can cooccur with a locative instance of da, as in (23): (23) Dar libar bo da der Gianni da hat gelek. the book that Part the G. there has put Without going into the details of the exact semantic function of the particle da, I will assume that it acts as a fore-grounding particle that is related to the topical properties of the clause and thus to the left-peripheral topic field.3 More important in the present context is the observation that in main clauses and fe-clauses, the particle da appears immediately after the finite verb: (24) a. Alle sunta hán-da gelaiitet die klokkng. every Sunday have-Part rung the bells b. Alle sunta laiitn-da die klokkng. every Sunday ring-Part the bells c. *Alle sunta da laiitn die klokkng. every Sunday Part ring the bells (Panieri, 2006, p. 317) (25) a. Dar Mario hatt khött ke alle sunta hân-da gelaiitet die klokkng. the M. has said that every Sunday have-Part rung the bells b. */ boas ke da khint di nona. I know that Part comes the granny Since the fronted adverbial in (24) suggests that the finite verb is in the V2 position in (24a/b), it looks like the particle da only attaches to the verb in V2 contexts. This assumption receives support from the observation that in az-clauses, da never follows the verb but enclitically attaches to the complementizer (recall that the complementizer bo belongs to the az-class): (26) a. I hân gèzt 'ζ proat, bo da mar hatt gètt dar nonno. I have eaten the bread where Part to-me has given the grandfather b. */ hân gèzt 'ζ proat, bo mar hatt-da gètt dar nonno. c. */ hân gèzt 'z proat, bo hatt da mar gètt dar nonno. Ί have eaten the bread that the grandfather has given to me.' (Panieri, 2006, p. 345) 3

Unlike Bayer & Suchsland (1997), w h o claim that the German correlate of da may be considered an expletive subject comparable to English there, I do not assume that the Cimbrian particle da can be interpreted on a par with the English expletive there. One of the reasons is that da not only cooccurs with preverbal DP-subjects in relative clauses and embedded interrogatives but also cooccurs with clitic and tonic subject pronouns.

On the Typology of Verb Second

79

We can conclude from the contrast between (24)/(25) and (26) that in az-clauses the finite verb occupies a lower position than in main and /ce-clauses. There is even reason to believe that in (24) and (25), the verb occupies a high position in the left periphery since, as we will see in the next section, the particle da precedes fronted topic and focus phrases. Moreover, the fact that da attaches enclitically to the verb in main and fe-clauses and to the complementizer in az-clauses seems to indicate that the position of the finite verb in main and £e-clauses is identical to the position of the complementizer in az-clauses. This hypothesis receives further support from the observation that clitics attach to the verb in main and i:e-clauses but to the complementizer in az-clauses: (27) a. I gloabe ke dar gebat-mar-s. I think that he gives-me-it b. *Dar hat-mar khött ke dar en sich morgn. he has-me said that he him sees tomorrow c. Dar hat-mar khött ke dar sich-en morgn. he has-to-me said that he sees-him tomorrow (28) a. Da they b. *Da they

soin are soin are

vortgont away-gone vortgont away-gone

ena without ena without

az-ta-s niamat barn. that-Part-it nobody noticed az niamat barn-da-z. that nobody noticed-Part-it

The hypothesis that the finite verb occupies a left-peripheral position in main and keclauses receives further support from the distribution of the expletive pronoun 'z ('it'). This expletive behaves like the German "prefield"-es ('it') and has the typical properties of CP expletives (no associate, no Case and (^-features). Consequently, the occurrence of this expletive correlates with "pure" EPP/Edge positions (ensuring the V2 configuration). We can therefore predict that the CP expletive disappears if the V2 requirement is satisfied by another XP. This prediction is borne out, as (29) demonstrates: (29) a. 'Z hán-da gelaiitet die klokkng alle sunta. it have-Part rung the bells every Sunday b. Alle sunta laiitn-da die klokkng. every Sunday ring-Part the bells (Panieri, 2006, p. 317) (29) can also be taken as further evidence for the above hypothesis that the particle da attaches to the element in V2 position. If this hypothesis is in fact correct, we can make a further interesting prediction. If da attaches to the verb in V2 position, then the fact that it also attaches to the verb in ^-clauses leads to the prediction that ^-clauses as opposed to az-clauses should be able to cooccur with the expletive pronoun 'z. This prediction is in fact borne out, as shown by the examples in (30): (30) a. Dar Mario hatt khött ke 'z hân-da gelaiitet die klokkng alle sunta. the M. has said that it have-Part rung the bells every Sunday

80

Günther

Grewendorf

b. *Dar Mario hatt geböllt az 'z hän-da gelautet die klokkng alle sunta. the M. has wanted that it have-da rung the bells every Sunday Our observations on the Cimbrian complementizer system imply that there is a specific interaction of verb position and complementizer system in the case of az-type complementizers. In fact, we can draw the conclusion that the finite verb is in complementary distribution with complementizers of the az-type. Cimbrian thus shows an asymmetry between main and embedded clauses that is characteristic of V2 languages in which the complementizer occupies the same position as the fronted verb in main declarative clauses. It also follows that the absence of this asymmetry in te-clauses shows that ke-type complementizers occupy a position different from the left-peripheral target position of the finite verb. As far as the asymmetry between main clauses and embedded clauses of the az-type is concerned, Cimbrian seems to be on a par with V2 languages like German. Should we therefore conclude that the V2 configuration is identical in the two languages? I will show in the next section that despite appearance, German and Cimbrian differ considerably in their V2 structures in that the V2 configuration is created by different areas in the left clausal periphery of the two languages.

3. V2 and the'Prefield' In order to derive the V2 configuration in languages like Cimbrian, I will make use of two theoretical assumptions. I will assume an extended left periphery of the clause along the lines of Benincà & Poletto (2004) and I will follow Kayne (1994) in assuming that adjunction is always to the left. We have already seen in the preceding section that there is evidence for the hypothesis that the sentence particle da is generated in a left-peripheral position. An additional argument for this hypothesis is provided by the observation that da is located in a position higher than topic and focus: (31 ) Dar libar bo da I in Gianni za on get. the book that Part I to-the-G. already have given (32) a. Dar libar bo da r IN GIANNI hat get. the book that Part-he TO THE G. has given b. *Dar libar bo IN GIANNI dar hat get If topics like in Gianni in (31 ) and focus phrases like IN GIANNI in (32) are located in left-peripheral positions, as argued for by Rizzi (1997) and Benincà & Poletto (2004), among others, then the surface position of the particle da must obviously also be in the left periphery. There is not only evidence for the left-peripheral position of the particle da, there is also reason to believe that clitics are likewise located in the left periphery. An interesting argument to this effect can be derived from an observation by Haider (2009). Haider has pointed out that scrambling an NP without Case morphology triggers a strong garden path effect, as is the case in (33):

81

On the Typology of Verb Second (33)

weil Marga Kollegen vorgestellt bekamen. since M . A C C colleaguesNoM introduced got

While (33) is very difficult to parse, there are no such problems if the preposed element is a clitic pronoun that is similarly ambiguous with respect to Case: (34)

weil es Kollegen vorgestellt bekamen. since it colleagues introduced got

To account for the contrast between (33) and (34), we can assume that there is a leftperipheral syntactic position which is specifically designed for pronouns and makes it easier to identify their syntactic function even in the absence of unambiguous Case morphology. If it is true that clitics and the particle da are located in the left periphery of the clause, the question arises as to whether they are generated there or have reached their surface position as a result of movement. In order to answer this question let us recall our assumption that right adjunction is not permitted. The answer then suggests itself if we have established where those elements are located to which clitic pronouns and the clitic particle da attach. Since it can be shown that complementizers are not generated in a position lower than Fin, 4 we can assume that those complementizers which are in complementary distribution with the finite verb are generated in the Fin 0 position. This assumption is in accordance with the finite properties of this position and traditional assumptions about mood indicating complementizers (cf. den Besten, 1983; Bayer, 2008). We can then conclude from the prohibition against right adjunction that clitics and the particle da must be generated in a position higher than Fin and that the complementizer raises from Fin to a higher position in the left periphery passing through the positions of clitics and da, where it picks up its enclitic elements. 5 Appealing to Baker's (1985) Mirror Principle, we can take the fact that da attaches closer to the complementizer than clitics as evidence that clitics must be located higher in the hierarchy of the left periphery than the particle da. If we assume that the latter is generated as the head of a topical projection GroundP and that clitics are generated as the head of a Wackernagel-projection WackP, then WackP must be higher in the left-peripheral structure than GroundP. The results that we have obtained so far are represented in the structure (35): (35)

[wackP

az clitics

[GroundP

da

[τορίςΡ

[FOCUSP

[FinP

OZ

[iP . . -

Note that the derivation in (35) already yields the correct prediction that in az-clauses, verb movement is blocked from reaching a left-peripheral head position. As far as main and fe-clauses are concerned, we have already seen that in these clauses, it is the finite verb that enclitically picks up clitic pronouns and the particle da. In accordance with 4

Among other problems, such an assumption would create problems for a theory of selection.

5

Independent evidence for complementizer movement can be found in Watanabe (1993), Browning (1996), Poletto (2000), Roussou (2000), Roberts (2004), Rizzi & Shlonsky (2007), Belletti (2008).

82

Günther

Grewendorf

the complementary distribution of αζ-complementizers and the finite verb, as observed in section 2, we can thus conclude that the finite verb undergoes movement through Fin 0 , Ground 0 and Wack 0 and finally reaches Force 0 , where it checks modal features of clausal typing. The latter assumption can be extended to complementizers of the az-class, since in view of their featural make-up, it is rather implausible to assume that they remain in the head position of WackP. Given that the "prefield"-expletive 'z precedes the finite verb in main and ^-clauses and that pure subordinators precede the CP-expletive we can derive the V2 configuration in (36), which significantly differs from the V2 configuration in languages like German: (36)

[subordP k e [ForceP

Ζ [ öz/Vfi„

[wackP

clitics

[GroundP

da

[jopicP

[FOCUSP

[finP a Z tlP · · ·

As for main and fe-clauses, the derivation in (36) implies the correct prediction that in both cases, the inflected verb can raise to Force yielding a V2 configuration with a prefield-expletive or an XP in SpecForceP. It also provides us with an adequate analysis of the enclitic attachments that the finite verb can undergo when it reaches its surface position. As far as az-clauses are concerned, derivation (36) enables us to account for the fact that the finite verb is blocked from reaching the left periphery in these clauses. As already pointed out, we thus derive the asymmetry of main and embedded clauses that is typical of V2 languages. The fact that the finite verb has to stay in a position within IP correctly entails that there is no clitic subject inversion nor any other cliticization of pronouns and left-peripheral particles to the finite verb. Finally, we can also capture the fact represented by the examples in (31) and (32) that the surface position of az-type complementizers precedes fronted topic and focus phrases. In this section we have seen that two different types of V2 languages can be distinguished. Put in topological terms, we can say that the "prefield" in V2 languages of the Cimbrian type is represented by SpecForceP while it is represented by SpecFinP in V2 languages of the German type. Let us call the former type "high V2 languages" and the latter type "low V2 languages". The claim that unlike high V2 languages, a V2 language like German localizes its CP-expletive in SpecFinP rather than SpecForceP receives independent support from the fact that the CP-expletive can be preceded by a left-dislocated phrase, which is not possible in a language like Cimbrian: 6 (37) [German] Den Studenten es hat den keiner gesehen. the student ACC it has him nobody seen

4. V2 and Information Structure In this section I turn to some more detailed properties of the left periphery in V2 languages. In particular, I want to consider the question how the V2 configuration interacts 6

Left-dislocated elements in German occupy the specifier of TopP, as shown in Grewendorf (2009) and Grohmann (1997).

On the Typology of Verb Second

83

with topicalization in the sense of left dislocation. There are some intricate phenomena here that, to the best of my knowledge, have not yet been investigated. The first observation concerns the fact that in V2 languages of the German type, a left-dislocated element cannot satisfy the V2 constraint: (38) [German] a. *Den the b. Den c. Den

Studenten studentACC Studenten, Studenten,

hat Maria den has M. him den hat Maria Maria hat den

geküsst. kissed geküsst. geküsst.

Example (38a) can only be interpreted as a yes/no question combined with left dislocation. As a declarative it is not well-formed. The same observation can be made with respect to Kashmiri: (39) [Kashmiri] a,*Con kalami dyut rameshan shiilayi sui. your pen gave Ramesh Sheila that 'As for your pen, Ramesh gave it to Sheila.' b. Con kalami SUi dyut rameshan shiilayi. your pen that gave Ramesh Sheila (Bhatt, 1999, p. 109f) The left-dislocated phrase Con kalam in (39a) is not a licit candidate for the "prefield" of the V2 clause. Unlike a left-dislocated phrase, a preverbal focus satisfies V2, as shown in (40): (40) [Kashmiri] CON KALAM dyut rameshan shiilayi. your pen gave Ramesh Sheila 'It was your pen that Ramesh gave Sheila.' (ibid.) In the V2 language Rhaetoromance, a similar situation obtains; a left-dislocated element cannot satisfy V2: (41) [Rhaetoromance] 7 a. *L giat, l'ai odu. the cat, it-have-I seen b. *De Giani, nen-ai bel baiè. of Giani, of-him-have-I already spoken (Poletto, 2002)

7

Note that the resumptive clitics in ( 4 1 ) form part o f the finite verb.

84

Günther Grewendorf

In view of these observations it comes out as a surprising fact that in Cimbrian, a leftdislocated element can occupy the "prefield" position and thus satisfy the V2 configuration: (42) [Cimbrian] Dar hatt-mar khött, ke in libar koaft-ar-en morgn. he has-me told that the book buys-he-it tomorrow The question may arise as to how we can be sure that the left-dislocated element in (42) is in fact located in SpecForceP rather than in SpecTopP (with the verb in Top 0 ). There is evidence that the cooccurrence of an XP in SpecTopP and a finite verb in Top is precluded on independent grounds (Rizzi, 1997). Furthermore, if the left-dislocated element in (42) were in SpecTopP and the verb in a position lower than Top, it should be possible for other XPs to intervene between the left-dislocated element and the finite verb. This prediction, however, is not borne out by the facts: (43)*Dar hatt-mar khött, ke in libar morgn/MORGN koaft-ar-en. he has-me told that the book tomorrow buys-he-it We will therefore continue to assume that in libar in (42) occupies SpecForceP. Notice that it is not the case that on independent grounds, a topicalized (left-disloated) element is prohibited from moving to the prefield position. The example (44b) shows that a leftdislocated element is not prevented from moving to the prefield of a higher clause: (44) [German] a. [Den Studenten]¿, Hans glaubt, [déni hat keiner gelobt]. the studentACC Hans thinks him has nobody praised b. [Den Studenten]¿ glaubt Hans, [deni hat keiner gelobt]. While in (44a) left-dislocation has taken place into the TopP of the matrix clause, (44b) can be analyzed as left-dislocation in the embedded clause followed by movement of the left-dislocated element from the embedded TopP to the matrix prefield (SpecFinP). The examples in (45) confirm that the constraint according to which the V2 restriction in German cannot be satisfied by a left-dislocated element only applies to the clause from which left-dislocation originates: (45) a,*[Den Studenten]i the studentACC b. *\Den Studenten]i the studentACC gelobt]]. praised

glaubt thinks glaubt thinks

Hans, [ HansN0M Hans, [ Hans N0M

hat keiner

deni gelobt].

h a s nobody N O M h i m p r a i s e d

meinte Maria [ thought Maria N0M

habe keiner den had nobody him

On the Typology of Verb Second

85

c. [Den Studenten]i glaubt Hans, [ meinte Maria, [ deni habe keiner the studentACC thinks Hans thought MariaN0M him has nobody gelobt]]. praised 'Hans believes Mary thought that nobody had praised the student.' Left-dislocation in (45a) and (45b) originates from the deepest clause (as shown by the resumptive pronoun) and even if the left-dislocated element is moved to higher clauses, it cannot satisfy the V2 configuration in the deepest clause. On the other hand, as can be seen from example (45c), if the V2 constraint is fulfilled by another XP in the deepest clause (the clause from which left-dislocation originates), further movement of the leftdislocated element may satisfy this constraint in higher clauses. Thus we face a paradox in view of the German examples in (38), (44) and (45): On the one hand, a left-dislocated element is precluded from satisfying the V2 constraint in a language like German; on the other hand, a left-dislocated element may satisfy this constraint if the relevant V2 configuration is not the one in the clause from which left dislocation originates. But even if we were able to solve this German-internal paradox, the contrast between the German example (38a) and the Cimbrian example (42) involves us in a further paradox: why is it that unlike German left dislocation, left dislocation in Cimbrian may satisfy the V2 constraint in the very clause in which left dislocation takes place? Let me first turn to the German-internal paradox. We have already seen that the "prefield" in German is represented by SpecFinP. It is a well-established fact that XPmovement to the German prefield is neutral with respect to information structure. Following Haegeman (1996), Roberts (2004) and Chomsky (2008), we can express this fact by saying that XP-movement to the prefield does not establish any Agree operation but is triggered by an EPP- or Edge feature. We can then analyze (44b) and (45c) along the following lines. Left dislocation takes place in the deepest clause with the resumptive pronoun satisfying the Edge feature in its prefield. The left-dislocated element "checks" its Topic feature in the lowest clause and then undergoes further movement and satisfies the Edge feature in the higher clauses. Obviously, as shown by (38a) and (45a/b), the left-dislocated element is not permitted to pass through the lowest SpecFinP on its way to SpecTopP thereby satisfying EPP in SpecFinP. This confirms our earlier assumption that the features associated with the finite verb in Fin are incompatible with the topic feature of the left-dislocated element. It is only AFTER this element has "checked o f f ' its Topic feature in the specifier of the TopP, that it is allowed to satisfy the EPP. This is the reason why the left-dislocated element can satisfy the V2 configuration in (44b) and (45c). The suggestion that an XP that has moved to SpecTopP can undergo further movement and is not frozen in place might, on principled reasons, give rise to the objection that topic movement is criterial movement in the sense of Rizzi (2006) and thus subject to what Rizzi calls "criterial freezing". But recall that Rizzi himself advocates the view (Rizzi, 2004) that "topic is special" and possibly not an instance of criterial movement in the first place. If we take into account that topic movement does not exhibit any minimality effects nor does it trigger Weak Crossover effects, we have to conclude that it is different

Günther

86

Grewendorf

from familiar instances of criterial movement such as movement of arguments, quantifiers, modifiers etc.. Let us therefore stay with the hypothesis that topics may leave the position where their topic feature is "checked". However, this option has to be restricted to specific targets: topics may only move to higher positions which are not criterial positions but are "neutral" with respect to information structure. In other words, topics which have "checked off" their topic feature in a TopP can only move to EPP/Edge positions i.e. to the prefield in traditional terms. We are now in a position to tackle the second paradox. This paradox is concerned with the satisfaction of the V2 constraint by a left-dislocated element and arises from the contrast between (38a) and (42), repeated here as (46) and (47) (46) [Cimbrian] Dar hatt-mar khött, ke in libar koaft-ar-en morgn. he has-me told that the book buys-he-it tomorrow (47) [German] *Den Studenten hat Maria den geküsst. the student ACC has Maria him kissed The crucial question is why it is that a left-dislocated element can satisfy the V2 constraint in a V2 language such as Cimbrian (as shown by (46)) but not in a V2 language such as German (cf. (47)). In order to account for this difference let us first recall that we have already established that the "prefield", i.e. the XP-position in front of V2, is represented by SpecForceP in Cimbrian and by SpecFinP in German. We have further seen that movement to the prefield is not triggered by features that are related to information structure but rather by an Edge/EPP feature that is neutral with respect to information structure. The variation in the structural representation of the prefield implies that the hierarchical order of prefield and TopP in Cimbrian is inverse to the structural relation between prefield and TopP in German. While in German, the TopP dominates the prefield, the prefield in Cimbrian is in a higher position than the Topic projection. In view of this difference and in view of what we said about the constraints on moving topics (left-dislocated elements), we can solve the second paradox along the following lines. In the German example (47), the left-dislocated element cannot fulfill the EPP/Edge feature in SpecFinP since it has not yet checked off its topic feature in the specifier position of the (higher) TopP and, as we saw above, an XP with an unchecked topic feature cannot pass through SpecFinP. Since the Topic projection in Cimbrian is lower than ForceP, the left-dislocated element in libar in (46) has already checked off its topic feature before undergoing movement to the prefield position SpecForceP. It is thus licensed to fulfill the "neutral" requirements of the EPP/Edge. It emerges from this analysis that the interaction of left dislocation and prefield movement in German and Cimbrian correlates in one important respect. The reason why a left-dislocated element in German can satisfy the V2 configuration only in clauses higher than the clause from which left dislocation originates is the same as the reason why a left dislocated element in Cimbrian can satisfy the V2 configuration within the clause where left dislocation takes place.

On the Typology of Verb Second

87

On the basis of what we have seen so far, we can draw the conclusion that there are two different types of V2 languages instantiated by Cimbrian and German respectively. The differences between the two kinds of V2 languages derive from the fact that different areas of the left clausal periphery may be activated in a V2 language with the consequence that V2 languages may differ with respect to the prefield position, the V2 position, the linear V2 restriction, and the asymmetry between main and embedded clauses. These differences are summarized in (48) and (49): (48) High V2 languages (Cimbrian a. b. c. d.

type)

The prefield position (EPP/Edge) is SpecForceP. The V2 position is Force. There is no linear V2 restriction. There is an asymmetry between main and embedded clauses only with a special type of complementizers.

(49) Low V2 languages (German type) a. b. c. d.

The prefield position (EPP/Edge) is SpecFinP. The V2 position is Fin. There is a linear V2 restriction (with only a few exceptions). There is an asymmetry between main and embedded clauses with all complementizers.

The asymmetry noted in (49d) is crucially related to the assumption that complementizers in languages such as German occupy the Fin position and are thus in complementary distribution with the finite verb (cf. (49b)). Notice that there is, in fact, independent evidence for the claim that complementizers occupy the Fin position in German. Although left dislocation is generally judged as illicit in German embedded clauses introduced by a complementizer, there is a clear contrast between sentences such as (50a), where the left dislocated element precedes the complementizer, and sentences such as (50b), where it follows the complementizer. (50) [German] a. ?£> he b. *Er he

hat has hat has

gesagt, said gesagt, said

den Studenten, dass the student a c c that dass den Studenten, that the student a c c

den jemand him somebody den jemand him somebody

geküsst kissed geküsst kissed

hat. has hat. has

The contrast between (50a) and (50b) can be attributed to the fact that a target position for the left dislocated element (TopP) is only available above the complementizer. We therefore predict that left-dislocated elements follow the complementizer in a language where the complementizer is in Force. The latter is the case with the Icelandic complementizer ad and, as shown by (51), the prediction is borne out.

88

Günther

Grewendorf

(51) [Icelandic] Jón segir [ad Pessum hring, honum hafi Ólafur lofad Mariu J. says that this ring d a t it d o t has O.nom promised M.dat (Thráinsson, 1979, p. 64, ("contrastive dislocation"))

].

At this point the question may arise as to why the linear V2 restriction is fulfilled in V2 languages like German. An answer to this question is provided by Roberts (2004) and Cardinaletti (2008), who assume that XP movement to the left periphery obligatorily passes through SpecFinP due to EPP requirements. Since FinP is taken to have only one specifier position, SpecFinP acts as sort of a low "bottleneck", which leads to the linear V2 restriction. Since left-dislocated elements are forced to bypass SpecFinP such that another XP has to fulfill the EPP in SpecFinP, we correctly derive the fact that only leftdislocated elements license a violation of the linear V2 constraint in V2 languages like German. As far as the linear V2 restriction in Cimbrian is concerned, we have seen that it is violated in fce-clauses and in clauses with preverbal definite subjects. The crucial question that we have not yet dealt with is the question whether sentences like (5), repeated here as (52), are instances of a V2 configuration in the first place. (52) [Cimbrian] a. Gestarn dar pua iz furse gestánt untar in reng. yesterday the boy is perhaps stood under the rain b. Pit fadige dar maurar hatt augemacht 'ζ haus. with difficulty the mason has built the house (Panieri, 2006, p. 307) There are some indications which cast doubt on the hypothesis that the examples in (52) really represent V2 clauses. We have seen in section 2 that enclitic attachment of the particle da to the finite verb is indicative of the verb occupying the V2 position Force. Against the background of this criterion it is a surprising fact that da cannot cooccur with the finite verb when the latter is preceded by a definite subject: (53) a. *Gestarn dar pua ist-da gestánt untar in reng. yesterday the boy is-Part stood under the rain b. *Di klokkng hân-da gelaiitet alle sunta. the bells have-Part rung every sunday It thus looks like definite subjects do not move to SpecForceP but stay within IP. Should this speculation turn out to be true, we can take the absence of V2 in these cases as indicating that Cimbrian is possibly in the process of losing its V2 nature and that this process is initiated by main clauses with preverbal definite subjects. More research is needed in order to work out an adequate analysis for sentences like (52) and establish speculations about the loss of V2 in Cimbrian. 8 8

For some suggestions and more empirical material see Grewendorf & Poletto (2010).

On the Typology of Verb Second

89

Leaving aside the problem of sentences with preverbal definite subjects and turning to the prefield in Cimbrian, we may ask whether SpecForceP has properties similar to those assumed by Roberts and Cardinaletti for SpecFinP. In other words, we may ask whether SpecForceP acts as a "bottleneck" in much the same way as claimed for SpecFinP, a question that would only be of interest if there is an XP position higher than ForceP. If such a position in fact exists and if the situation with SpecForceP is analogous to that with SpecFinP, we could make some interesting predictions depending on the properties of this higher position. If this position has topic features, we would predict that in case SpecForceP is occupied, only topic-like elements which have not yet checked off their feature can bypass SpecForceP, thereby licensing another violation of the linear V2 restriction. We will see in the next section that this kind of violation in fact exists and that we have strong reasons to assume that there is a topic-like XP position above ForceP.

5. V2 and Wh-Questions In this section I will primarily deal with embedded wh-questions in Cimbrian. As familiar from V2 languages like German, there seems to be an asymmetry in Cimbrian between main and embedded wh-questions. While the former show the V2 configuration with subject clitic inversion, there seems to be no subject clitic inversion in embedded whquestions. (54) [Cimbrian] a. Baz hàn-sa gimachtl what have-they done b. I boaz net, baz sa hân gimacht. I know not what they have done (Panieri, 2006, p. 345) The position of the particle da in direct wh-questions confirms the claim that the finite verb is in a high left-peripheral position: (55) Benn khìnta (khìnt+da) di nona? when comes-Part the grandmother (Panieri, 2006, p. 325) A closer look at embedded wh-questions reveals that the situation is more complex than indicated by examples such as (54b). In addition to the situation in (54b), there are embedded wh-questions in which the fronted wh-element occupies a position higher than ForceP and lower than the subordinator zega (literally 'to see'): (56) Se hanmar gevorst zega biavi geld 'z hàn-da vorbrennt die belesan they have-me asked Sub how-much money it have-Part burnt the Italian bonke. banks

90

Günther

Grebendorf

That the wh-phrase in (56) indeed occupies a position above ForceP follows from the fact that it appears in front of the CP-expletive, which we have shown to be located in the specifier position of ForceP. Enclitic attachment of the particle da to the finite verb provides further evidence for the high position of the wh-element. The claim that the finite verb is in fact in Force in certain embedded wh-questions seems to receive further support from the position of the clitic pronoun in (57): (57) Se hanmar gevorst biavi geld z'hat-as gevresst di bonka. they have-me asked how-much money it has-us eatent the bank On the other hand, the position of the particle da and the position of negation clearly show that in the embedded wh-question in (58), the finite verb does not occupy a left-peripheral position: (58) Bar bizzan ber da nèt âschauget di diarn. we know who Part not looks-at the girl Panieri (2006, p. 349) It should be obvious that (56) provides clear evidence for the hypothesis that the whelement is in a position higher than ForceP. On the other hand, the low position of the verb in (58) does not yet show that the wh-element is not in a position higher than ForceP. We know that the wh-element in (58) must be located in a position that is at least as high as the GroundP (the head of which hosts the particle) but we have no evidence to determine whether it is in the specifier position of the GroundP, in the specifier position of the ForceP, or in a position higher than ForceP. A similar remark is in order with respect to (54b). More tests are needed in order to establish which of these alternatives is in fact present. It might be revealing to investigate the interaction of wh-movement and left dislocation, but the relevant data is not yet available. Although it seems rather unlikely that the wh-elements occupy SpecForceP, since filling the prefield usually involves a filled head Force 0 , there is no syntactic evidence yet to rule this alternative out. It is tempting to draw a parallel between the contrasting verb positions with embedded wh-questions on the one hand and the contrasting verb positions in the case of ke-type and az-type clauses on the other hand. The fact that the verb does not reach the left periphery in clauses with az-type complementizers was attributed to the fact that these complementizers are generated in Fin and thus block verb movement to the left periphery. If we carry over this account to the contrasting verb positions with embedded wh-clauses, we would have to assume that there are embedded wh-questions which have a covert complementizer in the Fin position and thus block verb movement. Notice that the variation in embedded wh-questions is not due to the difference between complex and bare wh-elements so that we cannot say that it is bare wh-elements which act as complementizers and occupy the Fin position. As the example in (59) shows, there are also bare wh-elements that occur in a position higher than ForceP. (59) Bar bizzan nèt benn 'z khèmmen-da di khindar vo schual haiit. we know not when it come-Part the children from school today (Panieri, 2006, p. 347)

91

On the Typology of Verb Second

It is obviously not the morphological complexity of wh-elements that gives rise to the varying behavior of embedded wh-clauses. This is clearly reflected in the fact that one and the same bare wh-element may occur in embedded questions with the high verb as well as in embedded questions with the low verb: (60) a. Dar hat-mar gevorst zega

bas

ζ'hat-ta

gelaiitet.

he has-me asked Comp what it has-Part rung b. Dar hat-mar gevorst zega bas-da hat gelaiitet. he has-me asked Comp what-Part has rung On the other hand, complex wh-elements appear to obligatorily cooccur with the high position of the finite verb, as can be seen from the position of the particle da in (61a) and of the clitic dar in (61b): (61) a.*I I b. */ I

boaz know boaz know

nèt biavi not how many nèt biavi not how many

laiit people invitati guests

da sòin gestânt dahùampeng in reng. Part are stayed at home because of the rain dar on za gemacht a geschenk. you have already made a present

Interestingly, the high position of the finite verb is obligatory in embedded wh-questions even when the complex wh-element is clearly non-D-linked. The relevant tests are again provided by the position of clitic pronouns, negation, and the particle da (along with the CP-expletive). (62) a. I I b. */ I

boaz know boaz know

(63) a. Dar he b. *Dar he

nèt not nèt not

ber who ber who

hat-mar has-me hat-mar has-me

cavolo the hell cavolo the hell

gevorst asked gevorst asked

hat-dar get an has-you given a dar hat get an you has given a

zega Sub zega Sub

che what che what

cavolo the hell cavolo the hell

punk. punch punk. punch

du you du you

isst eat net not

net. not isst. eat

(64) a. Dar hat-mar gevorst zega ber cavolo 'ζ khin-ta a cioina. he has-me asked Sub who the hell it comes-Part for dinner b. *Dar hat mar gevorst zega ber cavolo da khint a cioina. Obviously, the variation that we find with embedded wh-questions cannot be explained with respect to the difference between D-linked and non-D-linked wh-elements. I will not go into the question as to how this variation might then be explained (see for some ideas Grewendorf & Poletto, 2009). The present paper is about the V2 phenomenon, and I have mentioned this variation because of its effect on the V2 configuration. This effect can be stated as the generalization in (65):

92

Günther Grewendorf

(65) Generalization on embedded

wh-questions

a. In embedded wh-questions of Cimbrian, wh-elements with a lexical restriction obligatorily occupy a position higher than ForceP with the finite verb being located in Force. b. Bare (pronominal) wh-elements have the two options: they can occupy the high position with the finite verb being in Force, or they cooccur with a low finite verb not located in the left periphery (with the exact left-peripheral position of the wh-element still to be determined). We may speculate that the low position of the verb with bare wh-elements is due to the presence of a covert complementizer in the Fin position. This suggestion would establish a connection between the variation of embedded wh-clauses and the variation of clausal complements as introduced by different types of complementizers. It would also draw a parallel to a familiar account of the asymmetry between main and embedded whinterrogatives in languages like German. As shown by the presence of the CP-expletive, the high wh-elements obviously do not pass through SpecForceP on their way to the high target position. This strategy of bypassing the prefield is reminiscent of what we observed with left-dislocated elements in German. In fact, there is reason to believe that the high position of the wh-elements has topic-like properties. This can be seen from the obligatory high movement of whelements with lexical restrictors. We can assume that some kind of topicality is always associated with the presence of an overt lexical restrictor. The optional high movement of pronominal wh-elements could then be attributed to the contextually determined presence of a covert restrictor. Non-topical pronominal wh-elements could then be analyzed as pure operators along the lines of Grewendorf (2010). The suggestion that the high target position of wh-elements may be represented as a topic-like functional projection receives independent support from Stowell & Beghelli's analysis (Stowell & Beghelli, 1994) of quantifier phrases with specific reference. Stowell/Beghelli show that such quantifier phrases covertly move to a left-peripheral position which they call "Referential Phrase" (RefP) and which they localize above the root CP. If specific reference of quantifier phrases correlates with the topical properties of whelements, then the Cimbrian data with high wh-movement can be taken to show that Stowell/Beghelli's covert movement of specific quantifier phrases is overtly realized in natural languages in the case of left-peripheral movement of wh-elements with topical properties. Other languages also provide independent evidence for the assumption that wh-elements may have topic properties and are then fronted to a higher position than non-topical wh-elements. As shown by Jaeger (2004), in colloquial Bulgarian, which has multiple wh-fronting in much the same way as Standard Bulgarian, the order of fronted (non-Dlinked) wh-elements is subject to a (selective) superiority constraint according to which a wh-subject has to precede wh-objects. Interestingly, if a wh-element is doubled by a clitic, which is possible in Bulgarian, it obligatorily precedes the other fronted wh-elements irrespective of whether this position results in a violation of the superiority condition

On the Typology of Verb Second

93

(Jaeger, 2004). In (66), the clitic doubled wh-object precedes the fronted wh-subject although the superiority condition would require the inverse order: (66) [Bulgarian] Kogo koj go obical whom who him loves 'Who loves whom?' (Jaeger, 2004, p. 214) (66) correlates with the observation that only topic phrases can precede fronted whelements in Bulgarian (67) and that clitic doubling is obligatory for fronted object topics (68):

(67) Ivan na kogo dade knigitel Ivan to whom gave books +DEF 'As for Ivan, to whom did (he) give the books?' (Jaeger, 2004, p. 209) (68) Marija *{ja) obicat decata. Maria her love3.PL childrenDEF PL 'As for Maria, the children love her.' (Jaeger, 2004, p. 215) As Jaeger (2004) points out, clitic doubling in Bulgarian indicates topicality (in the sense of 'aboutness'), and clitic doubling of wh-elements is crucially motivated by the indication of topicality. Fronting of clitic doubled wh-elements is thus subject to the same restrictions as fronting of topic elements in non-interrogative clauses. As I have shown for the high wh-elements in embedded interrogatives of Cimbrian, topicality is to be understood here as a contextual restriction that is stronger than D-linking. The Bulgarian examples provide independent evidence for the assumption that whelements can have specific topical properties and then occupy a higher syntactic position in the left periphery than pure wh-operators. Although we have not been able to figure out the exact left-peripheral position of non-topical wh-elements in Cimbrian embedded interrogatives, the evidence clearly suggests that there must be a high topic-like projection in the left periphery of the clause where topical wh-elements move to. We have seen that in a V2 language like Cimbrian, this position must be higher than ForceP and thus gives rise to a further configuration in which the linear V2 restriction is violated.

94

Günther Grewendorf

References Axel, Κ. (2007). Studies on Old High German Syntax - Left Sentence Periphery, Verb Placement and Verb-Second. (Linguistik Aktuell/Linguistics Today, vol. 112). Amsterdam: John Benjamins. Bayer, J. (2008). What is Verb Second? Manuscript. University of Konstanz. Bayer, J. & P. Suchsland (1997). Expletiva und leere Subjekte im Deutschen. Groninger Arbeiten zur germanistischen Linguistik 41. 12-38. Belletti, Α. (2008). Answering Strategies: New Information Subjects and the Nature of Clefts. Manuscript. University of Siena. Benincà, P. & C. Poletto (2004). Topic, Focus and V2: Defining the CP Sublayers. In: L. Rizzi (ed.) The structure of CP and IP. (Oxford Studies in Comparative Syntax). Oxford University Press. 52-75. den Besten, Η. (1983). On the Interaction of Root Transformations and Lexical Deletive Rules. In: W. Abraham (ed.) On the Formal Syntax of the Westgermania, Papers from the 3rd Groningen Grammar Talks. (Linguistik Aktuell/Linguistics Today, vol. 3). John Benjamins. 47-138. Bhatt, R. & J. Yoon (1991). On the Composition of COMP and Parameters of V2. In: D. Bates (ed.) Proceedings of the 10th West Coast Conference on Formal Linguistics. Chicago, IL: Chicago University Press. 41-53. Bhatt, R. M. (1999). Verb Movement and the Syntax of Kashmiri. (Studies in Natural Language & Linguistic Theory, vol. 46). Dordrecht: Kluwer. Bidese, E. (2009). Die diachronische Syntax des Zimbrischen. (Tübinger Beiträge zur Linguistik, vol. 510). Tübingen: Gunter Narr. Browning, M. (1996). CP Recursion and that-t Effects. Linguistic Inquiry 27. 237-255. Cardinaletti, Α. (2008). On a (Wh-) Moved Topic in Italian, Compared to Germanic. Manuscript. University of Venice. Chomsky, N. (2008). On Phases. In: R. Freidin, C. Otero, & M.-L. Zubizaretta (eds.) Foundational Issues in Linguistic Theory. Cambridge, MA: MIT Press. 133-166. Fuß, E. (2008). Word Order and Language Change. On the Interface between Syntax and Morphology. Habilitationschrift. University of Frankfurt. Grewendorf, G. (2009). The Left Clausal Periphery: Clitic Left Dislocation in Italian and Left Dislocation in German. In: B. Shaer, P. Cook, W. Frey, & C. Maienborn (eds.) Dislocated Elements in Discourse: Syntactic, Semantic, and Pragmatic Perspectives. London: Routledge. 49-94. Grewendorf, G. (2010). Wh-movement as Topic Movement. In: L. Brugè (ed.) Functional Heads. Oxford: Oxford University Press, to appear. Grewendorf, G. & C. Poletto (2009). Hidden Verb Second: The Case of Cimbrian. In: M. Putman (ed.) German-language Speech Islands: Generative and Structural Approaches. Somerville, MA: Cascadilla. to appear. Grewendorf, G. & C. Poletto (2010). Wh-Questions and Subject Inversion in Verb Second Languages. Manuscript, forthcoming. Grohmann, Κ. K. (1997). On Left Dislocation. Groninger Arbeiten zur germanistischen Linguistik 40. 1-33.

On the Typology of Verb Second

95

Haegeman, L. (1996). Verb Second, the Split CP and Null Subject in Early Dutch Finite Clauses. Geneva Generative Papers 4. 133-175. Haider, H. (2009). Gardenpaths mit Marga. Talk delivered at the workshop in honor to Marga Reis. Tübingen. Jaeger, T. F. (2004). Topicality and Superiority in Bulgarian WH-Questions. In: O. Arnaudova, W. Browne, M. L. Rivero, & D. Stojanovic (eds.) Annual Workshop on Formal Approaches to Slavic Languages. The Ottawa Meeting 2003, Ann Arbor: Michigan Slavic Publications. Kaiser, G. A. (2002). Verbstellung und Verbstellungswandel in den romanischen Sprachen. (Linguistische Arbeiten, vol. 465). Berlin: De Gruyter. Kayne, R. S. (1994). The Antisymmetry of Syntax. (Linguistic Inquiry Monographs, vol. 25). Cambridge, MA: MIT Press. Ohl, P. & A. Korn (2006). Performanzbasierte und parametrische Wandel in der linken Satzperipherie des Persischen: Der Subordinationsmarkierer ke und die Interrogativpartikel aya. Die Sprache 46. 137-202. Panieri, L. (2006). Bar Urnen ζ schraiba un zo reda az be biar. Grammatik der zimbrischen Sprache von Lusérn. Kulturinstitut Lusérn. Platzack, C. (1986). The Position of the Finite Verb in Swedish. In: H. Haider & M. Prinzhorn (eds.) Verb Second Phenomena in Germanic Languages. (Publications in Language Sciences, vol. 21). Dordrecht: Foris. 27-47. Poletto, C. (2000). The Higher Functional Field: Evidence from Northern Italian Dialects. (Oxford Studies in Comparative Syntax). Oxford: Oxford University Press. Poletto, C. (2002). On V2 and V3 Sequences in Rhaetoromance. Online Publication http://www.meertens.nl/books/synmic/pdf/poletto.pdf. Poletto, C. (2009). OV im Altitalienischen (und in anderen romanischen Sprachen). Talk delivered at the University of Frankfurt. Rizzi, L. (1997). The Fine Structure of the Left Periphery. In: L. Haegeman (ed.) Elements of Grammar. Dordrecht: Kluwer. 281-337. Rizzi, L. (2004). Locality and the Left Periphery. In: A. Belletti (ed.) Structures and Beyond - The Cartography of Syntactic Structures. (Oxford Studies in Comparative Syntax). Oxford: Oxford University Press. 223-251. Rizzi, L. (2006). On the Form of Chains: Criterial Positions and ECP Effects. In: L. L.-S. Cheng & N. Corver (eds.) Wh-Movement. (Current Studies in Linguistics). Cambridge, MA: MIT Press. Rizzi, L. & U. Shlonsky (2007). Strategies of Subject Extraction. In: U. Sauerland & H.-M. Gärtner (eds.) Interfaces + Recursion = Language? (Studies in Generative Grammar). Berlin: De Gruyter. 297-328. Roberts, I. (2004). The C-System in Brythonic Celtic Languages, V2, and the EPP. In: L. Rizzi (ed.) The Structure of CP and IP. (Oxford Studies in Comparative Syntax). Oxford University Press. 297-328. Robinson, O. (1997). Clause Subordination and Verb Placement in the Old High German 'Isiodor' Translation. Heidelberg: C. Winter. Roussou, A. (2000). On the Left Periphery. Modal Particles and Complementizers. Journal of Greek Linguistics 1. 65-94.

96

Günther Grewendorf

Stowell, T. & F. Beghelli (1994). The Direction of Quantifier Movement. Paper presented at the GLOW conference, Vienna. Taraldsen, K. T. (1986). On Verb Second and the Functional Content of Syntactic Categories. In: H. Haider & M. Prinzhorn (eds.) Verb Second Phenomena in Germanic Languages. (Publications in Language Sciences, vol. 21). Dordrecht: Foris. 7-25. Thráinsson, H. (1979). On Complementation in Icelandic. (Outstanding Dissertations in Linguistics). London: Taylor & Francis. Vance, B. S. (1997). Syntactic Change in Medieval French. (Studies in Natural Language and Linguistic Theory, vol. 41). Heidelberg: Springer. Watanabe, A. (1993). Agr-Based Case Theory and its Interaction with the Α-Bar System. Ph.D. thesis. MIT, Cambridge, MA.

Movement from Verb-Second Clauses Revisited Gereon Müller

1. Introduction Extraction from declarative complement clauses in German reveals a curious pattern: Whereas verb-final complement clauses headed by dass ('that') permit movement both into a higher verb-second and a higher verb-final clause, complement clauses in which verb-second has applied only permit movement into a higher verb-second clause again, but not movement into a higher verb-final clause. This systematic pattern has been addressed in a number of approaches of various kinds in the last decades, with varying degrees of success, but it seems fair to conclude that a conceptually simple analysis on the basis of minimalist program (see Chomsky, 2001, 2008) is still outstanding. The goal of the present paper is to show that such an analysis is readily available if certain insights of Staudacher's (1990) approach in terms of barriers are incorporated into the phase-based approach to Condition on Extraction Domain (CED) phenomena developed in Müller (2010), and if verb-second clauses are derived by reprojection movement. In a nutshell, the restriction on movement from verb-second clauses will be shown to follow from Chomsky's (2001) Phase Impenetrability Condition (PIC) because an edge feature that is needed to trigger movement from the verb-second complement to the next phase edge cannot be inserted on the matrix verb. I will proceed as follows. Section 2 introduces the relevant data and gives a very concise, and certainly incomplete, overview over the existing literature on the phenomenon. Section 3 sketches Staudacher's (1990) approach. Section 4 lays out the main assumptions of the phase-based analysis of CED effects developed in Müller (2010). Section 5 introduces an approach to verb-second movement that is based on the concept of reprojection; the approach combines features of the analyses in Fanselow (2008) and Georgi & Müller (2010). Section 6 then shows that the ban on movement from verb-second clauses into verb-final clauses can be derived given the assumptions laid out in sections 4 and 5, in a way that integrates some of Staudacher's original insights. Finally, section 7 addresses some further issues.

0

For listening to one of two ad hoc-presentations of the material here and then commenting on it, I am most grateful to Anke Assmann, Doreen Georgi, Fabian Heck, Johannes Hein, Stefan Keine, Marc Richards, and Philipp Weisser. In addition, this paper owes an obvious debt to Peter Staudacher's work on the topic.

98

Gereon Müller

2. The Problem 2.1 Data Two types of finite declarative clauses can be embedded under certain kinds of verbs (often bridge verbs, though the correlation is far from perfect) in German: (i) clauses headed by a complementizer dass ('that'); (ii) verb-second clauses with finite V in the C position and some XP in SpecC. Both types of complements as such appear to be transparent for w/i-movement to SpecC. Wî-movement from a dass clause may go to a dass clause or to a verb-second clause; see (1-ab).1 In contrast, as shown in (1-cd), whmovement from a verb-second clause may only end up in a verb-second clause again (see Tappe, 1981; Haider, 1984; Reis, 1985).2 (1)

a.

(Ich weiß nicht) [CPι wern (dass) du meinst [CP2 t¿ dass sie t¿ getroffen I know not whom that you think that she met hat]] has b. [CPi Weni meinst du [CP2 t' dass sie t, getroffen hat ]] ? whom think you that she met has c. [CP] Wenj meinst du [CP2 t- hat sie t¿ getroffen ]] ? whom think you has she met d. *(Ich weiß nicht) [CPi wen, (dass) du meinst [CP2 t¿ hat sie t¿ getroffen ]] I know not whom that you think has she met

The same restriction as in (1-d) holds when movement from SpecV2 to Specdass is followed by further vWi-movement, as in (2). (2)

*[CP 0

Weni glaubt er [cPi t" dass du meinst [cp2 t¿ hat sie getroffen ]]] ? whom believes he that you think has she met

Furthermore, the restriction is also active when the moved item is a topic or relative pronoun, as in (3-ab) and (4-ab), respectively.3

1

A complementizer of CPi in sentences like (1-a) must then be deleted in Standard German, but not in dialects and colloquial varieties. Following Pesetsky (1998), I assume that complementizer deletion is a PF phenomenon in languages like German and English, with a that/dass complementizer present in syntax proper.

2

The representations in ( 1 ) presuppose that long-distance movement proceeds successive-cyclically, via intermediate SpecC positions, and leaves traces (/) in the positions it targets on its way to the ultimate landing site. At this point, these assumptions are mainly for convenience; they are shared by Staudacher's approach sketched below, though.

3

(3-a), (4-a) and, perhaps to a lesser extent, (1-b) are marked, or even not possible at all, in some varieties of German. In what follows, I will only be concerned with varieties that permit extraction from dass clauses across the board.

Movement from Verb-Second Clauses (3)

a.

b.

(4)

a. b.

[cp„ Den the hat ]]] has *[CP0 Den the getroffen met die the *die the

Revisited

99

FritZi glaubt er [cp! t" dass du meinst [cp2 t' dass sie getroffen Fritz believes he that you think that she met

FritZi glaubt er [cp t t" dass du meinst [cp 2 t' hat sie Fritz believes he that you think has she ]]]

Frau [CP! die¿ (wo) du woman whom REL you Frau [CP, diet {wo) du woman whom REL you

meinst [cp 2 t': dass sie t¿ getroffen hat ]] think that she met has meinst [CP2 t· hat sie U getroffen ]] think has she met

2.2 Analyses The data have proven remarkably robust over the years, and many attempts have been made to account for the asymmetry involved. First, it has been suggested that illicit instances of movement from a verb-second clause into a verb-final dass clause reveal a locality effect. On this view, the verb-second clause acts as a barrier in ( 1-d) (as well as in (3-b) and (4-b)). This may or may not require some extra assumption about (1-c), where the barrier status of the verb-second clause seems to be voided. Proposals of this type include Staudacher ( 1990), Stemefeld (1989) on the one hand, and Reis (1996), Müller (2004), and Heck (2010) on the other. The former type of locality approach assigns verbsecond clauses barrier status if they are embedded by a verb-final clause (as in (1-d)) but not if verb-second movement takes place in the higher clause (as in ( 1 -c)). The latter type of locality approach treats verb-second clauses as barriers throughout (i.e, in (1-d) and ( 1 -c)), and attributes the well-formedness of the string in ( 1 -c) to an alternative option for analysis: (1-c) is assumed to involve a special kind of ("integrated") parenthesis (meinst du in (1-c); also cf. Kiziak (2007)). This option is assumed to be non-existent in (1-d). Second, it has been proposed that the asymmetry in (4) follows from directionality constraints on movement of the kind postulated in Kayne (1984) (via 'g-projections') and Köster (1987) (via 'global harmony'); see Müller (1989, ch. 6) and Haider (1993b)). The basic idea here is that the apparent clash in directionality of government (or selection) by an embedded V in a verb-second position and a matrix V in a verb-final position blocks movement. Third, the data have been approached in terms of constraints against improper movement; see Haider (1984), Stechow & Sternefeld (1988, ch. 11.7), Sternefeld (1992), Müller & Sternefeld (1993), Williams (2003). These approaches all differ substantially from one another (except perhaps for Sternefeld ( 1992) and Williams (2003), which both rely on the version of the cycle proposed in Williams (1974)), but they share a common core: SpecV2 and SpecC positions are viewed as sufficiently different to be able to block movement from one to the other as improper. Furthermore, an asymmetry must be built into the theory of improper movement to make a mixing of landing sites possible in ( 1 -b)

100

Gereon Müller

(movement may take place from SpecC to SpecV2) but not in (1-d) (movement may not proceed from SpecV2 to SpecC). Fourth, I have tentatively pursued an approach that relies on shape conservation (in the sense of Williams (2003)) in Müller (2003). The idea here is that there is a general but violable constraint that requires left edges of CP to have an identical shape. This constraint is satisfied in (1-a) and (1-c) but not in (1-b) and (1-d). Again, an asymmetry needs to be imposed, and this is achieved by local, cyclic, bottom-up optimization of CPs. In this approach, it turns out that the shape conservation constraint can be violated with verb-second movement in CPi in (1-b) because, given that CP2 has already been optimized, other options that would be shape-preserving (specifically, failing to carry out verb-second) violate higher-ranked constraints. In contrast, shape conservation cannot be violated in (1-d) because, given that CP2 (with verb-second movement) is already in place, verb-second movement can and must apply in CPi because this will give rise only to a violation of a lower-ranked constraint (the one that precludes movement which is not feature-driven). Finally, there is a strategy that solves the problem in a very simple way, and that could be referred to as data denial: Cavar (1996) and Fanselow & Mahajan (1996) claim that extraction from verb-second clauses into dass clauses as in (1-d) is possible after all. 4 This last view is at variance with the clear robustness that the effect has exhibited over the last thirty years or so. For this reason I will disregard this option in what follows. I cannot possibly make an attempt here to discuss all of these approaches in detail and highlight their merits as well as their shortcomings. I do believe, however, that virtually all of them rely on assumptions that are not really independently motivated, and that very often turn out to be construction-specific upon closer inspection. Furthermore, it can be noted that none of them is compatible with basic tenets of the minimalist program, where, e.g., there can be no designated locality constraints employing notions like barrier, no constraints that mention directionality, and no special constraints blocking improper movement; and where shape conservation effects must be treated as an epiphenomenon. Thus, if one adopts a minimalist perspective, as I will do here, it seems fair to conclude that the problem in ( 1 ) has not yet received a satisfying solution. That said, I think that many of the basic insights of Staudacher (1990) (and, to some extent, also of Sternefeld (1989)) can be integrated into a new phase-based approach to the ban on extraction from verb-second clauses into verb-final clauses in German that does without construction-specific assumptions and meets basic minimalist demands; and I will try to show this below. To this end, I first summarize the proposal in Staudacher (1990) in the next section.

3. Staudacher's (1990) Approach Staudacher (1990) presents two analyses, which share a number of properties (in particular, they are both based on the hypothesis that there is a barrier present in (1-d) that blocks 4

In a similar vein, Shin (1988, p. 253) claims that potential problems with the construction in (1-d) are not syntactic but pragmatic in nature.

Movement from Verb-Second Clauses Revisited

101

movement, and that verb-second in the matrix clause as in (1-c) opens up this barrier via transitivity of indexing) but differ in others (the first analysis, pp. 330-334, assumes the matrix VP to be a minimality barrier, whereas the second analysis, pp. 334-338, assumes the embedded verb-second clause to be a barrier because of non-selection). In what follows, I will focus on the second version of the approach since it is this version that the analysis to be developed below will resemble most. The first thing to note is that verb-second clauses that are embedded by a verb-final dass clause (as in (1-d)) are strong islands, in the sense that they uniformly block all kinds of extraction and do not distinguish between arguments and adjuncts; compare the cases of licit and illicit adjunct movement in (5-ab), which are fully parallel to the argument movement cases in ( 1 -c), ( 1 -d). (5)

a.

Ich I b. *Ich I

weiß know weiß know

nicht (χρ, wie, du meinst [cp2 V dass das U gehen soll ]] not how you think that that work should nicht [cpj wie,, du meinst [cp2 t,' soll das t¿ gehen ]] not how you think should that work

The Condition on Extraction Domain in (6) (CED; see Huang (1982), Chomsky (1986)) blocks both argument and adjunct movement from non-complements (where complements are sisters of lexical items). (6)

Condition on Extraction Domain (CED): a. b.

Movement must not cross a barrier. An XP is a barrier iff it is not a complement.

Assuming the CED to be responsible for the ill-formedness of (1-d), (5-b) and similar examples, we are led to looking for a barrier in these contexts. 5 A straightforward application of the CED in (6) to the data in (1) and (5) faces two basic problems. First, embedded verb-second clauses as in (1-c), (5-b)look like genuine complements of V - a verb like meinen ('think') subcategorizes for a complement and assigns a 0-role to it, and in the absence of an obvious alternative candidate, there is no good reason to assume that CP2 is not the complement that the matrix verb is looking for, irrespectively of whether verb-second movement has or has not applied in CP2. And second, given the notion of barrier in (6), it is completely unclear how verb-second movement in the matrix clause could remove the barrier status of the embedded verb-second CP. To solve the first problem, Staudacher proposes that the concept of barrier is to be refined in such a way that complement status alone does not suffice for an XP to avoid barrier status: Being a complement is viewed as a necessary but not yet sufficient condition for non-barriers. His revised notion of barrier is given in (7). 5

This version of the CED combines aspects of the proposals in Huang (1982), Chomsky (1986), Cinque ( 1990), and Manzini (1992). Staudacher ( 1990) does not actually invoke the CED; rather he embraces the more complex system incorporating the CED that Chomsky (1986) develops on the basis of Lasnik & Saito (1984). Where it does not affect the main points to be made, I tacitly simplify Staudacher's analysis in various respects, and also adjust it to more current terminology.

102 (7)

Gereon Müller Condition on Extraction Domain (CED, based on Staudacher (1990)): a. b.

Movement must not cross a barrier. An XP is a barrier iff it is not head-marked.

Head-marking is a somewhat stricter notion than being a complement because it requires that, in addition to being a complement of a lexical item, selection of (or co-indexing with) the head of XP must take place: 6 (8)

Head-marking (Staudacher (1990, 336)): a head-marks β iff a. b.

α is a lexical item, and α is a sister of β that selects or is co-indexed with the head of β.

The basic idea is that CPs with a complementizer dass are always head-marked by the matrix verb (because the verb selects the head of the dass clause), whereas embedded verbsecond clauses are not. Although embedded verb-second clauses are selected complements, their verb-second head, by assumption, is not selected by the matrix verb, and thus turns CP into a barrier. The assumption that heads of verb-second CPs are not selected receives independent justification - so Staudacher argues - from the observation that these heads are not filled by lexical material before verb-second movement applies. More generally, if dass complements are head-marked by V (via head-selection) and verb-second complements are not head-marked by V (via head-selection), this captures the insight that dass complements have a closer relationship with the matrix verb than verb-second complements, and it also captures the observation that embedded verb-second clauses, while having complement status as such, also exhibit several root properties, which also partially accounts for the conditions under which they are licensed in the first place. 7 Thus, a CP with a verb-second head is a barrier due to a lack of head-marking, in contrast to a CP with a lexical complementizer. This accounts for the contrast between examples like (1-a), (1-b) on the one hand (i.e., movement from dass clauses into dass clauses or into verb-second clauses), and examples like (1-d) on the other (i.e., movement from verb-second clauses into dass clauses). What remains to be explained is why verb-second movement in the matrix clause removes the barrierhood of an embedded CP with a verb-second head, as in (1-c). At this point, the option of bringing about head-marking by co-indexing (as an alternative to head selection) envisaged by (8) becomes relevant. After verb-second movement in the matrix clause, the following co-indexing relations obtain. First, since movement, by assumption, implies chain formation, and members of the same chain share an index, the verb-second head meinst and its trace in the V position are co-indexed. Second, for the same reason, the w/i-phrase wen is co-indexed both with its original trace in the object position of the embedded clause and with the intermediate trace in the embedded SpecC position. Third, 6

(8) is a simplified version of Staudacher's definition of head-marking, but it suffices for present purposes - the complications Staudacher introduces in the "proper" definition in Staudacher (1990, 336) (and also already in Staudacher (1987, 13)) do not bear on the issue of extraction from V/2 clauses.

7

On root properties of embedded verb-second clauses and similar constructions, see Hooper & Thompson (1973), Meinunger (2004), and Bentzen (2009), among others.

Movement from Verb-Second Clauses Revisited

103

and this is the crucial step in the argument, Staudacher assumes that there is a process of abstract agreement between specifier and head, and that this agreement relation is also indicated by co-indexing. Given that verb-second movement is movement to C, from this it follows that the indices of the moved wft-phrase in the matrix SpecC position and the moved verb in the verb-second position in the matrix clause are also identical. Fourth, by the same reasoning, the indices of the intermediate wA-trace in the embedded SpecC position and the embedded verb-second head are also identical as a consequence of abstract specifier/head agreement. And finally, by transitivity, this implies that the indices of the V trace of the matrix verb and the verb-second C head of the embedded clause are also identical in cases like (1-c). Consequently, joint verb-second movement and w/i-movement in the matrix clause leads to head-marking of the embedded verb-second clause, and hence to a removal of barrier status there. This explains why ( 1 -c) can respect the CED. Note that this way out is not available in (1-d) as long as it is assumed that co-indexing cannot arise accidentally: In (1-d), verb-second does not take place in the matrix clause, so the wA-item in the matrix SpecC position and the matrix verb are not co-indexed, and the transitivity chain is broken; as a result, matrix V cannot head-mark the embedded verb-second clause. Summarizing, the gist of Staudacher's proposal is that embedded verb-second clauses are barriers due to a lack of head-marking by head selection, but the barrier status can be removed by combined verb-second and wA-movement in the matrix clause because this results in head-marking by head co-indexing. I take this approach to be elegant and intuitively appealing, but it seems clear that it cannot be maintained as such under more recent minimalist assumptions as they are laid out in Chomsky (2001, 2008) and much related work. For one thing, on this view all constraints must either qualify as principles of efficient computation, or they must be motivated by properties of the interface, and the CED does not meet either requirement. To the extent that it makes the right predictions, the CED should therefore be derived as a theorem from more basic assumptions. This means that a different, arguably less stipulative reason must be sought that predicts barrier status of verb-second complement clauses in (1-d). For another thing, Staudacher's account of the removal of barrier status of verb-second complement clauses by verb-second movement in the matrix clause is not available under minimalist assumptions for a number of reasons. First, this account relies on indices and transitivity of co-indexing, but, following Chomsky's arguments, co-indexing is not a technical device that is legitimate under minimalist assumptions (let alone inhomogeneous co-indexing of the type that is needed in Staudacher's approach). Second, there is no room for a concept like specifer/head agreement if all Agree relations have their source on a lexical item (more precisely, a probe feature) and imply c-command by this item. And third, as it stands, Staudacher's account presupposes either a representational syntax or a derivational syntax that has look-ahead capacity (because the question of whether extraction from the lower clause is possible can only be answered once the edge domain of the matrix clause is reached); but phase-based minimalist analyses are characterized by the property of being (a) strictly derivational and (b) without look-ahead capacity. It is not clear whether the approach could be transferred into a derivational system without look-ahead (or back-tracking).

104

Gereon Müller

In view of this state of affairs, Staudacher's approach must either be abandoned or modified appropriately. In what follows, I will show that a new analysis that maintains main insights of Staudacher's approach suggests itself on the basis of the approach to CED phenomena in Müller (2010) (section 4) and a version of the approaches to reprojection movement in Fanselow (2008) and Georgi & Müller (2010) (section 5). More specifically, following Staudacher (1990), I will develop an analysis in section 6 that rests on the idea that embedded verb-second clauses are barriers because they are not as strictly selected as dass clauses, and that the barrier status can be removed by verb-second movement in the matrix clause because such movement in effect produces (more or less, as in the head selection/co-indexing disjunction in (8-b)) the same configuration as it is present with dass clauses from the start.

4. CED Effects: A Phase-based Approach 4.1 Background and Assumptions In Müller (2010), I argue that a version of the CED (which can be shown to be an empirically viable alternative to the one in (6)) follows from the Phase Impenetrability Condition (PIC) proposed in Chomsky (2000, 2001, 2008); in what follows I sketch the outlines of this approach. The PIC can be defined as in (9). (9)

Phase Impenetrability Condition (PIC): The domain of a head X of a phase XP is not accessible to operations outside XP; only X and its edge are accessible to such operations (where the edge of X includes specifier(s) of X and X).

Chomsky takes the PIC to contribute to efficient computation by reducing search space in derivations. Apart from that, one of the PIC's main effects is that it forces successivecyclic movement via phase edges; such movement is possible because edge features that drive it can be inserted on phase heads. It turns out that CED effects follow as PIC phenomena if the following four assumptions are made: First, all syntactic operations are driven by features of lexical items. Second, these features are ordered on lexical items. Third, all phrases are phases. And finally (and most importantly), edge features that trigger intermediate movement steps can be added (in minimal violation of Chomsky's (2001) Inclusiveness Condition) only as long as the phase head is still active. The version of the CED that can be derived from the PIC under these assumptions is (10). ( 10)

Condition on Extraction Domain (to be derived from the PIC): a. b.

Movement must not cross a barrier. α is a barrier if the operation that has merged α in a phase Γ is the final operation in Γ.

Let me briefly explicate (but not try to justify or provide background information for) the four assumptions that must be made. First, all syntactic operations are feature-driven.

Movement from Verb-Second Clauses Revisited

105

More specifically, there are two types of features that drive operations, viz., structurebuilding features (edge features, subcategorization features) that trigger (external or internal) Merge operations and are accompanied by bullets ([*F·]), and probe features that trigger Agree operations and are accompanied by asterisks ([*F*]). 8 Second, features on lexical items are ordered. For instance, if a head X subcategorizes for three items A, Β and C, where C is to become X's complement, Β is to become X's inner specifier, and A is to become X's outer specifier, X will inherently be equipped with a feature list [*C·] >- [ · Β · ] >- [·Α·]. Thus, structure-building features are located on a stack belonging to a lexical item. This does not only hold for subcategorization features that trigger external Merge; it is also the case with movement-inducing features that trigger internal Merge. For instance, if the lexical item X contains a fourth structurebuilding feature Ζ in addition which triggers movement of some designated item, this will show up most deeply embedded on the feature list: [*C·] >- [ · Β · ] >- [ · Α · ] χ [·Ζ·], In addition to the stack for structure-building features, a lexical item has a second stack for probe features. A Last Resort condition ensures that a syntactic (Merge or Agree) operation can only take place if it discharges (and thereby removes) a structure-building or probe feature; and only those features are accessible at any given step of the derivation that are on the top of a stack; see ( 11 ). (11)

Last Resort (LR): a. b.

Every syntactic operation must discharge (and delete) either [aFa] or [*F*]. Only features on the top of a feature list are accessible.

The third assumption is that all phrases are phases. As a consequence, w/z-movement must proceed via every XP edge domain on its way to its ultimate target position (the C[ eu ,h.] node that attracts it), given the PIC; and similarly for all other movement dependencies. This assumption is necessary to derive (10) in full; if one were to assume that only CP and vP are phases, only (last-merged) specifiers of CP and VP would be predicted to be barriers. Finally, a restriction needs to be imposed concerning the insertion of edge features that drive intermediate movement steps in Chomsky (2000, 2001). The first thing to note is that edge feature insertion cannot be free (or that having edge features freely available is an intrinsic property of phase heads; see Chomsky (2007, 2008)). Chomsky (2000, 2001) suggests that edge features do not come for free, and that edge features can only be inserted on a phase head after the phase is otherwise complete, i.e., after the phase head has become inert (by having triggered all the operations that it can trigger as a consequence of its inherent properties). Crucially, suppose that the opposite is correct: Edge features can only be inserted before the phase is otherwise complete, i.e., before the phase head has become inert (by having discharged all its structure-building and probe features). This can be formulated as the Edge Feature Condition (EFC) in (12-a). Given 8

The notation follows Heck & Müller (2007), which in tum combines various other notation systems that can be found in the literature. Note that whereas structure-building features for external Merge (i.e., basicgeneration) are categorial, structure-building features for internal Merge (i.e., movement) need not be, and often are not - thus, vWi-movement is effected by [ a w h a ] , topicalization by ( [ a t o p · ] , and so on.

106

Gereon Müller

reasonably standard assumptions about strict cyclicity, edge feature insertion can only go to the top of the existing stack of structure-building features, as in (12-b). This ensures a last-in/first-out property of feature stacks. ( 12)

Edge Feature Condition (EFC): An edge feature [ · Χ · ] can be assigned to the head 7 of a phase only if (a) and (b) hold: a. b.

7 has not yet discharged all its structure-building or probe features. [ · Χ · ] ends up on top of 7's list of structure-building features.

4.2 Analysis With these assumptions in place, it remains to be shown that the CED in (10) now follows as a theorem from the PIC. Ignoring probe features for Agree for the moment, the reason why last-merged specifiers act as barriers is, essentially, this: Suppose that some item a has made it to the edge domain of some phase β, and β is now merged as the last operation induced by a phase head 7 ' s structure-building features; suppose further that a eventually needs to undergo movement beyond 7 because the phase head that bears the feature [ · Ζ · ] which requires successive-cyclic movement of a is not yet part of the tree created so far. The dilemma that arises at this point is that, when β (including a in its edge domain) is merged, 7 ' s stack of structure-building features is empty - the phase head has become inert, and the EFC accordingly precludes edge feature insertion. This means that a cannot move from a β specifier to the next higher 7 specifier. Assuming a non-recursive definition of phase edges (such that the specifier of a specifier of 7 is not part of the edge of 7), subsequent extraction of a across the phase headed by 7 will invariably violate the PIC - a is too deeply embedded in the phase 7 (it is still part of an intervening phase β). The only conceivable way out of this dilemma would be for edge feature insertion on 7 to precede Merge of β (including a), so that 7 is still active (because it has not yet discharged its final structure-building feature for β). However, this also does not help: Either the newly inserted edge feature lands on top of 7 ' s stack of structure-building features; then, given LR, it is discharged again (attracting something from within the 7 phase established so far) before β is merged. Alternatively, the edge feature is inserted below the final inherent structure-building feature; however, this violates requirement (12-b) of the EFC. The three failed attempts of establishing an edge feature ([·Χ·]) on 7 to extract a out of a last-merged specifier β are illustrated in (13). ( 13)

Last-merged specifiers as barriers: a.

Edge feature insertion follows specifier feature discharge: r7: 7:

b.

[·β·] 0 [·Χ·]

violates (12-a)

Edge feature insertion precedes specifier feature discharge, version 1 :

107

Movement from Verb-Second Clauses Revisited

violates (12-b) c.

Edge feature insertion precedes specifier feature discharge, version 2: —> does not help because of ( 11 -b)

In contrast, complements do not have to be barriers. Suppose that the list of structurebuilding features is not yet empty when a subcategorization feature for a complement -

[·β·]

[·£·] [·Χ·]

violates nothing

ί·β·] 0

Note that because of the last-in/first-out property of the EFC/LR-based approach, intermediate movement steps to phase edges must take place before a (final) specifier is merged. This results in structures that look like (inherently acyclic) tucking in (Richards, 2001 ) has applied; but it has not: All movement steps extend the tree. This systematic effect (which I call the Intermediate Step Corollary) is illustrated for successive-cyclic movement of some DP 2 across a VP phase and a vP phase (both of which have DP specifiers) in (15).

108 (15)

Gereon Müller Intermediate movement steps: DP 2 ...

As argued in Müller (2010), this PIC-based approach to CED phenomena predicts that (last-merged) subjects are barriers both in Specv and in SpecC; that adjuncts are barriers (assuming that they are last-merged specifiers of special functional projections); and that indirect objects bearing dative are barriers (assuming that they are last-merged in SpecV or as the specifier of some projection between vP and VP). Furthermore, the approach predicts what I call a melting effect: Subjects, indirect objects and other categories that are normally barriers because they are last-merged in their respective phases should cease to be a barrier if the phase head has an additional structure-building feature [ · Ζ · ] triggering movement to an outer specifier. The reason is that in this situation, an edge feature should be insertable on the phase head after a specifier is merged, given that the phase head is still active because it has some feature [ · Ζ · ] left that will subsequently trigger internal Merge. I argue that melting effects do indeed occur with local scrambling to Specv and SpecV in German and Czech, and ensure transparency of otherwise opaque subjects and indirect objects. A German example that illustrates the transparency of an in-situ subject for wA-movement (here giving rise to was für split) after local scrambling of the object to an outer Specv position is (16-b); in contrast, (16-a) shows that the subject is opaque in its in situ position (and there is evidence that the subject must, or at least may, stay in situ, in vP, in this context). (16)

a. *WaSi what b. Was, what

haben [opk U für Bücher ] [opj den Fritz ] beeindruckt ? have for booksNOM the FritzACC impressed haben [qpj den Fritz ] [dp*, t i für Bücher] tj beeindruckt Ί have the FritzACC for booksNOM impressed

Still, the system developed so far raises an obvious question: How can complements (i.e., first-merged items) that are also last-merged in their phases evade barrier status? It is at

Movement from Verb-Second Clauses Revisited

109

this point that probe features of phase heads become relevant: Probe features can keep a head active, and accessible for edge feature insertion, in the same way as structurebuilding features do. Thus, suppose that a phase head 7, after having discharged its sole structure-building feature and thereby merged with a complement β, still has a probe feature [*F*] left that needs to be checked with a matching [F] feature that is either on β, or in some category included in β (evidently, this feature [*F*] could not be checked before β has entered the structure). In this case, an edge feature can be inserted after 7 has merged with β, and before [*F*] triggers an Agree operation with a matching feature in the complement β\ and this means that some a can be extracted out of β even though β is not just first-merged, but also last-merged in 7· 9 (17) illustrates why last-merged complements do not have to be barriers. They are barriers if there is an empty stack of probe features on 7 (see (17-a)), but they are not barriers if some probe feature is present (see (17-b)). ( 17)

Last-merged complements as non-barriers: a.

Edge feature insertion follows complement feature discharge, no probe feature: —> —

b.

7: Ύ7:

[·β·]

0 [·Χ·]

violates (12-a)

Edge feature insertion follows complement feature discharge, with probe feature: Ύ· Ύ· Ί'·

[·β·] [*F*] [*F*] [.Χ.] [*F*]

violates nothing

At this point the question arises which operation is carried out next with the phase head 7 in (17-b): Is it discharge of the probe feature or is it discharge of the structure-building feature? Note that what might at first sight perhaps look as the most obvious strategy viz., to immediately get rid of the edge feature that has just been inserted by carrying out an intermediate movement step of a out of β - is problematic because it incurs a violation of an elementary principle of derivational structure-building: the Strict Cycle Condition (SCC) (cf. Chomsky (1973)). A simple version of the SCC is given in (18).

9

Depending on whether Agree with a probe feature on 7 may also affect some other category included in β, or only (the head of) β, the PIC may or may not have to be relaxed for Agree operations (as proposed by Boskovic (2007)). Alternatively Agree might then be viewed as taking place successive-cyclically; see Legate (2005). Something to this effect would seem to be required independently for constructions like agreement of Τ with nominative objects in Icelandic and the analysis of long-distance agreement in languages like Hindi (but cf. Chomsky (2001) and Bhatt (2005), respectively, for alternatives that take these constructions to argue for a less restrictive notion of phase in general).

110

Gereon Müller

( 18)

Strict Cycle Condition (SCC): Within the current domain ζ, a syntactic operation may not exclusively apply to positions that are included within another domain π that is dominated by ζ.

If the phase head 7 is merged with a that has undergone movement out of β, with a therefore becoming 7 ' s specifier, subsequent discharge of the probe feature [*F*] on 7 with some item in β will have to violate the SCC because this latter operation exclusively applies to positions that are included in 7' (= π in ( 18)), which is dominated by the current domain 7 P including a (= ζ in (18)). An a priori possible way out might be to carry out Agree between [*F*] on 7 and a in the specifier position (or with some category included in a ) ; however, this would violate the c-command requirement on Agree operations (see Chomsky (2001, 2008)). We can therefore conclude that the next operation carried out with the phase head in (17-b) is discharge of [*F*] by Agree with some item in β, and this is then followed by movement of a from the edge of β to the edge of 7, which discharges [ · Χ · ] . Note that the discontinuous sequence of edge feature insertion and edge feature discharge thus required is unproblematic, given that structure-building features and probe features are on different stacks. Of course, it must now be ensured that the option for phase heads with empty stacks of structure-building features of being kept active by probe features is not available for last-merged specifiers. This follows without further ado from the interaction of the ccommand requirement on Agree and the Strict Cycle Condition. Probe features on a phase head can never remove barrier status from a last-merged specifier because (a) a probe feature cannot carry out Agree with (some item in) its specifier due to a lack of c-command, and (b) a probe feature cannot carry out Agree with (some item in) its complement after a specifier has been merged because of the SCC. To sum up, last-merged specifiers continue to be barriers; non-last-merged specifiers and complements are not barriers; and last-merged complements are not barriers if the phase head has one or more additional probe feature(s) for Agree with/into the complement. Given the vast number of (various kinds of) Agree relations that can (or, in fact, must) be postulated (involving features like person, gender, number, case, tense and other grammatical categories), complements, as a rule, will be transparent for extraction. Only when (a) a complement is last-merged and (b) there is no Agree operation with the head that it is a complement of, can a complement become a barrier. In section 6, I will argue that such an exceptional situation arises in the case of verb-second complements embedded by a verb-final clause. However, before that, something needs to be said about verb-second movement.

Movement from Verb-Second Clauses Revisited

111

5. Verb-second by Reprojection Verb-second movement in German is often conceived of as adjunction to C that proceeds by intermediate adjunction to Τ and v, as in (19). (19)

Verb-Second, a standard view: CP

C

TP

There are three potential problems with this view. The first problem is that there do not seem to strong empirical arguments for assuming verb-second movement to proceed from V to ν, then from ν to T, and finally from Τ to C (see Haider ( 1993a) and Roberts (2009, 2010), pace Sabel (1996)). And assuming that it does so nevertheless gives rise to various technical problems: Strictly speaking, C attracts Τ (not V), Τ attracts ν (not V), and only ν attracts V, so the existence of full verb-second movement of V to C emerges as a fortunate coincidence going back to a conspiracy of three separate movement rules. If the view embodied in (19) is correct, we might expect phenomena (in minimally differing varieties of German, e.g., earlier stages of the language, or regional variants) like bare T-to-C movement or V-to-v-to-T movement stopping there, for which there is no convincing evidence (see the above references for discussion). These considerations lead me to conclude that verb-second movement does not proceed via successive adjunction to higher heads. The second problem is that one never finds cases in German where there is a realization of C as dass together with an adjacent verb-second head, as in (20-a). Given that there are varieties of German that do not respect the Double Comp Filter (i.e., that allow a simultaneous realization of dass and some other item in the CP edge domain), this is unexpected. Furthermore, it seems that there are substandard instances of a simultaneous realization of dass and verb-second in Modem German after all (and, depending on the analysis, perhaps also in Old High German), but these look exactly like their Scandinavian counterparts, with a sequence C+topic+V2, as in (20-b) (see Lenerz (1984), Müller & Sternefeld ( 1993); and Axel (2007) and Frey wald (2009) for recent discussion concerning Old High German and Modern German substandard varieties, respectively).

112 (20)

Gereon Müller a.

*(lch I b. *!*{lch I

glaube,) [cp sie hati-dass [tp ihn getroffen t¿ ]] believe she has-that him met glaube,) [CP dass [ den Fritz hat [TP sie getroffen t¿ ]] believe that the Fritz has she met

The third problem concerns the nature of head movement as adjunction to another head in general. As is well known, this view of head movement creates several problems with respect to highly general (and independently motivated) constraints on movement, e.g., the c-command requirement for traces (which is a subcase of the SCC; cf. (18)); see Brody (2001), Mahajan (2001), Abels (2003), Müller (2004), and Matushansky (2006) (among others) for discussion. This problem can be solved by assuming that head movement is not adjunction but reprojection, in the sense that a head moves out of a phrase and remerges with it, projecting its label in the derived position. Reprojection movement of a finite verb is explicitly assumed for German verb-second clauses in Stechow & Sternefeld (1988), based to some extent on Reis (1985) who had argued that dass clauses and verb-second clauses should be treated as categorially different (essentially CP vs. VP). Both Stechow & Sternefeld (1988) and Sternefeld (1989) assume that treating verb-second as reprojection movement is a precondition for deriving the asymmetry between dass clauses (which are assumed to be CPs) and verb-second clauses (VPs) with respect to extraction into dass clauses. Subsequent approaches that analyse instances of verb movement (including, in some cases, verb-second movement) by reprojection include Holmberg (1991), Ackema, Neeleman & Weerman ( 1993), Koeneman (2000), Haider (2000), Hornstein & Uriagereka (2002), and Fanselow (2003, 2008). 10 In what follows, I adopt a reprojection analysis of verb-second clauses in German that combines aspects of the approaches to reprojection in Fanselow (2008), Georgi & Müller (2010) and other work just mentioned (but that nevertheless differs from all these approaches in certain minor respects). ' 1 The basic idea is that a head 7 may be equipped with a certain probe feature [*F*] that it cannot possibly check in situ, for the simple reason that there is no goal feature around that it might check it with. There are two possibilities: Either there is no matching goal feature in the c-command domain of 7 at all (recall that Agree requires c-command), or there is a matching goal feature that has already been checked with some other probe earlier in the derivation, so it is not available anymore for 7. In this case, 7 may, as a 10

As noted by Fanselow (2008), reprojection analyses of verb-second in German to some degree resemble the classic approaches developed in Bierwisch (1963) and Thiersch (1978), which had been superseded by the standard analysis in (19) going back to den Besten (1977).

" As a side remark, and anticipating the analysis of the data in ( 1 ) given in the next section, let me point out that, unlike Stechow & Sternefeld (1988) and Sternefeld (1989), I will not so much exploit reprojection movement and the categorial difference (CP vs. VP) between dass clauses and verb-second clauses that goes along with it to account for the difference in grammaticality between (1-a) and (1-d); i.e., the basic status of verb-second clauses as barriers will not necessarily be tied to their being VPs rather than CPs (although this will turn out to be one of several possibilities). Rather, reprojection movement will be required to explain the difference between (1-c) and (1-d); i.e., it will offer a simple account of the fact that verb-second movement in the matrix clause destroys barrierhood and turns an embedded verb-second clause into a transparent domain.

Movement from Verb-Second Clauses Revisited

113

last resort operation (a concept which is not to be confused with the Last Resort condition introduced in section 3 above), undergo reprojection movement in order to find a matching goal and discharge its probe feature [*F*] under c-command. [*F*] thus in effect acts as a Münchhausen feature. 12 The cases of reprojection movement investigated in Georgi & Müller (2010) are typically highly local; in particular, in that paper we are concerned with word order variation in NPs that is derived by extremely local reprojection movement of N. 13 In contrast, verbsecond movement by reprojection cannot be such an extremely local movement operation if verb-second clauses in German have a full TP-vP-VP structure, and if verb-second movement targets a position outside of TP. Indeed, there seems to be evidence in abundance for assuming that verb-second clauses in German have a full TP-vP-VP structure. For instance, in dass clauses as in non-subject initial verb-second clauses, only subjects can precede weak pronouns (except for items in the Vorfeld), which follows if there is a designated TP category whose specifier acts as a target for optional subject raising (as argued in Grewendorf (1989)), and scrambling cannot go beyond the vP/VP domain. Moreover, asymmetries between (in-situ) subjects and other arguments with respect to subcategorization by V, verb phrase topicalization, extraction, binding, and so on, are the same in dass clauses and verb-second clauses, and to the extent that these differences motivate a vP/VP distinction, they do so in both contexts. Similarly, it is clear that, since verb-second movement may end up in a position preceding a subject that in turn precedes a weak object pronoun (which in turn may precede a non-pronominal object), it must target a position outside of TP. The question then is: How does V get there? Here is a suggestion. Suppose that the Münchhausen probe feature on a V that is to undergo verb-second movement by reprojection is [*T*] (see Fanselow (2008, sect. 3.3)); such a feature is optionally instantiated on V in the numeration, and it indicates a special relation that certain kinds of V (viz., verb-second Vs) have to enter with a TP projection. Suppose further that, if V takes an object with clausal structure inside, [*T*] can never be discharged with a lower TP that it c-commands. This is obvious if the embedded clause is a verb-second clause itself (because in this case, the embedded verb-second head will have undergone Agree with TP already, and TP is not accessible for Agree anymore since it does not have an active goal). We may assume that it also holds in cases where a proper C (like dass) subcategorizes TP, perhaps because C also has a probe feature [*T*] in addition to the subcategorization feature [ · Τ · ] that triggers merge of C and TP; note that since a last-merged complement TP of C is transparent for extraction, the approach to CED 12

The name is self-explanatory: Baron Münchhausen escapes from a swamp (where he is trapped on the back of his horse) by pulling himself up by his hair. The use of the name 'Münchhausen' in syntactic theory arguably goes back to Sternefeld's (1991) characterization of an operation employed in Chomsky's (1986) theory of barriers: Here, VP is a barrier, but a V moved to I can belatedly justify its own (originally impossible) movement across the VP barrier by L-marking VP and removing barrierhood - clearly a case of pulling oneself up by one's own hair. (Incidentally, Staudacher's (1990) analysis in terms of transitivity of coindexing is not too dissimilar in this respect.) Fanselow (2003) applies the concept to reprojection movement ('Münchhausen-style head movement').

13

This in turn solves a number of problems that arise under the classic view that nominal projections are NPs, with DP as a specifier of N, rather than DPs, with NP as a complement of D. However, this issue is orthogonal to my present concerns, so I will continue to assume that nominal projections are DPs.

114

Gereon Müller

effects sketched in section 4 presupposes that there is some probe feature on C that keeps C active for edge feature insertion in order to move something out of TP's edge domain.If it is assumed that some other feature than [*T*] is involved in C-T Agree, this would make the same prediction as long as it is ensured that this Agree operation makes Τ an inactive goal for Agree with a higher V. Thus, irrespectively of whether there is an embedded Τ in the c-command domain or not, the only way to discharge a [*T*] feature of V is to carry out reprojection movement. Consequently, V[* Tt ] undergoes movement to the next phase edge, and from there to the next phase edge, until it finally reaches a position from which it can reproject a VP and discharge its Münchhausen feature [*T*] under ccommand of Τ - i.e., the movement operation as a whole is non-local, but like all other movement operations, long-distance reprojection movement is composed of a series of smaller steps, as required by the PIC. This derivation is shown in (21) (with o marking the intermediate positions successively occupied by V on its way to its reprojection position, and the dashed arrow from V to Τ indicating the Agree relation that provides the trigger for verb-second movement). (21 )

Verb-Second by reprojection: VP XP

V'

The assumption that verb-second in German looks as in (21) gives rise to various questions. 14 An obvious one concerns the Head Movement Constraint (HMC): This analysis 14

It also makes some further interesting predictions. Consider, for instance, the well-known fact that some morphologically complex verbs can occur in finite form in a verb-final position, but fail to participate in V/2 fronting (see Haider (1992) and Koopman (1995), among many others):

Movement from Verb-Second Clauses

Revisited

115

of verb-second in terms of reprojection movement of V to the T P domain takes place via local steps (viz., via all intermediate edge domains); however, it clearly violates the HMC because the intervening heads ν and Τ are skipped by the operation. Arguably, though, this is not a problem since the HMC can be shown to be too strict anyway; see in particular Roberts (2009, 2010), who argues that various kinds of head movement operations must be able to freely cross intervening head positions (among them long verb movement in Breton, V movement in predicate cleft constructions in Spanish and Hebrew, clitic-climbing in Italian and, incidentally, also verb-second movement in German). Second, we may ask whether the intermediate movement steps of V[»T„] are triggered by edge features, like all other cases of successive-cyclic movement. Indeed, there is every reason to assume that they are (note particularly that all extraction takes place from complements, so edge features can be inserted in line with the EFC in (12)). Third, why does V reproject, rather than just ending up as a further specifier of TP? A simple (and standard) answer here could be that if two items are combined, only one of them can have the feature that induces this operation, and the item that does will always be the one that projects. 15 Fourth, it is so far not clear how the specifier XP of the verb-second head V comes into existence that closes off the VP in (21). In the general type of approach adopted here, there must be a structure-building feature that triggers placement of some XP in the first position. There are various possibilities as to how this can be implemented in the present system. For concreteness, suppose that a V category that is equipped with the feature [*T*] inducing verb-second must (or may, depending on the analysis of apparent verb-first structures) also bear a structure-building feature creating a specifier in the reprojection position. Such a feature may trigger wft-movement ([*wh·]), as in the cases in (1), or topicalization ([»top·]), as in an example like (3), but typically not relativization ([»rei·]), since relative clauses in German standardly do not involve verb-second - cf. (4), but see also Gartner (2000) for some exceptions. In order to avoid a discharge of this structurebuilding feature while V is still in situ, it can be assumed that a feature like [*wh·] or (i)

a. b. c. d.

dass sie die Oper hier ur-auf-fiihren that they the opera here perform first *Sie führen die Oper hier ur-auf *Sie auf-fiihren die Oper hier ur *Sie ur-auf-fiihren die Oper hier

This is a priori difficult to account for if verb-second movement can uniformly be traced back to some attracting feature on an invariant C; in the present approach, one can simply assume a redundancy rule for certain complex verbs that systematically blocks the presence of [*T*] on such a verb. 15

Upon closer scrutiny, a bit more must be said, though. Why can't V stay in the specifier position of T, discharging its probe feature [*T*] and projecting there? If it does, the structure-building feature for T's DP specifier (the EPP property) can never be satisfied. So Τ must be able to remain in a position for a while from which checking would in principle be possible, and then move again to a further position outside of TP to actually carry out the checking (otherwise the SCC would be violated). Note that since this last operation technically goes to a position beyond the T P phase, the earlier intermediate step to a SpecT position is required. - That said, nothing in what follows would inherently be incompatible with an analysis of verbsecond as movement to a specifier of TP, with V' in (21) reinterpreted as T' or TP; to reach this result, one would have to give up the assumption that the label of a complex category is determined by the daughter that has contributed the operation-inducing feature that is responsible for creating that complex category.

Gereon Müller

116

[•top*] on V[ tX _] comes with a diacritic that makes its discharge dependent on a prior discharge of the probe feature [*T*] driving verb-second. This is shown in (22) for whmovement and topicalization respectively, on the basis of the entry for V in (21). 16 (22)

a. b.

V: [ . D P · ] >- [.wh.]([*T*]>, [*T*] V: [ . D P · ] V [.top«]([* T *l\ [*T*]

In the course of the derivation, a V as in (22-ab) first discharges [»DP·] by merging with a DP that becomes its complement; then it undergoes successive-cyclic movement steps via intermediate edge domains (forced by the PIC, and permitted by the EFC) until it can discharge its Münchhausen feature (the categorial probe feature [*T*] by reprojection, taking TP as its complement; and finally, this opens up the possibility of discharging the movement-inducing feature ([«wh·] or [*top·]) by creating a specifier; in a sense, [*T*] on V locks an operator movement feature, and the discharge of [*T*] unlocks it.17 Fifth, the question arises why C cannot embed a verb-second VP (abstracting away from marked configurations like the one in (20-b)). Standardly, the complementary distribution is derived by assuming categorial identity, which the present approach does not.18 In the present approach, a simple analysis suggests itself: C is equipped with a structurebuilding feature [·Τ·], but reprojection movement of V in verb-second clauses has created a VP, which C cannot subcategorize. Consequently, we are led to the conclusion that C only shows up optionally in a numeration (or subarray) - if it does in the presence of verb-second movement, the derivation will crash.19 16

I abstract away here from other possible probe features on V that it may check with its argument DP or some item included in it.

17

As noted, alternative approaches are possible. One could, for instance, assume that structure-building features and probe features do not show up on two separate stacks, but are actually ordered with respect to one another on a single stack. In the case at hand, an initial order ( » D P · ] >- [*T*] >- [ e w h · ] would make it possible to do without the diacritic on the movement-inducing feature, but such a theory is arguably more stipulative than the approach envisaged here because an order like [ » D P · ] >- [*T*] (or a generalized variant thereof) follows from the general make-up of the theory (a probe feature can only be discharged if there is some category present that provides a goal) and thus does not have to be stated as such. Furthermore, it might be that the present approach to filling the first position in verb-second clauses in German is not yet fine-grained enough. Frey (2004) and Fanselow (2003, 2008) argue that one should formally distinguish between unmarked realizations and marked realizations of the first position in verb-second clauses, with the former case involving, apparently, just items at the left periphery of the verb-second clause that would normally also show up at the edge of the TP or vP domain, and the latter case involving informationstructurally marked items, as, e.g., in »/¡-clauses and long-distance topicalizations. If this is right, one might want to address it by permitting an underspecified edge feature [ · Χ · ] as a third option for an intrinsic feature on a verb-second V: [ · Χ · ] discharge would then imply moving whatever happens to be the closest TP- or vP-internal item to the specifier of the reprojecting verb (given a constraint like the Minimal Link Condition; see Fanselow (1991), Chomsky (2001)), whereas [ * w h · ] (or (»top·]) discharge would imply moving the closest wh-item (or [top]-marked item).

18

However, note that categorical identity in the strict sense is not available under the standard approach either, given structures like (20-a); and also note the problem raised by (20-a) in this context. Furthermore, Stechow & Sternefeld ( 1988, 402-405) show that the idea of a complementary distribution of verb-second and dass is problematic to begin with.

19

A further derivation that must be ruled out has C merging with T P before reprojection movement of V. This follows if last resort operations like reprojection movement obey an earliness requirement, as assumed in Georgi & Müller (2010): The Münchhausen feature is discharged as soon as possible.

117

Movement from Verb-Second Clauses Revisited

Much more would ultimately have to be said about a reprojection approach to German verb-second clauses. However, I will refrain from doing so since the only thing that is of fundamental importance for the PIC-based approach to the data in ( 1 ) to be developed in the next section is really just the assumption (adopted from Fanselow (2008)) that a V that undergoes verb-second movement in the derivation has an additional operation-inducing feature that is absent when V stays in situ, in a dass clause.

6. A PIC-based Approach Now the system is basically in place that makes it possible to derive the the pattern in ( 1 ) from the PIC, in a way that incorporates Staudacher's (1990) hypothesis that verb-second clauses are barriers because they are not as strictly selected as dass clauses, and that the barrierhood is lifted by verb-second movement because this gives rise to a configuration that resembles the one found with dass clauses. There is just one more assumption that is needed: In the examples in ( 1 ), the most deeply embedded clause is both first-merged (i.e., a complement) and last-merged. This implies that the embedded clause can only be transparent for extraction if V is equipped with another operation-inducing feature that keeps it active (for edge feature insertion to be possible) after the clause is merged. As I will show momentarily, this state of affairs lies at the core of the 3-out-of-4 pattern in (1). However, the grammaticality status does not vary noticeably when an additional argument is added, as in (23-ab), which can be compared with ( 1 -a) and ( 1 -d), respectively (the added argument is underlined). 20 (23)

20

(i)

a. l(Ich weiß nicht) [ went (dass) du I know not whom that you getroffen hat ]] met has b. *(Ich weiß nicht) [ wenl (dass) du I know not whom that you getroffen ]] met

ihm gesagt hast [ t' dass sie t¿ him told have that she

ihm gesagt hast [ t' hat sie t, him told have has she

At least, there is no systematic improvement of the verb-final/verb-second combination; quite on the contrary, as a tendency the results get somewhat worse throughout. Incidentally, (23-b) is not quite fully well formed even if there is no movement from the embedded verb-second clause; cf. (i-a) vs. (i-b). a.

(Ich I b. IKIch I

weiß) [ dass du know that you weiß) [ dass du know that you

gesagt hast [ sie hat ihn said have she has him ihm gesagt hast [ sie hat him said have she has

getroffen ]] met ihn getroffen ]] him met

If this effect could be shown to be systematic, no additional assumption would be required for the cases in (23). However, there still seems to be a contrast between (i-b) and (23-b), and this implies (23-b) cannot solely be due to general non-licensing of embedded verb-second in the presence of a second argument in VP.

118

Gereon Müller

Against the background of the theory developed so far, the conclusion to be drawn from this is that clausal complements are always the only items merged in a VP; a further nonsubject argument will then be merged in some separate projection between vP and VP. Therefore, a structure-building feature for an indirect object can never make extraction from a CP possible by keeping the V head active. Possibly, this assumption should be generalized, such that indirect objects and direct objects are always merged in different projections, even when the direct object is non-clausal. On this view, extraction from a complement of V always depends on the availability of some additional probe feature on V. Here, then, is the analysis of the pattern in (1): Both dass complements and verbsecond complements enter the structure via a subcategorization feature on V ([*C·] and [ · ν · ] , respectively). However, the former clause type receives a special identification in the form of a probe feature on V that agrees with it; such a feature is absent from V if it embeds a verb-second clause. For present purposes, it does not really matter what exactly this additional probe feature on V that discriminates between dass clauses and verb-second clauses looks like. One might speculate that it is a case feature (assuming C to be nominal, CPs might need case, which verbal categories do not); but this question is discussed controversially in the literature. An alternative would be to postulate abstract φ features on a dass clause (but not on a verb-second clause). For now, I will simply assume that the probe feature in question is a categorial probe feature of just the type encountered in the previous section, the only difference being that it does not trigger reprojection movement because it can be checked by V under c-command. Thus, in the case of a dass clause, matrix V may bear a probe feature [*C*]; but there is no comparable categorial probe feature on matrix V for a verb-second clause. I take this to be a relatively straightforward and faithful transfer of Staudacher's (1990) distinction in terms of head-marking via head-selection (dass clauses are head-selected, verb-second clauses are not). 21 Given the approach to CED effects in section 4, this accounts for the transparency of dass clauses: At the point of the derivation where an edge feature needs to be inserted on the matrix V to make movement from the edge of CP to the edge of VP possible, V is active because it still bears [*C*]. The reason why dass complements of V are transparent for extraction (see (1-a), (1-b)) thus follows in exactly the way outlined in (17-b) above for last-merged complements in general. Consider first movement from a dass clause into a dass clause, as in (1-a). (1-a) is repeated here as (24). 22 21

Why can a V with a feature [*V·] not merge with a (finite or non-finite) bare VP with an in situ V head that is not equipped with [*T*1? Perhaps the most straightforward answer might be that it can, but that the resulting structure is uninterpretable since the lower "clause" does not have a subject (in addition, an object in the lower VP could not have its case feature checked).

22

Although there are many more intermediate movement steps, given the assumption of section 4 that all phrases are phases, I add here only the trace in the matrix VP position that signals that the movement step from the CP2 edge to the matrix VP edge is legitimate. - Note incidentally that I insert traces here only for convenience;

119

Movement from Verb-Second Clauses Revisited (24)

(Ich weiß nicht) [cP! wenl I

know not

(dass) du [vp t" meinst [cp2 t· dass sie t¿

whom that

you

think

that she

getroffen hat]]} met

has

The changes in feature-composition on the matrix V meinst

(25)

in (24) are illustrated in (25).

CP complements as non-barriers, matrix V in situ: V: [•c.] [*C*]

[*C*]

V: V:

violates nothing

[·Χ·]

[*C*]

After the edge feature [ · Χ · ] is inserted, [ * C * ] is discharged, triggering Agree, and finally, [ · Χ · ] is discharged, triggering extraction from the C P complement. The relevant part of the derivation of (24) is shown in (26). 2 3

(26)

CP complements as non-barriers: VP

wen

C

TP

Note that this account of dass clause transparency is completely independent of the issue of verb movement in the higher clause. Therefore, sentences like ( 1 -b), with movement from a dass clause into a verb-second clause, receive essentially the same treatment; ( 1 -b) is repeated here as (27) (with CPi replaced with V P i , in line with new analysis of verbsecond developed in section 5).

(27)

[vPi Wen, meinst du [vp t" [cp2 t¿ dass sie t¿ getroffen hat ]]] ? whom think

you

that she

met

has

copy theory, multidominance or radically traceless approaches to movement are also compatible with the present analysis. 23

Dashed lines indicate that the operation has not yet been carried out at this stage o f the derivation, and that the structure has not yet been created; the arrow y

signals edge feature insertion.

120

Gereon Müller

The only relevant difference to (24) is that matrix V has an additional probe feature, viz., the [*T*] feature that triggers reprojection movement. Given that probe features are ordered on their stack in the same way that structure-building features are on theirs, it must be assumed that the order is [*C*] (to be discharged with the CP complement by V in situ) >- [*T*] (to be discharged with the matrix TP after reprojection raising of V). Then again, no aspect of the present analysis really forces the view that probe features are ordered, too. In the present context, this assumption is mainly (and weakly) justified by considerations pertaining to homogeneity and uniformity of feature stacks. Thus, edge feature insertion can respect the EFC on matrix V, as indicated in (28). (28)

CP complements as non-barriers, matrix V undergoes V2: V: [ . C . ] [*C*] >- [*T*] V: [*C*] >- [*T*] V: [ · χ · ] [*C*] X [*T*]

violates nothing

Except for the additional [*T*] feature, the relevant part of the derivation of (27) is identical to what is shown in (26). Things are different with embedded verb-second clauses. Let us turn to movement from a verb-second clause into a verb-second clause first; the example in (1-c) is given again in (29) (with CP replaced by VP, and an intermediate trace in the edge domain of the matrix VP added). (29)

[vpj Wen, meinst du [vp t" [vp2 t' hat sie t, getroffen ]]] ? whom think you has she met

(29) is well formed, which implies that an edge feature can be inserted on matrix V for an item in the edge domain of its sole verb-second complement even though verb-second complements, by assumption, are not identified by a categorial probe feature (that would keep the V head active), like dass clauses are. It is at this point that the assumption made in section 5 becomes relevant according to which verb-second movement is brought about by a Münchhausen feature on V that triggers reprojection movement (viz., [*T*]). Crucially, this feature is present on matrix V in its in situ position in (29), and this is the reason why V is still active (after it has merged with the verb-second clause), and therefore accessible for edge feature insertion. The changing feature structure on matrix V in the relevant part of the derivation is shown in (30). (30)

VP complements as non-barriers, matrix V undergoes V2: V: [ . V . ] [*T*] V: [*T*] -i- V: [ · Χ · ] [*T*]

The decisive stage of the derivation itself is illustrated in (31).

violates nothing

121

Movement from Verb-Second Clauses Revisited (31)

VP complements as non-barriers: VP

wen

V

TP

In the three cases discussed so far, there is at least one probe feature on matrix V that keeps the head active and thereby permits edge feature insertion for movement out of the clausal complement: In (24), there is [*C*]; in (29), there is [*T*], and in (27), there is both [*C*] and [*T*]. However, in the case of movement from a verb-second clause into a dass clause as in (1-d), repeated here as (32) (with slightly modified labeling), there is no [*C*] on the matrix V (because the verb-second complement is not strictly selected, or "head-selected", in Staudacher's (1990) terminology), and there is no [*T*] on the matrix V either (because the matrix clause is not a verb-second clause). (32)

*(lch weiß nicht) [cPt wen, (dass) du meinst [vp2 t' hat sie t, getroffen ]] I know not whom that you think has she met

Consequently, the matrix V head becomes inert after it has discharged its subcategorization feature and merged with the verb-second complement: Edge feature insertion cannot apply, and a PIC violation therefore cannot be avoided once the derivation has proceeded beyond the matrix VP phase. This accounts for the ill-formedness of (32). The impoverished feature structure of matrix V in this derivation is shown in (33). (33)

VP complements as barriers, matrix V stays in situ: v

[·ν·]

[ · Χ · ] insertion impossible

The problematic step of the derivation itself is illustrated in (34).

122

Gereon Müller

This concludes the account of the pattern in (1) that is the subject of the present article: The restriction on movement from verb-second clauses to verb-final clauses can be derived if Staudacher's ( 1990) main assumptions are transferred into the PIC-based account of CED effects in Müller (2010) and enriched by a reprojection approach to verb-second movement that locates the operation-inducing feature on V, as in Fanselow (2008). The one thing that has been lost from Staudacher's original analysis in the course of doing so is transitivity of indexing as a means to remove barrier status from an embedded verb-second clause; since this device is either representational or demanding look-ahead capacity, and since its work can straightforwardly be done by the reprojection-inducing probe feature on matrix V against the background of the theory in Müller (2010), I take this loss to be unproblematic.

7. Some Further Issues Of course, the approach just sketched raises a number of further issues. I will confine myself to addressing three of them here. First, the question arises of what happens with periphrastic verb forms where, say, an auxiliary undergoes verb-second movement in the matrix clause and the main verb stays in situ. As shown in (35-ab), the pattern remains the same as in (1-c), (1-d) (= (29), (32)): (35)

a.

[vpt Went hast du [yp t g e m e i n t [γρ2 t' hat sie t¿ getroffen ]]] ? whom have you thought has she met b. *(Ich weiß nicht) [cpj wenl (dass) du gemeint hast [γρ2 t' hat sie t¿ I know not whom that you thought have has she getroffen ]] met

The challenge here is to account for the fact that raising of the higher auxiliary can lift barrierhood of the verb-second clause even though the main verb stays in situ in the matrix

Movement from Verb-Second Clauses Revisited

123

clause. There are several ways to solve this problem. 24 A first strategy might be to assume that in fact not all phrases are phases after all: In extended (verbal) projections, only the highest projection counts as a phase. This would derive the pattern in (35), but it would require further assumptions for extraction from dass clauses into dass clauses in contexts where there is an auxiliary in situ in the matrix clause, as in (36) (compare (1-a) = (24)). The reason is that it would be unclear why a probe feature on the lower V head (i.e., the main verb) could render an otherwise inactive higher V head (i.e., the auxiliary) active. (36)

{Ich weiß ni'c/ií) fcPi wetii (dass) du [vp t" gemeint hast [cp2 t- dass sie t, I know not whom that you thought have that she getroffen hat]]] met has

I will therefore pursue a slightly different approach here and postulate that an auxiliary (or modal) verb and its associate main verb are base-generated as a single complex head, and reprojection movement of the finite part of the complex head proceeds via excorporation, much as with particle stranding under verb-second. Second, throughout this article I have presupposed that complement clauses are merged in the canonical (left-peripheral) object position in German. However, finite complement clauses typically undergo extraposition; in fact, extraposition is obligatory in the case of verb-second complements (see Stechow & Sternefeld (1988), Heck (2010)), and also if extraction from a clause has applied (arguably a syntactic reflex of successive cyclicity in German; see Müller (1999), Lahne (2009)). On the view adopted here, obligatory extraposition is independent from the barrier status of a clausal complement; and one might argue that the transparency of dass clauses as in (1-a), (1-b) supports this view.25 Third, it can be observed that the present proposal is incompatible with one small piece of the analysis of CED effects I give in Müller (2010): There I suggest that the bridge verb/non-bridge verb distinction, as well as lexical variation with extraction from DP, can be tied to the presence or absence of an additional probe feature that may keep V active, and accessible for edge feature insertion; cf. (37) (lexically determined extraction from CP) and (38) (lexically determined extraction from DP). (37)

a.

(Ich weiß nicht) [cpj went (dass) du [vp t" meinst/glaubst [cp2 t¿ dass I know not whom that you think that du t¿ getroffen hast ]]] you met have

24

Closer inspection reveals that this problem shows up in a number of analyses of the pattern in (1), among them the one pursued by Staudacher (1990). In his system, one would have to assume that the auxiliary and the main verb are always co-indexed.

25

Heck (2010) assumes that verb-second clauses are barriers because they are obligatorily extraposed; but this then requires a heterogeneous approach to verb-second clause and dass clause extraposition (i.e., two extraposition rules with different properties) so as to ensure that extraposed dass clauses can still be transparent.

124

Gereon Müller b. '}*(Ich weiß nicht) [cPi wen,, (dass) du [yp t" bereust/weißt [cp2 t· dass I know not whom that you regret/know that du ti getroffen hast ]]] you met have

(38)

a.

(Ich weiß I know liest ] reads b. T*([ch weiß I know klaut ] steals

nicht) [ρρι worüber ] der Fritz [VP T• [DP ein Buch T¿ ] not about what the F r i t z s a book ACC

nicht) [pp¿ worüber ] der Fritz [VP t- [DP ein Buch t¿ ] not about what the FritZwoM a book ACC

According to the approach in Müller (2010), V may not have the special probe feature that is needed to keep the head active and permits edge feature insertion for extraction from the (CP or DP) complement in (37-b) and (38-b); i.e., on this view, t" in (37-b) and t' in (38-b) cannot be present, which gives rise to a PIC violation once the derivation moves on. This analysis cannot be maintained under present assumptions. The reason is that we would then expect verb-second movement to improve the examples in the same way that verb-second movement in the matrix clause makes extraction from an embedded verb-second clause possible. This would clearly not be a correct prediction; see (39-ab). (39)

a. ?*[CP, Wen, bereust/weißt du [vp t" [cp2 t· dass du t¿ getroffen hast ]]] ? whom regret/know you that you met have b. ?*[pp¡ Worüber ] klaut der Fritz [vp T\ [DP ein Buch t¿ ]] ? about what steals the Fritz N0M a bookACC

I conclude from this that the sentences in (37-b), (38-b) and (39) do not involve PIC violations; a categorial probe feature identifying the complement can be inserted on V in all cases. The deviance must then have a different source. This view might be supported by the fact that the grammaticality status of these examples is much more variable than that of examples like ( 1 -d), and subject to idiolectal variation and habitualization effects. An alternative account that would be compatible with the present analysis would be to treat phenomena like the distinction between bridge verbs and non-bridge verbs by postulating an empty operator in the latter case that gives rise to a minimality violation; see Manzini (1992,115), and also Roberts & Roussou (2002).

Movement from Verb-Second Clauses Revisited

125

References Abels, K. (2003). Successive Cyclicity, Anti-Locality, and Adposition Stranding. Ph.D. thesis. University of Connecticut, Storrs, Connecticut. Ackema, P., A. Neeleman & F. Weerman (1993). Deriving Functional Projections. In: Proceedings of NELS. (vol. 23). Amherst: GSLA. 17-31. Axel, K. (2007). Studies on Old High German Syntax: Left Sentence Periphery, Verb Placement and Verb-Second. Amsterdam: Benjamins. Bentzen, K. (2009). Embedded Root Phenomena, Assertion, Presupposition, and Main Point of Utterance. Ms., University of Troms0. den Besten, H. (1977). On the Interaction of Root Transformations and Lexical Deletive Rules. Ms., University of Amsterdam. Bhatt, R. (2005). Long Distance Agreement in Hindi-Urdu. Natural Language and Linguistic Theory 23. 757-807. Bierwisch, M. (1963). Grammatik des deutschen Verbs. (Studia Grammatica II). Berlin: Akademie Verlag. Boskovic, v. (2007). Agree, Phases, and Intervention Effects. Linguistic Analysis 33. 54-96. Brody, M. (2001). Some Aspects of Elegant Syntax. Ms., University College London. Cavar, D. (1996). Untitled. Ms., Universität Potsdam. Chomsky, N. (1973). Conditions on Transformations. In: S. Anderson & P. Kiparsky (eds.) A Festschrift for Morris Halle. New York: Academic Press. 232-286. Chomsky, N. (1986). Barriers. Cambridge, Mass.: MIT Press. Chomsky, N. (2000). Minimalist Inquiries: The Framework. In: R. Martin, D. Michaels & J. Uriagereka (eds.) Step by Step. Cambridge, Mass.: MIT Press. 89-155. Chomsky, N. (2001). Derivation by Phase. In: M. Kenstowicz (ed.) Ken Hale. A Life in Language. Cambridge, Mass.: MIT Press. 1-52. Chomsky, N. (2007). Approaching UG from Below. In: U. Sauerland & H.-M. Gärtner (eds.) Interfaces + Recursion - Language? Berlin: Mouton de Gruyter. 1-31. Chomsky, Ν. (2008). On Phases. In: R. Freidin, C. Otero & M. L. Zubizarreta (eds.) Foundational Issues in Linguistic Theory. Cambridge, Mass.: MIT Press. 133-166. Cinque, G. (1990). Types of A-bar Dependencies. Cambridge, Mass.: MIT Press. Fanselow, G. (1991). Minimale Syntax. Habilitation thesis, Universität Passau. Fanselow, G. (2003). Münchhausen-Style Head Movement and the Analysis of VerbSecond. In: A. Mahajan (ed.) Head Movement and Syntactic Theory. (Syntax at Sunset, vol. 3). Los Angeles & Potsdam: UCLA & Universität Potsdam Working Papers in Linguistics. 40-76. Fanselow, G. (2008). Bootstrapping Verb Movement and the Clausal Architecture of German (and Other Languages). Ms., Universität Potsdam. To appear in Proceedings of the Comparative Germanic Syntax Workshop 2007. Fanselow, G. & A. Mahajan (1996). Partial Movement and Successive Cyclicity. In: U. Lutz & G. Müller (eds.) Papers on Wh-Scope Marking. no.( 76 in Arbeitspapiere). Universität Stuttgart and Universität Tiibingen: SFB 340. 131-177.

126

Gereon Müller

Frey, W. (2004). The Grammar-Pragmatics Interface and the German Prefield. Sprache & Pragmatik 52. 1-39. Frey wald, U. (2009). V2 in German Complement Clauses Introduced by dass 'that'. Ms., Universität Potsdam. Gärtner, H.-M. (2000). Are There V2 Relative Clauses in German? The Journal of Comparative Germanic Linguistics 3. 97-141. Georgi, D. & G. Müller (2010). Noun Phrase Structure by Reprojection. Syntax 13. 1-36. Grewendorf, G. (1989). Ergativity in German. Dordrecht: Foris. Haider, H. (1984). Topic, Focus, and V-Second. Groninger Arbeiten zur Germanistischen Linguistik 25. 72-120. Haider, H. (1992). The Basic Branching Conjecture. Ms., Universität Stuttgart. Haider, H. (1993a). Deutsche Syntax- generativ. Tiibingen: Narr. Haider, H. (1993b). ECP-Etiiden: Anmerkungen zur Extraktion aus eingebetteten VerbZweit-Sätzen. Linguistische Berichte 145. 185-203. Haider, H. (2000). Branching and Discharge. In: P. Coopmans, M. Everaert & J. Grimshaw (eds.) Lexical Specification and Insertion. Amsterdam: Benjamins. 135164. Heck, F. (2010). Against Direct Recursion in Syntax. Ms., Universität Leipzig. Heck, F. & G. Müller (2007). Extremely Local Optimization. In: E. Brainbridge & Β. Agbayani (eds.) Proceedings of the 26th WECOL. Fresno: California State University. 170-183. Holmberg, A. (1991). Head Scrambling. Handout of talk; GLOW 1991 (Leiden). Hooper, J. & S. Thompson (1973). On the Applicability of Root Transformations. Linguistic Inquiry 4. 465-497. Hornstein, Ν. & J. Uriagereka (2002). Reprojections. In: S. D. Epstein & D. Seely (eds.) Derivation and Explanation in the Minimalist Program. London: Blackwell. 106-132. Huang, C.-T. J. (1982). Logical Relations in Chinese and the Theory of Grammar. Ph.D. thesis. MIT, Cambridge, Mass. Kayne, R. (1984). Connectedness and Unambiguous Paths. Dordrecht: Foris. Kiziak, T. (2007). Long Extraction or Parenthetical Insertion? Evidence from Judgement Studies. In: N. Dehé & Y. Kavalova (eds.) Parentheticals. Amsterdam: Benjamins. 121-144. Koeneman, O. (2000). The Flexible Nature of Verb Movement. Ph.D. thesis. Utrecht University. Koopman, H. (1995). On Verbs that Fail to Undergo V-Second. Linguistic Inquiry 26. 137-163. Köster, J. (1987). Domains and Dynasties. Dordrecht: Foris. Lahne, A. (2009). Where There is Fire There is Smoke. Local Modelling of SuccessiveCyclic Movement. Ph.D. thesis. Universität Leipzig. Lasnik, H. & M. Saito (1984). On the Nature of Proper Government. Linguistic Inquiry 15. 235-289. Legate, J. A. (2005). Phases and Cyclic Agreement. MITWPL 49. 147-156. Perspectives on Phases. Lenerz, J. (1984). Syntaktischer Wandel und Grammatiktheorie. Tübingen: Niemeyer.

Movement from Verb-Second Clauses Revisited

127

Mahajan, A. (2001). Word Order and Remnant VP Movement. UCLA, Ms. Manzini, R. (1992). Locality. A Theory and Some of Its Empirical Consequences. Cambridge, Mass.: MIT Press. Matushansky, O. (2006). Head-Movement in Linguistic Theory. Linguistic Inquiry 37. 69-109. Meinunger, A. (2004). Verb Position, Verbal Mood and the Anchoring (Potential) of Sentences. In: H. Lohnstein & S. Trissler (eds.) The Syntax and Semantics of the Left Periphery. Berlin: Mouton de Gruyter. 313-341. Müller, G. (1989). Barrieren und Inkorporation. Master's thesis, Universität Konstanz. Müller, G. (1999). Imperfect Checking. The Linguistic Review 16. 359-404. Müller, G. (2003). Local vs. Global Optimization in Syntax: A Case Study. In: J. Spenader, A. Eriksson & Ö. Dahl (eds.) Variation within Optimality Theory. Proceedings of the Stockholm Workshop. Stockholm University, Department of Linguistics. 82-91. Müller, G. (2004). Verb-Second as vP-First. Journal of Comparative Germanic Linguistics7. 179-234. Müller, G. (2010). On Deriving CED Effects from the PIC. Linguistic Inquiry 41. 35-82. Müller, G. & W. Sternefeld (1993). Improper Movement and Unambiguous Binding. Linguistic Inquiry 24. 461-507. Pesetsky, D. (1998). Some Optimality Principles of Sentence Pronunciation. In: P. Barbosa, D. Fox, P. Hagstrom, M. McGinnis & D. Pesetsky (eds.) Is the Best Good Enough? Cambridge, Mass.: MIT Press and MITWPL. 337-383. Reis, M. (1985). Satzeinleitende Strukturen im Deutschen. In: W. Abraham (ed.) Erklärende Syntax des Deutschen. Tübingen: Narr. 271-311. Reis, M. (1996). Extractions from Verb-Second Clauses in German? In: U. Lutz & J. Pafel (eds.) On Extraction and Extraposition in German. Amsterdam: Benjamins. 45-88. Richards, N. (2001). Movement in Language. Oxford: Oxford University Press. Roberts, I. (2009). On the Probable Existence of Head Movement and the Definite NonExistence of the Head Movement Constraint. Ms., University of Cambridge. Roberts, I. (2010). Agreement and Head Movement: Clitics, Incorporation and Defective Goals. Cambridge, Mass.: MIT Press. Roberts, I. & A. Roussou (2002). The Extended Projection Principle as a Condition for the Tense-Dependency. In: P. Svenonius (ed.) Subjects, Expletives, and the EPP. Amsterdam: Benjamins. 125-156. Sabel, J. (1996). Restrukturierung und Lokalität. Berlin: Akademie-Verlag. Shin, S. S. (1988). On Long-Distance Movement in German. Linguistische Berichte 115. 229-254. Staudacher, P. (1987). Extraktion ausdeutschen Verb-Zweit-Komplementen. Ms., Universität Regensburg. Staudacher, P. (1990). Long Movement from Verb-Second-Complements in German. In: G. Grewendorf & W. Sternefeld (eds.) Scrambling and Barriers. Amsterdam: Benjamins. 319-339. Stechow, A. v. & W. Sternefeld (1988). Bausteine syntaktischen Wissens. Opladen: Westdeutscher Verlag.

128

Gereon Müller

Sternefeld, W. (1989). V-Movement, Extraction from V/2 Clauses, and the ECP. Working Papers in Scandinavian Syntax 44. 119-140. Sternefeld, W. (1991). Syntaktische Grenzen. Chomskys Barrierentheorie und ihre Weiterentwicklungen. Opladen: Westdeutscher Verlag. Sternefeld, W. (1992). Transformationstypologie und strukturelle Hierarchie. Ms., Universität Tübingen. Tappe, T. (1981). Wer glaubst du hat recht? In: M. Kohrt & J. Lenerz (eds.) Sprache: Formen und Strukturen. Tübingen: Niemeyer. 203-212. Thiersch, C. (1978). Topics in German Syntax. Ph.D. thesis. MIT, Cambridge, Mass. Williams, E. (1974). Rule Ordering in Syntax. Ph.D. thesis. MIT, Cambridge, Mass. Williams, E. (2003). Representation Theory. Cambridge, Mass.: MIT Press.

Spurious Ambiguities and the Parentheticals Debate Andreas Pankau

Craig Thiersch

Kay-Michael

Würzner

1. Introduction There has been a long-standing debate in German linguistics regarding the proper analysis of sentences such as Wen glaubst du hat sie gesehen ? (Who do you think she saw?), which henceforth will be called - following Kiziak (2007) - the Critical Construction (CC). The debate is about the status of the part glaubst du, which from now on will be called the Verb First Part (VIP). Under one approach (Thiersch, 1978; Tappe, 1980; Grewendorf, 1988; Staudacher, 1990), the CC involves (i) the V I P as a main clause predicate, and (ii) long extraction of the wh-phrase out of an embedded V2 clause where the extraction crosses the VIP: (1)

[si Weni glaubst du [s 2 1'¿ hat sie t¿ gesehen]] who ACC think you has she seen

Under another approach (Reis, 1995, 1996; Kiziak, 2007), the CC involves (/') short extraction of the wh-phrase and (ii) the insertion of the V1Ρ as an integrated1 parenthetical into prefinite position: (2)

[s Went {glaubst du] hat sie t¿ gesehen]

In the course of the paper, we will call the first approach the crossing approach (CA), and the second approach the parenthetical approach (PA). 2 ' It is necessary to stress that the parenthetical is integrated since only those types of parentheticals are relevant for our discussion; cf. Reis ( 1995, 1996) for differences between integrated and non-integrated parentheticals. 2

Usually, the parenthetical approach is contrasted with an "extraction" (Tappe, 1980, p. 205; Reis, 1995, p. 31; Reis, 1996, p. 45; Reis, 2002, p. 3) or a "long extraction" approach (Staudacher, 1990, p. 320; Kiziak, 2007, p. 102). Both terms, however, are misleading. The first one because extraction is involved under a parenthetical approach as well, viz. short extraction of the wh-phrase; the second one because parentheticals are not restricted to host clauses with short extraction (the parenthetical expression is underlined): (i) Wen glaubst du hat Gisbert gesagt dass Peter gesehen hat? whoAcc think you has Gisbert said that Peter seen has 'Who do you think that Gisbert said that Peter saw?' What is crucial is that under the PA the extracted phrase did not cross the parenthetical expression (since it is not part of the sentence at the point where extraction applies), whereas it did so under the first approach;

130

Andreas Pankau, Craig Thiersch, Kay-Michael

Würzner

Both approaches have their respective merits. The PA approach is appealing mainly for empirical reasons. First of all, the CC behaves in certain respects identically to undisputed, viz. postfinite, parenthetical constructions (Reis, 1995, 1996)3: (3) a.

Wen hat glaubst du sie gesehen? whoACC has think you she seen b. Wen hat sie gesehen glaubst du? whoAcc has she seen think you (a&b) 'Who do you think that she saw?' c. Wem fragst du hat sie geholfen? whoDAT ask you has she helped (lit.)'Who do you ask that she helped?'

Examples (3-a) and (3-b) are undisputed parenthetical constructions because an approach that treats glaubst du as a part of the main clause would have to assume extraction of a non-constituent for (3-a), or clausal pied-piping for (3-b): (4) a. [s ([Wen [hat)i glaubst du ()¿ sie gesehen]]]4 b. [s, [s2 Wenk hat sie tfc gesehen]i glaubst du t, ] Extraction of a non-constituent is generally illicit, and wh-question extraction accompanied by overt clausal pied-piping lacks independent justification as an option for German. 5 (3-c) is undisputedly a parenthetical construction because fragen is not a bridge verb, and therefore does not license extraction. Second, the prefinite position is independently available for other types of parentheticals, viz. non-integrated parentheticals (cf. fn. 1): (5) a. Peter-wen wundert 's - ist immer noch mit Plato beschäftigt. Peter whoACC surprises it is always yet with Plato busy 'Peter - who'd be surprised - is still busy with Plato.' b. Wer - so fragte Peter - hat hier noch nie von Frege gehört? who so asked Peter has here yet never of Frege heard 'Who, asked Peter, has never heard of Frege here?' The CA approach is appealing mainly for theoretical reasons. First, it unifies the CC with extraction out of dasí-clauses in German: (6) a. Wen glaubst du hat sie gesehen? whoAcc think you has she seen hence the name. 3

To ease their identification, V I P s are typeset in bold face in some examples.

4

Extracting hat and wen separately would imply that two positions are available in prefinite matrix position; this however would violate the well-justified V2 constraint for matrix clauses in German. We disregard apparent counterexamples such as Gestern am Strand habe ich Maria gesehen. See discussion in Bildhauer & Cook (2010).

5

Clausal pied-piping has been proposed for German (Riemsdijk, 1982), but in the context of relative clauses, and not for wh-question extraction.

Spurious Ambiguities and the Parentheticals Debate

131

a'· [cp! Wen, glaubst du [cp2 t' [ c o hatk] sie t, gesehen tfc]] b. Wen glaubst du dass sie gesehen hat? whoACC think you that she seen has b'. [cpi Wen, glaubst du [cp2 t' [ c o dass] sie t¡ gesehen hat]] (a&b) 'Who do you think that she saw?' Both constructions are structurally almost identical, and differ only as to whether the embedded C° position is filled by a base-generated complementizer or by the verb extracted to the C° position. Second, given successive cyclic wh-movement, the restriction that the extraction out of the V2-clause must leave a gap in first position of the V2-clause (Reis, 1996, p. 50) is easily accounted for because movement in one fell swoop would violate that constraint. Although theoretically appealing, the CA approach is problematic for two reasons. First, it is empirically weakened by the apparently identical behavior of pre- and postfinite parentheticals since this parallelism is completely unexpected. Secondly, it is theoretically necessary to assume some notion of parenthesis due to the existence of undisputed, unintegrated parentheticals (cf. (5)). But then it is hard to see how to exclude this as an analytical option for the CC. To conclude, the PA approach seems to be preferable on both empirical and theoretical grounds.

2. Differences between the CC and True Parentheticals We will now present some data which show that the CC and true, postfinite parentheticals do not in fact behave identically. We will discuss in total five differences. First, the CC licenses interrogative wh-phrases in the VIP, postfinite parentheticals do not: (7) a. b.

y/ Wen glaubt wer hat sie gesehen? whoACC thinks whoNOM has she seen * Wen hat glaubt wer sie gesehen? whoACC has thinks whoNOM she seen * Wen hat sie gesehen glaubt wer? whoACC has she seen thinks whoNOM (a&b, lit.)'Whom does who think that she saw?'

Second, the modal particles eigentlich or denn are only licensed as part of the VIP in the CC, 6 but not in postfinite parentheticals:

6

Those two differences have already been independently observed by Hubert Haider (cited as a personal communication to Judith Berman in Berman, 2000, p. 109-110, fn. 4).

132

Andreas Pankau, Craig Thiersch, Kay-Michael

(8) a.

b.

Würzner

γ/ Wer glaubst du eigentlich hört deine Musik? who think you actually listens your music * Wer hört glaubst du eigentlich deine Musik? who listens think you actually your music * Wer hört deine Musik glaubst du eigentlich ? who listens your music think you actually 'Who do you actually think is listening to your music?' y Wer glaubst du denn hilft dir gerade? who think youNOM MP helps youDAT currently * Wer hilft glaubst du denn dir gerade? who helps think youNOM MP youDAT currently * Wer hilft dir gerade glaubst du denn? who helps youDAT currently think youNOM MP 'Who do you MP think is helping you?' [MP = modal particle]

Third, negation is licensed as part of the V I P in the CC, but not in postfinite parentheticals: 7 (9) a.

? Wen glaubst du nicht hat sie gesehen? whoACC think you not has she seen b. * Wen hat glaubst du nicht sie gesehen? whoACC has think you not she seen * Wen hat sie gesehen glaubst du nicht? whoACC has she seen think you not (a&b, lit.)'Who don't you think that she saw?'

Fourth, negation in the VIP licenses NPIs for the CC, but not in postfinite parentheticals: (10) a. ? Wen glaubst du nicht wird Peter damit sonderlich zufriedenstellen? whoACC think you not will Peter with-it particularly satisfy * Wen wird glaubst du nicht Peter damit sonderlich zufriedenstellen ? whoACC will think you not Peter with-it particularly satisfy * Wen wird Peter damit sonderlich zufriedenstellen glaubst du nicht? whoACC will Peter with-it particularly satisfy think you not 'Who don't you think that Peter will especially satisfy with that?' b. ? Woraus glaubst du nicht hat Peter einen Hehl gemacht? of what think you not has Peter a secret made * Woraus hat glaubst du nicht Peter einen Hehl gemacht? of what has think you not Peter a secret made * Woraus hat Peter einen Hehl glaubst du nicht? of what has Peter a secret made think you not 'What don't you think that Peter made a secret of?' 7

These judgments and those of (10) are only the judgments of one of the two native-speaker authors, since for the second one, extraction across negation is not well-formed.

Spurious Ambiguities and the Parentheticals Debate c.

133

? Was glaubst du nicht braucht Peter nochmal zu absolvieren? whatACC think you not needs Peter again to pass * Was braucht glaubst du nicht Peter nochmal zu absolvieren whatACC needs think you not Peter again to pass * Was braucht Peter nochmal zu absolvieren glaubst du nicht? whatACC needs Peter again to pass think you not 'What don't you think that Peter has to pass again?'

Fifth, the CC can host an infinitival clause, whereas postfinite parentheticals cannot: (11)

y Wen glaubst du vor dir zuhaben? whoACC think you in front of youDAT to have * Wen vor dir zu haben glaubst du ? whoACC in front of youDAT to have think you 'Who do you think is standing in fornt of you?'

Some remarks concerning the judgments of these sentences are necessary. First, the judgment for (7-a) is not the one standardly assumed in the literature on German syntax; we will return to this point in Section 3. Second, the examples in (9-a) and some of the examples in (10) are admittedly not fully grammatical, as indicated by the question mark. What is relevant, however, is the sharp contrast between the examples in (9-a) and (9b), and between the first sentences in (10-a) - (10-c) compared to the second and third sentences in (10-a) - (10-c). Third, sentences parallel to those in (8) are starred by Reis (1996, p. 64). 8 However, we consider these sentences fully grammatical; in addition a web search via Google for these sentences gave many hits. This means at least that the judgments of Reis are not shared by all speakers. We suspect that one reason for the controversial status of such structures lies in their requirement to be assigned an appropriate intonation. Stress on the wh-phrase or the verb of the V1Ρ makes the sentence sound very odd, whereas stress on the subject of the V1Ρ or the predicate belonging to the wh-phrase makes the sentences fully grammatical, illustrated with example (8-b) : (12) a.

* WER glaubst du denn hilft dir gerade ? * Wer GLAUBST du denn hilft dir gerade? b. Wer glaubst DU denn hilft dir gerade ? Λ/ Wer glaubst du denn HILFT dir gerade?

Given the general validity and attractiveness of the PA approach for the CC, it is hard to account for the differences between post- and prefinite VICs that we have just observed. The PA approach predicts all examples in (7) - ( 11 ) to be bad. The sentences in (7) (10) are predicted to be bad because they involve non-well-formed parentheticals, as can be determined by comparing the CC with other undisputed, postfinite parentheticals: (13) a. γ/ Die Chomsky-Hierarchie ist relevant glaubt Peter. the Chomsky hierarchy is relevant thinks Peter 8

Although she notes (1996, p. 82, fn. 26) that the prefinite versions are much better than the postfinite ones.

134

Andreas Pankau, Craig Thiersch, Kay-Michael b.

c.

d.

Würzner

* Die Chomsky-Hierarchie

ist relevant glaubt wer. ..., thinks who * Die Chomsky-Hierarchie ist relevant glaubt er eigentlich. ..., thinks he actually * Die Chomsky-Hierarchie ist relevant glaubt er denn. ..., thinks he MP * Die Chomsky-Hierarchie ist relevant glaubt er nicht. ..., thinks he not (a,b,c)'The Chomsky hierarchy is relevant,... ' * Die Chomsky-Hierarchie brauchen wir zu lernen glaubt er nicht. the Chomsky hierarchy need we to learn thinks he not (lit)'We don't have to learn the Chomsky hierarchy, doesn't he believe.'

We remain agnostic as to why integrated parentheticals with negation, modal particles, and interrogative wh-phrases are not well-formed; what is relevant for our discussion is the fact itself, which precludes an analysis of (7) - (10) along the lines of the PA.9 Although the sentences in (11) involve well-formed parentheticals, they are nevertheless predicted to be ungrammatical because the V I P is inserted into a non-well-formed host clause, viz. an infinitival, non-root clause: (14)

* Wen vor dir zuhaben? whoAcc in front of youDAT to have (lit.)'Who to be in front of you?'

3. An Unexpected Parallelism to Extraction Constructions Interestingly, though, the pattern of grammaticality in (7) - (11) is identical to the pattern of grammaticality in (15) - (16), i.e. to analogous sentences in undisputed extraction constructions such as extraction out of ¿toss-clauses (15) and wh-copying (16): 10 (15) a. \J Wen glaubt wer dass sie gesehen hat? whoACC thinks whoNOM that she seen has (lit.) 'Whom does who think she saw?' 9

Integrated parentheticals with interrogative wh-phrases are probably not well-formed because integrated parentheticals are declarative in nature. The ungrammaticality of integrated parentheticals with the modal particles might be related to the fact that the two modal particles under discussion are speaker-oriented (for denn: Bayer, 2008; for eigentlich: Nehls, 1989, p. 286); this, however, clashes with the function of the parenthetical to put the host clause into the perspective of the subject of the parenthetical (Reis, 1996, p. 47). We have no idea why the combination of parenthetical + negation is ungrammatical.

10

Most of the judgments for (15) are taken from the literature; the judgments in (16) were provided by the first author.

Spurious Ambiguities and the Parentheticals Debate

135

b. \J Wen glaubt er eigentlich dass sie gesehen hat? whoACC thinks he actually that she seen has y Wen glaubt er denn dass sie gesehen hat? whoACC thinks he MP that she seen has c. ? Wen glaubt er nicht dass sie gesehen hat? whoACC thinks he not that she seen has (b,c) 'Who does he (actually/MP/not) think she saw?' d. ? Wen glaubst du nicht dass er damit sonderlich zufriedenstellen wird? whoACC think you not that he with-it particularly satisfy will 'Who don't you think that he will particularly satisfy with that?' (16) a. yj Wen glaubt wer wen sie gesehen hat? whoACC thinks whoNOM whoACC she seen has (lit.)'Whom does who think she saw?' b. sj Wen glaubt er eigentlich wen sie gesehen hat? whoACC thinks he actually whoACC she seen has yj Wen glaubt er denn wen sie gesehen hat? whoACC thinks he MP whoACC she seen has c. ? Wen glaubt er nicht wen sie gesehen hat? whoACC thinks he not whoACC she seen has (b,c) 'Who does he (actually/MP/not) think she saw?'" d. ? Wen glaubst du nicht wen er damit besonders beeinflussen wird? whoAcc think you not whoACC he with-it especially influence will 'Who don't you think that he will especially influence with that?' The complementizer dass and wh-copying are confined to finite clauses only; we will therefore illustrate the grammaticality of extraction out of infinitival clauses with topicalization: (17)

[w Den Peter\i glaube ich [s2 t, erkannt zuhaben]] the PeterAcc believe I recognized to have Ί believe to have recognized Peter.'

Note that the judgments for the extraction constructions are completely parallel to the judgments for the a-examples in (7) - (11), i.e. for those examples for which the CA assumes a related structure. Some remarks concerning the judgments in (15) - (16) and our claim about the parallelism seem to be necessary by now. First, although we marked (15-a) as grammatical, there is actually speaker variation concerning the grammaticality of such structures. Some speakers consider these sentences ungrammatical (Grewendorf, 2002, p. 285), some speakers do not feel any deviance (Fanselow, 2004; Fanselow & Féry, ' 1 The literature contains the claim that wh-copying differs from extraction out of dass-clauses in German w.r.t. negation in matrix clauses (Reis, 2002, p. 395; Fanselow & Mahajan, 2000, p. 219). In a careful examination of wh-copying, Pankau (2007) could not replicate this claim. The speakers he consulted behaved consistently, i.e. they either licensed extraction over negation or not, independently of the specific extraction construction. This result suggests that the claim in the literature is due to mixing up judgments from various speakers.

136

Andreas Pankau, Craig Thiersch, Kay-Michael

Würzner

2008). Interestingly, we observe that there is no difference in grammaticality between the CC with a crossed wh-phrase in the matrix clause and the corresponding extraction constructions. This means: speakers that judge (15-a) (or (16-a)) as grammatical judge (7-a) as grammatical as well, and speakers that judge (15-a) (or (16-a)) as ungrammatical judge (7-a) as ungrammatical as well. 12 Second, the same idiolectal variation shows up with extraction crossing negation; the only difference is that this kind of extraction never gives rise to full grammaticality. And similarly, regardless of the specific judgments for extraction over negation, we observe that the judgments for the extraction construction parallel those for the CC. 13 After having discussed some possible objections regarding the judgments and some empirical observations about the judgments, let us clarify our main observation concerning the connection between (7) - (11) and (15) - (16). We hope that it has already become clear that we do not claim to have observed that all speakers will necessarily agree on the specific judgments they assign to each of the examples in (7) - (11) and (15) - (16). We accept that there is speaker variation concerning the judgments on extraction crossing negation, extraction crossing wh-phrases, or on the general availability of each specific extraction construction in the first place. What we do observe, however, is that the judgments for extraction constructions with (a) wh-phrases, (b) modal particles, (c) negation, (d) NPIs, and (e) infinitival clauses (cf. (17)) are always identical to the judgments given for the CC with (a) - (e). This simply means that (i) although there is variation in the judgments among speakers, there is (ii) also a clear pattern in the range of variation: the judgment given by each speaker for each extraction construction containing (a) - (e) is identical to the judgment given for the analogous CC containing (a) - (e) independent of the specific judgments itself.

4. Analysis In order to deal with the paradoxical observation that the CC behaves like a parenthetical construction in certain respects but like an extraction construction in other respects, we propose that the strings corresponding to CCs are in general structurally ambiguous, i.e. both structures (1) and (2) are in principle legitimate structural descriptions. Before we explicate our idea in more detail, we will clarify some terminology that is relevant for a proper understanding of our analysis. First, we make a distinction between a string and a structure. When we refer to a string, we simply refer to a concatenation of words. By structure (or structural description), we refer to a theoretical object that describes certain aspects of that string. Second, we distinguish between grammaticality and wellformedness. Grammaticality is a property of a string that indicates whether a string is part of the speaker's language: if a string is grammatical, then it is part of the speaker's language; if a string is ungrammatical, then it is not a part of the speaker's language. 12

Günther Grewendorf stars both examples (p.c., 23.01.2010), whereas Gisbert Fanselow considers both examples grammatical, although he slightly prefers ( 15-a) over (7-a) (p.c., 25.01.2010).

13

Therefore, the NPI data only apply to those speakers that allow extraction over negation.

Spurious Ambiguities and the Parentheticals

Debate

137

Well-formedness on the other hand is a property of a structure. A structure is well-formed if it violates no principle of the speaker's grammar; a structure is not well-formed if it does violate some principle of the speaker's grammar. 14 The connection between strings and structures is such that for each grammatical string, there exists at least one wellformed structure. The qualification 'at least' is necessary in order to cope with structural ambiguities. A string is structurally ambiguous if there are at least two structures which can be assigned to this string: ( 18) String: flying planes can be dangerous Structure 1: [s t [s2 Aying planes] [yp can be dangerous]] Structure 2: [S [NP flying planes] [vp can be dangerous]] Let us now turn to the analysis of the data discussed in Section 2. As we mentioned above, we assume that the strings corresponding to the CC are generally structurally ambiguous: ( 19) String: CA structure: PA structure:

wen glaubt er hat sie gesehen [s¡ went glaubt er [ S2 t',: hat sie t¿ gesehen]] —> well-formed [s went {glaubt er] hat sie t, gesehen] —> well-formed

grammatical Since neither the structure according to the CA nor the structure according to the PA violates any constraint, the string corresponding to the CC is grammatical. 15 Turning to the strings in (7) - ( 11 ), we generally assume that none of them is structurally ambiguous. The ambiguity is lost because no PA structure can be assigned to any of them. The failure to assign a PA structure is due to the constraints on parentheticals illustrated in (13) - (14). These constraints would be violated if a PA structure were to be assigned to them. We then note a distinction between the a-examples on the one hand, and the b-examples on the other hand. The a-examples are grammatical, the b-examples are ungrammatical. This distinction falls out easily in our analysis. Whereas there exists a well-formed structure for the a-examples, no such well-formed structure exists for the b-examples. The well-formed structure for the a-examples is the one according to the CA, cf. ( 1 ). No such structure, however, can be assigned to the b-examples because these structures would violate the constraints on extraction that we illustrated in (4). We will now exemplify our analysis by discussing each of the strings in (9) starting with (9-a).

14

We d o not use the acceptability/grammaticality distinction here because this terminology is mainly used in discussion revolving around the question as to which objects exactly belong to a language. Since we want to avoid misunderstanding, we did not employ this terminology.

15

In a detailed presentation of our analysis, w e would provide s o m e a r g u m e n t s as to w h y the structures we assume are well-formed, i.e. provide the reader with a f r a g m e n t of a grammar. Since a detailed presentation is beyond the scope of this paper, w e simply note that these structures are c o m m o n l y a s s u m e d in the literature for extraction and parentheticals.

138 (20)

Andreas Pankau, Craig Thiersch, Kay-Michael String: CA structure: PA structure:

Würzner

wen glaubt er nicht hat sie gesehen ^ wen,, glaubt er nicht fs 2 t'.( hat sie t,; gesehen]] —> well-formed [s wenL {glaubt er nicht) hat sie t, gesehen] ->· not well-formed

—» grammatical Of the two possible structures, only the CA structure is well-formed because it violates no principle of the grammar. The PA structure, however, is not well-formed because it violates the constraints on parentheticals that we established in (13) - (14). Since it is sufficient for a string to be grammatical if there is at least one well-formed structure that it can be assigned to, the string will be predicted to be grammatical, in accordance with the judgments. Let us now turn to first sentence in (9-b): (21)

String: CA structure: PA structure:

wen hat glaubt er nicht sie gesehen [s ([wen [hat)i glaubt er nicht ()¿ sie gesehen]]] -»• not well-formed [s weni hat (glaubt er nicht} sie t,: gesehen] —¥ not well-formed

—>• ungrammatical Of the two possible structures, neither is available. The CA structure is not well-formed because it violates the constraint on extraction structures according to which only constituents are targets for extraction. Since the PA structure still violates the constraints on parentheticals, the string can't be assigned a well-formed PA structure either. We therefore correctly predict the string to be ungrammatical. The same line of reasoning applies to the second sentence in (9-b): (22)

String: CA structure: PA structure:

wen hat sie gesehen glaubt er nicht [si [s 2 weni hat sie t¿ gesehen]k glaubt er nicht tfc] —¥ not well-formed [s weni hat sie t, gesehen {glaubt er nicht}] —• not well-formed

—>• ungrammatical There is no well-formed structure corresponding to the CA because it would involve clausal pied-piping, an option generally not available for German. And the PA structure suffers from the same defect as the previous ones, viz. that it violates the constraints on parentheticals. Again, we correctly predict the string to be ungrammatical. So far, we have only discussed three out of four possible cases: strings for which (i) both structures are well-formed (cf. (19)), (ii) neither structure is well-formed (cf. (21) and (22)), (Hi) only the CA structure is well-formed (cf. (20)). The fourth case, only the PA structure is well-formed, can be found in the sentences (3-a) and (3-b). Their respective structures are illustrated in (23):

Spurious Ambiguities

and the Parentheticals

(23) a. String: CA structure: PA structure:

Debate

139

wen hat glaubt er sie gesehen [s ([wen [hat)i glaubt er sie gesehen]]] —> not well-formed [s wenl hat {glaubt er} sie t ( gesehen] well-formed

—> grammatical b. String: CA structure: PA structure:

wen hat sie gesehen glaubt er [sj [s 2 wen, hat sie t¿ gesehen]^ glaubt er —> not well-formed [s wen, hat sie t¿ gesehen {glaubt er}] —¥ well-formed

grammatical In neither PA structure are any of the constraints on parentheticals violated, both CA structures, however, violate constraints on extraction. The grammaticality patterns of the strings containing NPIs (cf. (10)) requires some additional discussion. Note first that a matrix negation is a possible licensor for embedded NPIs: (24) a. \J Ich glaube nicht dass er damit sonderlich zufrieden I think not that he with-it particularly satisfied Ί don't think that he is very satisfied with that.' b. Ich glaube nicht dass er daraus einen Hehl gemacht I think not that he out of it a secret made Ί don't think that he made a secret out of it.' c. y/ Ich glaube nicht dass er das nochmal zu tun braucht. I think not that he this again to do needs Ί don't think that he needs to do that again.'

sein wird. be will hat. has

Second, it is the presence of negation that licenses NPIs; extraction of interrogative whphrases alone cannot license them: (25) a. * Womit ist er sonderlich zufrieden? with what is he particularly satisfied 'What is he very satisfied with?' b. * Woraus hat er einen Hehl gemacht? of what has he a secret made 'What did he make a secret of?' c. * Was braucht er zu tun? what A cc needs he to do 'What does he need to do?' Now, whatever factor is responsible for successful licensing of NPIs by negation in (10a), it is clear that this factor is confined to extraction constructions only (cf. (15-d) and (16-d)). For if it was active in parentheticals as well, then we would expect the strings in

140

Andreas Pankau, Craig Thiersch, Kay-Michael

Würzner

(10-b) to be grammatical. This contrast then shows - although only in an indirect manner - that a parenthetical structure for the strings in (10-a) must be unavailable. There is an unusual aspect of this analysis that makes it unique, and this is perhaps why it has not been pursued before in the literature: the structural ambiguity comes with no interpretational difference. Note that in (18), the structures not only differ syntactically but also semantically. According to (18-a), it is dangerous to pilot planes, whereas according to (18-b), planes that fly belong to the set of dangerous objects. Now, no interpretational difference is connected to the two structures in (19). Either one roughly means: which χ is such that he believes that she saw x. Ambiguities with no semantic impact are labeled spurious ambiguities (Karttunen, 1989, p. 57); spuriously ambiguous structures are said to be derivationally equivalent (Hepple & Morrill, 1989). The reputation spurious ambiguities have gained in syntactic theorizing ranges from being a standard assumption (Categorial Grammar) to being explicitly rejected (HPSG; e.g. Müller, 2008). There were however so far no clear empirical results that could let one decide. We hope to have provided some results that show that spurious ambiguities must not be excluded, at least for the syntax of German.

5. Predictions Our proposal makes two predictions: (26) Prediction 1 Strings corresponding to uncontroversial parenthetical constructions as illustrated in (4c) are always ungrammatical w.r.t. the factors discussed in Section 2 Prediction 2 Other forms of spurious ambiguities must exist in natural languages

5.1 Prediction 1 So far, we have talked about only one type of the CC, viz. the type involving verbs that belong to the class of so-called bridge verbs (such as glauben, 'believe'), i.e. verbs that license long distance extraction. There is however another type of CC that involves verbs such as fragen, 'to ask', as the predicate of the VIP: (27)

yj Wem fragt er hat sie geholfen? whoDAT asks he has she helped (lit.)'Who does he ask that she helped?'

Note that this type of the CC behaves like regular parentheticals as the VIP is not confined to the prefinite position:

Spurious Ambiguities and the Parentheticals Debate

141

(28) a. γ/ Wem hat fragt er sie geholfen? whocAT has asks he she helped b. y/ Wem hat sie geholfen fragt er? whoDAT has she helped asks he (a&b, Iit.)'Who does he ask that she helped?' fragen, however, differs crucially from verbs such as glauben in that it does not belong to the class of bridge verbs, i.e. it cannot function as matrix predicate in long distance extraction: (29) a. * Wem fragt er dass sie geholfen hat? whoDAT asks he that she helped has b. * Wem fragt er wem sie geholfen hat? whoDAT asks he w I i o d a t she helped has (a&b, lit.)'Who does he ask that she helped?' Since fragen is not licensed in uncontroversial extraction constructions, an analysis along the lines of the CA for (27) is unavailable because it involves a matrix predicate that cannot appear in a long distance extraction in the first place. We therefore predict that this type of CC will always be ungrammatical when combined with the factors negation, NPIs, wh-phrases, modal particles, and infinitival host clauses, irrespective of the position of the VIP. This prediction is confirmed: 16 (30) a.

Wh-Phrases * Wem fragt wer hat sie geholfen ? whoDAT asks whoNOM has she helped * Wem hat fragt wer sie geholfen ? whoDAT has asks whoNOM she helped * Wem hat sie geholfen fragt wer? whoDAT has she helped asks whoNOM (lit.)'Who does who ask that she helped?' b. Modal Particles * Wem fragt er denn hat sie geholfen? whoDAT asks he MP has she helped * Wem hat fragt er denn sie geholfen? whoDAT has asks he MP she helped * Wem hat sie geholfen fragt er denn ? whoDAT has she helped asks he MP (lit.)'Who does he MP ask that she helped?'

16

Cf. Wagner (2004, p. 205) for the same judgements on (30-b.)

142

Andreas Pankau, Craig Thiersch, Kay-Michael

Würzner

c.

Negation * Wem fragt er nicht hat sie geholfen? whoDAT asks he not has she helped * Wem hat fragt er nicht sie geholfen? whoDAT has asks he not she helped * Wem hat sie geholfen fragt er nicht? whoDAT has she helped asks he not (lit.)'Who doesn't he ask that she helped?' d. NPIs * Wem fragt er nicht braucht man das vorzulesen ? whoDAT asks he not needs one this read to * Wem braucht fragt er nicht man das vorzulesen? whoDAT needs asks he not one this read to * Wem braucht man das vorzulesen fragt er nicht? whoDAT needs one this read to asks he not (lit.)'Who doesn't he ask that one needs this to read to?' e. Infinitival Host Clauses * Wem fragt er geholfen zu haben? whoDAT asks he helped to have * Wem geholfen zu haben fragt er? whoDAT helped to have asks he (lit.)'Who does he ask to have helped?'

5.2 Prediction 2 The second prediction is also confirmed by certain string-vacuous instances of Heavy NP Shift (HNPS). HNPS is a construction in which a direct object is (optionally) put at the end of a clause if the object is heavy, i.e. either (i) long, or (ti) focused. If neither condition is fulfilled, the construction is ungrammatical: (31) a.

? Peter read [some boring articles about ancient Greek philosophy] yesterday y/ Peter read t( yesterday [some boring articles about ancient Greek philosophy]¿ b. y j I picked HIM/him up y/1 picked ti up HIM¿ * I picked tj up him, c. y j Peter introduced PTQ to students who soon became his colleagues * Peter introduced t¿ to students who soon became his colleagues PTQ¿

HNPS is not simply a late stylistic reordering of constituents that puts heavy NPs into clause-final position. It affects binding of anaphors (32), operator-variable binding (33), and NPI licensing (34):

Spurious Ambiguities and the Parentheticals (32)

(33)

Debate

143

a. y j Mary described [the old blind mani on the park bench] [to himselfi] ?? Mary described t¿ [to himselfi] [the old blind mani on the park bench], b. * John described [herselfi as a young girl] [to Maryi] y/ John described t¿ [to Maryi] [herselfi as a young girl], * I introduced [hisi partner] [to everyi student in the chemistry lab] 1 introduced t» [to everyi student] [hisi partner in the chemistry lab|,

(34) a. y / l revealed * I revealed b. * I revealed y j I revealed

[no secret] [to anyone] t¿ [to anyone] [none of the secrets you told me], [anything] [to no-one] t, [to no-one] [any of the secrets you told me]¿

Note that these effects are restricted to direct and indirect objects. HNPS never affects binding in case the subject is an anaphor (35-a), operator-variable binding between subject and direct object (35-b), or NPI licensing between subject and direct object (35-c): (35) a. "Himselfi described [the mani with big brown eyes and a mustache] to Mary 'Himself ι described t, to Mary [the mani with big brown eyes and a mustache]¿ b. *Hisi neighbor described [everyi student] [to the police]. *Hisi neighbor described t, [to the police] [everyi student with long hair],. c. * Anyone revealed [no secret] [to the congress]. * Anyone revealed t, [to the congress] [no secret that the government hides] j. Regardless of one's favorite theory for binding of anaphors, operator-variable binding, and NPI licensing, there is strong agreement among the frameworks that these phenomena are not licensed at the phonological or prosodie level. What the examples in (32) (35) then show is that HNPS clearly involves a distinct syntactic structure, and not late stylistic reordering of constituents, because late stylistic reordering is a phonological or prosodie operation qua definitionem. Accordingly, it is not unexpected that HNPS has an effect on all these phenomena since syntactic structures are what define the relevant domains for binding of anaphors, operator-variable binding, and NPI licensing. We can therefore safely conclude that a heavy direct object NP can appear in two distinct syntactic positions: in its base position or in a derived position which has the property of being linearized after clause-final adverbs and indirect objects. Then, however, nothing excludes these two options for sentences that lack both clause-final adverbs and indirect objects, i.e. sentences in which the base position of a heavy NP object is accidentally already clause-final: (36)

String: Peter read several boring articles about ancient Greek philosophy Structure 1 : Peter read [several boring articles about ancient Greek philosophy] Structure 2: Peter read t¿ [several boring articles about ancient Greek philosophy];

Note that in such cases of simple transitive sentences, no interpretational difference affecting binding of anaphors, operator-variable binding, or NPI licensing is expected. Irrespective of whether or not the sentence contains an anaphor, a QP, or an NPI, we have shown in (35) that HNPS never creates new interpretational options for subjects and ob-

144

Andreas Pankau, Craig Thiersch, Kay-Michael

Würzner

jects w.r.t. binding of anaphors, operator-variable binding, and NPI licensing. As no difference affecting subjects and direct objects can be found in structures with or without HNPS, we can conclude that HNPS in simple transitive clauses, i.e. in (36), involves a structural ambiguity to which no semantic effect is connected. Since this is another instance of a spurious ambiguity, our second prediction is confirmed as well.

6. Conclusion To sum up, we first have shown that there are contexts in which CCs behave like undisputed extraction constructions, and not like undisputed parenthetical constructions. Second, in light of examples that show the opposite, we argued that CCs involve a structural ambiguity to which no semantic ambiguity is connected, a so-called spurious ambiguity. We further argued that this ambiguity can get lost under specific conditions, which leads to the assignment of only one structure to a CC. We then discussed two predictions of this analysis. The first prediction we discussed is that we expect instances of the CC in which no structure can be assigned to it; we presented evidence that CCs involving non-bridge verbs under certain conditions are such a case at hand. The second prediction is that other instances of spurious ambiguities should be found; we argued that Heavy N P Shift in clauses with transitive verbs confirms this prediction.

Acknowledgements All three authors of this paper have known Peter Staudacher for many years at the University of Potsdam, variously as teacher, colleague, boss, and friend. I (Thiersch) also knew him in an earlier incarnation, as Professor in Regensburg. Many years ago, in the early 1980s, a group of linguists in Northern Germany and the Netherlands formed a small, informal group to meet from time to time (in the wake of Chomsky's Pisa Lectures) to discuss current topics in syntax and their own work. Meetings were quite free form, and coming from all over Germany and the Netherlands, we lodged at each others homes. Gradually the group extended southward via Frankfurt to Passau and Regensburg, where we were hosted by Peter (and his family). The atmosphere was stimulating but gemütlich. Peter kept us on our toes with probing questions, so there was never a lull in the discussions. What a surprise then, to discover after a hiatus of several years, that we were to be colleagues again, and Peter was to be my boss in Potsdam on a project dealing with mathematical properties of the minimalist program. Peter was an ideal boss, smoothing the ruffles of the Potsdam administration, always available for discussions, and a sympathetic friend as well - 1 have fond memories of Christmas parties at the Staudacher family home. His penetrating questioning again kept me and my colleagues hopping - 1 can remember in the later stage of our project a "short" discussion of remnant movement turning into a five day marathon, replete with handouts and meeting both in Potsdam and ZAS in Berlin. Often his observation sharpened my thinking on linguistics issues. And discussions about linguistics were always enlivened by his erudition in the classics and classical languages.

Spurious Ambiguities and the Parentheticals Debate

145

Both Andreas and Kay-Michael enjoyed and profited from his dedicated teaching in their seminars and the philosophical perspective he brought to the often intense discussion in and out of the classroom. We wish him well in retirement and hope that he enjoys having time to pursue his hobby of classical scholarship as a second career.

References Bayer, J. (2008). Wie erreicht denn denn die linke Satzperipherie. Talk at the Workshop Generative Grammatik des Südens (GGS 08). Berman, J. (2000). Topics in the Clausal Syntax of German. Ph.D. thesis. University of Stuttgart. Bildhauer, F. & P. Cook (2010). Mehrfache Vorfeldbesetzung. Handout, Lecture at University of Potsdam, Feb. 11, 2010. Fanselow, G. (2004). The MLC and Derivational Economy. In: A. Stepanov, G. Fanselow & R. Vogel (eds.) Minimality Effects in Syntax. (Studies in Generative Grammar, vol. 70). Berlin: De Gruyter. 73-124. Fanselow, G. & C. Féry (2008). Missing Superiority Effects: Long Movement in German (and other languages). In: J. Witkos & G. Fanselow (eds.) Elements of Slavic and Germanic Grammars: A Comparative View. (Polish Studies in English Language and Literature, vol. 23). Frankfurt: Lang. 67-87. Fanselow, G. & A. K. Mahajan (2000). Towards a Minimalist Theory of Wh-Expletives, Wh-Copying, and Successive Cyclicity. In: U. Lutz, G. Müller & A. v. Stechow (eds.) Wh-Scope Marking. (Linguistik Aktuell/Linguistics Today, vol. 37). Amsterdam: John Benjamins. 195-230. Grewendorf, G. (1988). Aspekte der deutschen Syntax. Eine Rektions-Bindungs-Analyse. (Studien zur deutschen Grammatik, vol. 33). Tübingen: Narr. Grewendorf, G. (2002). Minimalistische Syntax. (UTB, vol. 2313). Tübingen: Francke. Hepple, M. & G. Morrill (1989). Parsing and Derivational Equivalence. In: Proceedings of the Fourth Conference of the European Chapter of the Association for Computational Linguistics. Morristown, NJ: Association for Computational Linguistics. 10-18. Karttunen, L. (1989). Radical Lexicalism. In: M. R. Baltin & A. S. Kroch (eds.) Alternative Conceptions of Phrase Structure. Chicago: University of Chicago Press. 43-65. Kiziak, T. (2007). Long Extraction or Parenthetical Insertion? Evidence from Judgement Studies. In: N. Dehé & Y. Kavalova (eds.) Parentheticals. (Linguistik Aktuell/Linguistics Today, vol. 106). Amsterdam: John Benjamins. 121-144. Müller, S. (2008). Head-Driven Phrase Structure Grammar: Eine Einführung. (Stauffenburg Einführungen, vol. 17). 2nd modified edn. Tübingen: Stauffenburg. Nehls, D. (1989). German Modal Particles Rendered by English Auxiliary Verbs. In: H. Weydt (ed.) Sprechen mit Partikeln. Berlin: De Gruyter. 282-292. Pankau, A. (2007). Die w/i-Kopie-Konstruktion im Deutschen. Diploma Thesis, University of Potsdam. Reis, M. (1995). Wer glaubst du hat recht? On So-Called Extractions from Verb-Second

146

Andreas Pankau, Craig Thiersch, Kay-Michael

Würzner

Clauses and Verb-First Parenthetical Constructions in German. Sprache & Pragmatik 36. 27-83. Reis, M. (1996). Extractions from Verb-Second Clauses in German? In: U. Lutz & J. Pafel (eds.) On Extraction and Extraposition in German. (Linguistik Aktuell/Linguistics Today, vol. 11). Amsterdam: John Benjamins. 45-120. Reis, M. (2002). Wh-Movement and Integrated Parenthetical Constructions. In: J.W. Zwart & W. Abraham (eds.) Studies in Comparative Germanic Syntax, Proceedings from the 15th Workshop on Comparative Germanic Syntax. (Linguistik Aktuell/Linguistics Today, vol. 53). Amsterdam: John Benjamins. 3-40. Riemsdijk, H. v. (1982). Zum Rattenfängereffekt bei Infinitiven in deutschen Relativsätzen. In: W. Abraham (ed.) Erklärende Syntax des Deutschen. (Studien zur deutschen Grammatik, vol. 25). Tübingen: Stauffenburg. 75-98. Staudacher, P. (1990). Long Movement from Verb-Second-Complements in German. In: G. Grewendorf & W. Sternefeld (eds.) Scrambling and Barriers. (Linguistik Aktuell/Linguistics Today, vol. 5). Amsterdam: John Benjamins. 319-340. Tappe, H. T. (1980). Wer glaubst du hat recht? Einige Bemerkungen zur COMP-COMPBewegung im Deutschen. In: M. Kohrt & J. Lenerz (eds.) Sprache: Formen und Strukturen Akten des 15. Linguistischen Kolloquiums Münster, (vol. 1). Tübingen: Niemeyer. 203-212. Thiersch, C. (1978). Topics in German Syntax. Ph.D. thesis. MIT. Wagner, M. (2004). Asymmetries in the Syntax and Prosody of Verb-Initial Interpolated Clauses. In: S. Blaho, L. Vicente & M. de Vos (eds.) Proceedings of Console XII. University of Patras.

Part II

Semantics

On Squeamishness of the Royal Kind Christian Ebert

Cornelia

Ebert

1. Introduction In this paper we are concerned with the notorious lack of hair ascribed to the French monarch as well as with his interest in exhibitions as stated in the following pair of sentences. (1)

a. b.

The King of France is bald. The exhibition was visited yesterday by the King of France.

This pair exemplifies a peculiar contrast that has been discussed in the linguistic community at least since Strawson (1964). While hearers who are aware of the non-existence of a French king will feel squeamish about (1-a) they will judge (1-b) as plainly false. Strawson (1964) explains this contrast along the following lines. As starting point, definite descriptions such as the king of France are taken to introduce a presupposition asking for the existence of a suitable referent. 1 This contrasts with the Russellian treatment of definite descriptions, where this statement of existence is construed as part of the semantics proper. Under the presuppositional Strawsonian view, the hearer's squeamishness in case of (1-a) is expected: since his belief is contradictory to the presupposition he is not able to judge the sentence as either true or false. However, one would expect the same squeamishness in case of (1-b) along the same lines of reasoning. Strawson took the observed judgements as a direct indication of the presuppositionality of the description in question. Since (1-a) induces squeamishness and (1-b) is judged as false, he concluded that the first carries the (failed) existential presupposition while the latter does not carry that presupposition at all. He then resorted to the information structural notion of topicality to explain the presence/absence of the presupposition. Here topicality is understood in the aboutness sense, i.e. as marking that constituent which denotes an object about which the sentence makes a claim. Strawson held the view that only topical definite descriptions exhibit their presupposition. Non-topical definite descriptions on the other hand experience absorption by the predicate (thus forming a kind of complex predicate), a process that basically erases the presupposition. Von Fintel (2004) argued against both aspects of Strawson's explanation. First, truth value judgements cannot be directly taken as an indication of the presuppositionality of 1

We will ignore the presupposition o f uniqueness here and in the following.

150

Christian Ebert, Cornelia Ebert

a sentence, he argues. In particular, a sentence such as (1-b) can be judged false despite a failed presupposition. Von Fintel proposes that the judgement of falsity in these cases can be explained by a pragmatic principle of contextual revision (based on an earlier proposal by Lasersohn, 1993). The rough idea is, that a hearer might revise his beliefs about a failed existential presupposition introduced by a definite description if this helps to falsify the sentence on independent grounds. At this point he opposes the second aspect of Strawson's explanation (i.e. that topicality induces squeamishness) by listing sentences judged as false despite the fact that they contain topical definite descriptions. While we subscribe to von Fintel's argument that truth value judgements are not helpful for deciding questions of presuppositionality, we reject the second. In fact, we will defend the following hypothesis:

(2)

Squeamish Topic Hypothesis (STH) A topical definite description which is non-referring induces squeamishness.

To this end, we will show that all arguments from (Lasersohn, 1993; von Fintel, 2004) against this hypothesis are flawed by either misanalysis of the information structural findings or by misconstrual of the context of occurrence of the definite description. According to the hypothesis in (2), we expect squeamishness to arise with sentences that mark non-referring definites as topics. On the other hand, we do not exclude that nontopical definite descriptions may give rise to presuppositions. In this position we differ from Strawson (1964), who put forth that the presupposition of definites vanishes in the process of absorption, a position that cannot be upheld in light of the counterexamples of von Fintel (2004). The remainder of this article is as follows. In Section 2 we will take a close look at alleged counter-examples to the STH and we will argue that each one is flawed in one way or other. The third section will introduce the approach to topic interpretation put forth by Endriss 2 (2009), which gives an explanation for the observation stated in the STH. In the fourth section we will discuss von Fintel's contextual revision principle in light of the preceding discussion and Section 5 will conclude this article.

2. Topicality Induces Squeamishness In the following we will take a careful look at the data presented as counterevidence to the STH. We argue that most of the data do not actually threaten it but are misconceived w.r.t. the information structural findings. Let us start with the following example uttered by 'a speaker who points at an obviously empty chair' (Lasersohn, 1993, ex. (2)). (3)

F

The King of France is sitting in that chair.

Lasersohn observes that this statement in fact seems to be simply false, and he expresses doubt that this can be explained via a dependence on the topic-comment articulation of 2

w h o happens to be co-referential with the second author of this article.

On Squeamishness of the Royal Kind

151

the statement. We think that it is highly implausible that (3) is about the King of France in the given scenario in the first place. Since there is a highly salient, obviously empty chair pointed at by the speaker (3) is more likely about that chair. In fact, once we unambiguously mark the definite as topical, we arrive at a sentence that strikes us as much more squeamish than false. 3 To this end, we make use of the fact that left dislocation in German has exactly the desired effect of marking the dislocated constituent as aboutness topic (cf. e.g. Frey, 2004): 4 (4)

#Der König von Frankreich, der sitzt auf diesem Stuhl. the king of France RP sitPRES on this chair 'The King of France is sitting in this chair.'

It is even more plausible, however, that (3) constitutes an instance of a topic-less or thetic sentence. Such sentences are used to describe a state of affairs without being about an entity playing a role in the uttered proposition. Thetic sentences thus make good answers to questions like (5)

What's up?

or

What's

happening?

Indeed, (3) makes perfect sense as such a thetic description of a state of affairs, subsequent to a corresponding question as in (5). In fact, in German a paraphrase of (3) in form of a da (engl, there) sentence strikes us as much more natural in this context. 5 (6)

Da sitzt der König von Frankreich auf dem Stuhl. there sitPRES the king of France on the chair

Crucially, there sentences have been argued to be stereotypical instances of thetic sentences. So (3) cannot serve to counter the topic hypothesis: either it is read as a thetic statement without any topic marking on neither the definite nor the demonstrative, or it is read as being about, i.e. with topic marking on the demonstrative rather than the definite. The same criticism applies to the other alleged counterexamples from (Lasersohn, 1993, ex. (3),(4)): (7)

a. b.

F

The King of France (uttered in a situation F The King of France (uttered in a situation

is knocking on the door. where no noise has come from the direction of the door) ate that sandwich. where an obviously untouched sandwich is on the table)

First note that it would be rather peculiar to utter (7-a) out of the blue, i.e. when no noise has come from the door (in other words: when nothing has happened). If it is felicitous at 3

We abbreviate our descriptions of the facts by saying that a sentence is squeamish, instead of stating more precisely that it induces a feeling of squeamishness in a hearer on interpretation. We furthermore mark squeamishness by # and the judgement of falsity by F .

4

In the gloss, RP stands for resumptive

5

Note that a paraphrase in English in form of a there sentence is ungrammatical due to the definitness effect, i.e. the prohibition of definites in existential there sentences.

pronoun.

152

Christian Ebert, Cornelia Ebert

all in such a context, it certainly is not about the king of France but a thetic statement. The same holds for (7-b) although here the statement might also be about the salient sandwich. But again, it is rather unlikely that it is conceived as being about the king of France. Von Fintel (2004) also presents examples in favour of the view that not all topical definites lead to squeamishness as the following (8)

Let me tell you about my friend, the King of France.

First note that the second sentence of (8) sounds rather unnatural with the definite description instead of the pronoun, but let us assume for the moment that it is actually uttered in this way (we will come back to the matter of substituting the definite description by a pronoun below). Von Fintel notes that 'the second sentence [... ] is plainly false. Nevertheless, it would appear to be about the king of France, since it occurs in a discourse that is meant to be about the king.' But this is exactly the problematic aspect of (8). The context is not only meant to be about the king of France, it uses a definite description to introduce him into the discourse. Hence, the crucial sentence to look at is the first and not the second. In fact, if we firmly believed in the non-existence of the French king, we would object to this first sentence by means of a 'Hey, wait a minute' interruption and thus the question of truth or falsity of the second would not even arise. If, on the other hand, we accept the first sentence without comment, we can only do so by accepting the existence of a French king. But then this paves the way for the second sentence to make a false claim about that king. The following example (von Fintel, 2004, ex. 16) is flawed in the same way. (9)

A: B:

Have you heard anything about the king of France recently? I think he may be getting old and decrepit. Well. FBill Clinton had breakfast with him last week and he looked just fine I hear.

For (9) we are supposed to assume the role of an overhearer of a dialogue between A and B. As in the previous example, the definite description occurs in the initial part of the dialogue as part of A's question. Hence, the first and more important point is whether we accept A's question as felicitous. Once we do, it's no surprise that we judge B's utterance as false despite the fact that the king of France constitutes its topic. The same holds for the following example (von Fintel, 2004, ex. (16)), where the definite description is introduced in the first sentence. (10)

F

I had breakfast with the king of France this morning. scrambled eggs.

F

He and I both had

If the first sentence receives a truth value judgement (as false) then this requires that the existence of the king of France is acknowledged beforehand. For instance, von Fintel's contextual revision procedure (that we will discuss below) works by revising the interpreter's information state in a way that entertains the presupposition as a fact such that

On Squeamishness of the Royal Kind

153

the interpreter can arrive at a truth value judgement despite a prior presupposition failure. But against such a revised information state that includes the information that the king of France exists, the second sentence is also judged false. There is another problematic aspect with the examples above that stems from the fact that the definite description is used in the context preceding the sentence under consideration. Note namely that the sentences in (8) (in its more natural reading with a pronoun), (9), and (10) do contain a pronoun and not the definite description under scrutiny. Under the Ε-type view, where pronouns are resolved as definite descriptions themselves, one might predict the same presuppositions as for the antecedent. But then it is no surprise that no presupposition failure is observed on encountering the pronoun, as argued for the cases above. After all, if the sentence containing the antecedent is accepted, the presupposition is satisfied for any follow-up sentence containing a pronoun. However, under the presupposition as anaphora resolution view (cf. e.g. van der Sandt, 1992), all that the pronoun introduces is the presupposition that demands the existence of an accessible discourse referent to which it can be resolved. So under this view (8) in the pronoun variant, (9) and (10) do by no means introduce a presupposition that demands the existence of a French monarch and hence we would not even expect squeamish feelings to arise. So we conclude that all examples that have been put forth to counter a direct connection of topicality and squeamishness do not serve their purpose. Since none of them exhibits the required aboutness topicality on the definite in question, the STH can still be maintained. In the following section we will give an explanation for the observations predicted by the STH.

3. Topic Establishment as a Speech Act Endriss (2009) proposes a principle of topic interpretation that predicts the behavior stated in the STH. In more detail, she proposes that topic-marked DPs are interpreted via a separate speech act of topic establishment, resembling Strawson's ( 1964) speech function of identifying reference. This act of topic establishment has two essential functions. First it finds a suitable representative for the topic-marked DP. Starting from the assumption that all DPs are of generalized quantifier (GQ) type, she argues that one of the minimal (witness) sets of the involved GQ serves this purpose best, since it does not contain any 'superfluous' elements that are not characteristic of the GQ in question. The minimal (witness) sets of a generalized quantifier Q are defined as follows: (11)

MW(£) := \P.Ç(P)

Λ - 3 Q [ Q ( Q ) Λ Vx[Q(x) -»• P(x)}}

The second function of the act of topic establishment is the introduction of a discourse referent for the representative, which will then stand proxy for the original GQ in the originating speech act. In the following we will formally note this act of topic establishment for a topic marked constituent with semantics ψ as (12)

Topx()

154

Christian Ebert, Cornelia Ebert

where X is the discourse referent that is introduced by performance of the act. Crucially, the act of topic establishment is performed before the speech act under consideration. To illustrate this, an assertion of a simple predicative statement such as (13-a) with the topic marking as indicated will receive the analysis (13-b), where & marks speech act conjunction, i.e. subsequent performance of the two involved acts. (13)

a. b.

[JohnJx sleeps T o p x ( AP.P(john) ) & assert( sleep(X) )

Since the (unique) minimal witness set of the GQ AP.P(john)) is just {john}, the resulting interpretation in (13-b) has no visible semantic but only a discoursive effect. First the topic (which happens to be John) is introduced. Then it is asserted that the topic has the property of sleeping. It is in this sense that the topic interpretation of (Endriss, 2009) captures the aboutness character of the form of topicality we look at.6 We can now investigate what this approach to topic interpretation predicts for nonreferring topic-marked definite descriptions. To this end, we will compare its predictions w.r.t. a Strawson (i.e. presuppositional) and a Russellian (i.e. purely semantic) construal of the definite description. Consider a Strawsonian construal of the definite description the king of France first, formally noted by means of the i'oto-operator as Lx.kof(x). Following common practice, we assume that ix.P(x) denotes the (unique) object that has property P , if such an object exists. If such an object does not exist7, L X . P ( X ) is undefined. With this construal, the analysis of the classical example (1-a) comes out as follows, if we assume that the definite description the king of France is topic-marked. (14)

T o p x ( AP.P(ix.kof(x)) ) & assert( bald(X) )

Assuming that there is no king of France in the model and that tx.kof (x) is hence undefined, we end up with a presupposition failure. But crucially, this failed presupposition concerns the Top act of topic establishment and not the assertion. Hence, the topic act cannot fulfil its two functions (selection of a representative; introduction of a discourse referent) and therefore must fail. Interestingly, we get the same effect with a non-presuppositional Russellian construal of the definite expression as in (15-a). The analysis of (1-a) with topic marking on the definite is given in (15-b) for this case. (15)

a. b.

XP3x[kof(x) A P(x)} T o p x ( AP.3a;[kof(2;) Λ P(x)] ) & assert( bald(X) )

Assuming again that there is no object fulfilling the kof predicate, we note that the generalized quantifier in (15-a) is empty. This in turn makes it impossible to derive a suitable 6

The major theme of (Endriss, 2009) is a treatment of (truly) quantificational topic-marked DPs, where this approach is able to explain the exceptional wide scope behavior of indefinites, for instance. This constitutes a case where topic interpretation also has a semantic effect.

7

Again we ignore the issue of uniqueness here.

On Squeamishness of the Royal Kind

155

representative (i.e. to apply MW) and again the Top act of topic establishment necessarily fails. We argue that this is the reason for the felt squeamishness of an interpreter when confronted with non-referring topic-marked definite descriptions - the squeamish feelings result from a failed speech act rather than from a failed presupposition. Crucially, the failed speech act is one of topic establishment and not the assertion proper. This explains why in this case no repair strategy such as von Fintel's contextual revision can help. After all, the speaker tried to introduce a subject of conversation into the discourse, about which her assertion was supposed to be. But if there is no such subject it is hopeless to try to evaluate the assertion. This gives an explanation of the Squeamish Topic Hypothesis (2). As seen before, this approach to topic interpretation yields the same (squeamish) result no matter which analysis for the definite description is assumed. But it seems to be the case that nevertheless a presuppositional construal is necessary in order to explain the full range of data. The following examples (slightly adapted) from (von Fintel, 2004, p. 277f) show this. (16)

a.

b.

A: B:

What about this year 's Field Medal? Who was it awarded to ? It was awarded to the mathematician who proved the Goldbach Conjecture. A: Hey, wait a minute—I had no idea that someone proved the Conjecture. If this year's Fields Medal is awarded to the mathematician who proved Goldbach 's Conjecture, my friend James (who hopes on it himself) will be quite disappointed. F

In the first example (16-a), A's first question establishes the Field Medal as aboutness topic, while the wA-question induces a focus structure on B's subsequent answer, putting the definite description the mathematician who proved the Goldbach Conjecture in focus. A can in tum respond to B's assertion with a Hey, wait a minute reply, questioning the existence of a corresponding mathematician. This clearly shows that the non-topical, focussed definite description exhibits the existence presupposition we are interested in. The same point can be illustrated with (16-b). Here as well a corresponding Hey, wait a minute reply is felicitous, showing that the existence presupposition projects out of the (/-clause. Therefore it seems that we are faced with the following situation. Definite descriptions come with an existence presupposition. If this presupposition is not satisfied due to the lack of a referent, two situations may obtain. If the definite description is topical, the presupposition failure is non-recoverable due to the failure of the topic establishment act, and hence the sentence as such is felt to introduce squeamishness. If on the other hand, the definite description is non-topical, the presupposition failure concerns the originating speech act, e.g. the assertion. In this case the sentence can recover from the presupposition failure and some additional pragmatic principle can nevertheless help the interpreter to derive a truth value judgement. Let us illustrate these considerations again at hand of the classic examples in (1-a) and (1-b). Concerning (1-a), Strawson (1964) already observed that it induces squeamishness

156

Christian Ebert, Cornelia Ebert

on the side of the hearer when uttered out of the blue. We would like to qualify this observation. It is well known that grammatical subjects stereotypically are marked for topicality. Since the definite description is the grammatical subject in (1-a), it may indeed be read as being topic-marked by default, which in tum leads to a failure of the topic establishment act and hence squeamishness according to our explanation from above. But note that once the definite is non-topical, (1-a) may also be regarded as false. Strawson makes this point by regarding the following context for (1-a). (17)

Q: What reigning monarchs are bald? A: FThe king of France is bald.

Again, the question induces an information structure that makes the definite description non-topical. And indeed, the sentence is felt to be false. The same qualification concerns (1-b). Bearing the subject-as-topic preference in mind, the definite description the exhibition constitutes the topic in (1-b) by default. Assuming the existence of a suitable exhibition, the topic establishment act succeeds and it is asserted (about the exhibition) that the king of France visited it yesterday—which is indeed false, if there is no king of France. But once we change (1-b) so as to display the king of France as topic, we should observe squeamishness. At this point, it is important not to make the same mistake as exhibited by (8)—( 10) above. We may not precede (1-b) by any context that itself makes use of the definite description, e.g. by a question such as What about the king of France?. One way to avoid this pitfall is to rephrase (1-b) in active voice and rely on the subject-as-topic preference, as proposed by Strawson (1964) already. (18)

#The king of France visited the exhibition yesterday.

Indeed, (18) induces squeamishness as opposed to (1-b). This illustrates once more that the information structural findings are crucial to an explanation of the observed squeamishness. It remains to devise a principle that makes it possible to explain how an interpreter can derive at a judgement of falsity despite a failed presupposition, as in the many cases seen above. In the following we will take a closer look at von Fintel's proposal concerning such a principle.

4. Contextual Revision Von Fintel goes a long way to derive a principle of conversational revision that essentially allows a hearer to reject a proposition as false on grounds independent of a failed existence presupposition. Glossing over detail, the final version of his revision principle allows a hearer to revise his beliefs by removing propositions from a body of information D (i.e. his stock of knowledge), that (among others) were in D just because the falsity of the presupposition was in D. He should not remove, however, those propositions that

On Squeamishness of the Royal Kind

157

'could be shown to be true by examining the intrinsic properties of a contextually salient entity without at the same time showing that π [i.e. the presupposition] is false.' (von Fintel, 2004, p.290f) Eventually, the hearer should then add the presupposition to D and see whether the so revised D verifies or falsifies the original sentence. Von Fintel arrives at this formulation that mentions the examination of intrinsic properties because the mere presence of a contextually salient entity is not enough to prevent squeamishness. In his words,'the sentence has to make an independently falsifiable claim about the entity referred to.' (von Fintel, 2004, p. 289; von Fintel's emphasis). If we take aboutness as the defining characteristic of topicality, we arrive at the following paraphrase of von Fintel's observation: a sentence may be judged true/false in case it is possible to read it as containing an aboutness topic different from the presupposition failure inducing item. To exemplify, von Fintel argues that the squeamishness in case of (an out of the blue utterance of) (1-a) can be traced back to the lack of a contextually salient entity that a hearer could use as an independent foothold of rejection. However, the same sentence is felt to be false if presented as an answer to a question such as in (17). Von Fintel suggests that in this particular example a contextually salient entity needed for independent falsification is indeed present, namely the set of reigning monarchs, about which the sentence makes a false claim - hence the falsity judgement. But note that (1-a) is felt false in the same way in the context of the following question (19)

Q: Who is bald?

One would have to argue that in the case of question Q the corresponding contextual entity is something like the set of human beings, introduced by the wh-word. As mentioned in the previous section, examples like this rather point towards an information structural explanation. After Q, the king of France is focal and hence non-topical, which seems to facilitate a truth value judgement. Another example that illustrates the conversational revision procedure at work is (1-b). Here a hearer can find a salient entity, namely the mentioned exhibition. By examining its properties, in this case by inspecting its visitors, a hearer could derive at a falsity judgement despite the failed presupposition of the royal definite description. It is puzzling, however, why the same reasoning should not apply to ( 18).8 Since the former is only the passivized variant of the latter, it induces the same presupposition failure. Furthermore, it contains the same entity (viz. the exhibition) of which a hearer could examine intrinsic properties to eventually arrive at a falsity judgement. Therefore, the principle of conversational revision should work for both sentences exactly alike. But this is not bome out, as discussed above. Again, an information structural explanation seems to stand a better chance, since passivization crucially changes grammatical functions and hence indirectly information structure due to the subject-as-topic preference. If we furthermore take our observations in Section 2 seriously, sentences that express thetic statements also come out as false quite naturally. According to our intuitions, the very title of von Fintel's paper constitutes a further example in this respect. 8

Unfortunately, von Fintel does not discuss (18).

158 (20)

Christian Ebert, Cornelia Ebert Would you believe it? The king of France is back!

The initial question is again one that sets the stage for a thetic statement. And indeed, the second sentence strikes us as false rather than squeamish in this context. If we are correct, then the contextual revision procedure from above cannot explain the truth value judgement. Neither the question, nor the second sentence make available any contextually salient entity that could be used as an independent foothold for falsification. These cases of thetic (and thus topic-less) sentences together with the cases of nontopical definite descriptions from above lead to the following generalization: contextual revision is possible if the presupposition failure inducing item is not topical.

5. Conclusion The final picture is this: non-referring definite descriptions that are marked as aboutness topic inevitably lead to squeamishness without the possibility of contextual revision. Nonreferring definite descriptions that are not marked for topicality introduce a presupposition failure which can be overcome such that the sentence under consideration can receive a truth value nevertheless. In the first case, the failed presupposition concerns the speech act of topic establishment, which therefore also fails. It lies in the nature of this speech act that there is no possibility of recovery and thus squeamishness is inevitably induced. In the second case, the presupposition failure affects the assertion. But here some procedure of contextual revision is active and makes a truth value judgement possible. As we argued above, this revision seems to be possible in a much wider range of cases than suggested by von Fintel (2004). So it seems that Strawson (1964) was right all the way with his fundamental observation: the topical status of a non-referring definite description is the decisive category for the explanation of the squeamish feeling vs. truth value judgement contrast. And although his proposal for an analysis of the truth value judgement cases in terms of absorption cannot be sustained, aboutness topicality plays a role of major importance, contra to what his critics claim.

On Squeamishness of the Royal Kind

159

References Endriss, C. (2009). Quantificational Topics. A Scopai Treatment of Exceptional Wide Scope Phenomena. (Studies in Linguistics and Philosophy). Springer, von Fintel, Κ. (2004). Would You Believe It? The King of France is Back! (Presuppositions and Truth-Value Intuitions). In: M. Reimer & A. Bezuidenhout (eds.) Descriptions and Beyond. Oxford University Press. 315-341. Frey, W. (2004). Notes on the Syntax and the Pragmatics of German Left Dislocation. In: H. Lohnstein & S. Trissler (eds.) The Syntax and Semantics of the Left Periphery. Mouton de Gruyter. 203-233. Lasersohn, P. (1993). Existence Presuppositions and Background Knowledge. Journal of Semantics 10. 113-122. van der Sandt, R. (1992). Presupposition Projection as Anaphora Resolution. Journal of Semantics 9. 333-377. Strawson, P. (1964). Identifying Reference and Truth Values. Theoria 30. Reprinted in Steinberg and Jakobovits (eds.), Semantics, 1971.

Information Structure of schon Caroline

Féry

1. Introduction Two uses of schon have been identified in the literature. First, a temporal or phase adverb use which associates with a focus. Second a modal particle use, for which it is usually claimed that it does not have any influence on the truth-value of the sentence, but merely emphasizes its affirmative part and may have a concessive connotation. In this paper, it is proposed that the two uses of schon are two faces of the same coin: schon can be a focus particle or a free focus, depending on the information structure of the sentence as a whole.1 As a focus particle, it associates with a focused constituent and is unstressed, but as a free focus, it is itself a focused constituent and thus, it it does not associate and is usually accented. The meanings associated with the two variants appear to be very different, and, as a result, most authors writing on schon assume lexical separation. But then, the information structural facts and the associated accent pattern are unexplained and accidental. I propose in this paper that there is only one schon, and that the common interpretation of the two is to be found in their affirmative component. Let us begin with a short review of the two roles of schon and a first sketch of their interpretation. First, schon can be used as a short answer, both in its role as a focus particle and as a free focus. In (1), an example from Klein (2007, p. 5), schon is used as a focus particle. It is an elliptical version of Wir müssen schon gehen? 'We must already go?' In this answer, schon takes müssen gehen 'must go' as its associated focus. Klein (2007) comments that the answer schon cannot be taken as a denial of the fact that we must leave, and I fully agree with this judgment. (1) A: Wir müssen gehen. Β: Schon? A: 'We must go.' B: 'Already?' 0

This paper is dedicated to Peter Staudacher, with respect and friendship (and gratitude for the early SFB time and his invaluable help). I am aware that this work does not meet his semantic standards, but I hope that he will forgive me. Many thanks to the following colleagues: Gerrit Kentner, Cécile Meyer and Shin Ishihara for helpful comments of an earlier version of this paper, to Ede Zimmermann for several conversations, to Wolfgang Klein for the sending of his manuscript, to Gisbert Fanselow and Tom Hanneforth for their initiative and editor's activities, as well as Kay-Michael Würzner for converting the paper into IATgX. The usual disclaimers apply.

1

See Féry (2010) for a similar approach of the particles selbst 'self, even', wieder 'again' and auch 'also.'

161

Information Structure of schon

In (2), adapted from Klein (2007, p. 5), schon is used as a concessive affirmative word, a kind of ja 'yes'. It is typical for such an answer to be continued by a sentence beginning with aber 'but.' (2) A: Hättest du nicht anrufen können? Β: Schon. Aber es war mir nicht danach. A: 'Couldn't you have called?' Β: 'Sure. But I did not feel like that.' It is of course not an accident that only in (2), schon can be the (affirmative) answer of a yes-no question; it is a free focus. As a focus particle, as in (1) it needs an associate constituent, and cannot serve as a full answer. In (1) it is an elliptical question. Compare another pair of examples in (3) and (5), in which the wordings are identical, but the accent pattern differs. In (3-a), schon is a focus particle: the associate focused element Sommer 'summer' is accented, but schon is not. (3)

a. Es ist schon

SOMMER.

it is already summer 'It's already summer.' b. Es ist Sommer. As illustrated in (4), schon as a focus particle adds a dimension of earliness to a change or a state. The change expressed by the associated constituent of schon, here the passage to summer, has taken place earlier than expected. Without schon, the same sentence just indicates the season in which we are, see (3b). Both the change and the earliness of the change are absent from (3b). The dotted line in (4) shows the time of reference. As a first approximation, we can interpret (4) in the following way: the begin of summer has happened at the first full vertical line (a change from non-p to ρ in Löbner's 1989 terminology), but it was expected later, at the second full vertical line. Between the two full lines, there is a zone of penumbra, in which schon expresses affirmation + earliness. It always conveys the positive side of p. (4) Affirmation + 'earliness' as part of the analysis of schon2 Spring (not summer)

Begin of summer

Expected begin of summer 1 I 1

non-p

1 1

f

Ρ

time (succession of seasons)

The same sentence can be uttered with an accent on schon. In this case, schon is a free focus (and a modal particle). (5) Es ist S C H O N Sommer it is sure summer 'It's summer alright' 2

See Löbner (1989) for this kind of graphic illustration.

162

Caroline Féry

In this reading, schon expresses that on a scale of being in summer, we are in the positive part of it, but there is also some reservation. The scale stands for something completely different from the one in (4), namely for degrees of denial or affirmation (or of degrees of disagreement and agreement between the protagonists) about the topic of conversation. To the left of the first vertical line, speaker and hearer disagree, to the right of the second full line, they agree, but in between, in the zone of penumbra, the speaker expresses that she is not sure whether hearer and speaker fully agree on the topic at hand. It could be that we are in July and that martins are flying around, but that the temperature is too cold for summer. (6) schon Denial

Zone of penumbra

Affirmation

A third example comparing the two information structural roles of schon appears in (7). This sentence is ambiguous and can be understood first with schon as a focus particle and second with schon as a free focus. In the former case, its use is equivalent to (1), see the translation in (7a): the time of departure is earlier than expected. In the latter case, when schon is a free focus, the sentence is best translated as in (7b). The speaker expresses with schon that she does not agree with the departure of the hearer (even if she affirms that it takes place). Since schon is located in the prenuclear part of the sentence, the difference in accenting may be difficult to perceive. The reason is that there is a tonal movement on the verb complex in both cases, in (7a) because it is accented, and in (7b) because of the boundary tone associated with the end of the embedded intonation phrase. Intended is an unaccented schon in the meaning of (7a) and an accented schon in the meaning of (7b). (7)

Wenn Du schon gehen musst, nimm den Hund mit. If you already go must take the dog with a. 'If you must already go, take the dog with you.' b. 'It you really have to go, take the dog with you.'

The following sections elaborate on the two uses of schon, concentrating on the information structure, and compare this approach with some proposals of the literature. Section 2 examines schon as a focus particle and section 3 as a free focus. Section 4 contains a conclusion.

2. Schon as a Focus Particle 2,1 More Examples In its reading as a focus particle, schon can take all kinds of associated elements, as illustrated in (8) to (12). In this group of examples, schon expresses that something has definitely and positively changed and that the point attained so far is more (later) than could be expected, relatively to its associated element. This is called "early eventuation"

Information Structure of schon

163

by Michaelis (1996).3 The easiest cases are those in which passing of time implies a change that can take place early or late, but we will see in the next section that the passing of time - though very prominent in the interpretation of schon - is not compulsory. There are also cases where the scale induced by schon + associate element has no temporal connotation. In the following examples, time plays a role: a cow becomes heavier, a car drives faster, Peter will grow even taller, etc. Schon takes an associated constituent which is indicated with square brackets and a subscripted F (for focus) in the examples. (8) Die Kuh wiegt schon [600 KG]F the cow weighs already 600 kg 'The cow already weighs 600 kg.' (9) Wir fahren schon [ 160 K M / H ] F we drive already 160 km/h 'We are already driving 160 km/h.' (10)

Peter ist schon [GRÖSSER A L S SEIN V A T E R ] F Peter is already taller than his father 'Peter is already taller than his father.'

(11)

Maria hat schon [ G E T A N Z T ] F Mary has already danced 'Mary has already danced.'

(12) Maria ist schon [ M Ü D E ] F Mary is already tired 'Mary is already tired.' It has been noticed in the literature that temporal schon is ambiguous relatively to early or late time (see for example Kwon, 2005). In (13) something happened earlier than expected, and in this respect the example is comparable to the previous ones. But, at first sight, in (14), the perceived time is later than expected (see von Stechow's (2006) "type-2 puzzle", who uses Löbner's (1989) terminology). (13) Earlier: Maria ist schon [um N E U N ] P gekommen, obwohl wir sie erst um zehn Mary is already at nine come although we her only at ten erwartet haben. awaited have 'Mary already came at nine, although we only expected her at ten.'

3

Michaelis rejects early-eventuation as part of the analysis of schon. I agree with Krifka (2000) who claims that words like schon, noch and erst induce "the interpretation that the sentences express a deviation from expected values in a particular direction". Krifka posits that "these meaning components are conversational implicatures that arise from the fact that only such alternatives are constructed that can plausibly be entertained."

164

Caroline Féry

(14) Later: Es ist schon [ N E U N j p , wir sollten nach Hause gehen. It is already nine, we should to home go 'It is already nine, we should go home.' The puzzle is only apparent, and depends on the context. There is only one temporal schon (like in many other languages with exactly the same ambiguity). The combination of schon and a time (9 o'clock) is interpreted in the same way as before. The speaker may have had the impression that it is eight, but looking at the clock, she realizes that nine o'clock has come earlier than (psychologically) expected.4 In sum, schon as a focus particle is often sensitive to time, or to events taking place in time, like moving faster, becoming heavier or older, becoming tired, changing shape or colour and so on. Some events are more obviously time-oriented than others. In (15), it becomes clear that schon is to be interpreted in relation to the reported time. The speaker communicates that at the topic time t, Maria's meal will be located in the past. (15)

Maria wird schon [ G E G E S S E N ] F haben. Maria will already eaten have

2.2 Semantic Approaches Quite a few analyses of schon in the literature are concerned with its role as a "temporal" or "phase adverb". Löbner (1989, 1999) proposed a "dual" analysis, in which schon and noch, noch nicht and nicht mehr have complementary temporal readings, defined relative to a prior state and under a certain perspective. Basic schon (type S, see below) can be paraphrased along the following lines: schon(te,p), where te refers to a certain time and ρ to a proposition, meaning "atatime te, it is true that schon ρ". Schon(te,p) is equivalent to ~^noch{te, ->p),5 which explains the term "duality" (because it needs both an "internal" and an "external" negation). I do not have much to say about the duality analysis, or negation in general (but see van der Auwera, 1993; Mittwoch, 1993; Michaelis, 1996; Krifka, 2000; Klein, 2007 for comments and criticisms). Here I assume that the primary meaning of schon is to focus on the positive or affirmative part of a proposition.6 I rather concentrate on the elements of the analysis of schon having to do with information structure. For this reason, it is necessary to illustrate Löbner's typology with some examples. Löbner opposes four types of schon on the basis of the kind of associate focus and perfectivity of the sentence (see Comrie, 1976 for perfective vs imperfective sentences). 4

This ambiguity is not limited to temporal schon but extends to all scalar uses o f schon.

T h e c o w c o u l d have

w e i g h e d 6 0 0 kg before o n e notices, or it c o u l d have already attained a w e i g h t too large for selling. 5

6

From Krifka ( 2 0 0 0 ) : already(t, Φ):

assert:

Φ holds at t

Φ(ί)

presup:

->Φ w a s true before t

31' oc ί [ - > Φ ( ί ' ) ]

T h e other types of schon

are in a similar correspondence with erst.

A n d the fact that there is no unique

negative counterpart o f schon may be a natural c o n s e q u e n c e o f the intrinsic positive connotation o f this word.

Information

165

Structure of schon

Type 1 or S ("operator on the sentence focus in an imperfective sentence" = basic use), type 2 or F ("operator on a narrow focus in an imperfective sentence"), type 3 or

Tp¡

("operator focusing on a temporal adverbial in a perfective sentence") and type 4 or T l p ¡ ("operator focusing on a temporal adverbial in an imperfective sentence"). 7 In basic type S, the particle is associated with the natural focus o f the sentence. In other words, "the focus and the sentence accent are where they would be without the particle" (Löbner, 1999, p. 48). In ( 16) it is on an " o n " , but it also can be on Licht, or on ist, as far as I can see. 8 (16)

[ΑΝ]ρ.

Das Licht ist schon

the light is already on 'The light is already on.' In type F, schon takes an adjective or numeral as associated narrow focus. This is illustrated in (17). Fünf Kinder is conceived as potentially increasing with time. (17)

Sie hat schon

[FÜNF] K Kinder.

she has already

five

children

'She already has five children.' In type T, the focus lies on a time interval. Examples are shown in (18). In (18a), the sentence is perfective ( T p / ) , and in (18b) it is imperfective ( Τ φ / ) . (18)

a. Sie kommt schon

[MORGENjp an.

she comes already tomorrow

PART

'She already arrives tomorrow' b. Ich war schon I

[ G E S T E R N ] F da.

was already yesterday

there

Ί was already here yesterday.' A graphic illustration of the basic use of schon appears in (19), where ρ = "the light is on". According to Löbner, schon adds a "sense o f temporal dynamics" to the sentence, since it presupposes that the light was not on before te, in a relevant time interval. Compared to (4), the second full vertical line, which showed that the reference time was earlier than expected is missing. This is thus a very simple analysis. 9 (19)

Interpretation of schon (adapted from Löbner 1989, p. 173) te = Begin of non-p

7 8

ρ

schon Ρ

Types 1 to 4 are from Löbner (1989) and Types S, F and Τ from Löbner (1999). If the sentence accent (primary focus) is not on an element with which schon can associate (as for instance das Licht), schon + an must be either a secondary focus or a Second Occurrence Focus (Partee, 1999).

9

The idea of interpreting schon as an adverb of phase quantification, as Löbner proposes, is that time defines phases on a scale and that by expressing schon ρ, one implies that te ranks high enough. Its opposite noch nicht "not yet" implies that te does not rank high enough.

166

Caroline Féry

Some authors (König, 1977; Hoepelman & Rohrer, 1981; van der Auwera, 1993) consider that, additionally to the change expressed by schon, expectations of the protagonists should also be taken into consideration and should be part of the analysis. Not only prior non-instantiation of the state (which is present in all analyses, though in different guises), but also "earliness" of change is expressed by schon. In the approach proposed here, shown in (4), this part of the analysis of schon is crucial - beside the positive or affirmative meaning component of schon - since it is the one binding the two roles of this word. A further crucial aspect of Löbner's analysis is that ρ is opposed to an adjacent previous non-p, (to the left of te), see (19). This aspect has been criticized by several authors who argue that neither a previous non-p nor adjacency of non-p is necessary. Consider examples of what Michaelis (1996) calls (non-)"priority to process" which are supposed to refute the necessity of a period of non-p. The most cited example, reproduced in (20), comes from Mittwoch (1993). (20) "Person A tells person Β that she has applied for American citizenship, and person Β asks person A whether her husband has applied, too. Person A answers:" Er IST schon [Amerikaner]p, denn er ist in Amerika geboren. 'He is already American, since he was born in America.' It is conspicuous that the constituents er and Amerikaner have been mentioned previously and are thus given. But still, schon associates with Amerikaner as a secondary focus. The primary focus is on ist 'is' which provides a verum focus reading for the sentence (Höhle 1992), and cancels the presupposition that he is not American (see Löbner, 1989, p. 183). In this example, it is clear that schon is a focus particle. It can be translated as already in English or déjà in French (which it cannot when it is a focus itself)· Compare (20) with a continuation like (21), in which it is clear that schon associates with the nationality. 10 In this case, of course, Brite 'British' is new and gets the sentence accent. (21) Nee, er ist schon [BRITE] F . 'No, he is already British.' The interpretation that schon in (20) refers to a time at which A's husband was not yet born cannot be maintained (see van der Auwera, 1993, p. 622, for this interpretation). I propose instead that it refers to a potential time at which a change of nationality could have taken place. Relatively to this potential event, A's husband has previously acquired the property brought about by this event, which is earlier than expected by B. 11 Another interesting case of 'non-priority to process' is reproduced in (22), adapted from Michaelis (1996, p. 481). (22) The strawberry frappé has fewer calories. You don't have to put sugar into it because the strawberries are already sweet. 10

Thanks to Gerrit Kentner who discussed these examples with me.

" I do not think that a change of perspective from one speaker to the other, as postulated by Mittwoch (1993); Löbner (1989) is necessary.

Information Structure of schon

167

Here, too, there is no time when the strawberries are not sweet. I agree with Michaelis who claims that in such a dialogue the speaker does not refer to the time at which the strawberries are not yet ripe. Again, the strawberries already possess the sweet quality that a (potential) sugaring of the frappé would bring. Turning now to the necessity for non-p and ρ to be adjacent, Klein (2007) argues with example (23), referring to the year 1797 (Schubert's birth), that no adjacency of events is necessary.12 He claims that the sentence does not presuppose that Mozart lived up to 1797 or short before. However, I think it does induce adjacency, on its own scale. This sentence would be odd if Mozart was replaced by Cleopatra or the dinosaurs. (23) Mozart war schon [TOT]ρ, Haydn lebte noch. Mozart was already dead Haydn lived still 'Mozart was already dead, Haydn still lived.' Löbner's typology is based on a very restricted distribution of focus relative to schon. Type S allows all kinds of focus (the scope of schon is always the entire sentence), but Types F and Τ require a narrow focus. Moreover, Types F and Τ require a certain class of associated element, namely those with a scalar interpretation (Löbner, 1989, p. 184ff.)13 like a numeral, an adjective or a time. In my view, the distinction in types is dependent on the kind of sentence and on the context, and is not intrinsic to schon. In (16), no scalarity is possible, regardless of the focus: the light maybe on or off. But in (17) for a person with χ children, it is possible to have χ + 1 children. It is however easy to think of a context in which having 5 children may be interpreted in a non-scalar way. Imagine a society in which women climb the social ladder in steps like the following: they have to learn 3 languages, to dye their hair blue, to have 5 children, and to be good swimmers, in that order. In such a case, to have 5 children is just a step in this social scale. The sentence (17) itself is now non-scalar. Furthermore, the direction of the scale is also a matter of convention: depending on the perspective, it can be that the number of children is high (schon 5) or low (erst 5 "only 5 so far"). To make the same point more clearly, consider the example from van der Auwera (1993, p. 691) reproduced in (24). The sentence can be interpreted as non-scalar (the woman can not marry anymore as she is already married) or scalar (she has married at a young age, say 16). In both cases, the relevant factor is that schon is a focus particle and married is the associated focus. Whether it is scalar or not depends on the context of interpretation, but is not part of a type of schon. In both interpretations, schon expresses that she is now married, that she was not married before, and that this state has taken place earlier than expected. (24) Sie ist schon [ V E R H E I R A T E T ] F . she is already married 'She is already married.' 12

Mozart died in 1791, Haydn in 1809.

13

In Löbner (1989), the terminology scalar vs. non-scalar is rejected, but in Löbner (1989) it is part of the definition of the types.

168

Caroline Féry

Löbner's second distinction entering the types of schon is perfectivity vs. imperfectivity of the sentence. According to him, only Type Τ p f ('operator focusing on a temporal adverbial in a perfective sentence') can appear in a perfective sentence, all other types only take imperfective sentences. But as shown by Klein (2007), this distinction is not necessary either. First perfective verbs can appear with other types of schon, as in the following example from Klein with a perfective verb but no time adverbial. (25)

Der Ballon war schon [GEPLATZT]p. The balloon was already burst 'The balloon has already burst.'

Furthermore, there does not seem to be any necessary relation between perfective sentences and focus on the time adverbial (see Klein for such sentences). (26)

Sie kam gestern schon [in F R A N K F U R T ^ an. she came yesterday already in Frankfurt PART 'She arrived in Frankfurt already yesterday.'

And there are also imperfective sentences with a focus on the time adverbial (see below).

2.3 The Role of Information Structure In this section, it is shown that schon is a well-behaved focus particle which takes an associated element and adds a special meaning, exactly as other focus particles. The proposed meaning of schon is "it is affirmative and early relative to a scale". The accent on its associated element elicits a set of alternatives. Consider the variants in (27), where the sentence has different interpretations depending on the associated element. (27) a. Maria hat gestern schon [DREI ÄPFEL]Ρ geschält. Maria has yesterday already three apples peeled 'Mary has already peeled three apples yesterday.' b. Maria hat gestern schon [DRELJP (Äpfel) geschält. Maria has yesterday already three apples peeled 'Mary has already peeled three apples yesterday.' c. Maria hat gestern schon drei Äpfel [GESCHÄLT]P. 'Mary has already peeled three apples yesterday.' d. Maria hat schon [GESTERN]p drei Äpfel geschält. 'Mary has already peeled three apples yesterday.' e. Schon [MARIA]F hat gestern drei Äpfel geschält. 'Even Mary has peeled three apples yesterday.' In (27a), the accent on the object is the default sentence accent, and, as a result, it can elicit alternatives on different constituents, as for example on the kind and number of fruit, or on the whole VP, including gestern 'yesterday' or not. In each of these readings, the set of alternatives differs from the other ones. If schon associates with the direct object plus

Information Structure of schon

169

verb, the set of alternatives may include actions like the following ones, which are ranked on a scale: {sweep the kitchen, peel three apples, peel two pears, read a book}. At the same time, schon implies that the peeling of apples has taken place earlier than expected relative to this scale. In (27b), schon only associates with the numeral and the set of alternatives is {one apple, two apples, three apples . . . } . The speaker comments with schon that the action of peeling three of them has taken place earlier than expected. In (27c), the accent on the verb does not project further than itself (at least if the whole VP is new), in the same way as in (27b). The set of alternative may be {to buy, to peel, to bake, to eat}. Schon associates with the verb, and creates a scale of these acts. It says that the action of peeling has taken place earlier than expected. The narrow accent on the temporal adverb gestern plus schon in (27d) creates a scale like {yesterday, today, tomorrow} and asserts that the peeling took place yesterday, and that this day is early on the scale. Finally, in (27e), the scale contains persons, for instance {Johannes, Maria, Peter}, and it is asserted that as the person who did the peeling, Maria is early on this scale, maybe even unexpected. It is again the case that the accent on Maria cannot project further than itself. It could be that Maria is really clumsy and that Peter is not as clumsy, but he has only managed to peel two apples. In uttering (27e), the speaker may express that the number of apples peeled by Peter is small, as compared to the number managed by Maria. In this context schon resembles the scalar particle sogar, to the difference that sogar can be used in negative sentences. Schon Maria is a primary focus. It is left open whether the remainder of the sentence contains another focus, for example on drei 'three'. Since schon is a focus particle, it associates with a focus, and not with a topic. A sentence like (28a) in which the subject is clearly a topic and the focus is on another element, is not well-formed. However, as soon as the topic contains a focused part, as in (28b), schon is possible again. In this case, the associate focus constituent of schon is embedded into the topic. (28)

a. * Schon [ M A R I A ] t hat [DREI Ä P F E L ] F geschält. b. [Schon die [ J Ü N G S T E ] P Tochter}^ konnte [drei Äpfel pro T A G schälen]p. Already the youngest daughter could three apples per day peel 'Even the youngest daughter could peel three apples a day.'

And, of course, schon can associate with a constituent as a Second Occurrence Focus. We already saw such cases in (20) and also in the discussion of the alternative accent patterns of (16). In this case, there is no pitch accent on the associate constituent (at least when it is in the postnuclear position). To sum up so far, in its reading as a focus particle, schon emphasizes the affirmative part of the sentence and and denotes earliness on a scale. It takes all sorts of focused associate elements, which elicit scales relatively to which the sentence is interpreted. We have given an interpretation of the particle which is close to the one found in the literature. In line with Löbner's analysis, schon ρ may imply a dynamic temporal sense: there is then a preceding phase non-p, which may be only virtual (see (20) and (22)). Additionally (and in contrast to Löbner's proposal), there is an addition of earliness: "the change is

170

Caroline Féry

early relative to a scale". And, also in contradistinction to Löbner, there is only a single type of schon. The different readings identified by Löbner, as well as their associated perfectivity are a consequence of the context, at least as far as the information structure is concerned. Another difference is that the notion of change (the dynamic temporal sense) is not obligatory. In other words, I do not think that the temporal aspect, albeit pervasive, is an obligatory part of the interpretation of schon, see (27e).

3. Modal Particle: schon as a Free Focus Let us now turn to the "modal particle" use of schon (see for instance König, 1977; Jacobs, 1991). Löbner (1989, p. 167; 1999, p. 48) and Klein (2007) for instance consider it an entirely different word. 14 By contrast, I propose that the difference in meaning comes from the use of this word as a focus particle or as a free focus. In the latter reading it is usually accented and has no associate element. It can stand by itself, as in (29), see also (1).

(29)

A: Treibt Stella denn keinen Sporti does Stella then no sport 'Doesn't Stella do any sport?' B: (Doch) SCHON! Sure, she does

Here is my proposal in more detail. In its function as a free focus, there is no associated element to schon. As a result, the meaning component induced by schon + associate element is only partially present. The part asserting P: "it is true that P" is present both in the focus particle and in the free focus. And the reference to a scale is present as well. But in the focus particle reading, the interpretation of the scale comes from the associate element, as was illustrated in (27). Without associate element, the scale only comes from schon itself. As a result, the scale is always the same. In (6), reproduced in (30), it was shown that it stands for denial to affirmation (or disagreement to agreement). Schon elicits a zone of penumbra located between a region of denial or disagreement and a region of affirmation or agreement. The uncertainty expressed by schon can have different sources. In this use, schon has very often a concessive reading, as in a possible continuation of (29B) " . . . but only once in a month". (30)

schon

Denial 14

1

;

Zone of penumbra

ι

>

Affirmation

Ormelius-Sandblom ( 1997a,b) relates the two uses of schon with a process of grammaticalization. According to her, the modal particle has developed from the temporal adverb. But she does not show how the two meanings are related.

information Structure of schon

171

Let us examine further examples. In (31), speaker Β probably expects that speaker A thinks that Β does not like natto beans. The zone of penumbra corresponds to a domain in which the expectations of A and Β concerning B's liking of natto beans do not fit. (31)

A: Do you like natto beans'? B: Ja,

die

mag ich SCHON.

yes, them like I all-right 'Yes, I do like them.' In (32), Β is contradicting A. The contradiction goes through the addition of schon, rendering the contradiction softer than a simple denial or a simple verum focus as in B \ which sounds a bit rude in all variants. In a direct denial such as B', there is no zone of penumbra, but just ρ and non-p. By contrast, the first part of Β calls for a continuation like one of those offered in parentheses. (32) A: I thought that Mary is a vegetarian. B: Sie isst SCHON Fleisch she eats alright meat 'she eats meat alright' (but only little / but she does not like steakhouses / in fact she eats everything) Β':?Sie ISST Fleisch / Doch, sie ISST Fleisch / Nein, sie ist KEINE Vegetarierin. she eats meat / Sure she eats meat / no she is no vegetarian In (33), speaker Β may be unsure, and expresses a cautious affirmation. In (33B') the speaker expresses that until now she believed (or that she still believes) that Anna is married. In (33B), the speaker is little bit more sure. In fact (33B) is often used when speakers are dead sure of what they say. The addition of schon (and of a zone of penumbra) may be inserted for politeness. By answering A as in B, the speaker leaves open to A the possibility of disagreement. (33) A: Is Anna married? B: Das glaube ich SCHON that believe I alright Ί think so.' B':

Das GLAUBE ich

that believe I Ί think she is.' In (34), the role of schon is illustrated further. On a scale consisting of disagreement (A believes that Mary was not there, and Β believes that she was there), agreement (A and Β believe that she was not there), and zone of penumbra (B believes that Mary was there and lets open whether A believes that Mary was there), speaker Β chooses the zone of penumbra. By contrast, B' and B" are more direct, and elicit only non-p as a reaction to p. Observe that C is ill-formed as a reply to A. The reason is that a given element (da 'there' is anaphoric to auf der Party) is accented. In B, only schon (affirmation + scale) is new and accented.

172

Caroline Féry

(34) A: A pity that Mary was not at the party. Β: Aber Maria war SCHON da. But Maria was already there 'But Mary was there.' Β': Aber Maria WAR da. B": Aber Maria war DA. C: *Aber Maria war schon DA. In (35) and (36), the difference between focus particle and free focus schon is illustrated with pairs. In the Β versions, the focus particle use of schon is used, and in the C versions, the free focus version. 15 (35) A: Mary did not come yet. B: Doch sie ist schon [ D A J F yes she is already there C: Doch, sie ist S C H O N gekommen (aber sie ist gleich gegangen). yes she is alright come but she is immediately gone 'Well, she was there (but she left immediately).' (36) A: I'll do some shopping. Do we need milk? B: Nein, Maria hat schon [drei LLTER]P gekauft. no Mary has already three liters bought C:

Ja,

SCHON,

yes sure In the examples with schon as a free focus discussed so far, nothing new or highlighted is introduced into the Β sentences, containing schon. All constituents have been already mentioned in the previous sentences. There is thus no word which can carry the accent by default, except for schon. Since this word focuses the affirmative part of the scale (albeit the penumbrious one), it has an evident verum focus component. But this is not necessarily so. The next example shows that it is also possible to have another (primary) focus beside the one associated with schon, which is now a secondary focus (compare (37) with (28)). Due to the deaccenting of the postnuclear part of the sentence, which includes everything after Mary, it is difficult to get the reading of schon as a focus particle. But it is not impossible. The context could be a situation in which person A asks whether Peter and Maria already peeled three apples each, as planned. Person Ρ may utter (37) as an answer. In fact, as already discussed above, nur Maria cannot be a topic, which means that the preferred reading of this sentence is one in which it is the primary focus. (37) Nur MARIA hat schon drei Äpfel geschält. only Mary has schon three apples peeled O n l y Mary peeled three apples.' 15

True minimal pairs were introduced in (3) and (7).

Information Structure of schon

173

In the following examples, the difference between the two uses is also not evident. In (38), the adjective merkwürdig 'strange' is accented if new, but schon is easier to interpret as a free focus than as a focus particle taking the adjective in its scope. The latter is not impossible but implies that a scale is created. Being strange is then a position on this scale. I assume that the strong preference for the reading of schon in (38) as a free focus is due to the absence of such a scale. This example shows that to be a free focus does not necessarily means being accented. In fact, a stronger adjacent accent suppresses the one on schon (see Féry (2010) for such cases with other particles). (38)

Es ist schon [ M E R K W Ü R D I G ] F it is schon strange 'It is sure strange'

In both (39) and (40), the verb is accented and schon is not or less so, but still schon may be interpreted in its free focus role. The example (40) serves as an illustration of what may have happened in a grammaticalisation process from a focus particle reading to a free focus reading of schon. The preverbal position of schon is the default syntactic position for this word (see for instance Ormelius-Sandblom (1997b); Klein (2007)). If it takes the verb as an associate element, it says that the event has taken place, and that it took place earlier than expected. As shown in this paper, schon focuses on the affirmative part of the sentence. In its free focus reading, schon does exactly the same, except that it does not include in its meaning the part 'earlier than expected'. Thus both the focus particle reading and the free focus reading emphasize the affirmative part of the sentence. As a result, the free focus reading is a subset of the focus particle reading, and this latter reading (with the early component) needs an associate element to arise. It is easy to imagine situations where the scale is intended by the speaker (like not taken place, begun, taken place), but not understood by the hearer. In such a case, the free focus reading is created and only the affirmative part is perceived as focused. (39) Das schaffe ich schon. that manage I schon Ί can do that' (40) Schon passiert schon taken-place 'too late' Before closing this chapter, a last remark is in order. It has often been observed in the literature that schon as a focus particle can occupy more syntactic positions than as a free focus. The reason for this difference is that when it associates it has to show the element it associates with by adjacency, whereas as a free focus it is a sentence adverb, without much freedom in its location (see Ormelius-Sandblom ( 1997b,a) for a survey). A more complete survey would review the syntactic aspects in more detail. The relationship between information structure, prosody and syntax is left for future research.

174

Caroline Féry

4. Conclusion This paper has proposed that there is only one lexical item schon and that its two meanings are a consequence of the fact that it can be used as a focus particle or as a free focus. Both uses of schon emphasize the assertive or affirmative part of a sentence, and both introduce a scale. In its use as a focus particle (as a temporal adverb or phase adverb or aspectual particle), schon takes an associate focus element. The combination of schon and its associate element elicits a scale and conveys that the change/event/state expressed by the associate element is earlier on this scale than expected. In its use as a modal particle, schon is a free focus. It also asserts the sentence, and has thus as before a strong affirmative or verum component. In this case, there is no associate element, but still a scale is elicited by schon. This scale goes from denial (disagreement between the protagonists) to affirmation (agreement between them). Schon adds a zone of penumbra, leaving place for disagreement between speaker and hearer. The added nuance is often concessive or just polite.

References van der Auwera, J. (1993). Already and still: Beyond Duality. Linguistics and Philosophy 16. 613-653. Comrie, B. (1976). Aspect: An Introduction to the Study of Verbal Aspect and Related Problems. (Cambridge Textbooks in Linguistics). Cambridge: Cambridge University Press. Féry, C. (2010). Prosody and Information Structure of the German Particles selbst, wieder and auch. In: T. Borowsky, S. Kawahara, T. Shinya & M. Sugahara (eds.) Prosody Matters: Essays in Honor of Lisa Selkirk. (Advances in Optimality Theory). London: Equinox. Hoepelman, J. & C. Rohrer (1981). Remarks on noch and schon in German. In: P. Tedeschi & A. Zaenen (eds.) Tense and Aspect. (Syntax and Semantics, vol. 14). New York, NY: Academic Press. 103-126. Jacobs, J. (1991). The Semantics of Modal Particles. In: W. Abraham (ed.) Discourse Particles: Descriptive and Theoretical Investigations on the Logical, Syntactic and Pragmatic Properties of Discourse Particles in German. (Pragmatics & Beyond New Series, vol. 12). Amsterdam: John Benjamins. 141-162. Klein, W. (2007). About the German Particles 'schon ' and 'noch '. Manuscript. Nijmegen: Max Planck Institute for Psycholinguistics. König, E. (1977). Temporal and Non-Temporal Uses of schon and noch in German. Linguistics and Philosophy 1. 173-198. Krifka, M. (2000). Alternatives for Aspectual Particles: Semantics of 'still' and 'already'. Berkeley Linguistics Society Meeting 26. Kwon, M.-J. (2005). Modalpartikeln und Satzmodus. Untersuchungen zur Syntax, Semantik und Pragmatik der deutschen Modalpartikeln. Ph.D. thesis. Ludwig Maximilians-

Information Structure of schon

175

Universität, München. Löbner, S. (1989). German schon-erst-noch: An Integrated Analysis. Linguistics and Philosophy 12. 167-212. Löbner, S. (1999). Why German Schon and Noch are Still Duals: a Reply to van der Auwera. Linguistics and Philosophy 22. 45-107. Michaelis, L. (1996). On the Meaning of already. Linguistics and Philosophy 19. 4 7 7 502. Mittwoch, A. (1993). The Relationship Between schon!already and noch!still: A Reply to Löbner. Natural Language Semantics 2. 71-82. Ormelius-Sandblom, E. (1997a). Die Modalpartikeln'¡a, doch, schon. Zu ihrer Syntax, Semantik und Pragmatik. (Lunder Germanistische Forschungen, vol. 61). Stockholm: Almqvist & Wiksell. Ormelius-Sandblom, E. (1997b). The Modal Particle schon: Its Syntax, Semantics and Pragmatics. In: T. Swan & O. J. Westvik (eds.) Modality in Germanic Languages: Historical and Comparative Perspectives. (Trends in Linguistics. Studies and Monographs, vol. 99). Berlin: De Gruyter. 75-133. Partee, Β. Η. (1999). Focus, Quantification, and Semantics-Pragmatics Issues. In: P. Bosch & R. van der Sandt (eds.) Focus: Linguistic, Cognitive, and Computational Perspectives. (Studies in Natural Language Processing). Cambridge: Cambridge University Press. 213-231. von Stechow, A. (2006). Times as Degrees. Manuskript. University of Tubingen.

When-Clauses, Factive Verbs and Correlates Stefan

Hinterwimmer

1. Introduction This paper deals with the syntax and semantics of generic or adverbially quantified sentences like those in (la-c), where a right-adjoined when-clause is associated with the neuter pronoun it, which occupies an argument slot of the respective matrix verb: ( 1 ) a. Paul hates it when his colleague snores. b. Mary usually loves it when her daughter plays the piano. c. Jane usually likes it when the brass section is dominant in a Mahler

symphony.

What is interesting about these sentences is the fact that the right-adjoined when-clause fulfills two functions at the same time: on the one hand, it restricts the covert generic operator (see Krifka, 1995) or overt quantificational adverb, on the other hand, it (at least indirectly) provides an argument for the respective matrix verb . The intuitive meaning of the sentences can be paraphrased as in (2a-c): (2) a. Generally, for an event e, which is an event of Peter's colleague snoring, there is an associated event e' which is an event of Peter hating e. b. For most events e of Mary's daughter playing the piano, there is an associated event e' which is an event of Mary loving e. c. For all events e of the brass section being dominant in a Mahler symphony, there is an associated event e' which is an event of Jane liking e. Rothstein (1995b,a) provides an analysis which naturally accounts for these intuitions. She assumes that the right-adjoined wAen-clause is interpreted in the restrictor of the respective operator, while the neuter pronoun it provides a variable ranging over events which is bound by the operator: 0

An earlier version of this paper was presented at the Genericity Conference in Paris in May 2009 and at the University of Tübingen in June 2009. I would like to thank the audience of both events as well as Andreas Haida, Manfred Krifka and Sophie Repp for comments and suggestions. This research was funded by the Deutsche Forschungsgemeinschaft as part of the Sonderforschungsbereich 632 'Information Structure'. I would like to take this opportunity to express my deepest gratitude to Peter Staudacher, who has been a great teacher, advisor and colleague. I have learnt a lot from him, and I greatly enjoyed our discussions about linguistics, politics and Richard Wagner.

When-Clauses, Factive Verbs and Correlates

177

(3) a. Gen e [snore{e, colleague)] [3e'[o(e, e') Λ hate(e', peter, e)]] where o(e, e') means that the running times of e and e' overlap. b. Most e \play(e, daughter, piano)] [3e'[o(e, e') Λ love(e', mary, e)]] The problem with this analysis is that it leads us to expect that the variants in (4), where the when-clause has been left-adjoined, should be as good as the original examples, since in both cases the when-clause can be interpreted in the restrictor of the (covert or overt) Q-adverb (Johnston, 1994). This is not the case, however. While the sentences in (4a,b) are not ungrammatical, they are distinctly odd for most speakers I consulted and far less natural than the variants in (la,b) for all of them. The pattern in German is exactly the same, as shown by the contrast between (5a) and (5b), on the one hand, and (5c) and (5d), on the other: (4) a. HWhen his colleague snores, Paul hates it. b. llWhen her daughter plays the piano, Mary loves it. (5) a.

Paul hasst es, wenn sein Kollege schnarcht. Paul hates it when his colleague snores b. 11 Wenn sein Kollege schnarcht, hasst es Paul. c. Maria liebt es, wenn ihre Tochter Klavier spielt. Maria loves it when her daughter piano plays d. 11 Wenn ihre Tochter Klavier spielt, liebt es Maria.

The same contrast between left- and right-adjunction is found in episodic sentences: (6) a. Mary loved it when Paul suddenly kissed her at the party yesterday evening. b. 11 When Paul suddenly kissed her at the party yesterday evening, Mary loved it. Crucially, there is no comparable contrast in sentences where the when-clause is not associated with a neuter pronoun in the matrix clause, neither in generic or adverbially quantified sentences nor in episodic one: (7) a. b. c. d. e. f.

Peter is annoyed when his colleague snores. When his colleague snores, Peter is annoyed. Mary is usually happy when her daughter plays the piano. When her daughter plays the piano, Mary is usually happy. Mary was shocked when Paul suddenly kissed her at the party yesterday evening. When Paul suddenly kissed her at the party yesterday evening, Mary was shocked.

I will thus argue for an analysis that explains these contrasts as well as our intuitions concerning the interpretation of the sentences in (la-c) in section 3. But let us first consider some additional data that any successful analysis has to account for.

178

Stefan

Hinterwimmer

2. More Data The contrast between (1) and (4) is strikingly parallel to the one between (8) and (9): (8) a. Paul hates it that his colleague always snores. b. Mary loves it that her daughter plays the piano so well. (9) a. *That his colleague always snores, Paul hates it. b. *That her daughter plays the piano so well, Mary loves it. Note that the sentences in (9) are grammatical if the that-clause is separated from the matrix clause by an intonational break, and thus functions as a hanging topic. This possibility can be excluded by turning to the German translations. Since German is a V2-language, the difference between hanging topics and regularly fronted constituents is clearly visible in the syntax: while the latter immediately precede the finite verb, hanging topics are separated from the finite verb by another constituent (the subject being the default option), thus indicating their lack of syntactic integration into the respective sentence. Crucially, while the variant in (10a) with the f/iaf-clause functioning as the hanging topic is fine, the one in (10b), where the ί/ιαί-clause directly precedes the finite verb, is ungrammatical: (10) a. Dass sein Kollege immer schnarcht, Paul hasst es. that his colleague always snores Paul hates it. b. *Dass sein Kollege immer schnarcht, hasst es Paul. Before we move on, let me briefly summarize our observations so far: in both German and English, the neuter pronoun it/es may always be linked to a right-adjoined whenor that-clause. Concerning left-adjunction, the result is marked with vv/iew-clauses and completely ungrammatical with that-clauses. Let us turn to sentences where an ¿/-clause is associated with a neuter pronoun next. Pesetsky (1991), who discusses examples like the ones in (11) below, does not report any contrast in acceptability at all between the variants with right-adjunction and the ones with left-adjunction. He simply notes that according to his own intuitions as well as the ones of his informants there is a subtle difference in meaning between the two variants that is impossible to describe precisely, let alone pin down in terms of truth conditions. While I agree with the last point, I have the intuition that in both German and English the variants involving right-adjunction are slightly more natural than the ones involving leftadjunction, and this has been confirmed by the native speakers I consulted. Nevertheless, the contrast between left- and right-adjunction is nowhere near as sharp as with whenclauses. It is more like the one that we find in cases like (13a) vs. (13b), where the ¿/"-clause is not linked to a neuter pronoun. (11) a. / would like it if he played the violin right now. b. If he played the violin right now, I would like it. (Pesetsky, 1991, p. 65) (12) a. Ich würde es gut finden, wenn er jetzt Geige spielen würde. I would it good find if he now violin play would

When-Clauses, Factive Verbs and Correlates

179

b. (?)Wenn er jetzt Geige spielen würde, würde ich es gut finden. (13) a. I would be happy if he played the violin right now. b. If he played the violin right now, I would be happy. Finally, note that if the examples in (4) are read with the rise-fall-rise accent marking contrastive topics in English (Biiring, 1997, 2003) on some part of the w/¡en-clause (in combination with a focus-accent on some part of the matrix clause), they become fine. This intonational marking triggers the expectation that the speaker continues by uttering at least one sentence where the constituents marked as the contrastive topic and the focus in the preceding sentence have been replaced by "natural" alternatives, as shown by the continuations in brackets: (14) a. When his /COLLeagueV snores, Peter HATES\ it (while it does not bother him at all when his wife snores), b. When her /DAUGHteiVplays the piano, Mary usually LOVES\ it (while she usually hates it when her son plays the piano). The same is true for German (modulo the fact that contrastive topicality is marked by a simple rising accent in German), as shown by the examples in (15): (15) a. Wenn sein /KoLLEge schnarcht, HASS1\ es Peter. 'When his colleague snores hates it Peter' b. Wenn ihre /TOchter Klavier spielt, LIEBT\ es Maria. 'When her daughter piano plays loves it Maria' Even in cases like (16), where no constituent is marked as a contrastive topic via intonation, adding a continuation which makes clear that the time span during which the event introduced by the when-clause occurred is contrasted with some later time span, fronting the when-clause is acceptable (while the sentence in isolation is rather odd, cf. (6b) above). (16)

When Paul kissed her at the party yesterday evening, Mary loved it, but now she is ashamed.

Contrastive topic marking does not help in the case of left-adjoined ίΛαί-clauses which are linked to a neuter pronoun, however: (17) *That his /coLLEAGUEX/always

snores, Peter HATES\

it.

Taking all the facts discussed in this section together, we need an account that explains (a) the contrast in acceptability between left- and right-adjunction in sentences where a when-clause is associated with a neuter pronoun, (b) why the same effect occurs in stronger form in structurally similar sentences with ίΑαί-clauses, and is almost absent in ones with (/"-clauses, and

180

Stefan

Hinterwimmer

(c) why contrastive topicality cancels it in the case of left-adjoined when-clauses, but not in the case of left-adjoined ί/ιαί-clauses.

3. That- and When-Clauses Associated with it 3.1 77zfli-Clauses Let us concentrate on the sentences with that-clauses first. Neuter pronouns are sometimes obligatory and sometimes optional in sentences where a ί/ιαί-clause functions as an argument of the matrix verb, their presence being related to issues like factivity, givenness or specificity of the respective proposition (see Ross, 1967; Bolinger, 1977; Rothstein, 1995b for discussion). Now, in the sentences discussed in sections 1 and 2 the verbs are all factive in the following sense: negation does not affect the truth of their complements in the world of evaluation, i.e. if we negate the sentence in (8a), for example, it is still presupposed that Peter's colleague snores (see Vendler, 1967 and Peterson, 1997 for discussion): ( 18) Paul doesn 't hate it that his colleague always snores. It is thus not surprising that the presence of it is obligatory (at least as long as the CP is not fronted; see section 2 above): (19) a. Paul hates *(it) that his colleague always snores. b. Mary loves *(it) that her daughter plays the piano so well. Concerning the contribution of the neuter pronoun, there is a prominent line of thought according to which it does not have any semantic content, but is either an expletive element (Postal & Pullum, 1988) or the phonetic realization of the accusative case that the embedding verb has to assign to its complement (Authier, 1995). Rothstein (1995b), in contrast, assumes that the neuter pronoun denotes an event variable that in the cases where it is associated with a that-clause remains free, while it is bound by a generic operator or quantificational adverb in the cases where it is associated with a w/ien-clause (see section 1). Concerning the interaction of the neuter pronoun and the respective that-clause at the level of semantic interpretation, Rothstein assumes that the event argument of the CPinternal verb may optionally remain unsaturated, thus allowing the CP to be interpreted as an event predicate that is applied to the free event variable provided by it. The problem with the approaches advocated by Postal & Pullum (1988) and Authier (1995) is that they do not have anything to say about the link between the presence of it and factivity/givenness/specificity of the associated CP. Concerning Rothstein's (1995b) approach, I see the following major shortcoming: it is not only entirely stipulative to assume that the event arguments of the main verbs contained in ίΛαί-clauses may sometimes remain unsaturated, but it also leads us to expect that ίΛαί-clauses should be able to function as the restrictors of adverbial quantifiers over events, just as w/zen-clauses, which is

When-Clauses, Factive Verbs and Correlates

181

clearly not the case. Furthermore, as already mentioned in section 1, Rothstein's approach also runs into empirical problems in those cases where it is associated with a w/zen-clause. I therefore want to suggest a unified approach to both that- and wAen-clauses associated with neuter pronouns that avoids these problems, and which builds on ideas developed by Ross (1967), Kiparsky & Kiparsky (1973) and Chomsky (1973). I assume that the CPs functioning as the internal arguments of factive verbs are base generated within phonologically empty DP-shells, i.e. as the complements of a phonologically empty determiner. 1 Now, as the contrast between (20a) and (20b), on the one hand, and (20c) and (20d) on the other, shows, finite CPs functioning as the internal arguments of verbs need to be extraposed, no matter whether they are factive or not: (20) a. Paul hates it from the bottom of his heart that his colleague always snores. b. UPaul hates it that his colleague always snores from the bottom of his heart. c. Mary believes without any doubt that HPSG is superior to LFG. d. ??Mary believes that HPSG is superior to LFG without any doubt. I assume extraposition to be right-adjunction to vP or TP. Concerning the question of whether it is triggered by syntactic factors (as argued for by Stowell, 1981; Biiring & Hartmann, 1997) or by phonological ones (Goebbel, 2007), I remain agnostic, as it does not matter for the purposes of this paper. What is crucial, however, is the following assumption: as soon as the DP-shell has been emptied by moving the CP away from its base position, a phonological rule applies which causes the silent determiner to be spelled out as it. Note that a similar assumption is made by Elbourne (2001 ) in his analysis of donkey pronouns, which he assumes to be definite descriptions with an elided NP, where NP-elision causes the definite determiner to be spelled out as s/he, it, etc. With these assumptions in place, the ungrammaticality of examples such as (9a,b) (repeated here as (21a,b)) as well as of the corresponding German example given in (10b) above then follows from the same principle that blocks left-adjunction (in contrast to extraposition) of (headed) relative clauses: in order to leave the DP, the clause has to right-adjoin to it first, which blocks further movement to the left (cf. Müller, 1996 and Biiring & Hartmann, 1997). (21) a. *That his colleague always snores, Paul hates it. b. *That her daughter plays the piano so well, Mary loves it. Let us now tum to the semantic contribution of the empty determiner in the DP-shell associated with íAaf-clauses. I assume that it has a denotation comparable to that of the definite article: it takes the proposition denoted by the respective ί/ιαί-clause and returns a fact, where I assume (following Vendler, 1967, 1972; Kiparsky & Kiparsky, 1973; Peterson, 1997) that facts are semantic objects distinct from (though related to) propositions and events. More concretely, I assume that facts can be derived from propositions in the following way: they are abstract entities which make propositions true (see Peterson, 1997) 1

S e e Müller & S t e m e f e l d ( 1 9 9 5 ) for an analysis according to which all finite C P s are embedded within a (possibly covert) NP-shell.

182

Stefan

Hinterwimmer

in the world of evaluation. The silent determiner in DP-shells associated with ί/ιαί-clauses has the denotation given in (22): (22)

\DETjactl where

= Xp{s,t). u:[«(a;,p)(ii;o)], ρ) (tuo) means χ makes ρ true in wo.

The fact that results from applying the covert determiner to the CP functions as the internal argument of the respective matrix verb, i.e. extraposition does not have any semantic impact - either because the CP is reconstructed at LF, or because it is a PF-operation. A sentence such as (8b) (repeated here as (23a) is then interpreted as shown in (23b): (23) a. Paul hates it that his colleague always snores. b. Be [hate(e,paul,ix[9i(x, Xw'Me' [C(e')(w') ->• Be" [o(e, e')(w') A snore(e", colleague)(w')]])(w0)])(wo)]

3.2

When-Clauses

Let us now turn to the sentences where the neuter pronoun is associated with a whenclause. First, note that extraposition is obligatory in these cases, too: (24) a. Paul hates it from the bottom of his heart when his colleague snores. b. ΊΊΡαιιΙ hates it when his colleague snores from the bottom of his heart. I assume that in these sentences, too, the clause associated with it is base generated within a DP-shell, i.e. as the complement of a silent determiner, and that emptying the DP-shell by extraposing the when-clause causes the phonological rule introduced in the last section to apply. The unacceptability of the sentences in (4a,b) (repeated here as (25a,b)) can now be explained in the same way as the unacceptability of the ones in (8) (Recall, however, that it is not quite as strong. I will return to this issue in section 4.2). (25) a. V.When his colleague snores, Paul hates it. b. HWhen her daughter plays the piano, Mary loves it. On the interpretative side, the determiner in the DP-shell again has a denotation comparable to that of the definite determiner: it takes event predicates as its argument, and returns definite descriptions of events, which then function as the internal arguments of the respective matrix predicates (under the assumption that verbs like love and hate not only take concrete individuals and facts as their internal arguments, but also events). The pronoun it, being maximally underspecified (see Rothstein, 1995b for discussion), thus spells out (under the conditions discussed above) covert determiners that denote in different domains: in the domain of facts, and the domain of events. Now recall that in examples such as (la,b) (repeated here as (26a,b)), which contain (overt or covert) quantifiers over events, those quantifiers need event predicates as their restrictors, and there is a clear intuition that the right-adjoined w/ien-clauses provide those predicates, as revealed by the paraphrases in (2a,b) (repeated here as (27a,b).

When-Clauses, (26) (27)

Factive Verbs and

183

Correlates

a. Paul hates it when his colleague snores. b. Mary usually loves it when her daughter plays the

piano.

a. Generally, for an event e, which is an event of Peter's colleague snoring, there is an associated event e' which is an event of Peter hating e. b. For most events e of Mary's daughter playing the piano, there is an associated event e' which is an event of Mary loving e.

In principle, we have two options to accommodate this fact. According to the first one, extraposition does not have a semantic effect in the case of when-clauses, either, and the observed interpretation comes about in the following way: the restrictor of the respective operator is initially represented in the form of a free variable ranging over event predicates that is resolved on the basis of clause internal as well as contextual information and world knowledge (see Beaver & Clark, 2008 and the references cited therein for discussion). This variable is resolved to the semantic content of the w/ien-clause in the cases under discussion. According to the second one, the w/¡en-clause has a double function in the examples under discussion: the copy in right-adjoined position is interpreted as the restrictor of the respective operator (cf. Johnston, 1994), while the copy left behind in the base position is interpreted as the argument of the covert determiner. In order for this to go through, we have to assume that extraposition is not a PF-phenomenon, but a truly syntactic movement operation. Since I do not see any conclusive evidence against this assumption, let us adopt the second option for concreteness. Keep in mind, though, that this point is not really important for the purposes of this paper, and that the analysis could easily be recast in accordance with the first option. Now, in order for the required interpretation to come about, the events denoted by the combination of the covert determiner and the when-clause need to vary with the events quantified over by the generic operator/the Q-adverb. It is well-known that the uniqueness conditions of "ordinary" definite descriptions (i.e. ones denoting individuals) may be relativized to situations/events (see Elbourne, 2001, 2005; Biiring, 2004; Hinterwimmer, 2008; Schwarz, 2009 for discussion). Standardly, this is accounted for by assuming that the definite determiner combines with a silent pronoun introducing a free situation/event variable that is either resolved to the world of evaluation by default, or to a contextually salient situation/event, or turned into a variable that is bound by a situation/event quantifier under c-command. Now, in the cases under consideration, co-variation with the events quantified over by the respective operator cannot be achieved by simply introducing an (eventually bound) event variable, since the objects denoted by the definite descriptions are events themselves. Nevertheless, we get the right result by assuming (28) as the denotation of the covert determiner heading the DP-shells in whose complement position w/ien-clauses are base generated: (28)

{DETevent εηγ = Λ P . i e [ P { e ) { w 0 ) Λ o ( e , 5 ( e n ) ) ( « ; o ) ] , where o(e, g(eN))(Wo) means that the running times of e and g(en)

overlap in WQ.

Once the denotation of DETevent is applied to the denotation of the wAen-clause in its base position, the free event variable en can be bound by the Q-adverb - which I assume

184

Stefan

Hinterwimmer

to be adjoined to the matrix TP - under c-command. Technically, this is achieved by inserting an event variable binding operator η η , which turns every event variable bearing the index η into a λ-bound variable, and thus in effect into a variable bound by the respective Q-adverb, as shown in (29) (cf. BUring, 2004). (29)

| b „ X P p = Ae. [[XP]»[ n - e l(e)] where ηη is the event variable binding operator and g[n —> e] is the assignment function that (possibly) differs from the assignment function g insofar as it assigns the value e to all situation variables bearing the numerical index n.

Consider now in (30a) the LF-representation of (27a), which is interpreted as shown in simplified form in (30b): (30) a. [χρ [ when his c. snores] [TP Gene [ 7 [TP Paul hates [DP DETevent EI [ when his c. snores]]]]]] b. XP.XQMe [Q{e)(w0) Be" [e < e" Λ P(e")(w)\] (Xe.hate(e,paul, Le'[snore(e', colleague)(wo) A o(e, e\)(wo)) (Xe'".snore(e"', mileage) (WQ))] (WQ) = \/e[snore(e, colleague)(wo) 3e" [e < e" Λ hate(e", paúl, Le'[snore(e', coll.)(wo) A o(e',e")(w0)})(w0)i We now have an account that gives us an interpretation which is equivalent to the one sketched in the introduction, and at the same time explains why w/jen-clauses asscociated with it behave in (almost) the same way as í/iaí-clauses associated with it as far as the contrast between left- and right-adjunction is concerned. What still needs to be accounted for is (a) why left-adjunction is allowed in the case of structurally similar sentences with «/-clauses, (b) why contrastive topic marking allows left-adjunction in the case of whenclauses, too, and (c) why the contrast between right-and left-adjunction is stronger with that-clauses than with w/ien-clauses.

4. Why Left-Adjunction is Sometimes Allowed 4.1 //"-Clauses Associated with it Consider again the examples in (1 la,b) (repeated here as (31a,b)), where an «/-clause is associated with it, and where not only right-adjunction, but also left-adjunction is fine: (31) a. / would like it if he played the violin right now. b. If he played the violin right now, I would like it. (from Pesetsky, 1991, p. 65)

When-Clauses, Factive Verbs and Correlates

185

Building on the idea developed by Lewis (1973) that counterfactual conditionals 2 involve universal quantification over worlds that (simplifying somewhat) only differ from the world of evaluation insofar as the respective (/-clause is true in them, the intuitive meaning of the two sentences can be paraphrased as given in (32): (32) In all worlds that are minimally different from the actual world and where it is true that he plays the violin at the utterance time, I like the fact that he plays the violin at the utterance time. Now note that a derivation analogous to the one suggested for right-adjoined w/jen-clauses associated with it would not work for the sentence in (31a): it is well-known that i f clauses can only be interpreted in the restrictor of either an overt or covert modal operator or an overt or covert quantifier over situations/events (Kratzer, 1986), but not in the nuclear scope of an operator (Johnston, 1994). Consequently, a copy of the ¡/-clause that was base-generated within a DP-shell functioning as the internal argument of the matrix verb could simply not be interpreted in this position. Let us therefore assume that the «/-clauses in sentences such as (31a,b) are either leftor right-adjoined to the matrix TP, where they can be interpreted in the restrictor of the covert modal operator that I assume to be present in counterfactual conditionals (following Kratzer, 1986 and many others). Concerning it, I assume that it spells out the combination of the covert determiner turning propositions into facts (see (22) above) and a clause that only differs from the «/-clause insofar as the complementizer if has been replaced by that, and which accordingly has been deleted under (near) identity (see Pesetsky, 1991 for a slightly different, though related view). Consequently, (31a) gets the LF-representation in (33a), and (32b) the one in (33b), where MUST is the covert modal operator in whose restrictor the //-clause is interpreted. (33) a. [ T P [ T p MUST [ T P I would like [ D p D E T / a c t [cp that he played the violin right now]]]] [cp if he played the violin right now]] b. [TP [CP if he played the violin right now] [TP MUST [TP I would like [DP D E T / a c t [CP that he played the violin right now]]]]] In order to arrive at the correct interpretation, we have to assume that the free world variable introduced by DETfact is not resolved to the world of evaluation. Rather, it is turned into a variable bound by the covert modal operator via a binding operator that only differs from the event variable binding operator given in (29) above insofar as it turns free world variables instead of event variables into bound ones. The resulting interpretation is given in simplified form in (34): (34) Vw'[R(wo, w') Λ Be \play_violin(e, g(xi))(w')} —» 3e' [like(e',I, tx[SH(®, Xw"3e [piay_wzoftn(e,fl(xi))(«>")])K)])K)]] This corresponds to the paraphrase in (32), and is thus the desired result. 2

I use the term counterfactual conditional without committing myself to the view that conditionals of this type presuppose their antecedent to be false in the world of evaluation; see Bennett (2003) for detailed discussion.

186

Stefan Hinterwimmer

4.2 Mien-Clauses as Contrastive Topics Recall finally that the ungrammatical examples in (4) become acceptable if a constituent inside the respective when-c\a.use is marked as a contrastive topic, as shown by (14a,b) (repeated here as (35a,b)): (35) a. When his /COLLeagueV snores, Peter HATES\ it (while it does not bother him at all when his wife snores), b. When her /DAUGHterV plays the piano, Mary usually LOVES\ it (while she usually hates it when her son plays the piano). Now recall that our account of the unacceptability of examples like the ones in (4a,b) relies on the assumption that the when-clauses are base generated within DP-shells in argument position and thus cannot be moved to the left, but only to the right. Therefore, we have to assume that the when-clauses in (34a,b) are base generated in initial position. Concerning the semantic interpretation, the examples in (35a,b) only differ from the corresponding examples with right-adjoined when-clauses in (26a,b) above insofar as asserting them invokes a set of (relevant) alternative propositions that are true in the respective context, and that differ from the one asserted in both the constituent marked as the contrastive topic and the one marked as the focus (see the continuations in brackets). In other words, as far as their ordinary semantic value is concerned, the variants in (35a,b) are identical to the ones in (26a,b) - it is only with respect to their contrast semantic value (Büring, 1997) that they differ. Note first that applying an analysis based on Rothstein (1995b) to these examples, according to which the neuter pronouns function as event variables that are bound by the respective operator, while the when-clauses are base generated as TP-adjuncts, does not go through - independently of the fact that this would leave open the question of why this strategy is only allowed in the presence of contrastive topic marking. The problem is that applying standard mechanisms of pronoun binding would give us the wrong result: inserting the binding operator defined in (29) above, for example, would have the consequence of turning the (initially) free variable provided by the respective neuter pronoun into the same variable as the one ranging over the respective hating/loving events. In other words, (35a), for example, would receive a reading that can be paraphrased as given in (36) clearly the wrong result.

(36) Generally, for an event e, which is an event of Peter's colleague snoring, there is an associated event e' which is an event of Peter hating e!.

I therefore suggest that in cases like (35a,b) the following strategy is applied: two identical wAen-clauses are merged in different positions. The first one is base generated as a TPadjunct, which causes it to be interpreted in the restrictor of the respective operator, while the second one is inserted inside a DP-shell headed by the covert determiner DETevent introduced in section 3.2 above. Phonological identity with the TP-adjoined wAen-clause

When-Clauses, Factive Verbs and Correlates

187

causes the lower one to be deleted at PF, triggering the phonological rule discussed in section 3.1 and 3.2. Consequently, the ordinary semantic value of (35a) given in (37) is identical to the one given in (30b) above for the corresponding sentence in (26a) with the right-adjoined w/ien-clause. (37) Ve [snore(e,colleague)(wo) —>• 3e"[e < e" Ahate(e",paul,Le'[snore(e',coll.)(wo)

Ao(e',e")(t«o)])(wo)l]

Concerning the question of why contrastive topic marking is required in order for the strategy involving base generation of the wAen-clause in TP-adjoined position to be available, I assume that this is due to the following reason: as far as the ordinary semantic value is concerned, an equivalent result could have been achieved by simply inserting the respective when-clause as the complement of DETe,vent. Inserting an identical w/ien-clause in adjunct position is therefore a marked option which needs an independent justification. Now, the need to mark an element inside the respective when-clause as a contrastive topic provides such a justification, since there is a strong tendency for contrastive topics to occur in sentence-initial position (see Büring, 1997). With these assumptions in place, we can explain why sentences containing f/jaf-clauses associated with it are ungrammatical, why ones containing vWien-clauses are just very odd (in the absence of contrastive topic marking, that is): in the latter case, an alternative strategy involving base generation of the wftew-clause in left-adjoined position is in principle available. In the case of sentences with that-clauses, in contrast, employing this strategy would not give us an interpretable result, since there is no operator in whose restrictor the that-clause could be interpreted. We thus have an account that answers all the questions raised at the end of section 2, namely (a) why there is a contrast in acceptability between left- and right-adjunction in sentences where a when-clause is associated with a neuter pronoun, (b) why the same effect occurs in stronger form in structurally similar sentences with fAai-clauses, and is almost absent in ones with i/-clauses, and (c) why contrastive topicality cancels it in the case of left-adjoined when-clauses, but not in the case of left-adjoined f/ja/-clauses.

References Authier, J.-M. (1995). V-Governed Expletives, Case Theory, and the Projection Principle. Linguistic Inquiry 22. 721-740. Beaver, D. & B. Clark (2008). Sense and Sensitivity: How Focus Determines Meaning. Oxford: Blackwell. Bennett, J. (2003). A Philosophical Guide to Conditionals. Oxford: Oxford University Press. Bolinger, D. (1977). Meaning and Form. London: Longman.

188

Stefan

Hinterwimmer

Biiring, D. (1997). The Meaning of Topic and Focus - The 59th Street Bridge Accent. London: Routledge. Biiring, D. (2003). On D-trees, Beans, and B-Accents. Linguistics & Philosophy 26. 511-545. Biiring, D. (2004). Crossover Situations. Natural Language Semantics 12. 23-62. Biiring, D. & K. Hartmann (1997). Doing the Right Thing. The Linguistic Review 14. 1 —42. Chomsky, N. (1973). Conditions on Transformations. In: S. Anderson & P. Kiparsky (eds.) A Festschrift for Morris Halle. New York: Holt, Rinehart & Winston. 232-286. Elbourne, P. (2001). Ε-type Anaphora as NP-Deletion. Natural Language Semantics 9. 241-288. Elbourne, P. (2005). Individuals and Situations. Cambridge, MA: MIT Press. Goebbel, E. (2007). Extraposition as PF-Movement. In: E. Bainbridge & B. Agbayani (eds.) Proceedings ofWECOL 2006. California State University. 132-145. Hinterwimmer, S. (2008). Q-Adverbs as Selective Binders. Berlin: Mouton de Gruyter. Johnston, M. (1994). When-Clauses, Adverbs of Quantification, and Focus. In: M. S. Raúl Aranovich, Susanne Preuss & W. Byrne (eds.) Proceedings of the Thirteenth West Coast Conference on Formal Linguistics (WCCFL). Chicago: Chicago University Press. Kiparsky, C. & P. Kiparsky (1973). Fact. In: M. Bierwisch & Κ. E. Heidolph (eds.) Progress in Linguistics. The Hague: Mouton. 143-173. Kratzer, Α. (1986). Conditionals. In: P. F. Anne M. Farley & Κ. E. McCollough (eds.) Papers from the Parasession on Pragmatics and Grammatical Theory. Chicago: Chicago Linguistics Society. 115-135. Krifka, M. e. a. (1995). Introduction to Genericity. In: G. N. Carlson & F. J. Pelletier (eds.) The Generic Book. Chicago: The University of Chicago Press. 1-124. Lewis, D. (1973). Counterfactuals. Oxford: Blackwell. Müller, G. (1996). On Extraposition and Successive Cyclicity. In: U. Lutz & J. Pafel (eds.) On Extraction and Extraposition in German. Amsterdam: Benjamins. 213-243. Müller, G. & W. Sternefeld (1995). Extraction, Lexical Variation and the Theory of Barriers. In: P. E. P. Urs Egli & C. Schwarze (eds.) Lexical Knowledge in the Organization of Language. Amsterdam: Benjamins. 35-81. Pesetsky, D. (1991). Zero Syntax, Vol. 2: Infinitives. Unpublished Manuscript, MIT. (available from homepage). Peterson, P. L. (1997). Fact, Proposition, Event. (Studies in Linguistics and Philosophy 66). Dordrecht: Kluwer. Postal, P. & G. Pullum (1988). Expletive Noun Phrases in Subcategorized Positions. Linguistic Inquiry 19. 635-670. Ross, J. R. (1967). Constraints on Variables in Syntax. (Doctoral Dissertation). Cambridge, MA: MIT. Rothstein, S. (1995a). Adverbial Quantification over Events. Natural Language Semantics 3. 1-31. Rothstein, S. (1995b). Pleonastics and the Interpretation of Pronouns. Linguistic Inquiry 26. 499-529.

When-Clauses, Factive Verbs and Correlates

189

Schwarz, F. (2009). Two Types of Definite Determiners in Natural Language. (Doctoral Dissertation). Amherst: University of Massachusetts. Stowell, T. (1981). Origins of Phrase Structure. (Doctoral Dissertation). Cambridge, MA: MIT. Vendler, Z. (1967). Linguistics in Philosophy. Ithaca: Cornell University Press. Vendler, Z. (1972). Res Cogitans: An Essay in Rational Psychology. Ithaca: Cornell University Press.

The Proof Theory of Partial Variables Gerhard Jäger

1. Partial Variables: Motivation In Montague semantics, indefinites are interpreted as generalized quantifiers with an existential impact. Leaving aside issues of intensionality, the meaning of a phrase like (la) comes out as (lb), being of type ((e, t),t). (1)

a. b.

a man XP3x(Mx

A Px)

Post-Montagovian semanticists were not fully comfortable with this analysis. There is a common intuition that indefinites are, or can be, referring expressions, even if the referent is only known to the speaker, or not even to him. Formally speaking, indefinites "feel" like being of type e, despite all logical arguments to the contrary. Discourse Representation Theory implements this intuition to some degree by assuming that indefinites introduce a variable of type e into the semantic representation of the sentence or discourse where it occurs. However, strictly speaking, an indefinite like a man does not only introduce a free variable but an open formula, like M x , into the representation. So technically speaking, indefinites are of type t in DRT, which is even less intuitive than the Montagovian analysis. In a series of publications, Tanya Reinhart explored the empirical and logical consequences that result if indefinites are in fact taken to have the logical type e (Reinhart 1997; see also Winter 1997; Kratzer 1998 and much subsequent work). If one maintains the uncontroversial assumption that the head noun (plus its modifiers) in an indefinite DP expresses a property (type (e, t)), the indefinite article must have type ((e, t), e). If one grants the rather trivial assumption that the descriptive part of an indefinite should describe the object that is referred to, this entails that the indefinite article denotes a choice function. While this idea is intuitively appealing and led to insightful analyses of many empirical phenomena, it faces various problems, as pointed out, inter alia, by Reniers (1997), Geurts (2000) and Endriss (2001). The most serious is the so-called empty set problem. A choice function also supplies a value if applied to the empty set. So the sentence (2)

A unicorn entered.

The Proof Theory of Partial Variables

191

may be true in the actual world, depending on which choice function is used as interpretation of the indefinite article. The fundamental intuition that ordinary indefinites have an existential impact is not captured under the choice function analysis. In Jäger (2007), I made a proposal for a compositional semantics of indefinites that preserves the fundamental assumptions of the choice function analysis but leads to Montague style truth conditions for sentences. There I make use of a new syntactic device, partial variables. In the next section, I will briefly recapitulate this proposal. While Jäger (2007) was couched in type theoretic terms to allow a Montague style compositional treatment, partial variables are essentially first order devices. In the present paper, I will concentrate on the extension of classical first order logic with partial variables. I will present an embedding of this logic into standard first order logic. Since there is a sound and complete proof theory for the latter, this indirectly defines a sound and complete proof theory for the former as well.

2. Semantics The underlying idea of my proposal is inspired by DRT: Semantically, an indefinite noun phrase is an individual variable. So in a sentence like (2), a unicorn is to be translated as some variable χ of type e. On the other hand, I concur with the choice function proponents. This variable must refer to some unicorn. If the local assignment function maps χ to some non-unicorn, the NP does not refer. To carry this home formally, I extend the syntax of the logical representation language with the following clause: Definition 1. If χ is a variable of type e and ψ a formula of type t, [χ \ φ\ is a term of type e. (This could be generalized to variables of arbitrary type, but in the present context type e will do.) These terms are called partial variables. The formula is called its restriction. An indefinite like a unicorn would be translated as a term [x|C/i]. This term should behave like an ordinary variable, but it should fail to denote if x, without the restriction, would refer to some non-unicorn. So here is a first stab at the semantics of these terms:

A syntactically complex expression is assumed to be undefined (under some model and some assignment function) iff at least one of its sub-expressions is undefined. The only exceptions are the rules for the quantifiers. I only give the semantics of the existential quantifier here; the treatment of the universal quantifier is analogous.

w^wf

1 iff for some ο : \\φ\\^[χ/α] = 1 0 iff for all α : if \\• Fy) 3x(Bx Λ \/y(Gy Λ Myx Λ Lyx Fy)) 3x(Bx Λ Lyx Λ Vy(Gy Λ Myx Fy))

We are interested in the reading where the indefinite has scope over the entire sentence, which corresponds to the formula in (5b). The sub-formula GyAM(y, [x\Bx Λ Lyx]) —>• F(y) is only defined under g if χ is interpreted either as ± or as a boy who is loved by g(y). My{Gy Λ M (y, [x\Bx Λ Lyx}) F (y)) hence comes out as true if g(x) = _L and and every girl who met X fainted, or if g(x) is a boy and every girl who met and likes g(x) fainted. So the entire sentence would get the interpretation indicated in (5c). The clause Lyx here ends up as restriction of the universal quantifier. Such a reading does not exist - if the indefinite has a specific reading, this clause must be part of the restriction of the existential quantifier, and the occurrence of y in this clause must be free (as indicated in (5d)). To generalize this point, we want the following equivalence to be valid regardless of free variables in φ that appear to be bound inside φ: 3χψ{[χ\φ\)

·*=> 3χ(φ Λ φ(χ))

This has two consequences. First, free variable occurrences inside the restriction of a variable should not be visible from the outside while the variable that is being restricted is free. Second, as soon as a partial variable becomes bound, the free variable occurrences in it should become visible again. To take an example, in (6)

3y3x3yP([x\Rxy})

the occurrence of y inside the restriction of χ should be • free in Rxy, • not free in

P(\x\Rxy]),

• free in 3x3yP([x\Rxy\), • bound in

and

3y3x3yP([x\Rxy]).

The intuitive idea to make this work can be explained by talking about stacks of variables. If y occurs in the restriction of x, what is seen from the outside is a stack with χ on top and y underneath. Therefore, y is not visible for binding. If χ is bound by a quantifier though, it is taken off the stack, and y is visible again/can be bound. Formally, stacks are implemented as non-empty sequences of variables. So an assignment function is a function from V+ to the universe (where V is the set of variables).

194

Gerhard Jäger

Here is the final version of the semantics. For the time being, I only consider the treatment of partial variables in the context of first order logic.1 Definition 3. Let M = {E, F) be a first order model (where E is the universe and F the interpretation function). _L G E is a designated object, and E — {_L} φ 0. g, gx and g{x) are assignment functions, ie. a functions from V+ to E. _L is a designated constant. We define that for all ν G V*, χ, y G V with χ φ y, and formulas φ: 9Χ(ΰ) g{x}(x+) g{x}{x*yv) F{L) g(x)if IMIJÍ = 1 g(x)if\\3ί'[£' < £ A [£' < η A Mozart finishes his requiem at £']] [[Mozart starb [bevor [er sein Requiem vollendete]]]™ = \t[t < η A Mozart dies at £ A ->3£'[£' < £ Α [£' < η Λ Mozart finishes his requiem at £']]]

Applying the assertion operator, which crucially applies only at this stage and not already at the level of the clause Mozart starb, yields the following result (here somewhat simplified): (38)

[ASSERT]"([[Mozart starb [bevor [er sein Requiem vollendete]]]") = asserted: 3t[t < η A Mozart dies at ί Λ ->3t'[t' < t A Mozart finishes his requiem at ί']]

This is what we expect: At the time of Mozart's death t (situated before the speech time) there is no prior time at which Mozart finished his requiem. It follows from our background knowledge, as one cannot finish one's requiem after one's death, that there is also no later time at which Mozart finished his requiem.

3.2 The Implicature of Likelihood The current proposal seems insufficient, as we would use (34) only if the bevor clause describes a proposition that is likely at the time of the evaluation of the main clause. For this reason, the following example is odd: (39) #Mozart starb bevor er ein Oktett für Streicher und Helikopter komponierte. 'Mozart died before he composed an octet for string instruments and helicopters' To explain this, Heinämäki (1972), Ogihara (1995) and Beaver & Condoravdi (2003) have proposed modal accounts of before (which is similar to bevor) which require that non-factual before clauses are interpreted at alternatives of the real world that are likely at the time at which the main clause is evaluated. But I think going modal is unnecessary. The meaning component in question can be derived from pragmatic principles of informativity: A sentence [A before B] is only informative in contexts where it is considered reasonably probable that Β is true some time after the time of interpretation of the main clause. If the common ground carries the information that Β is highly unlikely at any time, then the statement that Β is not true at any time before a time at which A is true is an unmotivated restriction. More formally: If the a-priori likelihood of 3 t \ B ( t ) } is close to 0, then the a-priori likelihood of [A before B] is close to 1, and asserting it would violate the maxim of relevance. Consequently, when uttering a sentence [v4 before B], the speaker creates the implicature that the a priori probability that 3£[B(£)] is substantially greater than 0.

222

Manfred Krifka

3.3 The "Factual" Interpretation of bevor and the Concept of Reified Implicatures There is a general tendency that bevor/before-clauses are interpreted factually. That is, a sentence [A bevor Β] does not only imply that Β is likely, but that it is, in fact, true. Consider the following sentence: (40) Herr Maier lernte Frau Schmidt kennen, bevor er Herrn Schmidt 'Mr. Maier met Mrs. Schmidt before he met Mr. Schmidt.'

kennenlernte.

By default, we assume that Mr. Maier met Mr. Schmidt as well. A question meaning 'When did Mr. Maier meet Mr. Schmidt?' is fully appropriate after (40). But this inference can be cancelled: (41 ) Tatsächlich hat er Herrn Schmidt nie kennengelernt, denn die Schmidts haben sich kurz darauf getrennt. 'As a matter of fact, he never met Mr. Schmidt because the Schmidts separated soon after.' Cancellable inferences are conversational implicatures, according to Grice (1975). The implicature in question arises from general conversational principles, as follows: The sentence [A bevor Β] states that Β was not the case before a time t at which A is true. It competes with the stronger statement - Έ , that Β never was the case. The usual mechanism for scalar implicatures is set in motion: The hearer can infer that the stronger statement ->B does not obey the maxim of quality (typically, because it is false), as otherwise the speaker would have made that stronger statement, following the maxim of quantity. This means that, typically, the negation of the more general statement ->-iB, which is equivalent to B , is implicated. (42)

|ASSERT[A[bevor Β]]ψ asserted: 3 í [ [ A f ( í ) Λ t'[t' < t Α implicated: 3t"[\B\n{t")}

[ΒΓ(ί')]]

Together with the core meaning of [A bevor Β] that Β was not the case before a time t at which A is true, it is implicated that Β is true at some time after t: (43) implicated, given what is asserted: 3í[[[.4] n (¿) Λ 3t'[t < t' A [-g]"] As we do expect from conversational implicatures, this meaning component is partly independent of the lexical items involved, in particular from bevor, as it also occurs with expressions like ehe that have the same truth-conditional meaning. But it arises quite standardly whenever the meaning of bevor / ehe is applied, and hence can be difficult to distinguish from the lexical meaning of these expressions (except, of course, for the fact that it is cancellable). The meaning component is a generalized implicature in the sense of Grice (1975), or a default implicature in the sense of Levinson (2000).

How to Interpret "Expletive " Negation under bevor in German

223

For the following discussion I would like to "reify" this implicature as a meaning component of bevor that originates from general conversational principles but is "folded into" the meaning of a lexical item. It is a "hard-wired" implicature. Being part of the meaning of bevor, it is conventional, but it still can be cancelled; hence it is different from conventional implicatures in the sense of Grice (1975), and recently discussed in Potts (2005). Reified conversational implicatures will be implemented in form of two-dimensional semantic representations (α, β), where a is the core meaning, and β the implicature. This is similar to the treatment of scalar implicatures introduced by number words or Boolean disjunction in Landman (2000) and Chierchia (2004), who assume that those implicatures are built "in tandem" with truth conditions. The implicature is cancelled if it is contradicted by the core meaning in the context in which it is interpreted. If not cancelled, the overall meaning is the conjunction of a and β. In the case at hand, the adverbial clause [bevor Β] and the proposition expressed by [A bevor Β] are interpreted as follows, where I underline the implicature part: (44)

(45)

{before B\n = (Aí-at'[í' < t a lB}n{t%xt3t'[t

\t[lAr(t)A3t'[t3t[t 3í[t < η A Maria falls asleep at tA ->3t'[t' < t A ~{t' < η A Hans is home at £']]

The core meaning reduces to truth, as λέ[Maria falls asleep at iA_L] = λί[1_], and —[_L] = T. This means that the core meaning is always satisfied; it is a tautology. The only meaning contribution is due to the implicature, which states that there is no t at which Maria fell asleep that is followed by a time at which Hans was not home. This turns out to capture the truth conditions, as the diagram (74) shows: (74) Diagram for Maria schlief nicht ein bevor nicht Haus zuhause war Hans is not home

τ for t there is a time t\ t Maria falls asleep at í]¡ This formal rendering can be seen as rather direct representation of the near-paraphrase If Hans is not yet home, Maria doesn 't fall asleep. It represents Weisgerber's intuition that bevor sentences with "expletive" negation have a conditional interpretation. We also can explain why negation under bevor appears to be "expletive" - that is, why we can drop it without obvious change of meaning. The assertion part of (70), which is the only relevant meaning as the implicature is cancelled, is similar in meaning to the implicature part of (74), which is the only relevant meaning as the core meaning is a tautology. The truth conditions of these two readings are extremely similar, even though the ways by which we arrive at them are quite different. For this reason, both the version with "expletive" negation and the one without can be considered equally complex, and hence the version without negation cannot block the one with negation. Semantically, the two versions differ only insofar as Maria schlief nicht ein bevor Hans zuhause war excludes that Maria falls asleep at the moment when Hans comes home (or any moment before

232

Manfred

Krifka

that), whereas Maria schlief nicht ein bevor nicht Hans zuhause war allows for Mary to fall asleep at the moment when Hans arrives (cf. diagram (75)). This is a vanishingly small difference, especially in a model with a dense structure of time.

6. The Obligatoriness of Negation and the Conditional Interpretation Negation in the main clause is crucial for expletive negation under bevor. Without it, we get a clause that is necessarily false, due to the falsity in the second conjunct of the core meaning: (78) * Maria schlief ein bevor nicht Hans zuhause war. = asserted: 3t[Maria falls asleep at t Λ _L], ( = _L) implicated: 3ί[Maria falls asleep at t Λ 3t'[t < t' Λ ->[Hans is home at t'\\ However, we found examples without negation in the main clause, as in (25). But notice that this is a generic sentence that involves a quasi-universal quantification, where the bevor clause is in the restrictor of the quantifier. This results in a plausible interpretation. (79) Bevor das Kind nicht sitzen kann, fühlt es sich im Liegen am Mt[\bevor das Kind nicht sitzen kannj" (t) —> les fühlt sich im Liegen am wohlsten\'1 {t)\

wohlsten.

I have represented the generic force here by a universal quantifier, and the bevor clause as the restrictor of this quantifier. The meaning of the bevor clause is as follows: (80)

\bevor das Kind nicht sitzen kann¡n t A --[the child can sit up at t'), At3t'[t < t' Λ -ι [the child can sit up at £']]) = (At[_L], Xt3t'[t < t' Λ -.[the child can sit up at t'}})

= (Λí-at'[í'
[the child feels most comfortable lying down at ί]](= T ) , Vt[3t'[t < t' Λ -ι[the child can sit up at t']] —• [the child feels most comfortable lying down at t}])

Again, the core meaning is a tautology, for all plausible models, and all the information is carried by the implicature part. It states that for all times t that are followed by a time at which the child (still) cannot sit up, the child feels most comfortable lying down. This is the right interpretation. The assertion of (81) is the conjunction of both the core meaning

How to Interpret

"Expletive " Negation under bevor in German

233

and the implicature, which is equivalent to the implicature. We see that the occurrence in the restrictor of a universal quantifier leads to a situation in which the core meaning of the bevor clause is "neutralized", and only the implicature part survives. Of course, bevor clauses as restrictors of conditionals need not contain a negation. The following statement has (nearly) the same truth conditions, as predicted by our formalization: (82)

|[CP[ bevor das Kind sitzen kann] [c fühlt [IP es sich im Liegen am

wohlsten]}]}11

The bevor clause receives the following interpretation: (83)

\bevor das Kind sitzen kann\n = (Λί—>3£'[í' < t A [the child can sit up at £']], \t3t'[t < t' A [the child can sit up at t]], ( = A¿[T]))

Now the implicature reduces to tautology, for all plausible models: If we assume that the child learns to sit up at some time (and never looses this ability), then the set of times t that precede a time at which the child can sit up is the set of all times. With this, the representation of (82) is as follows: (84)

(Ví[->3í'[£' < t A [the child can sit up at i']] —> [the child feels most comfortable lying down at t], Vt[3t'[t < t' A [the child can sit up at ¿']] ->• T], ( = T ) )

Now it is the implicature that reduces to tautology, and the semantic weight is carried by the core meaning. It states, correctly, that for all times t at which the child still cannot sit up, he feels most comfortable lying down 10 . 10

(i)

It should be noted here that a proper representation of bevor clauses in quantificational contexts is in need of certain refinements of the representations considered here, which were simplified by the fact that we considered stative clauses. Consider the following generic sentence, and the representation we would currently assign to it: Die Glocke läutet bevor die Sonne aufgeht. 'The bell rings before the sun raises.' (Ví[-.3í'[í' < t Λ [the sun rises at i']] [the bell rings at t]], Ví[3í'[í < t' A [the sun rises at i']] [the bell rings at i]]) Even if we concentrate on a single sunrise at a time fo, the core meaning and the implicature state something much too strong - that the bell must ring at all times before ίο. One way to avoid this is to assume that bevor clauses specify stretches of times. For example, bevor Peter ging 'before Peter left' specifies the stretch of time from the beginning up to the point where Peter left. The main clause is then said to be true at some part of this time, making the bevor clause effectively a way to identify, or restrict, the reference time in the sense of Klein (1994). Quantification over bevor clauses then would be quantification over sets of such maximal times, e.g. in the case of quantification over all maximal times before each sunrise. Still, a number of things would have to be worked out, like guaranteeing that each sunrise is associated with its own bell ring (cf. de Swart, 1993), guaranteeing that the bell ring is temporally close to its sunrise, and taking care of cases which allow for simultaneous event times, as in Before a man is buried, a bell rings. The solution to these problems will have to wait for another occasion.

234

Manfred Krifka

7. The Syntax of Propositional vs. Assertive Negation As stated in the introductory section, it has been observed that the two types of negation differ in their syntactic behavior (cf. Weisgerber, 1960; Weiß, 2002; Schwarz & Bhatt, 2006). In particular, definite DPs (except pronouns) do not have to scramble over propositional negation, and do not have to result in negative determiners like kein with indefinites: (85) a. Bevor nicht Hans zuhause ist, schläft Maria nicht ein. b. Bevor er nicht zuhause ist.../ * Bevor nicht er zuhause

ist...

(86) Bevor nicht eine Nachricht von Hans gekommen ist, schläft Maria nicht ein. It should be mentioned that the negation with these syntactic features occurs in several other environments in addition to bevor-clauses, like in biased polarity questions and counterfactual conditional sentences: (87)

Wohnt nicht Peter in Berlin?

(88)

Wenn nicht ein Wunder passiert, sind wir verloren.

In contrast, regular or "assertive" negation often is preceded by definite NPs, and combines with indefinite determiners to kein-. Otherwise, we do get a contrastive interpretation. (89)

Weil Hans nicht zuhause ist / # Weil nicht Hans zuhause

ist...

(90) Weil keine Nachricht gekommen ist / # Weil nicht eine Nachricht gekommen

ist...

Given that there are these two kinds of negations, an obvious question is why they differ in their syntactic behaviour in this way. In particular, if regular negation indeed is associated with an illocutionary function, and even if it just existentially closes the time variable, we would expect it to reside higher in the syntactic tree than propositional negation, which needs a semantic representation with a disclosed time variable. A plausible reason for this behavior of regular negation is the following: If it is indeed related to illocutionary force, then it should be sensitive to the current common ground. In particular, it should indicate the givenness status of constituents, and this is exactly what its placement indicates, as given constituents have to scramble left of it, and non-given indefinites in its scope are marked by kein- forms. In contrast, the need to mark givenness is relaxed for propositional negation, as it does not relate the proposition to the common ground (see also Klein (submitted) for a slightly different motivation for the syntactic position of negation). There are other kinds of negations that are not propositional, but assertive in nature in particular, the clausal negation es ist nicht der Fall, dass . . . 'it is not the case that and rejecting or denying negation, as marked by keineswegs. Due to their assertive nature, they cannot be used as "expletive" negation under bevor either:

How to Interpret "Expletive" Negation under bevor in German

235

(91 ) *Maria schlief nicht ein bevor keineswegs Hans zuhause war. However, if these negations are assertive, why can they not serve as negations in the upstairs clause either (cf. (23) and (24))? These negations presuppose that a similar sentence without the negation was uttered or can be inferred from the conversational situation, and they are used to deny that target sentence. This makes them not suitable for our case, as their target sentence would be ungrammatical. In concluding, I would like to turn to the question why modern English does not exhibit "expletive" negation under before. I would like to propose that this is related to the fact that English expresses negation by a separate head (the inflected form do + η 't), whereas German expresses negation primarily with a modifier (nicht). Under this assumption the English expression, being a head, is necessarily tied to assertion, different from the German modifier negation. Hence, English does not have "expletive" negation under before because it does not have a propositional negation expressing the meaning \p\t^p(t), but only an assertive negation, Xp-u, = 1 iff for some a € D, \p{x)f¡(^/a) a. b.

= 1, where either

(opaque, default:) D = [fi]® w or (transparent, value loaded:) D is non-empty, and d G D iff the proposition {w : \R{x)\ä[ *Jd'1} is presupposed by the speaker.

The value loaded reading of sentence (3) then implies that among the elements in D are those for which the speaker knows they are witches. This does not imply that the speaker knows the actual extension of witch because D could be a subset of witches. As for (6), 3

(i)

Note that in "ordinary" sentences like The woman loves every

VfB-player.

the speaker may also pick out a certain set which he characterizes as VfB-players. But this is not part of the unmarked reading of (i), which only says that if someone is an VfB-player, the woman loves him. Value loading in (i) would be possible (and perhaps be part of pragmatics, but it would not change truth conditions in (i). It does change truth conditions in the examples under discussion. 4

Perhaps there are other ways of picking out a certain set and name it; some of these options are described below when we discuss the restriction of indefinite NPs.

242

Wolfgang

Sternefeld

D contains at least Amy, Berta, and Carla. The reading therefore does not imply that John wants to marry just someone (unspecifically), or that he wants to marry any sister of mine. On the other hand, as the only restriction on D is that it is not empty, there might only be one single individual in D, but in that extreme case the reading would no longer be called non-specific. W e might also consider several weaker readings, e.g. by replacing " i f f " in (14-b) by " i f " . The set D would then effectively be determined by the possible values for χ in the relevant possible worlds that satisfy the scope of the quantifier. The essential property of these readings is that, in a sense, they are all about a particular set of alternative individuals, chosen as values for the existential in possible words, but having the property described by the restriction in the actual world. 5 Another much weaker construal could be paraphrased as: There is a certain property Ρ such that John thinks that someone with this property is after him, and the extension of Ρ in the actual world is the (actual) set of witches. This reading would no more be about individuals chosen in the attitude worlds, as the choice of an individual there is no more restricted by the actual set of witches. 6 Due to the great variety of different construals and some disagreement and vagueness in the literature about which readings are possible, I will not pursue the matter any further. Rather, I will stick to (14-b) as an interesting relevant construal (one that, as far as I know, has not yet been formalized in the literature), but much of what is being said in what follows will also apply to these alternative construals. (14-b) states that value loading is always induced by the speaker, this, however, does not hold in general, but seems to be due to the simplicity of our previous examples. So let's consider a more complicated case: (15)

John suspects that Mary believes that a witch is after her.

Here is another value loaded reading which, for ease of exposition, is again approximated in Ty2: (16)

suspect(u;o,i, (Aw)(believe(u;, m, ( A w ' ) ( ( 3 a ; ) ( w i t c h ( w , χ) Λ i a h «

ζ))))))

(16) makes it clear that the value loading is not imposed by the speaker; this follows since witch is not evaluated at wq. Rather, value loading with w as shown in (16) suggests that we are considering x's such that John suspects them to be witches. According to my intuitions, however, sentence (15) might also imply that John would refer to the values of χ as witches, not that he suspects these to be witches. What is involved is a referential presupposition

on John's part (cf. also Cresti (1995) and Romoli

and Sudo (2008) for the presuppositional nature of value loading). On the other hand, consider (17)

John doubts that Mary believes that a witch is after her/him.

5

So these readings are in a sense "de re", but see fn. 2 above.

6

In fact, such a reading would no more be "de res".

243

Wide Scope in situ

Isn't there also a reading where John doubts that whatever is believed to be after her/him is a witch? I must admit that my intuitions are shaky, but some of my colleagues convinced me that such a reading indeed exists (informants also found a difference in acceptability between her and him, which I must ignore). This amounts to saying that D is pragmatically determined either (a) by the "referential beliefs" (beliefs about certain res) of the speaker or (b) of previously mentioned holders of propositional attitudes, or (c) by the propositional attitudes and their bearers mentioned in the contexts that embed the quantifier. We therefore have to modify (14-b) by adding further alternatives for all these possible readings. As a formal implementation we suggest the following. We interpret the restriction of a quantifier with respect to a context set C. This is a set of sets C 0 , Ci, C2 . . . o f disambiguations that will be built up during the evaluation of a formula. Co is the set of referential beliefs of the speaker, ie. a set of propositions presupposed by the speaker (Mary is a witch, Berta is a witch etc.), and initially C = {Co) Whenever we encounter a situation where we have to evaluate a propositional attitude, eg. John doubts that p, C can be enlarged and we interpret ρ with respect to a (potentially) larger set C'. There are two ways of enlarging C (by adding two new sets of propositions, possibly empty, if irrelevant for the further evaluation of p). We may add C¿ as a set of "referential beliefs" of John's, and we may add Cj as a set of propositions John doubts (e.g. if he doubts that Mary is a witch). We now interpret the transparent construal with respect to one of those sets Ct(18)

for some a e D, \p{x)\9$a)'C

[(3®)[R :: p(a:)]]?;£ = 1 a. b.

= 1, where either

(opaque, default:) D = or (transparent, value loaded:) D is non-empty, and for some C¿ e C, d e D iff the proposition {w :

is in C¿.

The construction of C is defined more formally in (19). Each time we evaluate a propositional attitude (with a as its subject) we may add to C either some of a ' s de re beliefs or some of the relevant attitudes (or both), depending on their potential usefulness in order to satisfy the presuppositions that might come up with the evaluation of p: (19)

[attitude-verb (w, a,p) = 1 iff ([ .

e·

Ι ( ν χ ) Φ ] ? , „ = 1 iff for all a , [ < E > ] f j x ' a > .

The notation g + (χ, a) of course symbolizes the incremental growth of g. We now reformulate (22-b) in terms of "choice functions" or "winning strategies", the latter term being perhaps a bit misleading because there is only one player in this game, namely the one that interprets the indefinite NP. We nonetheless use the terminology from g a m e theory in order to avoid confusion with the choice functions in Reinhart ( 1997). We define: (23)

A winning strategy 5 is a function that assigns an individual β to a variable a and an assigment g.

Intuitively, S assigns a value to the variable on the basis of the information about previous value assigments for other variables encoded in g. The standard truth condition for existential quantification can then be reformulated as: 7

Strictly speaking, this is not correct, but not even the standard interpretation of predicate logic is compositional. A s is standard practice, the denotations of the parts are relativized to assignments, but the interpretation of the quantifier requires the evaluation of different denotations, varying with the assignments. S o it is not only one denotation of an expression that determines the denotation of the whole, thus rendering the computation of meaning non-compositional. Of course, there are ways to make such a system compositional by making value assignments part of the denotation, cf. Bennett ( 1 9 7 9 ) (who refers back to Tarski and Vaught (1957)) or Stemefeld (2006), so we will not bother with compositionality any further.

8

Alternatively, we could stipulate that α may occur only once in the sequence, reducing sequences to ordinary sets. We will see later, however, that even if a occurs only once, the ordering in the sequence is important.

246 (24)

Wolfgang Sternefeld [(3ζ)Φ]? ι Μ , = 1 iff there is an S such that [ Φ ] ^ , 5 ( χ ' β ) > .

Note that S assigns a value to χ on the basis of the previous "shorter" assignment g, whereas Φ is evaluated on the basis of the "longer" assignment that provides a value for χ. The next step is to introduce scope independence. This means that the choice made by S does not depend on the entire (but still finite or partial) g, but only on a subsequence of 9(25)

Let g\a be the subsequence g' of g up to and excluding the first occurence of a.

Intuitively, the notation ( 3 x \ y ) Φ is intended to express that Φ is evaluated not with respect to S(x, g) but S(x, g\y). In particular, (3χ·\«;)Φ will say that the choice of a value for χ does not depend on the current evaluation world w. Given that in complex situations we must have access to any world previously considered in an evaluation, we have to resort to Ty2 in order to express independence of world variables. Moreover, we assume that the translation algorithm uses "new variables" when embedding propositions (or quantifiers). This is required to avoid confusion of variables. Eg., if two existential NPs A and Β are to have wide scope over C, then using identical variables for A and Β would imply that S chooses the same individual for A and B, which is of course not intended and would not represent a possible reading. This of course corresponds to the new variable restriction for indefinites in a Heimean file semantics, cf. Heim (1982). In the first clause of the recursive definition of truth we want to say that the matrix sentence is evaluated w.r.t. the actual world @ which, by convention, is the value of the variable wo(26)

[ Φ ] , is true in the actual world @ iff there is an S such that [ Φ ] } ^ 0 ' ® ^ = 1 (ie. S and the initial assignment g that interprets the free world variable WQ satisfy Φ with respect to the interpetation I).

The tripartite structure of universal quantification is interpreted as before: (27)

I(Vx)[ñ ::p(iu,x)]J? as in (13).

= 1 iff for all α G D: b ( w . z ) l f s X ' a > = 1, where £> is

s

As for existential quantification, we now have to account for the scope independent construals. As above, we consider two cases: (a) the case where R is interpreted in situ, which is usually associated with the opaque reading of R, and (b) the one where R is value loaded. Both cases work the way as we already discussed in section 4: (28)

[ ( 3 x \ a ) [ R :: p(w, \p{w,x)]fí a. b.

xMx

= 1 iff S(x, g\a) '*

a)) C

'

G D,

= h where either

(opaque, default:) D = or (transparent, value loaded:) for some C¿ G C , the proposition {w :

is

in Q .

Wide Scope in situ

247

(28) implies that we not only take into account the "referential beliefs" of a subject; examples like (29-a) (with stress on witch) might have a specific reading with witch being transparent with respect to Mary and opaque with respect to doubt; the reading is paraphrased in (29-b): (29)

a. b.

John doubts that Mary believes that a witch is after him. John doubts that it is a witch that Mary believes is after him.

6. Discussion 6.1 Scope Interaction Consider: (30)

Everyoney believesWo thatw ax witch blighted their mares. . . . and if they ever find out who she is, they'll try to catch her (loup (1977))

The continuation with she is meant to guarantee that the reading is specific in that a x has wide scope over believe. But it still seems to be possible to get both wide and narrow scope for 3 w.r.t. V, even if witch is interpreted as opaque, ie. in the scope of thatw . A disambiguated simplified representation of these readings in Ty2 is (31): (31)

a. b.

(Vy)believe(?i;o,2/, (Aw)(3a;\w;)[witch d e / a u ; t :: b t m ( w , x ) ] ) (\fy)believe(«;o, y, ( A w ) ( 3 x \ y ) [ w i t c h d e / n u i t :: b t m ( w , x)])

(31-a) says that every believer has a specific but possibly different person in mind that is supposed to be a witch haunting him, whereas (31 -b) implies that this person is the same for any believer. Note that scope independence from y automatically induces independence from w, a matter to which we return immediately. That the formulas in (31 ), together with the conditions in (28), correctly yield specific opaque readings should be obvious. 9

6.2 Perçus' Problem Theories that represent the loaded interpretation in Ty2 by simply chosing a different variable for the restriction seem to overgenerate if the choice of variables is totally free. For example, one can allow for (32-a) and (32-b), but (32-c) must be blocked: (32)

9

a. b. c.

(Aiu)((3x)(witch(u;, χ ) Λ iah (io, χ ) ) (Aw)((3:r)(witch(u/, χ ) Λ i a h ( w , χ ) ) (Aio)((3:r)(witch(w, χ ) Λ i a h ( u / , χ ) )

loup (1977) discusses a similar interaction, claiming that the reading she has in mind cannot be represented in first order logic. Unfortunately, I was not able to pin down what exactly her intuitions are, so there might be more involved here than can be captured by (31).

248

Wolfgang Sternefeld

This problem, discussed in Perçus (2000) and Romoli and Sudo (2008) does not come up in the present framework: loading is limited to the restriction of a quantifier, and this is directly implemented in the interpretation of the tripartite structure. No free choice of variables is permitted.

6.3 The Linear Scope Restriction Next consider a ditransitive structure and the scope possibilities for a book: (33)

everyx man bought everyy girl az book. a. b. c. d.

Wide scope: (3z\x)(book :: buy-for(x, y)) There is a certain book such that every man bought it for every girl Intermediate scope: (3z\t/)(book :: buy-for(:r, z, y)) For every man there is a book and he bought it for every girl Narrow scope: (3z)(book :: buy-for(x, 2, y)) For every man and girl there is a book he bought for her "Mixed" scope: For every girl there is a book such that every man bought it for her

The mixed scope can be generated by QR (May 1977) or by scope reversal in Categorial Grammar (Barker 2002). However, such a reading seems extremely difficult to construct; I therefore assume, contrary to the predictions of the theories mentioned above, that the mixed scope reading does not really exist and should therefore be excluded. This indeed follows from our theory: The mixed reading would result from leaving ζ scope-dependent on y but making it scope-independent from x. In the format chosen above, however, making 2 scope independent from χ implies scope independence from y because y follows χ in g and is therefore cut away by cutting away x. This built-in feature of the theory I dub the Linear Scope Restriction. The LSR blocks "mixed" readings. The mixed reading could nonetheless be expressed by modifying the mechanism of "cutting away" in the obvious way. However, such a move would be questionable for independent reasons: A more liberal mechanism not obeying the LSR would also allow for so-called branching constructions. Branching, however, seems to me a logician's invention; following others, most recently Szymanik (2009), I claim that branching is non-existent in the grammar of natural language.

6.4 Reinhart's Problem As noted by Reinhart (1997) the specific reading of (34) (34)

If we invite a (certain) professor, John will be upset.

cannot be expressed by simply giving wide scope to the existential quantifier: (35)

#(3:r)((invite(we, x0) Λ prof(x)) —> Jwbu)

Wide Scope in situ

249

These truth conditions are too weak: as soon as a non-professor (Reinhart's Donald Duck) is invited, the sentence comes out true. In our own analysis, the specific reading could only be expressed by (36): (36)

((3x\wo)(Prof :: i n v i t e l o , we, a:)) ->• Jwbu(w 0 ))

It seems that we get the wrong result in case we interpret D as the default. On the other hand, S would now have to choose an individual for χ and the empty assignment. The quesion then arises whether this should be admissible at all. If so, one might still impose the special condition that in this case the presuppositionally loaded interpretation (with respect to the speaker) is the only available reading. If not, the analysis of the conditional cannot be based on material implication. Material implication should indeed be replaced by a modal analysis. But in that case, scope independence is no problem, as the usual truth conditions make us evaluate the antecedent with respect to the closest possible worlds that render the antecendent true·, so there is no way to get the implication true by looking at a false antecendent. Unfortunately, however, Reinhart's problem rears its ugly head again in other downward entailing contexts: (37)

Every woman who loves a (certain) professor will be unhappy.

Again, a specific opaque reading yields the wrong truth conditions. We therefore need a revision of the above truth conditions, to the effect that every specific reading is in a sense presuppositional. This amounts to saying that formally there is no in situ interpretation of the restriction.

6.5 Specific Readings are Presuppositional We must conclude, therefore, that the specific reading is always one that does not allow for a simple in situ interpretation of the restriction. What does this imply for the bona fide cases of opaque readings? Consider: (38)

John doubts that a witch is after him.

Let us assume the reading is specific, and the restriction is not interpreted by the speaker. Then it is John who might have doubts concerning the witchhood of the person he thinks is after him. But now consider: (39)

John doubts that every man who met a (certain) witch will be unhappy.

This approximately translates as: (40)

John doubts Au;(V:r)[(A.z)(man(u>, z)A(3y\x) [witch :: met (tu, 2, y)]) :: unhappy(w, χ)]

The predicate witch in (40) seems to allow for the following disambiguations: (41)

a.

John believes that there is a witch and he doubts that every man who met her will be unhappy.

250

Wolfgang Sternefeld b.

John believes that there is a certain entity of which he doubts that it is a witch and he doubts that every man who met her will be unhappy.

(41-b) seems a little bit strange, but doubts concerning the witchcraft of a certain person are of course possible when considering yet another reading of (39): (42)

John believes that there is a certain entity of which he doubts that it is a witch but he does not doubt that every man who met her will be unhappy.

All these readings have in common that the restriction is interpreted in a way already proposed in (28) where the individual chosen by S is a witch either in John's belief worlds as in (41-a) or in the worlds compatible with what he doubts, as in (41-b) and (42). That is, the individual chosen by S must in fact be one that is a witch in any of the Q ' s that disambiguate the sentence. In other words, the reading called transparent in (28-b) is not necessarily "transparent" (I apologize for this terminological hardship), since the attitude verb we consider might be the one that immediately embeds the indefinite. In case of upward entailing contexts, the value loading is almost equivalent to the opaque reading, but here it is not: in the peculiar kind of "reading" (which is generated as one of the combinatory options the system allows for), it is presupposed that John doubts that a certain individual is a witch, and in (41-b) it is asserted that John doubts that every man who met her will be unhappy. The more natural reading (42) should also be accounted for, but here the problem is orthogonal to the one at hand. In order to grasp the difference between this reading and the ones discussed hitherto, compare (42) with the reading of (43) with heavy stress on him: (43)

John doubts that a witch is after HIM.

The preferred reading of (43) is one in which John does not doubt that a witch is after someone else, he only doubts that he himself is chased. The additional problem is that attitudes are focus sensitive—a topic beyond the scope of the present article. In general, then, we propose that the specific reading is always value loaded. Strictly speaking, there is no normal (opaque in situ) interpretation of the restriction, as already evidenced by Reinhart's Donald Duck example. Formally, this observation leads to the following update: (44)

Revision of (28): Clause (28-a) must be dropped from the definition of ( 3 x \ q ) in (28).

Summarizing so far, I think that specific readings always imply some pragmatically determined variation for the interpretation of the restriction; this can depend either on the speaker's referential beliefs or the subject's beliefs or their attitudes under consideration. As the same variability for interpreting the restriction was observed for the in situ interpretations of quantifiers, the general pattern is that the interpretations of the quantifiers and the restrictions are in fact logically independent, thus corroborating the thesis that the Theory of Generalized Quantifiers is on the wrong track.

Wide Scope in situ

251

7. Further Applications Note that the format of tripartite quantification in (1) is asymmetric: The quantifier does not relate two properties, but rather binds a variable in an open proposition. This asymmetry is not accidental; it reflects my general conviction that quantifiers in NL correspond to real quantifiers in logic, and that the scope of a quantifier is an open proposition. As discussed in extenso in Sternefeld (2006) Vol. 2,1 assume that NL-predicates correspond to open propositions rather than to complicated "Curried" functions. Such predicates contain "pointers" (the free variables) that link to a position in the syntactic tree where the respective variable must be bound. The pointers correspond to occurrences of theta roles; they form theta grids that are projected and "discharged" in the syntactic tree. Moreover, the system can be implemented compositionally by making variable assignments (or pointers) part of the ontology, as discussed in Sternefeld (2001). As a result, binding does not rely on QR, and there is no reason to resort to otherwise unmotivated syntactic operations. This works because NL quantifiers are real quantifiers, rather than relations between sets. That the latter is false has just been demonstrated: there's no way in such a theory to get wide scope for the quantifier without also getting wide scope for the restriction. Another property of the system is that S-functions, although they could be built up incrementally (as is the case with value assignments g), are still global in that the information encoded in S is available at any stage of the computation. This makes it possible to evaluate pronouns quasi dynamically, as in (45): (45)

Everyx man loves ay\x woman. Shey is pretty.

When encountering shey, S provides a unique value for y so that the pronoun can be interpreted properly. If a woman is interpreted scope dependently, no unique value exists, and anaphora is ungrammatical. This is of course as one would expect. A more interesting application would be so called Hob-Nob-anaphora discussed extensively in Haas-Spohn (1995): (46)

Hob believes thatw a witch is after him and Nob suspects thatw she might kill him.

In our previous definitions, we used the global function S only for interpreting scope independent construals. If we extend the use of S also to certain cases of "ordinary" (scopedependent) quantification, we might overrule the above restrictions on the use of new variables and choose the same variable w for both intensional contexts believe(u,'o, h, Λ w . . . ) and suspecter), n,Xw ...). This implies that S interprets she across intensional contexts so that the individuals chosen in the believe contexts are the same as that in the suspect context. Whether this line of thought is feasible is left to further research; one of the reasons I did not interpret ordinary quantification via S is that I wanted to avoid technical complications that arise with existentials having narrow scope with respect to negation. Specific readings as discussed here always scope over negation, and that's what the function S also does.

252

Wolfgang Sternefeld

8. Summary We finally summarize the motivation for designing the structure of quantification as we did by embedding the present approach into a broader program that comprises the following thesis: • • •

• • • •

There is no QR for quantifiers in object position (quantifiers bind variables in open propositions). There is no QR for binding pronouns (quantifiers are quantifiers rather than second order relations). Since all verbs have the same logical type, namely that of open propositions, operators like verb negation, which tends to occur as close to the verb as possible in German, can retain their simple proposition-negating semantics (without enforcing otherwise unmotivated "pseudo-scrambling" operations). There is no need for choice functions à la Reinhart (1997). There is no movement of the restriction or of the quantifier, everything is interpreted in situ. There are no special locality restrictions that block mixed readings. There are no further syntactic constraints for the use of world variables, as have been proposed in Perçus (2000).

Hence, the picture developed here differs in many details from that described in Romoli and Sudo (2008), and it was precisely this difference that prompted the present paper. Apart from that, however, the general program outlined above has nothing to say about an in situ interpretation of the surface form 3V when the universal quantifier seemingly gains wide scope over the existential quantifier. I claim that, at least in German, such a reading is available only in certain contexts that allow the indefinite to be "located" in some way, preferably in time and space as is normally the case with inversed linking. For me, such an inverted reading is impossible in simple sentences like a man loves every woman (ie. its German equivalent). It might come into reach only when V can quantify (post festum) over contexts like place and time. An in situ treatment of such constructions is left to future research.

Wide Scope in situ

253

References Barker, Chris (2002): 'Continuations and the Nature of Quantification', Natural Language Semantics 10, 211-242. Bäuerle, Rainer (1983): Pragmatisch-Semantische Aspekte der NP-Interpretation. In: M. Faust, ed., Allgemeine Sprachwissenschaft, Sprachtypologie und Textlinguistik: Festschriftför Peter Hartmann. Narr, Tübingen, pp. 121-131. Bennett, Michael (1979): Questions in Montague Grammar, mimeo. Indiana University Linguistics Club. Cresti, Diana (1995): 'Extraction and Reconstruction', Natural Language Semantics 3, 79-122. Fodor, Janet (1979): The Linguistic Description of Opaque Contexts. PhD thesis, MIT, Cambridge. Gallin, D. (1975): Intensional and Higher-Order Modal Logic. North-Holland, Amsterdam. Haas-Spohn, Ulrike (1995): Versteckte Indexikalität und subjektive Bedeutung. Akademie Verlag, Berlin. Heim, Irene (1982): The Semantics of Definite and Indefinite Noun Phrases. PhD thesis, University of Massachusetts, Amherst. Published by Garland, New York. Heim, Irene and Angelika Kratzer (1998): Semantics in Generative Grammar. Blackwell, Maiden, Mass. Hintikka, Jaakko and Esa Saarinen (1979): Game Theoretical Semantics. Reidel, Dordrecht. Ioup, Georgette (1977): 'Specificity and the Interpretation of Quantifiers', Linguistics and Philosopy 1, 233-245. May, Robert (1977): The Grammar of Quantification. PhD thesis, MIT, Cambridge, Mass. Montague, Richard (1973): The Proper Treatment of Quantification in Ordinary English. In: J. Hintikka and P. Suppes, eds, Approaches to Natural Language: Proceedings of the 1970 Stanford Workshop on Grammar and Semantics. Reidel, Dordrecht, pp. 221— 242. Perçus, Orin (2000): 'Constraints on Some Other Variables in Syntax', Natural Language Semantics 8, 173-229. Reinhart, Tanya (1997): 'Quantifier Scope: How Labor is Divided Between QR and Choice Functions', Linguistics & Philosophy 20, 335-397. Romoli, Jacopo and Yasutada Sudo (2008): De Re/De Dicto Ambiguity and Presupposition Projection. Sinn und Bedeutung, September 30, 2008. Saarinen, Esa (1979): Essays in Honour of Jaakko Hintikka. Reidel, Dordrecht. Schwager, Magdalena (t.a.): Speaking of Qualities. In: Proceedings from Semantics and Linguistic Theory (SALT) 19. CLC Publications, Ithaca, NY. Sternefeld, Wolfgang (2001): Semantic vs. Syntactic Reconstruction. In: C. Rohrer, A. Roßdeutscher and H. Kamp, eds, Linguistic Form and its Computation. CSLI Publications, Stanford, CA, pp. 145-182. Sternefeld, Wolfgang (2006): Syntax. Eine morphologisch motivierte generative Beschreibung des Deutschen. Band 2. Stauffenburg Verlag, Tubingen.

254

Wolfgang

Sternefeld

Szabó, Zoltán (t.a.): Specific, Yet Opaque. In: Proceedings of The 17th Amsterdam Colloqium. Springer, Berlin. (Lecture Notes in Computer Science). Szymanik, Jakub (2009): Quantifiers in Time and Space. ILLC Publications, Amsterdam. Tarski, Alfred and Robert Lawson Vaught (1957): 'Arithmetical Extensions of Relational Systems', Compositio Mathematica 12, 81-102.

What it Takes to be Missing Thomas Ede Zimmermann

At first glance, the predicate [be] missing appears to be an ordinary verb of absence to be analyzed along the lines of Quine (1956) or more recent implementations and alternatives. 1 On closer inspection, though, this particular specimen turns out to give rise to some surprising semantic complexity - which is what the present note is about. As always with referentially opaque verbs, one may distinguish between specific (or 'objectual') and unspecific (or 'notional') readings, the latter being typical of indefinites whereas proper names tend to be restricted to the former. Thus, whereas (1) does not seem to be ambiguous in the relevant sense, (2) may but need not be used to report about a specific screw: (1)

Peter is missing.

(2)

A screw is missing.

Although specific, (1) is heavily underspecified: just what Peter is missing from - an event, a group, an institution, a representation, . . . - is left open; it is something the context would have to make sufficiently clear lest the utterance of ( 1 ) should be felicitous. More likely than not, ( 1 ) is elliptic, and an appropriate prepositional phrase of the form from X is missing - from ( 1 ), that is - whereupon a general resolution procedure sees to it that the utterance context provides a suitable referent for X ? Analogous remarks apply to (2), and indeed to any underspecified uses of [be] missing, including ones in which the predicate occurs in an embedded position. In the following, I will therefore assume that missing expresses a binary relation (in intension) the second argument of which will be represented by a variable (of type e) 3 : (3)

M¿(p, ι )

(3) is but a first approximation to the semantic analysis of (1) in that the predicate is represented by a constant M (of type s(e(et))) the interpretation of which still needs 1

Cf. Montague (1969, 1970, 1973); Zimmermann (1993); Moltmann (1997); Forbes (2006, 2008).

2

Cf. Saeb0 (1996) for the kind of mechanism involved.

3

Until further notice, all formulae are meant to be terms of Gallin's (1975, §8) two-sorted type theory; i, j , k etc. stand for variables of type s, referring to points in Logical Space, a.k.a. indices, with i denoting the (default) point of evaluation. In case the argument is a variable, functional application is frequently expressed by subscripts instead of parentheses.

256

Thomas Ede Zimmermann

to be specified. Given that χ is the object Peter is missing from according to (1), (3) should imply some kind of incompleteness of x. Which kind depends on the nature of χ and, once more, contextual factors: completeness may relate to features as diverse as proper functioning, accuracy, or even memorability - if χ is, respectively, an appliance, a ceremony, or a party. In general, completeness amounts to having a property alluded to by the speaker and understood by the audience. In the following, these contextual factors will not play a role, so that completeness may be indicated by a mere constant C (of type s(et)). Given this caveat, (3) ought to imply: (4)

-nCi(®)

To be sure, (3) ought to say more than (4). According to (1), whatever Peter may be said to be missing from ( = x) is not only incomplete - it is incomplete without Peter. Hence, in addition to (4), (3) would also have to imply: (5)

-li(p,a:)

- where the constant I (of the same type as M) expresses the relation of being part of, taking part in, being in, being with - or whatever may be appropriate in the context at hand. Again, the exact nature of this relation will not be my concern here. (3) is clearly stronger than the conjunction of (4) and (5). After all, Peter cannot be said to be missing from the German government χ just because it happens to have a vacancy (4) and Peter does not happen to be a minister (5). On top of (4) and (5), Peter's missing from χ means that z's completeness requires Peter's participation, i.e. that χ can only be complete with Peter. This modal fact may be captured by some standard representation of counterfactual reasoning 4 . For current purposes, a constant t> of type e(s(st)) as characterized by postulate (6), comes in handy: (6)

j >x i = [fi-Qr) fe (Vy)[I¿(y,x) r(3k

-» lj(y,x)\

x -accessible from i (for given x) just in case (a) χ is complete at j without (b) any parts or participants present in i lacking (i.e., being substituted), and (c) otherwise only minimally differs from i. One may thus think of such j as specifying a completion of χ relative to i. Given this notation, (3) may be expanded as:5 (7)

b I ¿ ( p , x ) & ( V j >x ΐ)Ι,·(ρ,®)]

4

Cf. Stalnaker (1968); Lewis (1973), etc. The pertinent relation of comparative similarity will be denoted by a constant < of type s ( s ( s t ) ) ) ; k [υ Λ I ( p , χ ) ] ] ) ) ( [ > Ι ) ( Λ - ι Ι ( ρ , χ ) ) , where ρ and υ are variables of type s i and t, respectively, and constants equivocate, using Gallin's (1975, p. 61) *-operator.

258

Thomas Ede Zimmermann

If we take (12) to be the representation of the truth conditions of (1), it will still give rise to the above specific reading of (2), because the existential generalization (13) of (12) is equivalent to (8): (13)

(3y)[Si{y) k -íi(y,x)

k (Vj >x i)

lj[y,x)\

However, (13) allows for a straightforward swap of operator scopes, resulting in a constellation that is not equivalent to any of (9) - (11), viz.: (14)

(Vj t>x i) (3y) [Sj (y) k - I , (y, χ) k I, (y, x)]

According to (14), any completion of χ relative to i must contain a screw that is not already present at i. In view of the definition (6) of the accessibility relation underlying the modal, this seems to correctly capture the truth conditions of (2) on its unspecific reading. According to the latter, the completeness of χ requires there to be at least one screw that is not already present in i, i.e. a screw on top of the ones (if any) χ has in i. In fact, in i such a screw does not even have to exist, let alone be a screw, as might be the case if its particular make-up, size or form does not match any existing screws, or if the total number of screws in i does not suffice to make χ complete. The above observations suggest that [be] missing is referentially opaque and expresses a modal operator to be combined with a nominal quantifier. The lexical analysis (15) is a straightforward Montagovian implementation of this idea: (15)

[be] missing' =

o * i)

k

Ij{y,x)])

The unspecific reading (14) of (2) ensues by applying the operator in (15) to the intension of the existential quantifier expressed by the indefinite subject: (16)

(A5».(V¿ t) &j{Xy. h l i f o , ® ) &Ij(i/,a;)])) (Xj.XP.(3y)[Sj(y) k P(y)])

[ ξ (14)]

The specific reading (13) is obtained by raising the existential above the modal operator, using a general scoping mechanism (like quantifier raising):9 (17)

(XP.(ßy)[Si(y) L· P(y)]) (\y.{\