Topic Drop and Null Subjects in German 9783110446173, 9783110444131

Citation preview

Ewa Trutkowski Topic Drop and Null Subjects in German

Linguistics & Philosophy

Edited by Günther Grewendorf, Wolfram Hinzen, Hans Kamp, Helmut Weiß

Volume 6

Ewa Trutkowski

Topic Drop and Null Subjects in German

ISBN 978-3-11-044413-1 e-ISBN (PDF) 978-3-11-044617-3 e-ISBN (EPUB) 978-3-11-043724-9 ISSN 2198-2104 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliografische Information der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über http://dnb.dnb.de abrufbar. © 2016 Walter de Gruyter GmbH, Berlin/Boston Druck und Bindung: CPI books GmbH, Leck ∞ Gedruckt auf säurefreiem Papier Printed in Germany www.degruyter.com

Foreword This is a revised version of my doctoral dissertation submitted to the University of Potsdam in May 2014. My reviewers were Prof. Dr. Peter Staudacher (University of Potsdam) and Prof. Dr. Helmut Weiß (Goethe University Frankfurt am Main). The date of the oral examination was 21 November 2014. First and foremost I would like to thank Peter Staudacher and Helmut Weiß for encouraging and inspiring me, personally and scientifically. Thank you very much for your generous support and our intensive discussions in which you let me participate in your immense knowledge. I am also grateful to André Meinunger and Werner Frey for valuable comments, discussions and, most of all, for their questions. Possibly, this book would not exist without the scholarship I received from the Hans Böckler Stiftung. Thanks also to Prof. Dr. Hartmut Schröder in his function as Vertrauensdozent of the HBS and to the people from the Stiftung. In addition to Werner and André I would like to thank my former colleagues from the Zentrum für Allgemeine Sprachwissenschaft (ZAS) in Berlin, Jörg Dreyer, Clemens Mayr, Kerstin Schwabe, Barbara Stiebels and many others not only for scientific exchange but also for the nice time we spent at the ZAS. Particular thanks go to Fabienne Fritzsche with respect to our collaboration on the (experimental) surveys herein and to Manfred Krifka for associating me with the ZAS. A lot of thanks also to my colleagues from my Alma Mater, the University of Potsdam, and the Goethe University Frankfurt am Main, especially to Leah Bauke, Andreas Blümel, Jutta Boethke, Gisbert Fanselow, Günther Grewendorf, Erich Groat, Andreas Haida, Jan Köpping, Andreas Pankau and Ede Zimmermann as well as to the audiences of talks and conferences, in particular to Josef Bayer, Valéria Molnár and George Walkden. Many thanks also to my informants and everybody who participated in the questionnaires. I am grateful to the editors and Dr. Rafael Hüntelmann from Ontos De Gruyter for accepting this dissertation in the “Linguistics and Philosophy” series and to Maik Bierwirth and Olena Gainulina, both from De Gruyter, for their kind help and technical assistance with the publication. Many thanks also to my English language copy editors, most notably to Manfred “Schimmel” Momberger, but also to Gillian Fellows. I would also like to thank my family and friends for supporting me during the whole time I worked on my doctorate, above all my beloved and loving mother, who was always there for me and who gave everything to fulfil my hopes and wishes, my brother Dominik, my aunts Marysia, Ela and Ewa, and

VI | Foreword

my father for providing me with the technical equipment. A lot of thanks also go to Volker, Yvonne, Lidija, Clivia, Gabi-Maria, MK, Elisabeth and Hartmut together with my Berliner Bank Tennisdamen and all I have not mentioned here. All errors within this book are, of course, mine. Frankfurt am Main, 31 March 2016 E.T.

Contents 1 1.1 1.2 1.2.1 1.2.2 1.3

Overview | 1 What this Book is About | 1 The Syntax and Semantics of Topic Drop | 1 The Syntax of Topic Drop | 2 The Interpretation of Topic Dropped Elements | 6 Null Subjects in German | 9

2

Topic Drop in German: (The Lack of) Identity between Antecedent and Gap | 15 Introduction | 15 Topic Drop is Restricted to the Prefield | 16 General Conditions on Antecedents of Topic Dropped Elements | 19 Verbatim and Non-verbatim Topic Drop: Two Kinds of Topic Drop in German | 25 A First Approach to Verbatim Topic Drop (VTD) and Non-verbatim Topic Drop (NVTD) | 25 VTD is not ‘Stripping in Disguise’ | 33 An Empirical Basis for the VTD/NVTD Distinction | 37 Structural Well-Formedness Conditions on VTD and NVTD | 52 Topic Drop and the C- and S-selectional Properties of Predicates | 53 On the Correlation between Case, Theta Roles and Verbal Semantics | 56 A Comparison of Synonymous Verbs with Different Case Assignment Properties | 62 Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs | 79 Empirical Findings, Hypotheses and Theoretical Assumptions | 79 Investigation of Events Denoted by Nom/Dat Verbs (in Contrast to Events Denoted by Nom/Acc Verbs) | 83 Investigation of Events Denoted by Nom/Gen Verbs and Nom/AccREFL/Gen Verbs | 88 Investigation of (Events Denoted by) Three-Place Verbs | 94 Topic Drop and the Representation of +/-Complex Events | 100

2.1 2.1.1 2.1.2 2.2 2.2.1 2.2.2 2.3 2.4 2.4.1 2.4.2 2.4.3 2.5 2.5.1 2.5.2 2.5.3 2.5.4 2.6

VIII | Contents

2.6.1 2.6.2 2.6.3 2.6.4 2.7 3 3.1 3.2 3.2.1 3.2.2 3.2.3 3.2.4 3.2.5 3.3 3.4 3.4.1 3.4.2 3.4.3 4 4.1 4.2 4.2.1 4.3 4.3.1 4.3.2

The Relation between C-selection and S-selection in the Context of +/-Oblique Case Assignment | 100 The Syntax of Oblique Case Assignment – Some Introductory Notes | 102 +/-Oblique Case Assignment (and Topic Drop) | 105 The Syntactic Representation of +/-Oblique Case Assignment | 112 Appendix: Topic Drop of Expletive Elements | 120 Topic Drop in German: The Interpretation of the Gap | 122 Introduction: (Im-)possible Interpretations of VTD and NVTD Gaps | 122 Two Different Kinds of Binding in Topic Drop Dialogues | 128 The Target Binder is a Referential Expression | 129 The Target Binder is a Non-referential Expression | 138 Interim Summary: Referential vs. Non-referential Binding | 141 The Interpretation of the Reflexive Pronoun sich | 142 The Interpretation of 1st and 2nd Person (Indexical) Pronouns | 145 Survey on Bound Readings of the Gap: A (Better) Empirical Basis | 153 The Representation of the Gap | 165 Binding in Topic Drop Dialogues: Empirical Findings | 165 Binding in Topic Drop Dialogues: Some Theoretical Considerations | 170 The Identification and Licensing of pro | 175 Null Subjects at the Syntax-Pragmatics Interface | 185 Introduction: (Antecedent-Independent) Null Subject Omission in German | 185 Well-Formedness Conditions on Antecedent-Independent Null Subjects | 195 Empirical Investigations of German and Other (Non-)Null Subject Languages | 198 More on the Identification/Licensing of Null Subjects (in German) | 218 Independent Evidence for the Assumed Identification/Licensing Conditions on Null Subjects | 218 Positional Licensing of ‘Out of the Blue Drop’ (OBD) and the Empty Category Associated with OBD | 222

Contents | IX

4.3.3 4.3.4 4.3.5

OBD in Embedded V2 Clauses | 223 Non-Prefield Null Subjects in German | 228 German – a Partial Null Subject Language (?) | 234

References | 237 Index | 245

1 Overview 1.1 What this Book is About This book deals with two different kinds of ellipsis, (i) topic drop and (ii) null subjects. In both cases an argument is omitted (“dropped”) in the sentenceinitial position of a declarative verb-second (V2) clause in German. As for the description of these constructions I refer to standard notions known to those familiar with the German topological field model (Topologisches Feldermodell) and a standard version of Generative Grammar (Theory). Besides this overview (constituting chapter 1) the book contains three more chapters. Chapter 2 and chapter 3 deal with topic drop, i.e. with antecedent-dependent subject/object omissions as exemplified by (1): (1)

A: B:

Wo ist dein Ring? Where is your-sg ring-NOM? _ Hab ich verkauft. [ACC] (= My ring) have I sold (‘I have sold it.’)

Whereas chapter 2 focuses on a syntactic issue, namely the case features of the gap and its antecedent, chapter 3 is concerned with the interpretation of topic dropped elements. Chapter 4 is about subject omissions which are independent of the presence of an antecedent / which can take place ‘out of the blue’, cf. (2): (2) ∅ Bin dann mal weg! I am then PRT1 away (‘I’m off then.’) In the following I sum up the main findings.

1.2 The Syntax and Semantics of Topic Drop As mentioned above, in chapter 2 and chapter 3 I discuss German topic drop. Thus, the greatest part of this book deals with the topic drop construction.

|| 1 PRT means ‘particle’.

2 | Overview

Chapter 2 is about the syntax and semantics of (structural vs. oblique) case. Following Blume (1998, 2000) I elaborate an event-related syntax for obliquely cased topic dropped arguments. In sum, I argue that oblique case is the overt marker of an implicit subevent denoted by the verb. Before arriving at this hypothesis, I present a small survey on the acceptability of structurally and obliquely cased gaps and examine whether seemingly synonymous (two- and three-place) predicates selecting structural vs. oblique case for their (in)direct argument differ with respect to their (fine-grained) semantics. Further, I show that dative gaps behave in-between with respect to topic drop, i.e. dropping dative arguments is not as unconstrained as the drop of structurally cased (nominative, accusative) arguments, but subject to much looser conditions than the drop of genitive cased arguments and prepositional phrases (PPs). Thus, in the light of chapter 2 topic drop is elaborated as “another instance that separates nominative and accusative on the one hand and dative on the other” (Haider 2010:269). In chapter 3 I focus on the interpretation of topic dropped elements. I show that two kinds of binding have to be distinguished: ‘referential binding’ vs. ‘non-referential binding’ (that is, ‘bound by a referential term’ vs. ‘bound by a non-referential term’). Whereas a non-referential binding relation can be established despite phi-feature mismatches between the context binder and the nonreferential target binder, this does not hold with respect to the former. I.e., referential binding of the bindee part of the gap is only possible when context binder and referential target binder bear the same phi-feature specifications (or when context bindee (which is the antecedent of the bindee part of the gap) and target binder are PF-compatible2). As will be shown in detail, the differentiation between non-referential and referential binding allows us to distinguish relations which cannot be reduced to coreference (= “real” binding relations) and relations that are ambiguous between coreference and binding on a morphosyntactic basis.

1.2.1 The Syntax of Topic Drop Besides the fact that subjects and objects can be equally well topic dropped, cf. (3)-(4), other, non-nominal, elements can be dropped as well (see Fries 1988), cf. (5). || 2 This special case arises due to the occurrence of syncretisms within the paradigm of possessive pronouns.

The Syntax and Semantics of Topic Drop | 3

(3) A: B:

(4) A: B:

(5) A: B:

Heute hab ich den Hans gesehen. Today have I the Hans-ACC seen Echt? _ Ist mir ja schon seit Ewigkeiten nicht mehr begegnet. Really? [NOM] is me PRT already for ages not came-across Heute ist mir der Hans begegnet. Today is me the Hans-NOM came-across Echt? _ Hab ich ja schon seit Ewigkeiten nicht mehr gesehen. Really? [ACC] have I PRT already for ages not more seen Was machst du heute? What make you-sg today? _ Mach ich mal gar nichts. [ADVERB] make I PRT at-all nothing

Here and in the following data examples the antecedent of the gap is marked bold. As case features of antecedent and gap will play a major role in the following, I indicate them by SMALL CAPS. Furthermore, the gap is represented by an underline ‘_’ or by the underlying (crossed-out) PF-form (both variants stand for an (internally structured) pro-category, see chapter 3.4). Although expressions of different grammatical categories can be dropped in the German prefield (= the position in front of the finite verb in V2 clauses), cf. Fries (1988), I will concentrate on dropped arguments. In particular, I argue for a decomposition of topic drop into non-verbatim topic drop (NVTD) and verbatim topic drop (VTD). I show that NVTD gaps are limited to structural case (nominative, accusative) and that they allow case/theta role mismatches between antecedent and gap, i.e., they can be embedded under different predicates, cf. (3)/(4) and (6a). VTD gaps, on the other hand, are possible with all cases as long as predicates in context and target are semantically identical, cf. (6b)/(6c). Thus, case features of antecedent and gap can depart from each other in VTD environments, cf. (6c), whenever antecedent and gap bear the same finely granulated (micro) theta role (to be discussed in connection with examples as (7)-(9)). Of course, the structurally case marked gap in (6b) is also analysable as an NVTD instance: (6) A: a.

B:

Der Hans hat die Oma gestern beim Einkaufen getroffen. The Hans has the grandma-ACC yesterday at shopping met _ Hat der Otto heute zum Flughafen gefahren. NVTD [ACC] has the Otto today to-the airport driven

4 | Overview

b.

B:

c.

B:

_ Hat der Otto heute beim Bäcker getroffen. VTD (and NVTD) [ACC] has the Otto today at-the baker(y) met _ Ist der Otto heute beim Bäcker begegnet. VTD [DAT] has the Otto today at-the baker(y) come-across

The need to define VTD in terms of an identical predicate constraint rather than in terms of identical case features becomes more obvious when we look at nominal phrases (NPs) which are obliquely case marked. Whereas micro theta role identity does not play a role when the gap is structurally case marked, cf. (3)/(4), example (7) shows that identity of micro theta roles seems to be a wellformedness condition on obliquely cased gaps (and PPs as well). Data as (7) are not well-formed because the respective micro theta role assigned to antecedent and gap in context and target (respectively) is not identical (enough). In (8) the identical micro theta role condition is (trivially) satisfied by the presence of identical predicates in context and target. (9) is well-formed, because it is an instance of NVTD (although the gap’s antecedent is a genitive marked NP, the gap itself is structurally case marked). I.e., micro theta role identity is not a necessary condition on the well-formedness of nominative or accusative marked gaps: (7) A:

Der Hans gedenkt der Rosa Luxemberg. The Hans commemorates the R.L.-GEN B: * _ Schämt sich manch ein Politiker. [GEN] ashamed REFL some a politician

(8) A: B:

(9) A: B:

Der Hans gedenkt der Rosa Luxemberg. The Hans commemorates the R.L.-GEN _ Gedenkt der Otto auch. [GEN] commemorates the Otto as-well Der Hans gedenkt der Rosa Luxemberg. The Hans commemorates the R.L.-GEN _ Kennt der Otto gar nicht. [ACC] knows the Otto at-all not

*NVTD

VTD

NVTD

Thus, the first thing that is worth noting is the fact that there are two kinds of topic drop, NVTD and VTD. Whereas structurally cased gaps can undergo NVTD as well as VTD, obliquely cased gaps cannot undergo NVTD but are only wellformed in a ‘verbatim environment’, where context and target predicate are

The Syntax and Semantics of Topic Drop | 5

semantically identical. The only way to drop an obliquely cased argument is by making sure that the micro theta role that an obliquely cased gap receives from the target predicate is the same which its antecedent received from the context predicate. In chapter 2 it is shown that “the more obliquely case marked a gap is” (on the Keenan/Comrie hierarchy), the stricter are the well-formedness conditions on topic dropped elements – in particular: semantic identity between context and target predicate becomes a necessary condition. Due to the fact that topic drop is sensitive to the distinction between structural and oblique case (cf. the literature from Ross 1982 to Haider 2010 and the results of a little survey that I have conducted on 60 native speakers), the question which factors keep structural and oblique cases apart will be discussed in great detail. In order to answer this question I refer to the +/-complexity of events that are denoted by +/-oblique case assigning verbs (in the spirit of Engelberg 1995, Lenz 1997, Blume 1998, 2000) and maintain that an obliquely case marked element indicates the presence of a complex event, e.g. in the case of helfen (to help) the dative object is – next to its patient/theme properties within the main event – a secondary agent in a (preceding) subevent. I.e., I assume that for actants of verbs the involvement within complex events implies that they have to bear certain proto-properties (in the sense of Dowty 1991) and that oblique case assignment constitutes the overt marker of these (additional) proto-properties. In particular, I claim that oblique case marking is the spell-out of a certain semantic content. Due to a general constraint that meaningful content cannot be absorbed/has to be spelled out at PF, these specific properties have to be spelled out as well – which, however, is impossible under topic drop. Therefore, the “meaning” of an obliquely cased dropped NP (that is ‘marking the presence of a complex event’) has to be ascertained/reconstructed via semantic identity with its antecedent/(the micro theta role assigned by) the context predicate. I.e., obliquely case marked gaps evade the (in principle necessary) obligation to be spelled out by establishing a semantic identity relation with this/these element(s) which can ensure the reconstruction of the features that the gap is lacking at PF (Phonetic/Phonological Form). This hypothesis is confirmed by independent evidence (relative clause data from German dialects, the cessation of micro theta role identity when the additional event is spelled out).

6 | Overview

1.2.2 The Interpretation of Topic Dropped Elements Chapter 3 is on the interpretation of topic drop gaps. There I show that the bindee part of a gap resulting from topic drop can be bound under predicate identity/synonymy (in the sense of semantic equivalence) in context and target, cf. (10). Gaps in NVTD environments do not display the bound (sloppy) reading, instead the gap can only be interpreted strictly, cf. (11): (10) A: B:

(11) A: B:

Der Hansi hat gestern seineni/k Prof getroffen. The Hans has yesterday his prof-ACC met _i/k/m Hat der Ottom heute auch getroffen. [ACC] has the Otto today also met Der Hansi hat seinemi/k Prof gestern beim Umzug geholfen. The Hans has his prof-DAT yesterday at moving-house helped _i/k/*m Hat der Ottom heute im Supermarkt getroffen. [ACC] has the Otto today in-the supermarket met

Thus, the occurrence of different readings supports splitting up topic drop into VTD and NVTD also from a semantic point of view. As mentioned above, I argue that one has to differentiate whether the binder is a referential NP or not – depending on whether this condition is met or not, two different instances of binding can be postulated in the context of topic drop. These two instances of binding differ with respect to the +/-necessary phifeature compatibility of context bindee (which is the antecedent of the bindee part of the gap) and target binder. When the target binder is a referential NP, (PF-related) phi-feature compatibility is necessary, (12a)/(12b), otherwise the gap remains unbound, cf. (12c). However, when the target binder is a nonreferential NP, the phi-features of context bindee and target binder can depart from each other, cf. (13). (12) A: a.

B:

b.

B:

c.

B:

Die Elterni mögen ihreni/k Hund. The parents like their dog-ACC Ihreni/k Hund / Ihrenm Hund mögen die Tantenm auch. Ref. binding [ACC] like the aunts as-well Ihreni/k Hund / “Ihren”m Hund mag die Mariam auch. Ref. binding [ACC] likes the Maria as-well Ihreni/k Hund / Seinen*m Hund mag der Peterm auch. No binding [ACC] likes the Peter as-well

The Syntax and Semantics of Topic Drop | 7

(13) A: B:

Hansi mag seineni/k Vater. Hans likes his father-ACC Seineni/k Vater / Ihrenm Vater mögen fast allem Jungs. Non-ref. binding [ACC] like nearly all boys

I.e., in case of referential binding the target binder must either have the same phi-feature specifications as the context binder, or phi-feature specifications that are PF-compatible with the phi-feature specifications of the context bindee. In (12a) the context binder (die Eltern), the context bindee (ihren) and the target binder (die Tanten) all have the same feature specifications, namely 3rd person plural (gender forms being uniform in the plural). In (12b) the feature specifications of context binder and target binder differ from each other, however the surface form of the context bindee (ihren), which is bound by the context binder (die Eltern), is PF-compatible with the phi-feature specifications of the new target binder (Maria). When the feature specifications of context binder and target binder are not the same and the feature specifications of the context bindee are not PF-compatible with those of the target binder, referential binding is not possible, and hence the gap lacks a sloppy reading, cf. (12c). Besides other requirements (as e.g. the above mentioned PF/phi-feature identity obligation in referential binding contexts), the crucial condition for bound readings in topic drop dialogues is the presence of a verbatim environment, which ensures that context and target are semantically parallel. In addition, the gap in the target can only be interpreted sloppily when the relation between the dependent and the independent element in the context is a “real” (e.g. quantifier) binding relation or a relation that is ambiguous between coreference and binding. (14) exemplifies a case where this condition is not met: (14) A: a.

B:

b.

B:

Irenei mag seine*i/k Schwester. Irene likes his sister-ACC Seinek/*s Schwester mag der Hanss auch. [ACC] likes the Hans as-well Seinek/*r Schwester mag jederr Junge. [ACC] likes every boy

No coreference/no binding No binding

(15) summarises (further) necessary conditions for bound readings in topic drop dialogues:

8 | Overview

(15) Conditions on sloppy gap readings in topic drop dialogues (a) The antecedent of the gap must be a non-rigid designator such that its reference can vary across context and target (b) Syntactic conditions on binding must be met (cf. c-command) (c) Presence of a binding/coreference relation in the context (cf. two coindexed elements) (d) Identical/synonymous predicates in context and target (= verbatim environment) The differentiation between “real” binding (= non-referential binding) on the one hand and referential binding on the other hand is further supported by a small survey conducted on 60 native speakers. In chapter 3 I also discuss the interpretation of topic dropped 1st/2nd person objects. I show that in fact there is nothing wrong with topic drop of indexical pronouns (contra Ross 1982, Cardinaletti 1990, Rizzi 1994, Thrift 2003, Steinbach 2007). The (only) problem is the shifting operation that speakers have to carry out in order to get the possible/intended interpretation – which is often difficult because of the indexical character of 1st/2nd person pronouns (moreover, this difficulty combines with the fact that in German 1st/2nd person object pronouns serve to express both a reflexive and a non-reflexive meaning). Generally, a dropped indexical object pronoun can have two different interpretations: A (with respect to the reference of antecedent and gap) non-shifted (strict) interpretation which involves a different form of antecedent and gap in context and target, cf. (16a), and a (with respect to the reference of antecedent and gap) shifted (sloppy) interpretation under an identical form in context and target, cf. (16b): (16) A: a.

B:

b.

B:

Wir mögen uns. We like us/ourselves-ACC Euch mögen wir auch. [You-pl-ACC] like we as-well Uns mögen wir auch. [Ourselves-ACC] like we as-well

Thus, dropped indexicals (which are directly referential elements) can be bound by an element that is an indexical pronoun as well, under phi-feature identity between context binder and target binder (as known from referential binding, see above). By adapting Kaplan’s (1989a,b) view on indexicals, the strict reading, (16a), is referred to by character conversion under content identity, and the

Null Subjects in German | 9

sloppy one, (16b), by content conversion under character identity. A third option, namely content conversion under character conversion is only available when the target binder is a non-referential element and binds a (possessive) indexical in determiner position (at LF), cf. (17), but not when the indexical is a ‘pure indexical’, cf. (18): (17) A: B:

(18) A: B:

Wiri mögen unserei Oma. We like our grandma-ACC Eurei Oma / Ihrer Oma mögen fast aller (Menschen). [Your-pl grandma-ACC/Their grandma-ACC] like almost all (humans) Wiri mögen unsi. We like us/ourselves-ACC Euchi / Sich*r mögen fast aller (Menschen). [You-pl-ACC/REFL-ACC (= themselves)] like almost all (humans)

Finally, I show that the differences between non-verbatim topic drop and verbatim topic drop (with/without a sloppy interpretation of the gap) can be captured by assuming different extensions of the Parallelism Domain as argued for by Takahashi & Fox (2005) with respect to elliptical constructions that involve rebinding.

1.3 Null Subjects in German In chapter 4 I investigate 1st/2nd (vs. 3rd) person null subjects in German. I focus on referential (thematic) null subjects that can be dropped independently of the presence of a discourse antecedent and contrast the occurrence of null subjects in German with the occurrence of null subjects in German dialects as well as in other languages / language types. I show that in contrast to 3rd person subject gaps, 1st and 2nd person null subjects occurring in the German prefield, i.e. in the Spec-CP position of finite V2 clauses, are grammatical out of the blue, cf. (19): (19) a. b.

Ø Komme/Kommst/*Kommt leider immer zu spät. [I/You-sg/He, she, it] come(s) unfortunately always too late Ø Kommen/Kommt/*Kommen leider immer zu spät. [We/You-pl/They] come unfortunately always too late

10 | Overview

No contextual antecedent, whether linguistic or non-linguistic, is needed to identify the omitted 1st/2nd person referent. By using several diagnostics for that claim, I show (following Trutkowski 2011) that this out of the blue-drop (OBD) of 1st/2nd person subjects is neither topic drop nor diary drop, but an instance of inflection-dependent subject omission. I claim that in German 1st/2nd and 3rd person null subjects are syntactically and pragmatically two different phenomena. In particular, I argue that 3rd person subject omissions are an instance of antecedent-dependent topic drop, whereas 1st/2nd person referential null subjects are licensed and identified by discrete inflectional endings notwithstanding apparent syncretisms within the German verbal inflectional paradigms, cf. (20): (20)

Verbal inflectional paradigms (indicative present/preterite active of the verbs können (can), sagen (to say), kommen (to come), tragen (to wear), sein (to be)); syncretisms are marked bold

Praeteritopraesentia

Weak conjugation

1 sg

kann/konnte

sage/sagte

2 sg

kannst/konntest sagst/sagtest

3 sg

kann/konnte

1 pl

können/konnten sagen/sagten

2 pl

könnt/konntet

3 pl

können/konnten sagen/sagten

sagt/sagte sagt/sagtet

Strong conjugation

Strongumlauting conjugation

Suppletive conjugation

komme/kam

trage/trug

bin/war

kommst/kamst

trägst/trugst

bist/warst

kommt/kam

trägt/trug

ist/war

kommen/kamen tragen/trugen

sind/waren

kommt/kamt

seid/wart

tragt/trugt

kommen/kamen tragen/trugen

sind/waren

The reason why null subject licensing in German is independent of syncretisms between the 1st/2nd and the 3rd person is due to the fact that the licensing of 1st/2nd person null subjects takes place in a different discourse domain than the licensing of 3rd person null subjects: Whereas 1st/2nd person null subjects are licensed (and identified) out of the blue, 3rd person null subjects can only be licensed (and identified) by the presence of a discourse antecedent (i.e., they are an instance of topic drop), cf. table (21): (21)

Null subject licensing and identification in German (OBD and topic drop)

Number

Person

Discourse Domain/Licensing Mechanism

Construction Type

SINGULAR

1/2 sg

Antecedent-independent / out of the blue,

OBD

Null Subjects in German | 11

Number

Person

Discourse Domain/Licensing Mechanism

Construction Type

by non-syncretic inflection

PLURAL

3 sg

Antecedent-dependent

Topic Drop

1/2 pl

Antecedent-independent / out of the blue, by non-syncretic inflection

OBD

3 pl

Antecedent-dependent

Topic Drop

In pro-drop languages, on the other hand, not only 1st/2nd person null subjects are licensed via discrete inflections but 3rd person null subjects as well (even if the latter need a discourse antecedent for identification, see Samek-Lodovici 1996), cf. table (22) for an overview: (22)

Null subject licensing and identification in pro-drop languages (pro-drop + antecedentdependent identification of 3rd person null subjects)

Number

Person

Discourse Domain/Licensing Mechanism

Construction Type

SINGULAR

1/2 sg

Antecedent-independent / out of the blue, by non-syncretic inflection

3 sg

Non-syncretic inflection + antecedent-dependent identification

1/2 pl

Antecedent-independent / out of the blue, by non-syncretic inflection

3 pl

Non-syncretic inflection + antecedent-dependent identification

pro-drop (for all person and number combinations) . . . . . . . .

PLURAL

As a consequence of not having (3rd person) topic drop, in pro-drop languages as Italian, Spanish etc. syncretisms between the 1st/2nd and the 3rd person matter and cause (e.g.) a 1st person out of the blue dropped null subject in Spanish to be ill-formed (in contrast to German): (23) Juan y yo llegamos tarde. *Ø Tenía mucho que hacer. Juan and I came-1pl late. [I/he] had-1/3sg lot to do

12 | Overview

(24) Hans und ich kamen spät. Ø Hatte viel zu tun. Hans and I came-1pl late. [I] had-1(/3)sg a-lot to do On the basis of minimal pairs as (23)/(24) I conclude that in German (but not in e.g. pro-drop languages) we have to do with two different constructions (namely 1st/2nd person out of the blue dropped null subjects vs. 3rd person antecedentdependent topic drop) and that because of the independent co-existence of these two constructions syncretisms become neutralised and unique identification of a given referent is possible despite syncretic forms in a given verbal inflectional paradigm. To prove that a language licenses null subjects really out of the blue (that is, to exclude the presence of a possibly interfering (default) antecedent), I use the so-called ‘Coordinated Antecedents Test’ (CAT), cf. the examples (23)/(24) above. This test goes back to Cole (2009) and works as follows: When two XPs are coordinated, they constitute equally salient/non-salient antecedents for a subsequent null element. Furthermore, the presence of the two coordinated antecedents makes a default antecedent choice impossible, because antecedents that are located within a coordination are subject to some Coordinate Structure Constraint, as known from Ross (1967). As a consequence, in a CAT context none of the coordinated items can act as an antecedent for a subsequent null subject. Thus, a null subject that is licensed under the CAT, is not licensed by the presence of one of the coordinated antecedents but out of the blue, under its own steam – either by speaker/hearer features (as could be the case in Chinese) or by discrete inflectional endings at the finite verb (which will be shown to be (also) the case in German and some German dialects). If a null subject is ill-formed when embedded under the CAT, it may be the case (i) that this language does not license null subjects at all, or (ii), that the relevant (and otherwise active) licensing mechanism cannot apply (e.g. because inflectional endings are not discrete enough in a given inflectional paradigm, cf. the plural of Swabian, the Imperfect paradigm of Spanish or the Subjunctive paradigm of Italian). Note that a simple out of the blue-context does not yield the same output as the CAT. In particular, in an out of the blue-context the null subject in the Spanish sentence can be interpreted as 1st person singular, cf. (25), whereas this is not possible under the CAT, cf. (23) above. In German, the CAT, cf. (24), and an out of the blue-context, cf. (26), yield the same output: (25) Tenía mucho que hacer. [I] had-1(/3)sg lot to do

Null Subjects in German | 13

(26) Ø Hatte viel zu tun. [I] had-1(/3)sg a-lot to do I.e., we can assume that something like an out of the blue context does not really exist. Instead, to test ‘out of the blue’ null subject licensing (and identification) we must either use a non-accessible/not inferrable antecedent (as provided by the CAT) or an incompatible antecedent (i.e. one whose feature specifications depart from those of the finite verb). An example of the latter is provided by the context in (27): (27) Die Eltern waren gestern da. Ø Hatte viel zu tun.3 The parents were yesterday there. [I] had-1(/3)sg a-lot to do As to the syntactic location of OBD in German, I suggest that this kind of null subjects is confined to the prefield. I.e., they can only be licensed under a SpecHead configuration because the finite verb in C° carries the person/number specifications of the null subject (located in Spec-CP) and stands in an Agree relation with it. Other kinds of null subjects in German, namely those occurring in Wackernagel position, are discussed separately. In the (generative) literature different language types can be classified according to their property of (not) having null subjects, cf. (28), see e.g. Biberauer et al. (2010). According to such classifications I suggest that (Colloquial) German is a partial null subject language, or on the way to become one. (28) (Non-)null subject language types (i) pro-drop languages (e.g. Latin, Italian, Spanish, Greek, Polish) (ii) partial pro-drop languages (e.g. Finnish, Marathi, German) (iii) diary drop languages (e.g. English, Dutch (?)) (iv) radical pro-drop languages (e.g. Thai, Chinese) (v) “real” non-null subject languages (?) As will be shown, this classification of German is not only based on the fact that German has a prefield-version of (partial) pro-drop (that is, OBD) as well as the ability to topic drop elements of different grammatical categories in the prefield, but also a 2nd person singular (and marginally: 2nd person plural) pro-drop op-

|| 3 However, the CAT is a better choice as it captivates possible but not accessible antecedents within a coordination, whereas in contexts as given in (27) a further (pragmatically inferrable) (partial) antecedent could slide in.

14 | Overview

tion in Wackernagel position (which is independent of a Spec-Head configuration). This kind of subject omission, however, is hard to detect because in the singular the inflectional ending –st that licenses the null subject in (29a) could be regarded as a (reduced) variant of –ste, cf. würdeste. In the plural, on the other hand (where we do not find such overlapping), null subjects in Wackernagel position are very marginal, although one can find them on the internet (mainly in informal registers) quite frequently, cf. (29b): (29) a.

Was würdest Ø mir empfehlen? What would-2sg [you-sg] me recommend? b. (?) Was würdet Ø mir empfehlen? What would-2pl [you-pl] me recommend?

As will be shown at the end of chapter 4, the licensing/identification conditions for Wackernagel null subjects are much harder to fulfil than the licensing/identification conditions for OBD null subjects. In particular, for a Wackernagel null subject to be well-formed the licensing inflectional ending must be non-syncretic throughout all tenses of a given inflectional paradigm of a certain verb. As for the 2nd person singular, most verbs fulfil that requirement (despite verbs whose stem ends with an /s/4). With respect to the 2nd person plural, the relevant licensing condition can only be satisfied by a small group of verbs, namely by modal verbs and verbs belonging to the strong-umlauting conjugation as e.g. raten (to recommend), schlagen (to hit) or sehen (to see).

|| 4 Consider therefore the following minimal pair with an /s/-stem verb, hassen (to hate), and a non-/s/-stem verb, mögen (to like): (i) Wen magst / *hasst Ø am meisten? Whom like-2sg [you-sg] / hate-2sg/2pl [you-sg/pl] at most?

2 Topic Drop in German: (The Lack of) Identity between Antecedent and Gap 2.1 Introduction As is well known, topic drop (in German) refers to the antecedent-dependent omission of a constituent in the prefield of a declarative verb-second (V2) clause (cf. e.g. Ross 1982, C.T.J. Huang 1984, Fries 1988, Cardinaletti 1990, Sigurðsson 1993, Rizzi 1994, 2002, Y. Huang 2000, Ackema & Neeleman 2007, ErteschikShir 2007, Sigurðsson 2011). The examples in (1) and (2) show prototypical NP omissions (although elements with other syntactic categories (e.g. adjective phrases, VPs, etc.) can be omitted as well, see Fries 1988, I will mainly concentrate on NPs (and to a lesser extent: PPs)). (1)

SPIEGEL:

Herr Bundespräsident, besitzen Sie Zertifikate? Mister Federal President, own you-formal certificates-ACC? Horst Köhler: Nein, _ habe ich nie gezielt gekauft.1 No, [ACC] have I never purposely bought

(2) A: B:

Kennst du den Hans? Know you-sg the Hans-ACC? Na klar, _ is’ mein Nachbar. Of course, [NOM] is my neighbour

In the following, I will refer to the topic dropped element as the ‘gap’. In the respective data examples the topic dropped element is represented by an underline, [ _ ], as above. As the case features of the gap (and its antecedent) will play a major role in the further discussion, I indicate them within the interlinear glosses. In the glosses the gap is “represented” by squared brackets, bearing a case label which can be different from the case label of its antecedent. By A and B I refer to speaker A and speaker B who are involved in a given topic drop dialogue. English translations of the German examples will only be given when the glosses are insufficient to render the meaning of the respective dialogues.

|| 1 http://www.spiegel.de/spiegel/print/d-61086119.html (checked 15.05.2014).

16 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

Previous syntactic analyses of topic drop were mostly concerned with questions, such as e.g. what kind of empty category “fills” the gap, why topic drop is only possible in the prefield, or which syntactic categories can be dropped. In this chapter I will not deal with these questions very extensively.2 Instead, the focus of my investigation is on the relation between antecedent and gap, i.e., I will try to examine the syntactic and semantic identity conditions which (have to) hold between antecedent and gap and, in a broader sense, between context and target, respectively. In the following, I will first introduce the phenomenon of topic drop by presenting some empirical facts and data – some of which are well-known, some are quite new and have not been noticed in the literature so far. The aim of this short introduction is to give the reader an impression/overview of some properties of topic drop in German.

2.1.1 Topic Drop is Restricted to the Prefield It should first be noted that according to standard assumptions the German prefield can host only one syntactic constituent.3 Whenever the prefield (i.e. the position in front of the finite verb in a V2 clause) is filled, topic drop becomes impossible, cf. (3) and (4). (3) A:

Ich mag den Hans. I like the Hans-ACC B: * Ich mag _ auch. I like [ACC] as well

(4) A:

Der Hans hat heute seinen Schlüssel vergessen. The Hans-NOM has today his key forgotten B: * Ja, heute ist _ wirklich ziemlich unkonzentriert. Yes, today is [NOM] really fairly unconcentrated

However, one could object that topic drop (of different kinds of XPs) can also take place in the German middlefield and support such a claim by examples as the ones in (5)–(9):

|| 2 For the licensing and identification of topic dropped elements see chapter 3.4. 3 Since multiple filling of the prefield is subject to very special conditions, we can neglect it for our purpose.

Introduction | 17

(5) Die Journalisten im ersten Stock des Kanzleramts schauen The journalists in-the first floor of-the chancellor-office look sich ungläubig an. Hat sie ihn wirklich rausgeworfen? Ja, sie hat [ ].4 each-other doubtfully at. Has she him really fired? Yes, she has [him fired] (6) Hatte sie gar ihre “Abscheu” vor den “Halbwesen” kundgetan, die aus Had she even her disgust for the “half-beings” declared, which from künstlicher Befruchtung entstehen? In der Tat, sie hatte [ ].5 in-vitro fertilization develop? Indeed, she had [her disgust for the “half-beings” declared] (7) A: B:

(8) A: B:

(9) A: B:

Ist das Bad frei? Is the bathroom free? Ja, is’ [ ]. Yes, is [free] Wann ist denn wieder Vollmond? When is PRT again full-moon? Gestern war [ ]. Yesterday was [full-moon] Wann hast du Geburtstag? When have you-sg birthday? Ich hatte gestern [ ]. I had yesterday [birthday]

Wilder (1996, 1997, p.c.) claims that the examples (5)–(9) are not instances of topic drop but incomplete VP deletions in which the finite verb has evaded PFdeletion by moving out of the VP into C°, leaving the PF-deleted rest of the VP in the middle field (and the right bracket): In (5) and (6) the NP and the non-finite part of the verb are the relevant (covert) residues, whereas in (7)–(9) it is the adjectival and nominal part of a predicative that is (covertly) left behind.

|| 4 http://www.spiegel.de/politik/deutschland/merkel-wirft-nach-nrw-wahl-umweltministerroettgen-raus-a-833604.html (checked 15.05.2014). 5 http://www.spiegel.de/kultur/literatur/lewitscharoff-rede-buechner-preistraegerin-zubefruchtung-und-onanie-a-957254.html (checked 15.05.2014).

18 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

Wilder’s suggestion that we are dealing with two different phenomena (topic drop vs. incomplete VP deletion) is endorsed by (yes-no) questions: Whereas topic drop is impossible in (yes-no) question contexts, cf. (10), incomplete VP deletions are well-formed, cf. (11) (the data examples are from Chris Wilder (p.c.); see also Cardinaletti 1990:76): (10) A: a.

B:

b.

B:

(11) a. b.

Guck mal, da ist der Johann. Look PRT, there is the Johann-NOM Kennst du *(den)? Know you-sg (him-ACC)? Woher kennst du *(den)? Wherefrom know you-sg (him-ACC)? Ich hab keine Angst. Hast du [ ]? I have no fear-ACC. Have you-sg [fear-ACC]? Ich hab heute Prüfung. Wann hast du [ ]? I have today examination-ACC. When have you-sg [examination-ACC]?

Moreover, verb-final clauses introduced by a complementiser further support Wilder’s claim: When the NP is not a predicative NP (in (12) it is one) but an argument of the verb as in (13), such dropping instances are ungrammatical:6 (12) A: B:

Warum ist heute nicht Vollmond? Why is today not full-moon? Weil gestern [ ] war. Because yesterday [full-moon] was

(13) A:

Warum ist Peter heute nicht da? Why is Peter-NOM today not there? B: * Weil [ ] gestern da war. Because [Peter-NOM] yesterday there was

Thus, by assuming that the above differentiation between topic drop and incomplete VP deletions is justified, I conclude that the restriction to the prefield (Spec-CP position) is characteristic for topic drop.

|| 6 Note that the base positions of genuine arguments and nominal parts of predicates differ: NPs that act as parts of predicates stand closer to the verb in dependent clauses (because they are part of the right bracket (within the topological field model)).

Introduction | 19

As to the reasons for this restriction, it has been argued that topic dropped elements must be located in the Spec-CP position/the prefield, because it constitutes a topic position and the dropped element is a topic (cf. Huang 1984). However, as will be shown in the following, the gap/its antecedent does not have to be a topic, but at least a salient and (somehow) given element.7 Moreover, the prefield is not a designated topic position and besides, although there is evidence for a designated topic position in the German middlefield, cf. Frey (2004), topic drop is (nonetheless) only possible in the prefield. Another line of argumentation, cf. Rizzi (1994), notes that the prefield guarantees optimal access to the discourse, which provides the antecedent for the gap. Although there are other kinds of ellipses where this obligation does not hold, with respect to topic drop such an explanation makes sense because a topic drop gap does not constitute a blind copy of its antecedent but can accommodate semantic and syntactic variation (different interpretations, case mismatches). Due to the latter fact, it is reasonable to assume that topic drop must be restricted to the prefield (to ensure optimal access to the antecedent).8

2.1.2 General Conditions on Antecedents of Topic Dropped Elements Besides the positional restriction to the prefield, the most important condition for a topic drop gap to be well-formed is that its antecedent is in some way given by the discourse context. I.e., the following salience/givenness condition on topic drop antecedents must be satisfied, cf. (14): (14) Salience/givenness condition on topic drop antecedents The referent of a topic dropped element must be inferrable/contextually given (via some antecedent expression). With respect to the question how to define salience, I will adapt the following definition from Chiarcos et al. (2011:2): Salience defines the degree of relative prominence of a unit of information, at a specific point in time, in comparison to the other units of information.

|| 7 Steinbach (2007:59) notes that in topic drop dialogues “the null pronominal is [ ] a prominent discourse referent.” 8 As to the restriction of topic drop to the prefield see also chapter 3.4.3.

20 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

Generally, the above definition is not very useful in order to determine what counts as a salient element. Nevertheless, I will assume ‘salience’ as a cover term for referents that are (highly) accessible in a current discourse, but not necessarily topical. If (14) is not fulfilled, a gap cannot be identified/interpreted by the addressee. This is shown by the data in (15) and (16): In (15) no context is given, i.e. the sentence containing the gap is uttered out of the blue. In (16) the verbal inflection is marked for 3rd person plural whereas the only possible antecedent for the gap has 3rd person singular features. Thus, topic drop is impossible when no antecedent is present, cf. (15), or when there is a mismatch between the features of the (subject) gap/its antecedent and the feature specifications at the finite verb in the target sentence, cf. (16):9 (15) * _ Kommt nachher vorbei.10 [NOM] comes later along (16) A: B:

Wo ist denn der Hans? Where is PRT the Hans-NOM? *_ Kommen nachher vorbei. [NOM] come-(1/)3pl later along

With respect to the form of a topic drop antecedent, (17) shows that it does not have to be linguistically given, i.e. topic drop is also well-formed when the gap is derived from a non-linguistic antecedent, cf. Ackema & Neeleman (2007:100): (17) [Pointing to someone wearing a T-shirt in the middle of winter] _ Denkt zeker dat het zomer is. [NOM] thinks certainly that it summer is

|| 9 As will be seen in the following (this section), there are exceptions to this statement. In particular, when the antecedent is singular but denotes a plurality, as e.g. die Abiklasse (the Alevel class), das Volk (the nation). See Volodina & Weiß (to be published) for such mismatches in omissions occurring in Middle High German and Early Modern High German. 10 Examples in which the gap is not well-formed are marked as ungrammatical [*], instead of pragmatically not well-formed [#]. Though one could argue that the respective gap is fine in another context and therefore the ‘#’-marking should be used, I hold the view that the ‘*’marking should be used as long as the respective sentence is considered in this and only this context (in which it is ungrammatical). I will only use the hash ‘#’ to indicate that the given context is inappropriate (e.g. due to reasons of information structure).

Introduction | 21

Thus, independently of the phonetic realisation of the antecedent, a dialogue that contains a topic dropped element is only felicitous when the interlocutor is able to identify the entity the speaker is referring to within a given discourse. Merchant (2013) makes an equivalent point with regard to the kind of givenness that antecedents in ellipses have to obey, cf. datum (18), where the context containing the antecedent for the ellipsis site is a Greek sentence, whereas the target is an English sentence: (18) A: B:

Pires tin tsanda mazi su? Took-you-sg the bag-ACC with you? Yes, I did.

(from Merchant 2013:80)

Merchant (2013:27–28) states that “we must assume that the English ellipsis in B’s response in ([18]) is sensitive not to the overt form of the antecedent in Greek, but rather to more abstract properties not immediately obvious in the surface form of the Greek.” 11 While (17) and (18) show that topic drop/ellipsis antecedents do not have to occur linguistically or literally in order to license a gap/ellipsis, (19) illustrates that, firstly, the antecedent of a topic drop gap can be syntactically very deeply embedded, i.e. it can e.g. be a genitive attribute, and, secondly, that topic drop antecedent and topic drop gap do not need to have the same grammatical function/case features in context and target, respectively (a point which will be subsequently discussed in great detail): (19) A: B:

Wie war denn die Griechenlandreise der Abiklasse? How was PRT the Greece-trip-NOM of-the A-level class-GEN? _ Ist ganz begeistert zurückgekommen. [NOM] is very impressed came-back ‘How was the Greece trip of the A-level class? It (= the class) came back quite impressed.’

Example (20) gives further evidence for the supposition that the structural relationship between antecedent and gap is a very loose one:

|| 11 Weiß (2002) reports a similar example with respect to VP deletion (and negation): (i) Im Garten war niemand und im Haus auch nicht [jemand]. In-the garden was nobody and in-the house also not [someone] In this case the indefinite part of niemand (nobody) serves as the antecedent of the ellipsis.

22 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(20) A: B:

Wie war denn die Griechenlandreise der Abiklasse? How was PRT the Greece-trip-NOM of-the A-level class-GEN? _ Sind ganz begeistert zurückgekommen. [NOM] are very impressed came-back ‘How was the Greece trip of the A-level class? They (= the participants) came back quite impressed.’

In (20) topic drop is licensed under the assumption that the antecedent semantically entails the (sum of) individuals to which the gap refers. In this example the referent which is represented by the gap and to which the B-speaker refers, is neither the Greece trip nor the A-level class, but the people participating in the trip (the pupils/teacher(s) of the A-level class). This is signalled by the verbal inflection marked for 3rd person plural. Thus, in (20) (partial) semantic identity (but not syntactic identity) between antecedent and gap is given. Examples (21) and (22) show that the entities a topic dropped gap has to refer to in order to get identified, do not have to be mentioned before. Instead, the gap’s reference can also be resolved by a ‘post’cedent (see also Siewierska 2004:7), i.e. a subsequent antecedent (and a matching photograph): (21) _ Kennt die Straßen im Karlshorster Kiez wie seine [NOM] knows the streets in Karlshorst’s neighbourhood like his Westentasche: Ortschronist Günter Bergner ist häufig auf Achse. vest pocket. Local-chronicler G.B. is often on tour [photograph]12 (22) _ Kennt die Spielregeln seit seiner Kindheit: Gerd Poppe von [NOM] knows the rules since his childhood: Gerd Poppe from Bündnis 90/Die Grünen landete in der Wertung auf Platz sechs. Alliance ‘90/The Greens landed-up in the evaluation on position six [photograph]13 The following datum in (23) shows a further property of topic drop, namely that even an element whose reference is (yet) unknown can act as a topic drop antecedent. Thus, it is not a well-formedness condition for a topic drop gap that its antecedent is assigned a referential value/a fixed reference:

|| 12 DWDS Corpus, Berliner Zeitung, 04.05.1995, p. 23. 13 DWDS Corpus, Berliner Zeitung, 27.11.1995, p. 17.

Introduction | 23

(23) A: B:

Mit wem warst du gestern in der Oper? With whom-DAT were you-sg yesterday in the opera? _ Kennst du nicht. [ACC] know you not

What is given by the context in (23A) is the presupposition of the existence of a person X/some persons X’ to which the antecedent (i.e. the w-phrase) refers and for which it might be true that X/X’ was accompanying the speaker to the opera. Thus, (23) shows that even in cases in which the referent of the antecedent is unknown, the gap can refer back to the entity/entities the antecedent PP mit wem (with whom) relates to. With the above remarks on information structural properties, givenness and referentiality of topic drop gaps and their antecedents I intended to point out that the givenness/salience condition in (14) is presumably correct. However, as will be seen, salience of the antecedent alone is not a sufficient well-formedness condition for topic drop gaps and, besides, further syntactic and semantic peculiarities have to be fulfilled. Before starting the syntactic discussion, there is one more point to be examined in this introduction, namely the question/the problem of the choice of an antecedent for the dropped element. Sometimes this question is resolved by “external” factors, e.g. by the fact that the gap’s antecedent is (i) a nonreferential element as in (24), or (ii) a 1st or 2nd person object pronoun as in (25) and thus, it can hardly be dropped (cf. Fries 1988 for the first claim and Ross 1982, Cardinaletti 1990, Rizzi 1994 and Steinbach 2007 for the second one): (24) A:

Heute in der Vorlesung hat fast jeder geschlafen. Today in the lecture has almost everyone-NOM slept B: * _ Hat gestern im Kurs auch geschlafen. [NOM] has yesterday in class also slept

(25) A:

Mögen sie dich in deiner neuen Klasse? (from Steinbach 2007:59) Like they you-sg-ACC in your-sg new class? ‘Do they like you in your new class?’ B: * _ Kann keiner leiden. [ACC] can nobody like ‘Nobody likes me.’

24 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

In the examples (24) and (25) the context provides an antecedent the gap cannot/can hardly refer back to.14 However, there are also cases where a context offers two or even more alternative antecedents, each of them being characterised as a possible representative of the gap with regard to its syntactic and semantic characteristics, cf. the following data in (26)–(27):15 (26) A: B:

(27) A: B:

Die Maria hat den Hans geheiratet. The Maria-NOM has the Hans-ACC married Oh Gott, [?die/den] würd ich nicht mal geschenkt nehmen. Oh God, [her-ACC/him-ACC] would I not even as-a-present take Den Hans hat die Maria geheiratet. The Hans-ACC has the Maria-NOM married Oh Gott, [den/?die] würd ich nicht mal geschenkt nehmen. Oh God, [him-ACC/her-ACC] would I not even as-a-present take

The examples (26) and (27) show that a 3rd person object gap can hardly be derived from a 3rd person subject: Instead, there is a clear preference to pick out the antecedent that displays the same grammatical function as the gap (when such a possibility exists). This also holds for the data in (28) and (29): Here the preference is directed to the subject. (28) A: B:

(29) A: B:

Die Maria hat den Hans geheiratet. The Maria-NOM has the Hans-ACC married Toll, [sie/?er] wird jetzt sicher glücklich sein. Great, [she-NOM/he-NOM] will now surely happy be Den Hans hat die Maria geheiratet. The Hans-ACC has the Maria-NOM married Toll, [?er/sie] wird jetzt sicher glücklich sein. Great, [he-NOM/she-NOM] will now surely happy be

|| 14 However, note that neither 1st/2nd person pronoun drop (to be discussed in detail in chapter 3) nor the drop of QPs is necessarily ill-formed: (i) A: Hans mag fast jede Oper. Hans likes nearly every opera-ACC B: _ Mag der Egon auch. [ACC] likes the Egon as well 15 For reasons of clarity the possible gap representatives are crossed out (which – in these examples – is equivalent to the underline-notation, cf. ‘_’).

Verbatim and Non-verbatim Topic Drop: Two Kinds of Topic Drop in German | 25

I.e., when one has to decide between two elements that have the same referential properties (here: being a specific and definite NP) there exists a preference to pick out that element as a gap antecedent that has the same grammatical function as the gap. Of course, such a preference can be overridden by discourse factors: In (26)/(27) this may be the sex/sexual preference of speaker B. However, as long as “external” (= discourse) factors do not intervene (as in (28) and (29)), the most probable antecedent is the element that has the same grammatical function as the gap. Incidentally it should be noted that from the minimal pairs (26) vs. (27) and (28) vs. (29) we can also conclude that (shorter) distance between antecedent and gap is not a relevant factor with regard to the choice of an antecedent for the gap.

2.2 Verbatim and Non-verbatim Topic Drop: Two Kinds of Topic Drop in German 2.2.1 A First Approach to Verbatim Topic Drop (VTD) and Non-verbatim Topic Drop (NVTD) Here and in the following sections I will mainly discuss topic drop of nominal and prepositional phrases (= NPs16 and PPs) and examine the syntactic and semantic (non-)identity relations that must/can hold between the antecedent in the context (uttered by speaker A) and the gap in the target (uttered by speaker B). As the data below show, topic drop can affect arguments with structural case, (1)–(2), as well as arguments which bear oblique/lexical case,17 (3)–(4). PPs can be dropped as well, cf. (5), see also Fries (1988). (1)

A: B:

Da vorne steht der Chef. Over there stands the boss-NOM _ Steht da schon seit Stunden! [NOM] stands there already for hours

|| 16 If not indicated otherwise, I will use the notions NP and DP equivalently. 17 I will use the terms ‘oblique’ and ‘lexical’ case equivalently and follow the tradition of generative grammar in considering dative and genitive as oblique/lexical cases and nominative and accusative as structural cases.

26 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(2) A: B:

(3) A: B:

(4) A: B:

(5) A: B:

Kennst du den Hans? Know you-sg the Hans-ACC? Ne, _ kenn ich nicht. No, [ACC] know I not Gestern bin ich der Dekanin begegnet. Yesterday am I the dean-fem-DAT came-across _ Bin ich heute auch begegnet. [DAT] am I today also came-across Hast Du jemals der Entrechteten gedacht? Have you ever the disenfranchised-GEN commemorated? Ne, _ hab’ ich noch nie gedacht. No, [GEN] have I yet never commemorated Neapel liegt am Meer. Naples lies by-the sea-PP _ Liegt Ostia auch. [PP] lies Ostia too

Note first that, due to a register mismatch, some speakers find that example (4) sounds odd: This is because in German the genitive case is mostly used in formal registers whereas topic drop belongs to rather informal registers. Later we will see that in examples in which the use of the genitive is not perceived as formal a genitive-marked topic drop gap is well-accepted. In all the examples (1)–(5) the dropped elements in the target sentences show the same grammatical function/case features as their antecedents in the contexts. Moreover, predicates in context and target are identical too. For the moment I will refer to such an instance of topic drop as verbatim topic drop, VTD.18 This characterisation will be further explored and adjusted. In contrast to the examples (1)–(5), in (6)–(8) we have different predicates in context and target and the grammatical function/case feature of the dropped element can differ from its antecedent’s grammatical function/case feature:

|| 18 The term ‘verbatim drop’ relates to Bayer et al. (2001). However, in contrast to me, Bayer et al. (2001) employ a different analysis (in terms of a deleted d-pronoun) for topic drop. Due to the fact that topic dropped elements can receive a sloppy reading (to be discussed within chapter 3) such an analysis cannot be correct.

Verbatim and Non-verbatim Topic Drop: Two Kinds of Topic Drop in German | 27

(6) Kuttes Kran ist nichts passiert. _ Hat nur ein bißchen gewackelt.19, 20 Kutte’s crane-DAT is nothing happened. [NOM] has only a bit rocked (7) A: B:

(8) A: B:

Was war mit dem Fahrrad? What was with the bicycle-PP? _ Haben wir auch in den Brunnen geworfen.21 [ACC] have we also in the well thrown Wie waren denn Marias Abi-Noten? How were PRT Maria-GEN A-level-grades? _ Hat insgesamt eine 1,2 bekommen. [NOM] has overall a 1.2 received

Case mismatches between antecedent and gap, like in the topic drop data (6)– (8), will be referred to as instances of non-verbatim topic drop, NVTD. In contrast to these gaps (which bear structural case), lexical case gaps as in (9) and (10) or a dropped PP as in (11) can hardly be inferred from an antecedent which displays a different case/grammatical function than the dropped element: (9) A:

Hans wollte mir heute Aktien verkaufen. Hans-NOM wanted me today stocks sell B: ? Lass es, _ kannst du nicht vertrauen. Let it (be), [DAT] can you-sg not trust

(10) A:

Der Hans tut mir leid. The Hans-NOM is by-me felt-sorry B: ?* Ja, _ hab ich mich neulich auch mal erbarmt. Yes, [GEN] have I REFL recently also PRT taken-pity

(11) A:

Ich liebe die Ostseeküste. I love the Baltic-sea-coast-ACC

|| 19 DWDS Corpus, Berliner Zeitung, 08.01.1994, p. 57. 20 Sören Schalowski (p.c.) suggested that the underlying form could also be [Es] (= it) hat nur ein bißchen gewackelt. However, for me this interpretation is not possible in the given context. Nonetheless, if he was right, the gap in (6) could simply refer back to an (omitted) expletive pronoun that denotes a (topical) situation as allowed for by the aboutness topic definitions of Reinhart (1981) and Falk (1993). 21 Inspektor Barnaby, “Kind des Todes”, 16.01.2011, ZDF (TV broadcasting).

28 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

B: * _ Liegt ja auch Heringsdorf.22 [PP] lies PRT also Heringsdorf

[PP] = ‘At the Baltic Sea Coast’

The above data reveal (so far) that case mismatches between antecedent and gap are fine when the gap bears structural case but ungrammatical when the gap is a PP or when it bears lexical case. Thus, according to these data contrasts we can state that structurally cased elements can undergo VTD as well as NVTD, whereas PP gaps and non-structurally (i.e. lexically/obliquely) cased elements can only undergo VTD. This first approximation is illustrated in (12): (12)

A first approximation to VTD and NVTD

Kind of topic drop

Antecedent

Gap

VTD

case x

case x

NVTD

case x

(structural) case y

In essence, the syntactic differentiation between VTD and NVTD was already established by Fries (1988). Fries (1988:31; 47) notes the following: Insgesamt scheint für lexikalisch zugewiesene Kasus sowie für Dativ-NPs eine strukturelle und semantische Konstanz der Folgesätze Voraussetzung für “Pronoun Zap” zu sein. […]; [wobei Antezedent und gedropptes Element] in ihren jeweiligen Sätzen dieselbe grammatische Funktion besitzen. TRANSLATION: In case of lexically assigned case as well as dative NPs a structural and semantic consistency seems to be a prerequisite for “Pronoun Zap” […]; [the antecedent and the dropped element, however,] have the same grammatical function in their respective sentences.

|| 22 With regard to datum (11) Sören Schalowski (p.c.) suggested that the PP is droppable when (11B) is replaced by (11B’): [An der Ostseeküste] bin ich noch nie gewesen ([At the Baltic Sea Coast] am I PRT never been), which might be the case for some speakers. However, then (11B’) is probably no instance of a topic dropped PP, but originates from another register-specific construction (e.g. in “Kiezdeutsch”) in which time/place designations lack a preposition, as e.g. in (i): An der [Ostseeküste] bin ich noch nie gewesen, or (ii): Ich bin jetzt am Mehringdamm/Hauptbahnhof (I am now at Mehringdamm/main station). Thanks to André Meinunger for pointing that out to me.

Verbatim and Non-verbatim Topic Drop: Two Kinds of Topic Drop in German | 29

Besides, the authors of the Grammatik der Deutschen Sprache, cf. Strecker et al. (1997:576), observe that case identity between antecedent and dropped element is not required, when transitions proceed from: (i) (Nominative) subject … to … (accusative) object, (ii) (Accusative) object … to … (nominative) subject, (iii) (Dative/PP) object … to … (nominative) subject. However, a transition from: (iv) *(Nominative) subject … to … (dative/PP) object is said to be ungrammatical. Although the IDS grammarians (cf. Strecker et al. 1997) neither present an exhaustive treatment of the data nor offer any generalisation, e.g. they fail to recognise that in essence all lexical NPs and PPs can act as topic drop antecedents as long as the gap in the target sentence bears structural case (Nom, Acc), cf. the data in (6)–(8), they support our intuition to split up topic drop into VTD and NVTD, respectively. In the following, I will further differentiate the particular properties of VTD. Consistently with our preliminary characterisation of VTD, (13), (14) and (15) show that for lexical case gaps and PP gaps case matching alone is not sufficient. Instead, predicates in context and target have to be identical, cf. the respective (a) vs. (b) answers: (13) A: a. b.

Ich bin vorhin dem Hans begegnet. I am just the Hans-DAT came-across B: ? _ Folge ich jetzt auf Twitter. [DAT] follow I now on twitter B: _ Bin ich heute auch begegnet. [DAT] am I today also came-across

(14) A: a.

b.

Maria bedarf dieser Droge ganz akut. Maria requires this drug-GEN very urgently B: ?*_ Erfreue ich mich auch. [GEN] please I myself as-well ‘This drug delights me as well.’ B: _ Bedarf ich auch. [GEN] require I as-well

30 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(15) A: a.

b.

Neapel liegt am Meer. Naples lies at-the seaside-PP B: * _ Lebt es sich auch schön. [PP] lives EXPL REFL indeed nicely ‘The seaside is a nice place to live’ B: _ Liegt Ostia auch. [PP] lies Ostia as-well

However, data like (16)–(18) weaken the above mentioned requirement of predicate identity (and case matching), showing that identity of (the underlined) predicates does not seem to be a necessary condition for topic drop of lexically cased NPs and PPs: (16) A: B:

(17) A: B:

(18) A: B:

Gestern hab ich Maria in der Mensa getroffen. Yesterday have I Maria-ACC in the canteen met _ Bin ich letzte Woche im Kolloquium begegnet. [DAT] am I last week in-the colloquium came-across Maria braucht eine Gehhilfe. Maria needs a walker-ACC _ Bedarf Martin auch. [GEN] requires Martin as-well Neapel liegt am Meer. Naples lies at-the seaside-PP _ Befindet sich Ostia auch. [PP] located REFL Ostia too

Thus, in accordance with the data (13)–(18), I modify the view I have argued for within the above paragraph: I.e., I will state that it is not the case that VTD is only licensed under case and/or predicate identity. In particular, the data (13)–(15) show that case identity between antecedent and gap is not a sufficient condition in order to license a VTD gap. On the other hand, (16)–(18) show that identity of context and target predicate is a too strong requirement for VTD. Furthermore, if we look at the examples in (19)–(21), we can note that identity of theta roles, assuming “standard” notions of theta

Verbatim and Non-verbatim Topic Drop: Two Kinds of Topic Drop in German | 31

roles,23 seem to impose a too weak condition on VTD as well, at least for genitive cased gaps and PP gaps: Waldi folgt dem Hans ja gar nicht. Waldi (a dog’s name) follows the Hans-DAT PRT at-all not B: ? _ Hat er noch nie gehorcht. [DAT] has he yet never obeyed Theta roles of antecedent/gap: co-agent

(19) A:

(20) A:

Maria wurde des Mordes angeklagt. Maria was the murder-GEN accused B: ?* _ Wurde Martin auch bezichtigt. [GEN] was Martin as-well incriminated Theta roles of antecedent/gap: theme

(21) A:

Ich wohne am Meer. I live at-the seaside-PP B: * _ Liegt übrigens auch Ostia. [PP] lies likewise also Ostia Theta roles of antecedent/gap: position

Thus, I will suggest that instead of case identity (of antecedent and gap) or case identity together with predicate identity (of context and target predicate) the crucial condition for the well-formedness of VTD is the identity of finely granulated (micro) theta roles assigned to antecedent and gap, respectively, which is ensured when context and target predicates are “highly” synonymous (i.e. semantically equivalent). Note that pure PF identity of context and target predicate is not relevant.24 I regard a ‘micro theta role’ as a finely granulated/fine-grained, semantically enriched kind of theta role (in contrast to standard notions of theta roles (see e.g. Bornkessel et al. 2006 for different approaches to argument structure and semantic/theta roles)). I will return to the notion of a micro theta role after

|| 23 The respective theta roles in (19)–(21) are classified according to Primus (2012:56). 24 As (i) shows the gap and the overt form are both pragmatically marked: (i) A: Dem Hans drohte die Pleite. The Hans-DAT threatened the bankruptcy (= ‘to be on the rocks’) B: # _ / # Ihm drohte aber auch die Mafia. [DAT]/Him-DAT menaced however also the Mafia

32 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

having investigated the relation between topic drop antecedents and their respective gaps more precisely. For now I state the following, cf. (22) (22) Micro theta role condition In a given verbatim context two predicates count as equivalent if their arguments have the same micro theta roles. The concept ‘micro theta role’ should be understood as a notion that allows to capture particular instances of topic drop that demand stricter identity between antecedent and gap / between the respective predicates in the context and target than others. Later, I will make clear how the concept ‘micro theta role’ may be related to the c- and s-selectional properties of predicates. However, until having elaborated finer instruments and notions to predict/determine the wellformedness of different types of topic drop, I will assume the following characterisations/definitions of NVTD and VTD, (23): (23) Well-formedness conditions on NVTD and VTD a. NVTD: When predicates in context and target are not semantically equivalent, i.e. lacking identity of micro theta roles, the gap resulting from topic drop must be structurally cased in order to be wellformed. b. VTD: When predicates in context and target are semantically equivalent (to the degree that context and target predicates possess identical micro theta roles), a topic drop gap bearing any case is wellformed. The hypothesis that I assume in (23) is linked to the traditional distinction between structural and lexical cases which can be made explicit by quoting Chomsky (1981:171): “Structural Case in general is dissociated from theta-role; it is a structural property of a formal configuration. Inherent Case is presumably closely linked to theta-role.” The summary given in (24) is a revised version of (12) and illustrates the differences between VTD and NVTD:

Verbatim and Non-verbatim Topic Drop: Two Kinds of Topic Drop in German | 33

(24)

Simplified well-formedness conditions on VTD and NVTD

Kind of topic drop VTD

Antecedent … context predicate

Gap … target predicate

Micro theta role x

Micro theta role x

Predicate x

Identical/synonymous predicate x/x’

--- No reference to case features of antecedent/gap --NVTD

(Micro) theta role x

(Micro) theta role y

Predicate x

Predicate y

No reference to case features of antece- Gap bears structural case dent (Nom or Acc)

2.2.2 VTD is not ‘Stripping in Disguise’ As VTD requires identity/synonymy of predicates in context and target, one could ask whether VTD is really an instance of topic drop or rather ‘stripping in disguise’.25 If VTD, cf. (25), was a subcase of stripping (i.e. ‘stripping in disguise’), cf. (26), the realisation of the identical lexical verb in the target sentence would constitute an additional insertion option, i.e., VTD could be understood as an (optional) violation of the Gricean Maxim of Quantity: (25) VTD A: Der Hans mag den Otto. | B: [Den Otto] mag der Olaf auch. The Hans likes the Otto-ACC | [The Otto-ACC] likes the Olaf as-well (26) Stripping A: Der Hans mag den Otto. | B: [Den Otto] mag der Olaf auch. The Hans likes the Otto-ACC | [The Otto-ACC] [likes] the Olaf as-well Incomplete elliptical deletions often lead to ungrammaticality; thus, in order to prevent their application in grammar some authors (cf. Takahashi & Fox 2005 (with reference to Merchant unpublished), Merchant 2008, Hartman 2011) have suggested the principle of MaxElide, which can roughly be paraphrased as the

|| 25 Note that stripping is said to occur in coordinations as in Manny plays the piano and Anna __, too (cf. Winkler 2006). Nonetheless, I will refer to the above dialogues where all material except the target binder and the element auch (too) is omitted as stripping, too.

34 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

general preference to elide a larger constituent over a smaller one. However, the following data suggest that VTD is a construction on its own: VTD (27) A:

Der Hansi hasst sichi. The Hans hates REFL-ACC B: (?) _i/m Hasst der Olafm auch. [ACC] hates the Olaf as-well

VTD (28) A:

Der Hansi erholt sichi. The Hans relaxes REFL-ACC B: * _i/m Erholt der Olafm auch. [ACC] relaxes the Olaf as-well

Stripping26 (29) A: Der Hansi hasst sichi. The Hans hates REFL-ACC B: e*i/m Der Olafm auch. [e] the Olaf as-well Stripping (30) A: Der Hansi erholt sichi. The Hans relaxes REFL-ACC B: e*i/m Der Olafm auch. [e] the Olaf as-well (27)–(30) show that, besides the fact that VTD with inherently reflexive verbs is ungrammatical, stripping and VTD show different interpretations, respectively: Most speakers can interpret the VTD gap in (27) strictly. This, however, is not possible at all for the stripping construction in (29) where the gap can only be interpreted sloppily. With respect to speakers who consider (27) as marginal, one could argue that the gapped constituent in (27) is a reflexive pronoun and as such it can only be positioned in the prefield when it receives a (focal) accent – which is impos-

|| 26 In order to avoid suggesting that the same empty category is involved in VTD and stripping, I use the underline notation for VTD (as before) and a small ‘e’ (standing for (some) empty category) for a stripped constituent.

Verbatim and Non-verbatim Topic Drop: Two Kinds of Topic Drop in German | 35

sible under deletion. However, if this was the reason for a potential marginality of (27), it would be very probable that the stripped XPs in (29)/(30) are not located in the prefield but in the middlefield. I.e., the stripping examples in (29)/(30) should rather have a (surface) structure as indicated in (31) and (32): Stripping (31) A: Der Hans mag den Otto. | B: Der Olaf mag [den Otto] auch. The H. likes the Otto-ACC | The Olaf [likes] [the Otto-ACC] as-well Stripping (32) A: Der Hans hasst/erholt sich. | B: Der Olaf hasst/erholt [sich] auch. The H. hates/relaxes REFL-ACC | The Olaf [hates/relaxes] [REFL-ACC] as-well When stripping involves the elision of a reflexive pronoun, the elided constituent can hardly be interpreted strictly, cf. (29)–(30). Moreover, when considering data that involve a dropped reflexive pronoun and a person-switch (i.e. a situation where the respective binders in context and target (cf. Olaf, ich) have different person features), stripped elements, cf. (33), become very marginal whereas VTD gaps are still fine (even if the sloppy interpretation is hard to get), cf. (34):27 Stripping (33) A: Der Olafi hasst sichi. The Olaf hates REFL-ACC B: e*i/?*m Ichm auch. [e] I as-well VTD (34) A: B:

Der Olafi hasst sichi. The Olaf hates REFL-ACC _i/?m Hass’ ichm auch. [ACC] hate I as-well

However, there are occurrences of stripping in which both, the sloppy and the strict reading, seem to be possible, cf. (35):

|| 27 See chapter 3 for details regarding the interpretation of topic dropped elements and some notes on stripping.

36 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(35)

Figure 1: Advertisement of the Frankfurter Sparkasse 1822. Seen: 15.11.2014

Even if (35) allows both, the strict and the sloppy reading, the sloppy reading under which the costumer should (together with the savings bank) help the company Corpus Sireo to sell their property obviously makes no sense. Note further that in a topic drop version of (35), namely (36b), also both readings are available, however (as above), the strict reading seems to be easier to get in topic drop constructions than in stripping contexts: (36) a. b.

Sie möchten ihre Immobilie verkaufen? e Wir auch! You-formal want your property-ACC sell? [e] we as-well Sie möchten ihre Immobilie verkaufen? _ Möchten wir auch verkaufen! You-formal want your property-ACC sell? [ACC] wish we as-well to sell

A further difference between stripping and VTD is illustrated by the contrast of (37) vs. (38) which shows that with some quantified phrases stripping is better than VTD: (37) Stripping A: Ich esse nur wenige Speisen. | B: e Ich auch. I eat only few dishes-ACC | [e] I as-well

An Empirical Basis for the VTD/NVTD Distinction | 37

(38) VTD A: Ich esse nur wenige Speisen. | B: *_ Ess’ ich auch. I eat only few dishes-ACC | [ACC] eat I as-well Note that the dissimilarity between VTD and stripping can also be determined by idioms: under topic drop the idiomatic meaning is lost, (39a), whereas it is not lost under stripping, (39b), cf. Gisbert Fanselow (p.c.) for this test: (39) A:

a.

B:

b.

B:

Gestern hab ich dem Hans Honig um den Mund geschmiert. Yesterday have I the Hans honey-ACC around the mouth spread ‘Yesterday I was buttering up Hans’ _ Hab ich dem Udo heute auch um den Mund geschmiert. [ACC] have I the Udo today also around the mouth spread e (Hab) ich dem Udo heute auch. [e] (have) I the Udo today also

Thus, the above minimal pairs clearly suggest that VTD is not ‘stripping in disguise’. Even if the two constructions share some properties, it is no option to maintain that topic drop is stripping ‘minus’ verb ellipsis. Besides, VTD involves (only) semantic identity of the predicates in context and target whereas stripping requires PF-identity of the respective predicates.

2.3 An Empirical Basis for the VTD/NVTD Distinction In order to provide an empirical basis for the VTD/NVTD data, I conducted a small survey in which 60 subjects were asked to judge 30 different VTD/NVTD sentences. The data were presented acoustically (spoken by two different speakers), online and in a randomised order. Subjects were between 16 and 73 years old. 40 of them were female, 20 were male. The mean age was 34 years. The scale included the following values [-2 | -1 | 0 | 1 | 2] where [-2] was the value corresponding to the lowest acceptability and [2] the value corresponding to the highest acceptability. All subjects were native speakers of German. The instructions given to the subjects were the following, cf. (1): (1)

Bitte lesen Sie folgende Anweisungen genau durch! Sie werden Dialoge hören, die Sie als Ganzes bewerten sollen. Es soll nicht darum gehen, ob die Dialoge im Sinne der Schulgrammatik richtig oder falsch sind, sondern einfach darum, ob “man das so sagen kann”. Vertrauen Sie bei der Auswahl Ihrer Antworten einfach ihrem

38 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

Bauchgefühl. Versuchen Sie nicht, irgendwelche speziellen Regeln für Ihre Antworten auszuarbeiten. Uns interessiert wirklich nur Ihre Intuition. Achtung! Wenn sie einmal Ihre Antwort gegeben haben, können Sie nicht mehr zurückgehen und die Antwort revidieren. Bitte versuchen Sie nicht, auf den “Zurück”-Knopf in Ihrem Browser zu drücken: das wird zum Absturz des Programms und zum Verlust Ihrer Daten führen. Sollten Sie feststellen, dass Sie aus Versehen die falsche Antwort angeklickt haben, probieren Sie, sich den entsprechenden Dialog zu merken. Am Ende des Experiments haben Sie die Möglichkeit, uns Ihre Bemerkungen zum Experiment mitzuteilen – dann können Sie uns auf Ihren Fehler hinweisen. Wichtig: Für die Befragung benötigen Sie Kopfhörer oder Lautsprecher. Bitte setzen Sie die Kopfhörer jetzt auf bzw. schalten Sie die Lautsprecher an. (1’)

[Translation] Please read the following instructions carefully! You will hear a dialogue that is to be evaluated en bloc. The task is not to decide whether the dialogue is correct or false according to grammatical rules, you are just supposed to decide whether “one can say it this way or not”. Please trust your personal intuition. Do not try to establish specific rules for your answers. We are only interested in your intuition. Attention! The answer cannot be revised after submission. Please do not try to push the “return” button on your keypad. This will lead to the program’s crash and the loss of your data. In case you realise that you have entered the wrong answer, try to remember the dialogue. At the end of the experiment you get the chance to give comments – there you can tell us about your mistake. Important: For the survey you need headphones or speakers. Please put on the headphones now, respectively turn on your speakers.

In an abstract manner the table in (2) represents the items the subjects were asked to evaluate. Thereby e.g. ‘Nom  Acc’ means that the antecedent was nominative case marked and that the gap was marked for accusative case. Instances as e.g. ‘Nom  Nom’ mean that both, antecedent and gap were nominative case marked and that the predicates in context and target were (semantically) different. Instances as e.g. ‘Nom  Nom (i)’ refer to examples in which antecedent and gap were both marked for nominative case and the predicate was identical in the context and target sentence.

An Empirical Basis for the VTD/NVTD Distinction | 39

(2)

Test items VTD/NVTD

Nom target

Acc target

Dat target

Gen target

PP target

Nom  Nom Nom  Nom (i) Acc  Nom Dat  Nom Gen  Nom PP  Nom

Nom  Acc Acc  Acc Acc  Acc (i) Dat  Acc Gen  Acc PP  Acc

Nom  Dat Acc  Dat Dat  Dat Dat  Dat (i) Gen  Dat PP  Dat

Nom  Gen Acc  Gen Dat  Gen Gen  Gen Gen  Gen (i) PP  Gen

Nom  PP Acc  PP Dat  PP Gen  PP PP  PP PP  PP (i)

Being aware of the fact that I tested 30 different conditions and every condition consisted of only one single item, this survey can hardly be called an experiment. This is so, even if I tried to impose some “control” by constructing test sentences which are the best possible for each condition. Nonetheless, this survey may provide a better empirical basis for the proposal to split up topic drop in VTD and NVTD, respectively. In what follows I will present the results arranged in a way that corresponds to my hypotheses (partly) elaborated so far. The respective hypotheses are given in (3a/b): (3)

VTD/NVTD hypothesis a. Obliquely cased (dat, gen) NPs and PPs can only be dropped in a verbatim environment (where predicates in context and target are identical/at least synonymous), but not in a non-verbatim environment. b. Structurally cased (nom, acc) NPs can be dropped in a verbatim as well as in a non-verbatim environment (without any reference to predicate identity/synonymy in context and target).

First, I will present those context/target pairs in which the case features of the gap and its antecedent are identical. Two cases have to be differentiated with regard to these pairs: (i) (ii)

Identical case features of antecedent and gap and identical predicates in context and target Identical case features of antecedent and gap, but different predicates in context and target

40 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

According to what has been said before, I regard (i) but not (ii) as instances of VTD (this is also indicated in the subsequent table in (4)). In the following I present the respective test items: Nom|Nom|identical David weiß so viel über das Alte Rom. David-NOM knows so much about the Old Rome _ Weiß aber nicht, wie man einen Autoreifen wechselt. [NOM] knows however not, how one a car tyre changes Nom|Nom|different Daniel kommt auch auf die Party. Daniel-NOM comes also to the party _ Wird aber sicher früher gehen – wie immer! [NOM] will however surely earlier leave – as always! Acc|Acc|identical Gestern beim Vortrag hab ich Peter Staudacher kennengelernt. Yesterday at-the talk have I P.S.-ACC become-acquainted-to _ Hab ich mal auf einer Tagung kennengelernt. [ACC] have I once at a conference become-acquainted-to Acc|Acc|different Ich kenne den “Prozess” in- und auswendig. I know “The Trial”-ACC inside out _ Hab ich auch schon zwei Mal gelesen. [ACC] have I also already two times read Dat|Dat|identical Am meisten vertraue ich Ärzten. At most (= The most) trust I doctors-DAT Echt? _ Vertraue ich am wenigsten. Really, [DAT] trust I the least Dat|Dat|different Ich habe dem Hans heute zu seinem neuen Job gratuliert. I have the Hans-DAT today to his new job congratulated Schön. _ Bin ich seitdem leider noch nicht begegnet. Nice. [DAT] am I since-then unfortunately PRT not come-across

An Empirical Basis for the VTD/NVTD Distinction | 41

Gen|Gen|identical Ich harre seiner Ankunft schon seit Tagen. I expect his arrival-GEN already for days _ Harre ich schon seit Wochen. [GEN] expect I already for weeks Gen|Gen|different Martin rühmt sich ständig seines Wissens. Martin praises himself perpetually of-his knowledge-GEN _ Bemächtige ich mich aber ganz gerne. [GEN] usurp I myself however quite readily PP|PP|identical Neapel liegt am Meer. Naples lies at sea(side)-PP _ Liegt Ostia auch. [PP] lies Ostia as-well PP|PP|different Jan hat seinen Verkaufsstand unter der Brücke. Jan has his sales stand under the bridge-PP _ Muss er Gottseidank aber nicht wohnen. [PP] must he thanks-god however not live The table in (4) gives the respective mean values and standard deviations for the above instances of topic drop. The first part of the table represents the values of those items where (next to case identity) predicates in context and target are identical, i.e. these results give the values for instances of what might be called “real” instances of VTD. The second part of the table represents the values of these items which are characterized by the fact that antecedent and gap bear the same case, however predicates in context and target are not identical and not synonymous, i.e. these results give the values for some instances of NVTD.

42 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(4)

Results of the survey (part 1)

Mean values | standard deviations for test items (identical case and +/-identical predicate), across 60 subjects Identical case of antecedent and gap + Identical predicates

Nom|Nom identical

1,167 | 1,152

Acc|Acc identical

1 | 1,221

Dat|Dat identical

0,467 | 0,455

Gen|Gen identical

-0,45 | 1,455

PP|PP identical

1,233 | 1,125

Identical case of antecedent and gap + Different predicates

Nom|Nom different

1,667 | 0,681

Acc|Acc different

0,6 | 1,487

Dat|Dat different

0,117 | 1,451

Gen|Gen different

-1,35 | 1,039

PP|PP different

-0,35 | 1,436

1,083 0,417 0,008

1,133 -0,528 -0,617

By looking at table (4) we can observe that the average subject rated the data more or less coherently with our intuitive grammaticality judgements. It shows that independently of the case features of antecedent and gap, the verbatim pattern is rated better than the non-verbatim pattern, except in the nominative case, where Nom|Nom different was rated better than Nom|Nom identical. However, the latter result is of minor interest, because according to our hypothesis structurally cased gaps can pattern verbatim as well as non-verbatim. What is important here, and this is confirmed by the 60 informants, is that topic drop of obliquely cased NPs and PPs is by far better rated when the predicates in context and target are identical than when they are different. The diagram in (5) illustrates the individual values for all conditions: (5)

2 1 identical predicate

0

different predicate

-1 -2 Nom

Acc

Dat

Gen

PP

An Empirical Basis for the VTD/NVTD Distinction | 43

When we compare those items which – according to our hypotheses in (3a/b) – have to pattern verbatim (i.e. obliquely cased NPs and PPs) with those which can pattern verbatim and non-verbatim (i.e. structurally cased NPs) in a verbatim and non-verbatim environment, respectively, we see that in a verbatim environment all instances of topic drop received mean values in the plus range (except the genitive – which is grouped together with the dative in the following chart), whereas in a non-verbatim environment only those test sentences where antecedent and gap were structurally case marked received values in the plus range. This is illustrated by the diagrams in (6): (6)28 1,5

Nom/Acc

1

Dat/Gen

0,5 0

PP

-0,5 -1

identical predicate

different predicate

By comparing the mean values in table (4) and by looking at the two charts in (5) and (6), respectively, we can observe that the minus mean values for obliquely cased NPs are mostly due to the low acceptability of dropped genitive NPs.29 In contrast, dropped dative NPs received mean values in the plus range, even when occurring in an ‘identical case + different predicates’ environment (which I consider as a non-verbatim environment). Thus, for those who assume that dative case in German is not necessarily idiosyncratically assigned (e.g. Wegener 1991, Blume 1998) the results concerning the dative may constitute evidence that in (some) two-place predicates dative case is also assigned structurally (or, at least, regularly). As for ditransitive predicates with canonical nom/dat/acc order of arguments as e.g. schenken (to donate) there is more agreement: Most authors assume that in such constellations dative case is assigned regularly (e.g. Fanselow 2000:180), constituting a predictable and thus “inherent”/non-idiosyncratic case, cf. Woolford (2006:112) || 28 Note that the chart contains less data for PPs than for structurally and obliquely cased NPs, respectively. 29 As already mentioned above, genitive case belongs to formal speech, whereas topic drop is attributed to rather informal speech. Thus, besides the grammatical conditions that we are investigating, we are also confronted with a (sociolinguistic) register mismatch.

44 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

and Haider (2010:254). I will later come back to the question whether the dative should be treated as a regular or a non-regular (and thus, idiosyncratic) case. I will discuss that issue in great detail (especially with respect to the occurrence of the dative in two-place predicates). As for now, we can empirically state that the dative exhibits some in-between status (oscillating between structural and lexical case). In the subsequent data set which was judged by native speakers, all test sentences were NVTD instances. In contrast to the above ‘test battery’ not only predicates in context and target but also case features of antecedent and gap were different. The research question was the same as above, i.e., is it true that only structural cases can pattern non-verbatim? Or, in other words: is the hypothesis in (3), repeated here for convenience, correct? (3)

VTD/NVTD hypothesis a. Obliquely cased (dat, gen) NPs and PPs can only be dropped in a verbatim environment (where predicates in context and target are identical/at least synonymous), but not in a non-verbatim environment. b. Structurally cased (nom, acc) NPs can be dropped in a verbatim, as well as in a non-verbatim environment (without any reference to predicate identity/synonymy in context and target).

In order to check whether (3) makes the correct predictions with respect to the NVTD data given below, consider the table in (7) which represents the mean values and standard deviations of the ratings of the following examples. Acc-Nom Ich würde gerne den Ingo kennenlernen. I would gladly the Ingo-ACC get-to-know Kein Problem – Is’ mein Nachbar. No problem – [NOM] is my neighbour Dat-Nom Gestern hab ich Klara bei den Hausaufgaben geholfen. Yesterday have I Klara-DAT with the homework helped Sehr gut, _ würde ja sonst wieder sitzenbleiben. Very good, [NOM] would PRT otherwise again have-to-repeat-a-class

An Empirical Basis for the VTD/NVTD Distinction | 45

Gen-Nom30 Heutzutage bedarf fast jeder Patient einer Zusatzversicherung. Nowadays requires nearly every patient a complementary-insurance-GEN _ Wird aber nicht von jedem Patienten abgeschlossen. [NOM] will however not by every patient be-effected PP-Nom Ich bin jedes Jahr auf Rügen. I am every year on Rügen-PP _ Ist auch eine wunderschöne Insel. [NOM] is indeed a wonderful island Nom-Acc Da vorne steht der Hans. Over there stands the Hans-NOM Echt? _ Hab ich im ersten Moment gar nicht erkannt. Really, [ACC] have I at the first moment at-all not recognised Dat-Acc Gestern bin ich der Maria begegnet. Yesterday am I the Maria-DAT came-across Toll. _ Hab ich schon seit Wochen nicht mehr gesehen. Great. [ACC] have I already for weeks not PRT seen Gen-Acc Gestern bei der Demo haben wir nicht der Rosa Luxemburg gedacht. Yesterday at the march have we not the R.L.-GEN commemorated.

|| 30 Due to a mistake I had to collect the evaluations for this condition one more time. Again the item was presented online and acoustically (spoken by two different speakers), and as a first item of a further experiment. Subjects were non-linguists and native speakers of German. The 60 participants were between 13 and 70 years old and their mean age was 35.8. 42 persons were female, and 17 were male, one person declared to be of gender ‘X’. As will be seen in the results, the (re-)collected data are somehow useless because they can hardly be compared to the other (NVTD) data which were collected within the first experiment. Thus, we could as well say that we have a lack of data with respect to the condition Gen-Nom (which is rated unexpectedly lower than the – according to hypothesis (3) – comparable conditions PP-Nom or Dat-Nom, respectively). Thus, in the following all NVTD data (especially the Nom-NVTD data) are a bit confounded. According to my expectations, NVTD should be rated a bit higher than it is.

46 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

_ Werden wir beim nächsten Umzug erwähnen. [ACC] will we at-the next demonstration mention PP-Acc Wir müssen noch was mit dem Fahrrad machen. We must PRT something with the bicycle-PP do _ Werden wir auf dem Flohmarkt verkaufen. [ACC] will we at the flea-market sell Nom-Dat Gestern war Hanna mal wieder in Berlin. Yesterday was Hanna-NOM PRT again in Berlin Schade, _ hätte ich gern mal wieder Hallo gesagt. Its-a-pity, [DAT] had I gladly PRT again ‘hello’ said Acc-Dat Letzte Woche hab ich mal wieder Steffi getroffen. Last week have I PRT again Steffi-ACC met _ Hab ich neulich bei ihrer Hausarbeit geholfen. [DAT] have I recently at (with) her term-paper helped Gen-Dat Ich habe mich neulich meiner Schüler erbarmt. I have REFL recently my pupils-GEN take-pity-on Ach, _ ist doch eh nicht mehr zu helfen. Ah, [DAT] is however anyway no longer to help PP-Dat Anna fährt dieses Jahr wieder mit Sven in Urlaub. Anna goes this year again with Sven-PP on holidays Ne, ne, _ hat sie bis auf weiteres nichts mehr zu sagen. No, no, [DAT] has she for the time being nothing more to say Nom-Gen Mir gefällt Martin so gut. Me appeals Martin-NOM so good _ Konnte ich mich bislang auch nicht erwehren. [GEN] could I REFL until-now also not fend-off

An Empirical Basis for the VTD/NVTD Distinction | 47

Acc-Gen Ich hab Vera das Geburtstagsgeschenk anvertraut. I have Vera the birthday-present-ACC entrusted-with _ Hat sie sich aber ganz schnell entledigt. [GEN] has she REFL however very fast get-rid-of Dat-Gen Ich werde Petra morgen beim Aufräumen helfen. I will Petra-DAT tomorrow with cleaning-up help _ Hab ich mich neulich auch mal erbarmt. [GEN] have I REFL recently also took-pity-on PP-Gen Ich vertraue auf Inas Zuverlässigkeit. I trust in Ina’s trust-worthiness-PP _ Vergewissere ich mich lieber zwei Mal. [GEN] make-sure I REFL better two times Nom-PP Da vorne steht ein tolles Motorrad. Over there stands a great motorcycle-NOM _ Kann ich nichts anfangen. [PP] can I not make-use-of Acc-PP Ich habe Werner gestern zum Kaffee eingeladen. I have Werner-ACC yesterday for coffee invited _ War ich neulich auch Kaffee trinken. [PP] was I recently also coffee drink Dat-PP Nina ähnelt so sehr meiner Mutter. Nina resembles so much my mother-DAT Ne, _ hat sie gar keine Ähnlichkeit. No, [PP] has she no resemblance Gen-PP Ich habe mich des Alkohols entwöhnt. I have REFL the alcohol-GEN weaned-off

48 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

_ Hatte ich noch nie zu tun. [PP] had I PRT never to do (7)

Results of the survey (part 2)

Mean values | standard deviations for test items (different case and different predicates), across 60 subjects Acc|Nom

1,6 | 0,572

Dat|Nom

0,933 | 1,133

Gen|Nom

-0,41731 | 1,279

PP|Nom

1,883 | 0,324

Nom|Acc

1,6 | 0,616

Dat|Acc

1,083 | 1,013

Gen|Acc

0,417 | 1,453

PP|Acc

0,767 | 0,184

Nom|Dat

0,717 | 1,354

Acc|Dat

0,15 | 1,246

Gen|Dat

-1,15 | 1,191

PP|Dat

-1,583 | 0,869

Nom|Gen

-1,15 | 1,205

Acc|Gen

-1,25 | 1,068

Dat|Gen

0,117 | 1,391

PP|Gen

-0,817 | 1,371

Nom|PP

-1,417 | 0,996

Acc|PP

-1,583 | 0,809

Dat|PP

-0,383 | 1,508

Gen|PP

-1,483 | 0,892

Nom-NVTD: 1

Acc-NVTD: 0,967

Dat-NVTD: -0,467

Gen-NVTD: -0,775

PP-NVTD: -1,217

|| 31 Note that due to the second collection of judgements with respect to the Gen-Nom item, “later” data points (including the results of the Gen-Nom item) are confounded. However, this little lack of data (Nom-NVTD) is not particularly crucial for my argumentation (according to my expectations, the Nom-NVTD column should be a bit higher in the chart in (8)).

An Empirical Basis for the VTD/NVTD Distinction | 49

Clearly, the 60 informants confirmed the intuition that NVTD is more acceptable with structurally cased NPs than with obliquely cased NPs or PPs, respectively. This is illustrated by the diagram in (8): (8)

2 1 0 -1 -2 Nom-NVTD

Acc-NVTD

Dat-NVTD

Gen-NVTD

PP-NVTD

Let us now look at a diagram which describes the results for structural cases on the one hand and oblique cases and PPs on the other; cf. (9):32 (9)

Nom/Acc 1 0,5 0

Dat/Gen PP

-0,5 -1 -1,5

In sum, the 60 speakers’ judgements undoubtedly show that “the less structurally cased” an NP is, the less it can be dropped in a non-verbatim environment. I.e., the results hint at the supposition that in non-verbatim environments the topic drop gap can hardly be an obliquely cased NP or a PP, but is restricted to represent a structurally cased NP, cf. Bayer et al. (2001), Steinbach (2007)33

|| 32 Again, as above, the graph describes less data for PPs than for structural NPs and oblique NPs, respectively. 33 Steinbach (2007:59) gives the following dative drop example (which he judges ungrammatical, contrary to my intuition): (i) A: Und wieso hat den Asylanten keiner geholfen? And why has the asylum-seekers-DAT nobody helped? B: * _ Hilft hier doch nie einer. [DAT] Helps for-all-that never one

50 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

and Haider (2010) who claim that topic drop is only possible with structurally cased elements. Hence, speaker’s attempts to interpret obliquely cased NPs/PPs in non-verbatim environments as pro are predicted to fail, resulting in degraded acceptability/grammaticality judgements when compared with the mean values of structurally cased gaps in non-verbatim environments. According to the above results the following summary can be given, cf. (10): (10) Empirical facts: Topic drop in non-verbatim and verbatim environments a. Structurally cased topic drop gaps are well-formed independently of additional semantic requirements (cf. predicate equivalence). b. For obliquely cased gaps to be well-formed, additional requirements (cf. semantically equivalent predicates in context and target) are necessary. Topic drop of dative NPs is subject to less strict semantic identity conditions (i.e. case identity may be sufficient) than topic drop of genitive34 NPs and PPs35 (micro theta role identity being necessary). In order to explain the findings in (10) it seems that we have to account for the presence of structurally and non-structurally cased gaps in two different ways: The fact that VTD is possible with all cases could hint at the supposition that VTD gaps are more or less blind copies of their antecedents, being insensitive to case as long as identity between context and target predicate is warranted. However, as already noted, data as (11)–(13), repeated here for convenience, speak against such a “blind” copy and deletion approach to VTD. Instead, something we have referred to as ‘micro theta role’ may be involved in the wellformedness conditions on VTD: (11) A: B:

Gestern hab’ ich Maria in der Mensa getroffen. Yesterday have I Maria-ACC in the canteen met _ Bin ich letzte Woche im Kolloquium begegnet. [DAT] am I last week in-the colloquium came-across

|| 34 Note, that the degraded acceptability of dropped genitive NPs is possibly confounded with a sociolinguistic factor, i.e. the register-specific, formal use of the genitive case can hardly be kept apart from a violation of a (core) syntactic/semantic well-formedness condition. 35 In the following PPs will be mostly left out from the discussion of VTD. The reason is that they are barely comparable to nominative, accusative, dative or genitive gaps, as not only the NP as such has to be (re)constructed by the gap, but the preposition as well.

An Empirical Basis for the VTD/NVTD Distinction | 51

(12) A: B:

(13) A: B:

Maria braucht eine Gehhilfe. Maria needs a walker-ACC _ Bedarf Martin auch. [GEN] requires Martin as-well Neapel liegt am Meer. Naples lies at-the seaside-PP _ Befindet sich Ostia auch. [PP] located REFL Ostia too

Moreover, when we suppose that some instances of topic drop (those of obliquely cased NPs) have to be differentiated from others (those of structurally cased NPs), suggesting that the latter are represented by a genuine null pronoun (pro) while the former are instances of a different kind of ellipsis (e.g. copy deletion/‘stripping minus verb ellipsis’) we will be faced with the severe problem of what to do with the dative case which behaves in-between. I.e., the main argument that speaks for a unified analysis of NVTD and VTD is the behaviour of dative gaps. Both, a classification as VTD instance as well as a classification as NVTD instance is a suboptimal solution for dative gaps. Hence, I will look for a solution which is better for unifying NVTD and VTD. The chart in (14) summarises all results in toto: (i)

Identical predicate relates to those items where predicates in context and target were identical (and hence, case features of antecedent and gap were identical too). (ii) Identical case + different predicate relates to those items where predicates in context and target were different, but case features of antecedent and gap were identical. (iii) NVTD relates to these items where predicates in context and target were different and case features of antecedent and gap were different too.

Note that I also consider instances of identical case + different predicate as NVTD (non-verbatim topic drop). However, in the following chart NVTD only refers to data where both, case features and predicates in context and target, were different.

52 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(14)

2

Identical predicate

1 0

Identical case +

-1

different predicate

-2

NVTD

Nom

Acc

Dat

Gen

PP

Of course, due to the fact that only one item per condition was tested, in certain points the picture appears somewhat inexact; however, the main findings are indisputable, namely (i) non-verbatim topic drop, cf. NVTD, is only possible with structurally cased NPs and marginally possible with dative NPs, but impossible with genitive cased NPs and PPs, and (ii) the possibility of verbatim topic drop (identical predicates in context and target), cf. Identical predicate, is (more or less) independent of a gap’s case features (modulo the genitive drop instances which may be marginal due to a sociolinguistic register mismatch). Moreover, (iii), the effect of case identity under a different predicate, cf. Identical case + different predicate, is best observed with respect to dropped dative NPs.

2.4 Structural Well-Formedness Conditions on VTD and NVTD In the following, I will investigate the well-formedness conditions on topic drop gaps, differentiating between structurally cased gaps and lexically cased ones and their well-formedness in non-verbatim environments vs. verbatim environments.36 In particular, I will try to find out by which properties dative gaps can be kept apart from structurally cased gaps on the one hand and genitive (and

|| 36 Haider (2010:269) notes that “[t]opic drop in German is another instance that separates nominative and accusative on the one hand and dative on the other.” He gives the following examples, (i) and (ii), under the context Wo ist Max? (Where is Max?) (i) _ Habe ich soeben zu dir geschickt. [ACC] have I just to you sent (ii) * _ Habe ich soeben den Weg gezeigt. [DAT] have I just the way shown

Structural Well-Formedness Conditions on VTD and NVTD | 53

PP) gaps on the other hand. Thus, I will look for answers to questions as e.g. why are the well-formedness conditions for dative gaps less strict than for genitive gaps, i.e., why is case identity between antecedent and gap (in most cases) sufficient to identify a dative gap, but not sufficient to identify a genitive gap (whose well-formedness depends on semantic identity of the context and target predicate)? Furthermore, I will examine why structurally cased gaps (in contrast to obliquely cased ones) are exempt from the obligation to pattern verbatim. And more general: Which (semantic) properties decide on the circumstance that a verb assigns structural or oblique case, respectively, to its nonsubject argument?

2.4.1 Topic Drop and the C- and S-selectional Properties of Predicates In order to find out which factors have an impact on the NVTD/VTD split, I will try to make the notion of the above mentioned concept of a ‘micro theta role’ clearer. Therefore, I will give a short characterisation of theta roles as well as the (thereby involved) c-selectional and s-selectional properties of predicates. My outline is mainly based on Bierwisch (2006) and standard views from the literature. According to standard assumptions (see e.g. Grimshaw 1990:46f) theta roles are mostly related to verbs, but they are not restricted to verbs. By the contrast (1a) vs. (1b) Bierwisch (2006:89) shows that “the relation between Brecht, Villon and adaptation in ([1]a) is semantically parallel to that between the corresponding elements in the morpho-syntactically different verbal construction ([1]b)”: (1)

a. b.

Brecht’s adaptation of Villon Brecht adapted Villon

In the following I will only discuss theta roles as entities which a verbal predicate assigns to its nominal arguments (i.e. NPs). In order to capture the differences between structurally and obliquely cased gaps which we have observed in our survey, I will distinguish between cselectional and s-selectional properties of predicates and examine in which particular way they might be related to each other (in the context of topic drop). I will adopt Bierwisch’s understanding of c-selectional and s-selectional properties, cf. (2):

54 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(2) c-selection and s-selection, according to Bierwisch (2006:91): Categorial or c-selection determines the syntactic and morphological requirements that optional or obligatory complements of an expression must meet, while Semantic or sselection specifies the corresponding semantic constraints.

With regard to synonymous verbs like fürchten (to fear) and grauen (to shudder) Bierwisch (2006:91) speaks of “similar” theta roles, which is exactly what I will refer to by the term ‘identical micro theta role’. He notes that in an example as fürchten/grauen, both verbs “select a person and a content of the emotional attitude, but they differ with respect to their syntactic properties, as fürchten is either a standard transitive verb, or requires a reflexive pronoun, while grauen c-selects what might be called an oblique subject.” With this quote and the (already familiar) examples (3)–(5) in mind it is clear that a micro theta role is primarily a semantic entity. In case a gap is obliquely case marked, it must obligatorily share particular s-selectional properties with its antecedent, otherwise (verbatim) topic drop is not well-formed. Hence, a micro theta role should be described by reference to (a predicate’s) sselectional properties. In particular, I assume that a micro theta role can be equated with the s-selectional properties that are shared by a topic drop gap and its antecedent in a verbatim environment. (3) A: B:

(4) A: B:

(5) A: B:

Gestern hab’ ich Maria in der Mensa getroffen. Yesterday have I Maria-ACC in the canteen met _ Bin ich letzte Woche im Kolloquium begegnet. [DAT] Am I last week in-the colloquium came-across Maria braucht eine Gehhilfe. Maria needs a walker-ACC _ Bedarf Martin auch. [GEN] Requires Martin as-well Neapel liegt am Meer. Naples lies at-the seaside-PP _ Befindet sich Ostia auch. [PP] Located REFL Ostia too

With regard to the c-selectional properties Bierwisch (2006:116) notes the following:

Structural Well-Formedness Conditions on VTD and NVTD | 55

[L]anguages may distinguish different syntactic and morphological categories by means of formal features, where the computational content of syntactic features like ±N or ±V determines, among others, the organization of possible ASs [Argument Structures, ET], while morphological features for categories like Case, Number, Person etc. are crucially involved in c-selection associated with positions in AS.

Bierwisch explicitly includes case features into c-selectional properties, a view I will adopt in the following. However, by discussing c-selectional properties of a (verbal) predicate, I will refer (unless otherwise indicated) to the case features assigned to its argument(s) – and not to e.g. category selection or other features that are c-selected. For the moment I will not discuss additional features contained within the c-selectional properties of (verbal) predicates, however I certainly assume that c-selection hosts more features than case features, also following Bierwisch (2006). With respect to a general definition of theta role(s), I assume that a theta role is first and foremost a theoretical construct mediating the bundle of the respective c-selectional and s-selectional properties that a predicative element subcategorises for. Bierwisch (2006:90) notes: Theta-Roles must be able to reconcile the same semantic relation with alternative grammatical conditions and also the same morpho-syntactic properties with different semantic relations. To this effect, they must have access to semantic as well as syntactic and morphological information, participating in at least two levels of representation. In this sense, Theta-Roles are (part of) the interface mediating between formal or morpho-syntactic and semantic or conceptual aspects of linguistic expressions.

This is in accordance with e.g. Dowty (1991:548) who states that theta roles are “obviously creatures of the syntax-semantics interface, and thus require a sound semantic theoretical basis as well as a syntactic one (and these must be mutually consistent)” or Jackendoff (1990:46) who assumes that “thematic roles are part of the level of conceptual structure, not part of syntax”. I will neither argue for – what Bierwisch (2006:98) refers to – an ‘Intrinsic View’ nor for an ‘Extrinsic View’ on theta theory. The latter assumes the existence of a presumably universal and desirably restricted and ordered set of theta roles, the former “considers the relations in question as an inherent aspect of the items they rely on, with no independent status, content, or ranking outside these structures.” Instead, I will rather stick to approaches which assume that verbal arguments are associated with proto-properties, in the spirit of Dowty (1991). Note that the notion of a micro theta role is a description for those properties which are essential to establish (contextually dependent) semantic synon-

56 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

ymy between two (verbal) predicates in order to ensure an obliquely cased topic drop gap to be well-formed. I.e., I will not assume a separate role for each predicate (which would make the notion of a micro theta role void). Instead, I suppose that the information contained within a micro theta role is linked to some more coarse-grained properties, namely to an argument’s proto-properties as well as to particular combinations of these proto-properties. Besides, I will hypothesise (following and extending approaches such as Dowty 1991 or Ackerman & Moore 2001) that an argument’s specific proto-properties are linked to grammatical function and case – although I am aware of the fact that attempts to link case to a specific semantics are faced with several and serious problems, cf. the discussions centering on synonymous nom/acc vs. nom/dat verbs as e.g. beglückwünschen (to felicitate) and gratulieren (to congratulate), cf. e.g. Haider (1993), (2010) or McFadden (2006). I will argue (and show) that it generally holds that verbs which select oblique case have a more specific semantics than verbs that select structural case for their nonsubject argument. I will mainly discuss two-place predicates when comparing accusative and dative as alternative object case realisations, however, dative and genitive assigning three-place verbs will be discussed as well.

2.4.2 On the Correlation between Case, Theta Roles and Verbal Semantics Within this section I hope to shed some light on the questions already mentioned above, reformulated in (6) and (7): (6)

Apart from the syntactic differentiation of structural vs. non-structural cases: What are the semantic differences between (verbal) predicates assigning structural and non-structural case to their nonsubject argument(s) (observed in topic drop contexts)?

The null hypothesis is: There are none. The differences between structural and non-structural case assigning predicates are solely syntactic. (7)

With regard to the different behaviour of predicates assigning lexical case: What are the semantic differences between predicates assigning dative vs. genitive case (vs. a PP) to their nonsubject argument (observed in topic drop contexts)?

Structural Well-Formedness Conditions on VTD and NVTD | 57

The null hypothesis is: these differences are idiosyncratic and do not follow any regularity. As to the questions in (6) and (7) I will not adopt the null hypothesis, but see whether the different behaviour of structural and non-structural case assigning predicates can be alternatively explained (with respect to topic drop). As I aim to show that case features are linked to particular semantic properties of a verb’s arguments or, respectively, to the thematic relations that they bear in dependence of particular (case/theta role assigning) verbs, and since I do not believe that recourse to standard theta labels is a promising way to account for such a hypothesis, I will use Dowty’s proto-role concept, cf. Dowty (1991). Instead of assigning discrete theta roles to a predicate’s respective argument(s), Dowty (1991:547; 552) assumes [...] a theory in which the only roles are two-cluster-concepts called PROTO-AGENT and PROTO-PATIENT, each characterized by a set of verbal entailments [...] That is, a role type like ‘Agent’ is defined semantically as whatever entailments of verbs about NP referents are shared by the verbal argument-positions that we label with the term ‘Agent’ (and excludes whatever is entailed for those arguments that differs from one verb to the next).

According to such an assumption, verbal arguments bear a certain (theoretically freely combinable) amount of proto-agent and proto-patient properties. In (8i/ii), Dowty’s (1991:572) characterisation of the proto-agent and proto-patient role, respectively, is given: ([8i]) Contributing properties for the Agent Proto-Role: a. volitional involvement in the event or state b. sent[i]ence (and/or perception) c. causing an event or change of state in another participant d. movement (relative to the position of another participant) (e. exists independently of the event named by the verb) ([8ii]) Contributing properties for the Patient Proto-Role: a. undergoes change of state b. incremental theme c. causally affected by another participant d. stationary relative to movement of another participant (e. does not exist independently of the event, or not at all)

Dowty’s idea can be best subsumed by rendering his ‘Argument Selection Principle’ and the respective corollaries (Dowty 1991:576): ([9]) ARGUMENT SELECTION PRINCIPLE: In predicates with grammatical subject and object, the argument for which the predicate entails the greatest number of Proto-Agent

58 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

properties will be lexicalized as the subject of the predicate; the argument having the greatest number of Proto-Patient entailments will be lexicalized as the direct object. ([10]) COROLLARY1: If two arguments of a relation have (approximately) equal numbers of entailed Proto-Agent and Proto-Patient properties, then either or both may be lexicalized as the subject (and similarly for objects). ([11]) COROLLARY2: With a three-place predicate, the nonsubject argument having the greater number of entailed Proto-Patient properties will be lexicalized as the direct object and the nonsubject argument having fewer entailed Proto-Patient properties will be lexicalized as an oblique or prepositional object (and if two nonsubject arguments have approximately equal numbers of entailed P-Patient properties, either or both may be lexicalized as direct object). ([12]) NONDISCRETENESS: Proto-roles, obviously, do not classify arguments exhaustively (some arguments have neither role) or uniquely (some arguments may share the same role) or discretely (some arguments could qualify partially but equally for both protoroles).

Note that here and in the following +/-oblique case is related to the Keenan/Comrie (1977) grammatical function hierarchy: SU > DO > IO > OBL and the well-known case markedness hierarchy: Nom/Abs > Acc > Dat > PP/oblique (see e.g. Meinunger 2007). Before going on, I would like to point out that I do not believe that a protoagent or proto-patient property is crucial for an element to be dropped. Instead, what seems much more important to me is that NPs that are “embedded” under predicates whose case assignment properties are predictable (i.e. structurally determined) are easier to drop than NPs that are “embedded” under predicates whose case assignment properties are unpredictable or seemingly idiosyncratic.37 The assumption that atypical/non-standard distribution and arrangement of proto-properties, e.g. being a non-subject argument with proto-agent properties (cf. Blume 1998 for (dative) arguments of interaction verbs), correlates with atypical thematic relations (and hence, oblique case assignment) is consistent with the fact that topic drop of obliquely cased NPs is less accept-

|| 37 Cf. Haider (2010:252) who notes the following: “In German there are three morphologically different ways of specifying the licensing relation for an object in the argument structure. First, the argument may be unspecified for a specific case in the lexical argument structure. This is what we refer to as a structural case relation. Second, it may be specified for a specific case. This is the lexical case, and it is invariant. Third, the argument may be lexically determined as a category with a specific case licenser. This is what we are used to calling a prepositional object. The preposition is determined by the selecting verb.”

Structural Well-Formedness Conditions on VTD and NVTD | 59

able/subject to stricter well-formedness conditions than topic drop of structurally cased NPs. If we find some (probably semantic) prerequisites for lexical case assignment, we will be in a situation in which we can explain and predict the reason why and when lexically cased topic drop gaps are well-formed (or not). I.e., by determining (semantic) reasons for a predicate’s particular case selection properties (e.g. genitive case instead of dative case or structural case), “idiosyncratic” case will turn into regular (i.e. predictable) case, irrespective of the fact how it is assigned, cf. structurally (on syntactic grounds) or lexically (on semantic grounds). Thus, I intend to link the particular case assigning property of a (verbal) predicate to its semantics – according to standard assumptions this link is mediated by a thematic relation between the predicate itself and its (internal) argument(s), cf. the discussion above and Bierwisch (2006). Consequently, I suppose that the reason for the different behaviour of dative and genitive assigning verbs in topic drop contexts (i.e. more/less “identity” of the context and target predicate as a well-formedness condition on the gap) is not immediately linked to these predicates’ case assignment properties, but due to their semantics which is reflected via their particular thematic relations (by which dative vs. genitive (vs. PPs) assigning verbs are characterised). I hypothesise that case assignment is a secondary and mediated reflection of the semantics of verbs, mediated through the particular proto-properties assigned to an argument. Although, as is common knowledge, such a link seldom follows a one-toone correspondence,38 I will proceed by elaborating the idea that case assignment is a reflection of the amount of proto-agent and proto-patient properties of

|| 38 McFadden (2006:55f) summarises the main point of the problem (by discussing a synonymous nom/acc vs. nom/dat verb pair): “While there are clear patterns that need to be accounted for, it has proved extremely difficult to come up with a single semantics that would apply to all and only dative arguments in a language like German. For example, there are wellknown pairs of (nearly) synonymous verbs which differ in their case assignment. So both gratulieren and beglückwünschen mean ‘congratulate’, but the former takes a sole dative object, while the latter takes a normal accusative direct object. It proves quite difficult to show that the objects of these verbs receive clearly distinct semantic roles, and even more difficult to show that the relevant distinction consistently patterns with case-marking with other verbs in the language. It was for this reason that researchers shied away from tying inherent case to specific θ-roles in GB. E.g., Haider (1985:80) claims that ‘lexical Case on arguments is not tied to unique thematic values’. He explicitly contrasted this with the semantic case on non-argument DPs like the German accusative of extent or the various local cases in a language like Finnish, which are tied to specific roles.”

60 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(verbal) arguments, which, in turn, is a consequence of the event structure denoted by a (verbal) predicate. Thus, I will examine whether the differences between topic drop of structurally cased and obliquely cased gaps are a reflex of the semantic properties that a predicate assigns to its arguments: In particular, I will investigate a hypothesis according to which verbs that assign +oblique case to their arguments have a more complex (and hence: more specific) meaning than -oblique case assigning verbs. If this is true, it should be reflected by the fact that, with regard to synonymous verb pairs, nom/acc variants are possible in more (syntactic and semantic) contexts than nom/dat or nom/gen variants and which is – as I will subsequently show – indeed the case.39 I will argue that the relevant level at which ‘complexity of meaning’ and ‘markedness of case’ is determined, is event structure. I.e., I will claim that (structural vs. lexical case assigning) predicates can be differentiated with respect to the complexity of events they denote, cf. Blume (1998, 2000)40 as well as Ramchand (2008) who observes and discusses such a correlation as well. Consequently, I do not believe that the acceptability differences between structurally and lexically cased topic drop gaps should be explained by different (case-specific) syntactic licensing conditions (which are as such meaningless). Instead, I assume that acceptability differences between structurally and obliquely cased topic drop gaps emerge from additional, semantic specifications of obliquely cased arguments in contrast to structurally cased arguments. In particular, I suppose that the obligation for obliquely cased gaps to pattern verbatim is due to the fact that the only way to identify (and “spell out”) the specific semantics of the gap is via reference to the contextually given antecedent/context predicate, thereby ensuring semantic identity of the here (context) and there (target) involved thematic relation (cf. micro theta role identity). In order to capture the topic drop data (in semantic terms), I will try to establish a link between the respective argument’s case features, the (combination of) proto-role(s) as well as the involved verbs’ event semantics (in the spirit of Blume 1998, 2000). I will assume a correlation of case/case properties and verbal semantics as indicated in the following table/scale, cf. (18):

|| 39 See Ackerman & Moore (2001) for similar tests (minimal pairs) examining the proto-patient properties of direct, indirect and oblique objects. 40 Thanks to Leah Bauke who drew my attention to Blume’s work.

Structural Well-Formedness Conditions on VTD and NVTD | 61

(18)

Correlations: case – proto-role – event structure

Nom/Acc

Dat

Gen

verb assigns -oblique case to nonsubject

...

verb assigns +oblique case to nonsubject

the nonsubject’s proto-properties are primitive

...

the nonsubject’s proto-properties are combined/complex

verb denotes a simple event

...

verb denotes a complex event

By showing that the meanings that are associated with predicates that assign structural case are more prototypical than those of verbs that assign case lexically/idiosyncratically we can e.g. explain why Funktionsverbgefüge (‘function verb structures’) in German occur predominantly with predicates that assign structural case to their direct arguments, e.g. einen Antrag stellen (to apply for; literally: to put a proposal), eine Entscheidung treffen (to take a decision; literally: to meet/hit a decision), cf. Winhart (2002:18): In der Regel trägt die Nominalphrase den Kasus Akkusativ, aber einige Kombinationen aus FV [= Funktionsverb(en), ET] mit Nominalphrasen in Dativ, Genitiv und Nominativ werden in der Literatur zumindest als Randfälle ebenfalls zu den FVG [= Funktionsverbgefüge, ET] gezählt: [Translation: Normally, the noun phrase is assigned the accusative case. However, some combinations of FV [= function-verb(s), ET] with noun phrases in dative, genitive and nominative case are included as peripheral cases to the FVG [= function-verb structure, ET] in the literature.] ([19]) a. b. c.

Wir unterziehen den Doktoranden einer Prüfung. We subject the PhD-student a test-DAT Dieses Thema bedarf noch einer genaueren Untersuchung. This topic requires still a more-detailed investigation-GEN Zwischen den Del[e]gierten besteht keine Übereinstimmung Between the delegates exists no consensus-NOM

I.e., according to our hypothesis we would predict that predicates that can act as Funktionsverben (light verbs) are characterised by a non-idiosyncratic semantics, i.e. by a verbal semantics that is semantically so underspecified that it can host a very wide semantic range of arguments.

62 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

2.4.3 A Comparison of Synonymous Verbs with Different Case Assignment Properties The main idea of Dowty (1991) has been presented in the above chapter. Ackerman & Moore (2001) discuss Dowty’s corollaries and show (within a couple of unrelated languages) that, given two alternating object realisations, it is always the case that the more obliquely marked variant involves less proto-patient properties. Smith’s (1987) and Blume’s (1998) idea goes in another (however similar) direction: They both claim that the dative argument of nom/dat verbs is typically characterised by the fact that it displays some proto-agent properties (which goes back to Wegener 1985:309: “Die Dativform enthält [nur] das semantische Merkmal /mitwirkend am Vorgang/.”41). In the following I will focus on the +proto-agent properties of nonsubject arguments (instead of focussing on their ‑proto-patient properties as Ackerman & Moore 2001 do). I will illustrate this idea by a quote from Smith (1987:378f) and Blume (1998:269), respectively: [the dative marked entity, ET] is not construed as totally passive or patient-like in the conceived situation, but rather as actually or potentially potent (secondary agent) in its own right. In other words, DAT makes salient not only the passive aspects of the entity in question, but its implied subject-like quality as well, thereby evoking the notion of bilateral involvement. (Smith 1987) These verbs [i.e. ʻinteraction verbsʼ, ET] are analyzed as expressing two subevents, at most one of them being implied. Each of the two participants of these verbs is engaged as an agent in one of the subevents. […] the event denoted by ‘help’ was analyzed as decomposed into two subevents where an agent A with the proto-agent property “sentience” is presupposed to strive consciously for some aim (s1) [subevent 1, ET] while a proto-agent B performs an unspecified act that contributes to the achievement of A’s aim (s2) [subevent 2, ET]. (Blume 1998).

Blume (2000:158) gives the following definition of interaction verbs: Als Interaktionsverben bezeichne ich Verben, die Situationen bezeichnen, in denen zwei Agens-Partizipianten unabhängig voneinander aktiv sind. [Translation: Interaction verbs are verbs that denote situations in which two agent participants are active, independently of each other.]

|| 41 Translation: The dative form contains [only] the feature /participating in the event/.

Structural Well-Formedness Conditions on VTD and NVTD | 63

Note that Blume (1998) explicitly argues that the event the subject and the nonsubject argument participate in consists of two subevents (where the nonsubject argument of nom/dat interaction verbs is active/engaged in the first event – which (typically) does not hold for nonsubjects of nom/acc verbs). This follows somehow naturally: When a nonsubject argument bears secondary proto-agent properties then the event denoted by the verb is necessarily complex (i.e. internally structured), because secondary proto-agent properties (require and) induce a structural representation in which the action of its bearer (= the obliquely case marked nonsubject argument) can be described (within the additional subevent). However, before turning to the events denoted by nom/acc vs. nom/dat verbs, I will present some examples of synonymous verbs and show that in most cases the semantics of the more oblique variant (nom/dat), (nom/gen) of a synonymous pair displays some aspect of meaning that the less oblique variant (nom/acc) is unspecified for. Besides, I will investigate whether the objects of nom/obl verbs are indeed more agentive (in terms of animateness) than the objects of nom/struct verbs (cf. the hypothesis of Smith 1987 and Blume 1998, 2000) Let me start by differentiating helfen (a nom/dat verb) from unterstützen (a nom/acc verb). First, note that symmetry in (verbatim) topic drop contexts is given: (19) a. b.

(20) a. b.

Gestern hab ich dem Hans beim Putzen geholfen. Yesterday have I the Hans-DAT at cleaning helped _ Hab ich heute beim Abspülen unterstützt. [ACC] have I today at washing-up supported Gestern hab ich den Hans beim Putzen unterstützt. Yesterday have I the Hans-ACC at cleaning supported _ Hab ich heute beim Abspülen geholfen. [DAT] have I today at washing-up helped

However, even if helfen and unterstützen count as synonyms, the following examples show that their (non)subject arguments are subject to different animacy restrictions:

64 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

* Dieses Buch unterstützt mich.42 This book supports me-ACC b. Dieses Buch hilft mir. This book helps me-DAT

(21) a.

(22) a.

Dieses Buch unterstützt meine Arbeit. This book supports my work-ACC b. * Dieses Buch hilft meiner Arbeit. This book helps my work-DAT

(23) a. b.

Ich unterstütze den Hans. I support the Hans-ACC Ich helfe dem Hans. I help the Hans-DAT

(24) a.

Ich unterstütze diese Publikation. I support this publication-ACC b. * Ich helfe dieser Publikation. I help this publication-DAT

The behaviour of unterstützen and helfen with regard to animacy/agentivity is illustrated in table (25): (25)

unterstützen vs. helfen (animacy/agentivity)

unterstützen (nom/acc)

*

helfen (nom/dat)

Subject

Object

-animate

+animate



Subject

Object

-animate

+animate

|| 42 Note that similar examples which contain a subject that refers to entities which include +animate/+human elements are well-formed: Diese Organisation (this organisation) / ?Dieses Projekt (this project) unterstützt mich. This speaks for a finer grained differentiation of animacy/agentivity properties of verbal arguments. Moreover, the contrast between unterstützen and helfen becomes also evident in sentences as (ia) and (ib), where die Bank (the bank/bench) is +/-ambiguous, cf. Helmut Weiß (p.c.): (i) a. Die Bank unterstützt mich. The bank/*bench supports me-ACC b. Diese Bank hilft mir schon mal. This bank/bench helps me-DAT once a-time

Structural Well-Formedness Conditions on VTD and NVTD | 65

unterstützen (nom/acc)

helfen (nom/dat)



-animate

-animate



+animate

+animate



+animate

+animate



+animate

-animate

*

+animate

-animate

*

-animate

-animate

The relevant generalisation for unterstützen and helfen can be formulated in terms of proto-agent properties of the respective arguments: helfen demands its nonsubject (i.e. dative) argument to display proto-agent properties, cf. Blume (1998). As the data (22b) and (24b) do not meet this demand, they are ungrammatical. With regard to unterstützen the generalisation must be: The subject is not allowed to display less proto-agent properties than the direct object. If it does, the sentence gets ungrammatical, cf. (21a). Consider now the behaviour under (recipient) passive: both arguments (that act as nonsubjects in the respective active sentences) can be passivised. (26) a. b.

Der Hans bekam geholfen. The Hans-NOM get helped Der Hans wurde unterstützt. The Hans-NOM was supported

With regard to further semantic differences unterstützen and helfen are difficult to keep apart. Consider e.g. depictive interpretations of secondary predication (cf. Haider 1997, Maling 2001, McFadden 2006). For two-place nom/acc verbs the interpretation of secondary predicates is ambiguous, i.e. subject- or objectoriented, whereas it is obligatorily subject-oriented in two-place nom/dat verbs, cf. (27): (27) a. b.

Hansi malte siem nackti/m. (possible readings: H./she was naked) Hans-NOM painted her-ACC naked Hansi half ihrm nackti/*m. (only reading: H. was naked) Hans-NOM helped her-DAT naked

However, looking at the pair unterstützen/helfen as well as at another pair, namely beglückwünschen/gratulieren, we see that in the nom-acc variant the ambiguity (nearly) disappears: in both cases the secondary predicate relates to the subject (as it is obligatory in dative assigning two-place predicates):

66 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(28) a. b.

(29) a. b.

Hans unterstützte sie nackt. (preferred reading: H. was naked) Hans-NOM supported her-ACC naked Hans half ihr nackt. (only reading: H. was naked) Hans-NOM helped her-DAT naked Hans beglückwünschte sie nackt. (preferred reading: H. was naked) Hans-NOM felicitated her-ACC naked Hans gratulierte ihr nackt. (only reading: H. was naked) Hans-NOM congratulated her-DAT naked

The secondary predication data further support the assumption that an increased amount of proto-agent properties is not a peculiar property of nom/dat verbs, but can also constitute a (presumably atypical) property of nom/acc verbs – which may be one reason why it is so hard to separate typical nom/dat verbs from nom/acc verbs on (event-)semantic grounds. Blume (1998:255) notes that within the class of ʻinteraction verbsʼ that she has investigated43 some are nom/acc verbs (Blume 1998 mentions jagen, to hunt). Besides the secondary predication data it seems that, typically, nom/acc verbs involve less personal commitment/dedication and less immediate action of the subject argument than the respective nom/dat counterparts.44 This is illustrated by the following minimal pairs (which involve very subtle differentiations): (30) a. (?) Ich werde dir mit Geld helfen. I will you-sg-DAT with money help b. Ich werde dich mit Geld unterstützen. I will you-sg-ACC with money support

(impersonal commitment)

(31) a.

? Ich helfe dir durch ein / mit einem Projekt. (impersonal support) I help you-DAT through a/with a project b. Ich unterstütze dich durch ein Projekt. I support you-sg-ACC through a project

|| 43 Note that Blume’s (1998) corpus of interaction verbs includes data from several unrelated languages. I.e., the proto-agent property of dative arguments is not a peculiarity of German. 44 This is in contrast with what Smith (1987:389) found: “Native informants feel that helfen involves physical or financial help, whereas unterstützen stresses more emotional support, though it can also be used for financial help.” (underlining in Smith’s original).

Structural Well-Formedness Conditions on VTD and NVTD | 67

(32) a.

Ich kann dir unmittelbar helfen. I can you-sg-DAT immediately help b. (?) Ich kann dich unmittelbar unterstützen. I can you-sg-ACC immediately support

The above minimal pairs show that unterstützen and helfen indeed have a very similar meaning (cf. the secondary predication data). However, they also indicate that the dative object of helfen displays more proto-agent properties than the accusative object of unterstützen, cf. Blume (1998) and the summary in table (33):45 (33)

unterstützen vs. helfen (summary)

unterstützen (nom/acc)

helfen (nom/dat)

subject

+/-animate

+/-animate

object

+/-animate

+animate

passive

syntax ok; semantics ok

recipient passive

fine-grained semantics

No restrictions (possibly: no personal engagement of the subject argument)

+personal engagement of the subject argument (instead of abstract, e.g. sole financial support)

By looking at the pair gratulieren (to congratulate) and beglückwünschen (to felicitate) one could be tempted to see this synonymous pair as counterevidence for the hypothesis that the more oblique variant is possible in less contexts because, as it seems, in all contexts in which beglückwünschen is allowed, gratulieren is allowed as well – but not conversely. In particular, it seems that (more than gratulieren) beglückwünschen demands that its nonsubject argument has fulfilled something with a will of its own.46 So it is impossible to felicitate someone/somebody (beglückwünschen) on

|| 45 Consider also the following contrast, (ia) vs. (ib), involving a situation in which the object actant is not active: (i) a. # Ich half dem Hans, während er schlief. I helped the Hans-DAT while he was-sleeping b. Ich unterstützte den Hans, während er schlief. I supported the Hans-ACC while he was-sleeping 46 Consider Smith’s parallel finding (1987:388f): “Native speakers prefer the latter verb [beglückwünschen, ET] for use in the context of a real achievement. In other words, the verb beglückwünschen is felt to be more formal than gratulieren and thus evokes a less personal

68 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

his/her birthday, cf. (33b). On the other hand it is possible to felicitate someone on having a baby (which involves some engagement), cf. (34b): (33) a.

Ich gratuliere Dir zum Geburtstag. I congratulate you-sg-DAT to-the birthday b. * Ich beglückwünsche dich zum Geburtstag.47 I felicitate you-sg-ACC to-the birthday

(34) a. b.

Ich gratuliere Dir zur Geburt des Kindes. I congratulate you-DAT to-the birth of-the child-GEN Ich beglückwünsche dich zur Geburt des Kindes.48 I felicitate you-ACC to-the birth of-the child-GEN

Thus, sometimes only extremely subtle distinctions make the difference between nom/acc and nom/dat verbs. However, even if it is difficult to find some differences, and the predicate pair gratulieren/beglückwünschen can indeed hardly be distinguished with respect to the proto-agent properties of the object, one could try to distinguish it with respect to the proto-agent properties of the respective subject.49, 50 If this is true one would expect that nom/acc verbs whose nonsubject arguments are characterised by displaying a great amount of protoagent properties have subjects that compensate for this surplus. One could argue that otherwise (when such a surplus is not counterbalanced by the nominative subject) such verbs become dat/nom (experiencer) verbs. I will come back to this idea in a moment, when discussing the next verb pair. The next pair I will focus on is besitzen (to own/possess, nom/acc) vs. gehören (to belong, dat/nom).

|| reaction on the part of the person congratulated. To use beglückwünschen for a birthday would sound overly formal or hyperpolite.” (underlining in Smith’s original). 47 Note that a nominalised form is fine, cf. Geburtstagsglückwünsche. 48 Thanks to Lidija Marjanovic for providing me with this minimal pair. 49 Smith (1987) compares the minimal pair (i) Ich helfe dir beim Häuserbauen (I help you-DAT at house-building) and (ii) Ich unterstütze dich beim Häuserbauen (I support you-ACC at housebuilding) and notes: (on pp. 389-390) “The contrast between the use of helfen in ([i]) and the use of unterstützen in ([ii]) lies in the fact that in the former a cooperative physical activity is profiled, as opposed to the more one-sided situation in the latter, where the NOM entity supports or props up an activity (financially and/or emotionally) which might not continue otherwise.” (underlining in Smith’s original). 50 See also Blume (2000:62-67) for a discussion (which relates to Wegener 1985) on this correlation.

Structural Well-Formedness Conditions on VTD and NVTD | 69

First, with respect to this pair, the data in (36)/(37) confirm the assumption that not identity of case but identity of semantically enriched (micro) theta roles is the relevant factor with respect to the well-formedness of verbatim topic drop (of obliquely cased NPs): (36) A: B:

(37) A: B:

Der Hans besitzt diese Firma. The Hans-NOM owns this company-ACC _ Gehören auch noch andere Firmen. [DAT] belong also even other companies-NOM Dem Hans gehören viele Firmen. The Hans-DAT belong many companies-NOM _ Besitzt auch diese Firma. [NOM] owns also this company-ACC

Note that the well-formedness of data as (36)/(37) is not exceptional: (38) A: B:

(39) A: B:

Hans befürchtete schon den Super-GAU. Hans-NOM feared already the ultimate-MCA-ACC Ja, _ schwante im wahrsten Sinne des Wortes das Schlimmste. Yes, [DAT] sensed in the purest sense of-the word the worst-NOM Dem Hans schwante im wahrsten Sinne des Wortes das Schlimmste. The Hans-DAT sensed in the purest sense of-the word the worst-NOM Ja, _ befürchtete schon den Super-GAU. Yes, [NOM] feared the ultimate-MCA-ACC

Consider also possible restrictions with respect to the animacy/agentivity of the arguments of besitzen (nom/acc) vs. gehören (dat/nom): (40) a.

* Dieser Fernseher / Job besitzt den Hans. This TV/job-NOM owns the Hans-ACC b. * Diesem Fernseher / Job gehört der Hans. This TV/job-DAT belongs the Hans-NOM

(41) a.

Dieser Fernseher besitzt eine Fernbedienung. This TV-NOM owns a remote-control-ACC b. * Diesem Fernseher gehört eine Fernbedienung. This TV-DAT belongs a remote-control-NOM

70 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(42) a. b.

(43) a. b.

Diese reiche Dame besitzt einen Sklaven. This rich lady-NOM owns a slave-ACC Dieser reichen Dame gehört ein Sklave. This rich lady-DAT belongs a slave-NOM Der Hans besitzt eine Firma. The Hans-NOM owns a company-ACC Dem Hans gehört eine Firma. The Hans-DAT belongs a company-NOM

Again, as above, the respective animacy restrictions are summarised in a table: (44)

besitzen vs. gehören (animacy/agentivity)

besitzen (nom/acc) Subject

gehören (dat/nom) Object

Logical Subject

Logical Object

*

-animate

+animate

*

-animate

+animate



-animate

-animate

*

-animate

-animate



+animate

+animate



+animate

+animate



+animate

-animate



+animate

-animate

The results show that the logical subject of gehören (realised as dative argument) must be +animate, as expected. If this is not the case, ungrammaticality results, cf. (40b)/(41b). With respect to besitzen we can state that the subject argument is not allowed to display less proto-agent properties than the object argument (cf. also the pair helfen vs. unterstützen). If it does, the sentence gets ungrammatical, cf. (40a). In accordance with the above discussion on the exceptional case marking of gehören one could hypothesise that the switch from a nom/acc verb to a dat/nom verb might have taken place to indicate that the (internal) nonsubject argument is more agentive than the (external) subject argument (as there is no (other) way for the nominative subject to compensate for the surplus of the agentive properties of the (regularly accusative marked) nonsubject argument).51

|| 51 Note that a proper hypothesis about a “real” switch from a nom/acc to a dat/nom verb would make it necessary to know the form and the properties of the precursor. This, however,

Structural Well-Formedness Conditions on VTD and NVTD | 71

Note that recipient passivation of the dative object of gehören is at odds with the fact that this dative argument is the logical (however not nominative marked) subject of the sentence, cf. (45b): (45) a.

Dieser Sklave wurde von vielen reichen Damen besessen. This slave-NOM was by many rich ladys-DAT owned b. * Viele reiche Damen kriegen/bekommen von diesem Sklaven gehört. Many rich ladies-NOM get/become by this slave-DAT belonged

As to further arbitrary semantic differences we can state that in contrast to besitzen, gehören expresses ‘+emotional affinity’ with respect to the object which is possessed, cf. (46), whereas besitzen is better in formal contexts (e.g. court language) although, of course, in no way restricted to them: (46) a.

# Ich besitze meinen Bauch. I-NOM own my belly-ACC b. Mein Bauch gehört mir.52 My belly-NOM belongs me-DAT

Except for (46), we can conclude that (again) the dative variant of two synonymous predicates is slightly more restricted than the accusative variant, the latter being more neutral in contexts in which the dative variant can hardly occur, cf. animacy, fine-grained verbal semantics. See the general summary in the table beneath: (47)

besitzen vs. gehören (summary)

besitzen (nom/gen)

gehören (dat/nom)

Subject

+/-animate

+/-animate (logical object)

Object

+/-animate

+animate (logical subject)

Passive

syntax ok; semantics ok

*recipient passive

|| cannot be provided in our example. Nonetheless, it seems to be a property of dat/nom verbs that their logical subject (the dative marked argument) displays more proto-agent properties than the respective nominative argument in nom/acc verbs, cf. (41a) vs. (41b). One could speculate that such a ‘movement of case features’ takes place when nominative marking is not sufficient to indicate an extremely big amount of proto-agent properties. 52 Motto of a well-known campaign for the freedom of abortion, concerning §218 of the German BGB.

72 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

fine-grained semantics

besitzen (nom/gen)

gehören (dat/nom)

rather abstract possessorpossessum relation

+personal relation with respect to the thing that is possessed (instead of ‘pure’ physical possession)

The next synonymous pair I will discuss is brauchen/benötigen (to need, nom/ acc) vs. bedürfen (to require, nom/gen). As before, I first focus on the well-formedness in topic drop contexts and then on the animacy restrictions of the arguments of these two verbs. (48) A: B:

(49) A: B:

Oma braucht besondere Pflege. Grandma needs special care-ACC _ Bedarf Opa auch. [GEN] requires grandpa as-well Oma bedarf besonderer Pflege. Grandma requires special care-GEN _ Braucht Opa auch. [ACC] needs grandpa as-well

Thus, with respect to topic drop, we can conclude that there is at least one context, in which bedürfen and brauchen are symmetric. Let us now look at possible animacy/agentivity restrictions: (50) a. b.

(51) a. b.

(52) a.

Dieses Projekt / *Diese Publikation braucht mich. This project/This publication needs me-ACC Dieses Projekt / *Diese Publikation bedarf meiner. This project/This publication requires me-GEN Dieser Fernseher braucht eine Reparatur. This TV needs a repair-ACC Dieser Fernseher bedarf einer Reparatur. This TV requires a repair-GEN

Diese reiche Dame braucht einen Liebhaber. This rich lady needs a lover-ACC b. (#)Diese reiche Dame bedarf eines Liebhabers. This rich lady requires a lover-GEN

Structural Well-Formedness Conditions on VTD and NVTD | 73

(53) a.

Ich brauche diesen Fernseher. I need this TV-ACC b. # Ich bedarf dieses Fernsehers. I require this TV-GEN

The above data show that even if the genitive variant is equally restrictive with respect to the animacy of its subject/nonsubject arguments (when compared with the accusative variant), the accusative variant can be differentiated from the genitive variant since it is more neutral with regard to the semantic selection of the kind of nonsubject argument – the pragmatic anomaly of (53b) reflects this fact: As the genitive assigning verb bedürfen implies the existence of a real and urgent need it cannot be used in the context of a superficial desire, cf. (53b)/(52b) vs. (54b): (53) b. # Ich bedarf dieses Fernsehers. I require this TV-GEN (52) b. (#)Diese reiche Dame bedarf eines Liebhabers. This rich lady requires a lover-GEN (54) a. b.

Ich brauche diese Gehhilfe. I need this walking-frame-ACC Ich bedarf dieser Gehhilfe. I require this walking-frame-GEN

The animacy/agentivity restrictions of the arguments of brauchen and bedürfen are summarised in table (55): (55)

brauchen vs. bedürfen (animacy/agentivity)

brauchen (nom/acc)

/*

bedürfen (nom/gen)

Subject

Object

-animate

+animate

/*

Subject

Object

-animate

+animate



-animate

-animate



-animate

-animate



+animate

+animate

/(#)

+animate

+animate



+animate

-animate

/#

+animate

-animate

Thus, we can state that with regard to the arguments it takes, bedürfen is semantically more selective than brauchen. This confirms the supposition that the

74 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

nonsubject argument of bedürfen has a “more specific” semantics. For a summary, consider the table in (56): brauchen vs. bedürfen (summary)

(56)

brauchen / benötigen (nom/acc)

bedürfen (nom/gen)

subject

+/-animate

+/-animate

object

+/-animate

+/-animate

passive

syntax ok; semantics ok

No recipient passive of genitive objects

fine-grained semantics

no restrictions

subject argument is implicitly suffering; expression of a requirement for an apparently “needless” (object-)thing collides with verbal semantics

It is hard to find further nom/acc vs. nom/gen synonyms besides brauchen/ benötigen and bedürfen.53 One could try erinnern54 and sich erinnern. However, besides the restricted local use of erinnern with accusative, topic drop indicates that this pair is not very symmetric: (57) a.

Ich habe diesen Moment immer gerne erinnert. I have this moment-ACC always with-pleasure remembered b. ?* _ Hab ich mich auch immer gerne erinnert. [GEN] have I REFL also always with-pleasure remembered

(58) a.

Ich habe mich dieses Moments immer gerne erinnert. I have REFL this moment-GEN always with-pleasure remembered b. ?* _ Hab ich auch immer gerne erinnert. [ACC] have I also always with-pleasure remembered

|| 53 Note that bedürfen might be indeed special: Winhart (2002:18) notes that bedürfen is one of the rare cases of Funktionsverben (function verbs in the sense of light verbs) which are not nom/acc marked (see 2.4.2). 54 According to ProGr@mm (a reference site of the IDS Mannheim) erinnern as a nom/acc verb is used in some regions, especially in the North and in the Western part of the middle of Germany (“erinnern wird im Nord- und Westmitteldeutschen mit einer NP im Akkusativ im Sinne von sich erinnern verwendet: (i) Jetzt erinnere ich die Episode [‘Now I remember the episodeʼ, ET] (nach Rheinischer Merkur, 08.12.1989, S. 17)”).

Structural Well-Formedness Conditions on VTD and NVTD | 75

Both dialogues, (57) and (58), are rather marginal, i.e., the gap is not wellformed independently of whether it bears accusative or genitive case. Whereas the marginality of the genitive gap in (57) could be explained by a lack of predicate synonymy/identity, this explanation is not an option for the NVTD gap in (58), because a structurally cased gap can be derived from an antecedent with any case. Thus, we have to assume a general (semantic) incompatibility of erinnern and sich erinnern which means that these two verbs may have nonsymmetric meanings (at least in the context given here). However, it is quite obvious that the occurrence of a genitive object implies the presence of a complex event (consisting of a main event and an implicit subevent). Consider e.g. the contrast between (59) and (60): (59) Wir konnten die Probleme einfach nicht lösen. We could the problems-ACC simply not solve. (60) Wir konnten der Probleme einfach nicht Herr werden. We could the problems-GEN simply not master become (= handle). Note that the genitive marked version of ‘problems’, cf. (60), selected by the verb Herr werden (literally: to become master of something (= to have something under her/his control)), contains an agent-like component in the sense of ‘the problems overwhelmed us’ / ‘steadily some new problems arose’, which is not the case with respect to the accusative marked variant of ‘problems’ in (59). Let me close the investigation of synonymous verbs by concluding that there must be good reasons why (in contrast to nom/acc vs. nom/dat synonyms) there are nearly no nom/acc vs. nom/gen synonyms. This, I suppose, is due to the fact that nom/gen verbs have idiosyncratic meanings, and hence, their nonsubject arguments display unpredictable secondary proto-properties (e.g. in contrast to nom/dat verbs the nonsubject argument of nom/gen verbs is not designated as a secondary agent). This is also attested by the minimal pairs in (61) vs. (62) showing that nom/gen verbs are hardly combinable with respect to each other: (61) A: B:

Der Staatsanwalt klagte die Maria des Mordes an. The prosecutor charged the Maria the murder-GEN PRT ?*_ Bezichtigte/_ Beschuldigte der Richter die Maria auch. [GEN] incriminated / [GEN] accused the judge the Maria as-well

76 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(62) A: B:

Der Staatsanwalt beschuldigte die Maria des Mordes. The prosecutor accused the Maria the murder-GEN _ Beschuldigte / ?_ Bezichtigte der Richter die Maria auch. [GEN] accused / [GEN] incriminated the judge the Maria as-well

Although some of the comparisons between structural and lexical case assigning synonymous verbs discussed up to this point are merely slight and tentative indications, they nevertheless confirm the claims made by Smith (1987) and Blume (1998, 2000). Of course, there is a considerable amount of nom/acc predicates that behave like verbs with an obliquely cased nonsubject argument. I.e., some nom/ acc verbs involve a thematic relation which is a peculiar property of (Blume’s 1998) nom/dat interaction verbs. Even Blume herself (1998:255) notes that “not every interaction verb may select a /nom/dat or /abs/dat case frame”. Nom/acc verbs that can be exchanged with nom/dat verbs (cf. the pairs unterstützen/helfen (to support/to help), treffen/begegnen (to meet/to encounter) etc.) in topic drop contexts may have thematic relations of a very similar kind. However, it is probable that this irregularity does not stem from the properties of nom/dat verbs described so far but rather from “irregular” properties of nom/ acc verbs. The following passivation data show that nom/acc verbs that are exchangable with nom/dat verbs are atypical nom/acc verbs, insofar as their nonsubject arguments display proto-agent properties. Accusative arguments of such nom/acc (interaction) verbs, cf. e.g. hören (to hear), treffen (to meet), verstehen (to understand) can hardly be passivised, cf. (63) – in contrast to nom/acc verbs whose nonsubject argument does not display any proto-agent properties, cf. e.g. sehen (to see), quälen (to torment), cf. (64): (63) a.

? Der Hans wurde von Maria gehört. The Hans-NOM was by Maria heard b. * Der Hans wurde von Maria getroffen.55 The Hans-NOM was by Maria met c. ? Der Hans wurde von Maria verstanden. The Hans-NOM was by Maria understood

|| 55 Note that the sentence is well-formed when the (ambiguous) verb treffen means ‘to hit/ strike’.

Structural Well-Formedness Conditions on VTD and NVTD | 77

(64) a. b.

Der Hans wurde von Maria gesehen. The Hans-NOM was by Maria seen Der Hans wurde von Maria gequält. The Hans-NOM was by Maria tormented

Assuming that through passivation a nonsubject that becomes a subject receives additional agent properties through overt nominative case marking it is to be expected that an argument which already has secondary agent characteristics, cf. (65), cannot be passivised, cf. (63). Whereas, when a nom/acc verb is not an interaction verb, cf. (66), passivation is no problem, cf. (64). (65) a. b. c.

(66) a. b.

Die Maria hört den Hans. (H. is e.g. making some noise) The Maria-NOM hears the Hans-ACC Die Maria trifft den Hans. (H. must be active in order to meet M.)56 The Maria-NOM meets the Hans-ACC Die Maria versteht den Hans. (H. must be explaining something) The Maria-NOM understands the Hans-ACC Die Maria sieht den Hans. The Maria-NOM sees the Hans-ACC (H. can be passive/dead) Die Maria quält den Hans. The Maria-NOM torments the Hans-ACC (H. can be passive/almost dead)

Thus, the property of accusative arguments of particular (interaction) verbs to resist passivation could be attributed to semantic reasons and explained by the fact that these arguments already display a great amount of proto-agent properties. An analogous explanation could also be relevant for dative arguments which display a great amount of proto-agent properties and disallow (recipient) passivation as well, e.g. the nom/dat verbs dienen (to serve), gehorchen (to obey), folgen (to follow), danken (to thank). By contrast, dative arguments that display a relatively little amount of proto-agent properties, selected by verbs as amputieren (to amputate), widmen (to devote), die Leviten lesen (to lecture

|| 56 When contrasting the nom/acc verb treffen (to meet) with the near-synonymous nom/dat verb begegnen (to come-across), it turns out that the nonsubject argument of treffen is more proto-agent like than the nonsubject argument of begegnen, i.e., theoretically, treffen rather than begegnen should be dative marked (as the nonsubject argument of begegnen bears less proto-agent properties than the nonsubject argument of treffen).

78 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

somebody), gratulieren (to congratulate) etc. can easily receive nominative case in a recipient passive sentence. However, I will not go into further details of passivation and so I will remain a bit speculative at this point. In addition to the above comparison of synonymous verbs and the argumentation of Blume (1998, 2000) which will be discussed in more detail in the next chapter, Czypionka et al. (2012) and Czypionka (2013) conducted psycholinguistic research which shows that animate57 arguments are processed more slowly when the nonsubject argument is accusative marked than when it is dative marked. In particular, Czypionka et al. (2012) detect reading time differences with respect to the following data: Condition (1) a/b: accusative verb with (a) inanimate or (b) animate object NP Condition (2) a/b: dative verb with (a) inanimate or (b) animate object NP (67) Peter sagt, dass Studentinnen Peter says that students-fem,pl(nom) (a) Vorlesungen / (b) Professoren (a) lectures-pl(inanim,acc=dat) / (b) professors-pl(anim,acc=dat) begeistert (1) loben / (2) applaudieren, enthusiastically (1) praise / (2) applaud und Ida sagt das auch. and Ida says that too ‘Peter says that students enthusiastically (1) praise / (2) applaud (a) lectures / (b) professors, and Ida says so, too.’ Czypionka et al. (2012) note: “Self-paced reading times indicate differences in the effect of object animacy between verb classes, with reading times on the postverbal word ( und ) about 15 ms longer in 1(b) than in 1(a), but no difference between 2(a) and 2(b).” They conclude: “The interaction between a verb’s case assignment pattern and the animacy of its arguments causes significant behavioural and physiological effects, even when case marking is not morphologically overt.” We can interpret the findings of Czypionka et al. (2012) by saying that seemingly idiosyncratic/oblique case (dative) marking is associated with certain typical properties and a departure from oblique case marking is perceived as atypical, causing additional processing costs – costs which mirror the extra|| 57 Blume (2000:53f) (in my view) correctly notes that as a crucial criterion in the context of nom/acc vs. nom/dat patterns ‘animacy’ should be replaced by ‘agentivity’.

Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs | 79

effort that is needed to “assign” agentivity (which correlates with animacy) to accusative marked arguments. Given this result it is probable that case (even if not overtly marked) indicates that a nonsubject displays/lacks particular protoproperties.

2.5 Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs In order to find out why obliquely cased and structurally cased NPs behave differently in topic drop structures (cf. the results of the survey in 2.3), I will hypothesise that case and proto-role(s) of a nonsubject argument are linked to the kind of event (+/-complex) denoted by the verb. To present the whole idea ʻin a nutshellʼ I assume that for actants of verbs the involvement within complex events implies that they have to bear multiple proto-properties. The overt marker of these multiple proto-properties is oblique case assignment. I.e., oblique case marking is the spell-out of a certain semantic content. Due to a general demand/principle that meaningful content cannot be absorbed/has to be spelled out at PF, these specific properties have to be spelled out as well – which, however, is impossible under topic drop. Therefore, the semantic properties of obliquely cased dropped NPs have to be ascertained/ reconstructed via semantic identity with their antecedents/the context predicate. Thus, obliquely cased gaps evade the (in principle necessary) obligation to be spelled out by establishing a semantic identity relation with this/these element(s) which can ensure the reconstruction of the features the gap is lacking.

2.5.1 Empirical Findings, Hypotheses and Theoretical Assumptions In the following I will be mainly concerned with predicates that assign their arguments case on a (seemingly) idiosyncratic basis – i.e. two-place nom/dat and nom/gen verbs and three-place nom/acc/dat and nom/acc/gen verbs. I will contrast the characteristic properties of these verbs with the properties of verbs that assign their arguments case on a structural (or better: non-idiosyncratic) basis. The theoretical framework which the ideas presented here are based on is (mainly) the work of Smith (1987), Lenz (1997), Blume (1998), (2000) and Engelberg (1995), (2005) who builds up on Pustejovsky’s (1991) (sub)event semantics.

80 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

However, as will be seen, the event semantics I am proposing makes use of more primitive distinctions than Engelberg’s (and Pustejovsky’s). For a clearer idea of what is intended, consider the following (extensive) quotes from Engelberg (1995) and (2005) on the general framework (I will refer to in some parts), as well as a quote from Lenz (1997) on nom/acc and nom/gen verbs and one from Blume (1998) on nom/dat verbs: Engelberg (1995:40) An event structure is a representation of the verb’s meaning that integrates the thematic arguments of the verb into a description of the events that the verb can refer to. The main characteristics of an event structure are: a) verbs refer to complex situations that consist of subevents; b) these subevents are of different types (punctual events, durational events, states);58 c) the event participants corresponding to the thematic arguments of the verb are involved in different subevents; d) semantic functions relate participants and subevents. Engelberg (2005:263) Lexical entries of verbs have to reflect the fact that the participants of the event the verb refers to are not necessarily involved in all subevents, but only in some of them, and that they are involved in different subevents in a different manner. [According to Engelberg 1995:38 this finding goes back to Grimshaw (1990)]. Lenz (1997:22) Grundlegend unterscheiden sich Situationen dadurch, ob sie sich als komplex (aus mehreren Teilsituationen bestehend) oder als nicht-komplex (aus nur einer Teilsituation bestehend) beschreiben lassen. Im Bereich der genitivfähigen Verben finden sich komplexe Situationen nur bei G[enitiv]- und D[ativ]-Verben, allerdings gibt es auch in diesen beiden Gruppen Verben, die nicht-komplexe Situationen denotieren. Die A[kkusativ]- und P[P]Verben jedoch denotieren ausschließlich nicht-komplexe Situationen. [Translation: Situations basically differ in whether they can be described as complex (consisting of several situations) or non-complex (consisting only of one situation). In the field of genitive assigning verbs complex situations can only be found with [G]enitive- and [D]ative verbs. However, these two groups again include verbs that denote non-complex situations. The A[ccusative]- and P[P]-verbs however denote exclusively non-complex situations.] Blume (1998:254f) I will show that agentive /nom/dat verbs denote complex events consisting of more than one subevent — one of them typically presupposed — where each participant in the complex event is independently active in at least one of the subevents. Two-place (or more) verbs that denote such complex events will be called “interaction verbs”. The complex event denoted by helfen, for example, consists of two temporally overlapping subevents s1

|| 58 This differentiation does not play a role in my work.

Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs | 81

and s2: in s1 one participant is presupposed to strive consciously for a certain aim (where sentience is one of the properties that constitute the proto-agent role [...]) and in s2 the other participant performs an unspecified act that contributes to the achievement of the aim of the first one. It will be argued that all verbs with /nom/dat or /abs/dat case frames whose nominative absolutive argument is a proper agent belong to the class of interaction verbs. […] [It was] Engelberg’s (1996) finding that arguments of verbs that express complex events do not necessarily participate (or bear the same thematic relation) in each subevent.

I will start by presenting hypothesis (1) in which I argue (according to Lenz 1997 and Blume 1998) that +oblique case is linked to a +complex event structure (of course, the reverse holds as well, i.e. -oblique case is linked to a -complex event structure). Besides linking case and event structure, (1) makes a statement with respect to the well-formedness conditions on structurally and obliquely cased topic dropped arguments: (1)

Hypothesis on verbal meaning, event structure and case (in topic drop dialogues) a. The more complex verbal meaning is, (i) the more internally structured is the associated event, and (ii) the more atypical/manifold are the proto-properties of the nonsubject argument, and (iii) the more probable it is that the nonsubject argument receives oblique case. b. (Null) objects of (verbal) predicates which denote simple events do not bear any additional proto-properties. Therefore they receive structural case (which is meaningless and, hence, can be absorbed under passivation). Due to this, no reference to case/predicate identity (or predicate synonymy) is necessary in topic drop constructions. c. (Null) objects of (verbal) predicates which denote complex events bear multiple proto-roles (which allow a structural representation of the involvement of an actant within different (sub-)events). Therefore such (null) objects receive oblique case (which is meaningful and correlates with case conservation under passivation). Due to this, reference to case/predicate identity (or predicate synonymy) is necessary in topic drop constructions ‒ otherwise the meaningful content (spelled out as case feature) could not be reconstructed.

Thus, according to hypothesis (1), the event structure(s) of the (context and) target predicate determine(s) the proto-properties of the (omitted) argument. These proto-properties (in turn) are crucial/decisive with respect to the case features that such an NP receives and with regard to the fact whether this NP can be topic dropped or not. Topic drop (of obliquely cased NPs) is insofar inter-

82 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

esting as it constitutes a lackmus-test for the (lack of) correspondence of an antecedentʼs and a gapʼs proto-properties (where proto-roles can be understood as an instance mediating between case assignment and event denotation). By looking at the results of our survey (cf. 2.3) and extending our hypothesis in (1), we can predict under which conditions gaps with particular case features can/cannot be topic dropped, cf. (2a-c): (2)

Topic drop, case and the involvement of (multiple) proto-roles (a) Structurally cased (nom, acc) arguments/gaps are not linked to a specific proto-role (cf. case absorbtion under passivation). Thus, their semantic underspecification allows them to be derived from an antecedent with any case. (b) Dative arguments/gaps are linked to a specific proto-role (cf. no case absorbtion under passivation and the findings of Smith 1987 and Blume 1998). Thus, in order to be properly identified, the gap must match the semantic peculiarities borne by the antecedent. However, the predicates in context and target do not have to be identical, i.e. (dative) case identity between antecedent and gap is sufficient. The latter fact is indicative for the presence of a combined proto-role (according to Blume 1998: being an agent in subevent e and a patient in subevent e’) common to (ideally: all) dative cased arguments (which receive dative on a seemingly idiosyncratic/lexical basis). (c) Genitive arguments/gaps should represent a semantically meaningful proto-role as well (cf. no case absorbtion under passivation). However, the fact that topic drop of a genitive marked argument almost always requires the predicates in context and target to be identical (synonymy being rarely sufficient) speaks for the supposition that genitive case assigning verbs do not form a uniform/predictable semantic class. I.e., it is reasonable to assume that the proto-properties of genitive objects are not the same for all genitive case assigning verbs and rather verb-specific.

In the following I will show how the additional proto-properties Blume (1998) speaks of with respect to nom/dat verbs are spelled out within the events denoted by particular verbs. I will show that, generally, the hypothesis in (1) makes the correct predictions in linking thematic roles, event structure and +/-oblique case assignment. As already mentioned, I suppose that a particular (abstract) structuring of events demands a particular case frame (as case features reflect the verbʼs arguments characteristics with regard to bearing particu-

Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs | 83

lar proto-properties). Following Pustejovsky (1991:77) I assume that “[e]ach subevent must be associated with at least one argument position at lexical structure” which captures the obvious fact that for any event there must be an actant that is (actively or passively) involved in this event. I further assume that the reverse holds as well: I.e., every argument is necessarily associated with at least one event. As will be seen later, this also holds for the occurrence of seemingly meaningless expressions such as expletives. I start with examining whether Blume’s (1998:254f) findings/hypotheses with respect to the properties of interaction verbs, cf. (3), are correct: (3)

Event-related characteristics of interaction verbs59 (according to Blume 1998) (i) The dative actant bears proto-agent properties. (ii) The nom/dat verb denotes a complex event consisting of two overlapping subevents. (iii) The involved actants have a common aim.

Let us see whether the characterisation in (3) can cover the events denoted by nom/dat verbs and whether it holds exclusively for nom/dat verbs (or, at least, for interaction verbs). To test this, I will first investigate nom/dat verbs vs. nom/acc verbs.

2.5.2 Investigation of Events Denoted by Nom/Dat Verbs (in Contrast to Events Denoted by Nom/Acc Verbs) I start with examples of event denotations of verbs that display a nom/dat case frame:

|| 59 Blume (1998:254) defines interaction verbs as follows: “[...] agentive /nom/dat verbs denote complex events consisting of more than one subevent – one of them typically presupposed – where each participant in the complex event is independently active in at least one of the subevents. Two-place (or more) verbs that denote such complex events will be called ‘interaction verbs’.” See also Blume’s (2000:158) definition of interaction verbs at the beginning of chapter 2.4.3.

84 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(4)

Ich helfe dem Hans (I help the Hans-DAT)

helfen (to help) (e) Hans (Agent1) does something (that possibly requires help wrt a Theme/Patient) (e’) I (Agent2) help Hans (Patient=former Agent1)

(5)

Ich gehorche/diene… dem Hans (I obey/serve… the Hans-DAT)

dienen/gehorchen/nachgeben/klein beigeben (to serve, to obey, to relent, to (a little) give in) (e) Hans (Agent1) puts some pressure on me (Patient) (e’) I (Agent2=former Patient) do something that Hans (Theme=former Agent1) forced me to do

(6)

Ich folge dem Hans (I follow the Hans-DAT)60

folgen (to follow) (e) Hans (Agent1) moves (e’) I (Agent2) follow Hans (Theme61=former Agent1)

Note that event decompositions as in (4)–(6) can be assumed for further nom/dat verbs as well, e.g. verzeihen (to forgive): When X forgives Y, both, X and Y, wish that something (the thing that was done by Y and is now forgiven by X) has never happened. Inasmuch as this is true, Y bears proto-agent properties and X and Y pursue the same aim within the event denoted by verzeihen.

|| 60 Blume (2000:139) explicitly notes that the dative participant of folgen (to follow) bears proto-agent properties as it actively moves on its own: “Der Dativ Partizipant von folgen hat z.B. im vorliegenden Modell die Proto-Agens-Eigenschaft ‘Aktivität’, da durch die Verbbedeutung präsupponiert ist, daß er sich in dem vom Verb bezeichneten Ereignis selbständig bewegt, vgl. *Ich bin ihm den ganzen Abend gefolgt, denn er hat sich nicht von der Stelle gerührt [I followed him the whole evening because he did not leave this place, ET] und Ich bin ihm an diesem Abend nicht gefolgt, denn er hat sich nicht von der Stelle gerührt. [I did not follow him the whole evening because he did not leave this place, ET].” Note as well that the nonsubject argument of the nom/acc interaction verb verb jagen (to hunt) can be inanimate. In this case it displays no proto-agent properties, cf. Ich jage einen Schatz (I hunt a treasure-ACC), where the particular treasure is not moving from one place to another, but hidden (at a fixed place). 61 Note that patient is not the appropriate label – theme might fit better in this context. However, such detailed theta role labels do not really matter within my approach. In cases where neither proto-agent nor proto-patient properties of an argument prevail (mostly in circumstances where the respective argument in a particular subevent is not agentive, not affected and often not animate) one could assume that such an argument bears a theme role, see also Blume (2000:144).

Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs | 85

Equivalently, this holds for verbs as zustimmen (to agree), antworten (to answer), and many more nom/dat verbs.62 All the interaction verbs mentioned above (cf. Blume 1998:267) can be contrasted with “ordinary” two-place nom/acc verbs which can be characterised by the fact that their nonsubject argument does not take part within an implicit subevent and, hence, it does not display additional proto-properties, e.g. sehen (to see) quälen (to torment) etc.: (7)

Ich sehe den Hans (I see the Hans-ACC)

sehen (to see) (e) I (Agent) see Hans (Theme)

(8)

Ich quäle den Hans (I torment the Hans-ACC)

quälen (to torment) (e) I (Agent) torment Hans (Patient)

However, the situation is often not so clear-cut as above: In some cases the proto-agent property of the dative marked argument is less obvious, i.e. one could suggest that e.g. danken (to thank) is “exceptionally” dative marked: namely, although there is a complex event, it is unclear what the common aim of the agent and the secondary agent could be: (9)

Ich danke dem Hans (I thank the Hans-DAT)

danken (to thank) (e) Hans (Agent1) did something (such that someone (Patient1) is affected) (e’) I (Agent2) am thankful with respect to Hans (Patient2=former Agent1)

However, the subject referent could thank the nonsubject referent even when the latter has done something for the former without knowing her/him (e.g. composing a wonderful song or inventing the fridge etc.), but such an argumentation means stretching the point. Instead, it makes more sense to assume that from all characteristics in (3) the common aim of subject and nonsubject argument is the weakest condition and does not have to be (necessarily) fulfilled. This seems to be true for further predicates as well – applaudieren (to applaud)

|| 62 See Meinunger (2000) for an extensive list of nom/dat verbs in German; see also Mater (1971).

86 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

may also be a predicate which bears an event structure that is atypical for a nom/dat verb because the two (sub)events denoted by the verb are characterised by the fact that their (proto-)agents do not share a common aim but pursue different goals, cf. (10), where Ich applaudiere dem Hans (I applaud the HansDAT) involves an event e in which Hans did something and, as a result, the subject Ich shows him her/his admiration (= event eʼ). In order to not consider applaudieren as an exception of Blume’s (1998) principle (3iii), it could be argued that both actants pursue a common aim, namely the aim that applause is given/deserved. (10) Ich applaudiere dem Hans (I applaud the Hans-DAT) applaudieren (to applaud) (e) Hans (Agent1) did something admirable (such that someone/some patient is affected) (e’) I (Agent2) show my admiration for Hans (Patient=former Agent1)

However, it makes more sense to qualify Blume’s statement in (3iii), which says that the involved actants have a common aim. Such a relativization is also confirmed by the verb folgen (to follow) where it is quite impossible to assume that the person whom one follows shares a common aim with its follower. Besides, there are further verbs that apparently do not meet “prototypical” expectations, e.g. the nom/gen verb bedürfen (to require): As shown above, there are contexts in which bedürfen can simply be exchanged with brauchen (to need), even if bedürfen (to require) has a slightly more specific meaning than brauchen (namely, in contrast to brauchen, bedürfen implies the existence of a real and urgent need and it cannot be used in the context of a superficial desire (as has already been shown above), cf. (i) #Ich bedarf dieses Fernsehers (I require this TV-GEN)). I.e., the fact that a structurally cased NP (selected by brauchen) can serve as an antecedent for an obliquely cased gap (selected by bedürfen) shows that not all oblique case assigning (verbal) predicates are (equally) complex (i.e., bedürfen may possibly be simple(r), see also Winhart 2002). Before going on, I will discuss some possible objections against Blume’s criteria in (3) in the opposite direction, i.e. I will look at nom/acc verbs whose nonsubject argument bears (additional) proto-agent properties. Blume (1998) mentions the verb jagen (to hunt). Another such example is the verb hören (to hear):

Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs | 87

(11) Ich höre den Hans (I hear the Hans-ACC) hören (to hear) (e) Hans (Agent1) performs a perceivable sound (e’) I (Agent2) perceive the sound that Hans (Theme=former Agent1) produces

Further examples are verstehen (to understand), unterstützen (to support) or auslachen (to deride).63 Clearly, these verbs involve (i) the presence of two events and (ii) the presence of secondary proto-agent properties of the nonsubject argument. However, in contrast to most nom/dat verbs it is rarely the case that subject and nonsubject of a nom/acc verb share a common/the same aim cf. e.g. auslachen (to deride). Moreover, nom/acc verbs like bemitleiden (to pity), bedauern (to feel sorry for someone) or beobachten (to observe) seem to involve a complex event as well. However, with regard to these verbs there is no necessity to assume an event in which the respective nonsubject arguments are actively involved. The same holds for the seemingly complex verb verunglimpfen (to disparage). In this case too, the object which is disparaged does not have to be active in order to be disparaged. By contrast, the meanings of helfen (nom/dat) or gedenken (nom/ gen) involve some (prior) activity of the obliquely cased argument. Thus, some apparent counterexamples could be “explained away” by assuming that it is not a conditio sine qua non that subject and nonsubject share a common aim. However, as already pointed out by Blume (1998), there are interaction verbs with a nom/acc case frame (her example jagen (to hunt), but also verbs as treffen (to meet) or hören (to hear) etc.) where the patient displays proto-agent properties even though the verb assigns structural case to its nonsubject argument. I.e., sometimes the nonsubject arguments of nom/acc verbs display some proto-agent properties so that the event described by these verbs can be split up in two (sub)events. The aforementioned verbs unterstützen (to support), verstehen (to understand) or treffen (to meet) might be cases in point, as they meet the characteristics of interaction verbs as described/defined by Blume (1998), cf. (3), repeated here as (12) in a more abstract/universal manner (such that not only nom/dat verbs are covered): (12) Event-related characteristics of interaction verbs (according to Blume 1998) (i) The nonsubject actant bears additional proto-properties.

|| 63 For further examples of such verbs consider also the secondary predication data in 2.4.3.

88 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(ii)

The verb denotes a complex event consisting of two overlapping subevents. [(iii) The involved actants have a common aim.] Notwithstanding possible counterexamples, in most cases it proves to be right that verbs that assign structural case to their nonsubject argument, cf. nom/acc verbs as sehen (to see), quälen (to torment), füttern (to feed) etc., have less complex meanings / denote less complex events in the sense that no second (preceding) subevent exists, because the patient/theme actant (= the accusative marked nonsubject) is not active in any sense/not designated to be active. However, even when the nonsubject actant acts in an agentive way, it is less often the case that subject and nonsubject argument of a nom/acc verb share a common aim (when compared to nom/dat verbs). In the following I will look at nom/gen and nom/acc/gen verbs. After that I will shortly discuss ditransitive verbs, focussing on the proto-properties of the dative argument (i.e., I will examine whether the dative argument in three-place verbs also bears multiple/additional proto-properties (as dative arguments of two-place verbs)).

2.5.3 Investigation of Events Denoted by Nom/Gen Verbs and Nom/AccREFL/Gen Verbs In the above section I have given some evidence for the assumption that the proto-role that a (dropped) dative argument fulfils within the first subevent is a priori defined/determined (in the sense that the nonsubject argument displays additional/secondary proto-agent properties). Moreover, this property of dative arguments can (in most cases) be traced back by reference to case features (cf. the well-formedness of dat-dat topic drop data under different predicates, see the survey in 2.3). Both specifics described above do not hold for genitive marked NPs: notwithstanding the fact that nom/gen and nom/acc-REFL/gen verbs denote a complex event (to be shown subsequently), their secondary proto-role is subject to verb-specific variation and, hence, reference to case identity is not sufficient – as a consequence, genitive topic drop is only wellformed under (semantic) predicate identity which ensures that the proto-role that the genitive marked actant plays within the first and the second subevent, respectively, is the same in context and target. By looking at the following event decompositions one can observe that a verbal predicate that assigns genitive case to its nonsubject argument expresses

Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs | 89

two independent events. However, these two events are not directed towards the same/a common aim of the subject and the nonsubject ‒ moreover, as shown by the subsequent examples, subject and genitive object can never have the same aim (as illustrated by the exemplary event decomposition of the meaning of the verbs gedenken and harren): (13) Ich gedenke der Rosa Luxemburg (I commemorate the RL-GEN) gedenken (to commemorate) (e) RL (Agent1) fights for freedom and justice (e’) I (Agent2) actively recall RL (Theme/Patient1=former Agent1)

(13’) Ich gedenke der Rosa Luxemburg (I commemorate the RL-GEN) gedenken (to commemorate) (e) A conspirative group (Agent1) killed RL (Patient1) (e’) I (Agent2) actively recall RL (Theme/Patient2=former Patient1)

(14) Ich harre seiner (I await his-GEN), (He=genitive object) harren (to await) (e) He (Agent1) is on the way to me (Theme1) (e’) I (Agent2=former Theme1) am affected, nervous etc. because of awaiting him (Theme2=former Agent1)

Moreover, note that the subevent that is denoted by nom/gen verbs can be differently structured, cf. (13) vs. (13’): The genitive object can act as a proto-agent or as a proto-patient within subevent1, e. I.e., it is reasonable to suggest that, generally, genitive objects do not bear an a priori determined proto-role in the events they take part in. I will claim that it is this property which makes genitive case assignment – in contrast to dative assignment – (more or less) unpredictable/idiosyncratic. As the proto-role that a genitive object bears within the first subevent denoted by a nom/gen (or nom/acc-REFL/gen) verb is not particularly specified, genitive case assigning verbs cannot be characterised along the lines of particular secondary proto-properties of their nonsubject arguments. However, as will be shown subsequently in more detail, there are some common properties of genitive case assigning verbs, cf. (15):

90 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(15) Properties of nom/gen and nom/acc-REFL/gen verbs (i) The (genitive marked) nonsubject argument plays an additional role in a (prior) subevent. However, the shape/quality of this role is not a priori determined (can be agent or patient/theme). (ii) The events denoted by genitive case assigning verbs are very often related to each other by a cause-result relation (as e.g. in the case of sich schämen (to be ashamed of) or sich erwehren (to ward off) etc.). (iii) The participants involved within the two (sub)events do not share the same aim. I.e., one possible “function” of genitive marking could be – among others – to indicate that the aim of the subject actant (in the main event) is different from the aim of the nonsubject actant (in the subevent).64 The latter supposition becomes obvious in example (16) which exemplifies that genitive object and nominative subject cannot have the same aim (here: the aim to feel ashamed). Moreover, even if it is an aim of the nonsubject argument to make the subject argument feel ashamed, feeling ashamed cannot be the (simultaneous) aim of the subject argument: (16) Ich schäme mich seiner (I shame REFL his-GEN), (He=genitive object) sich schämen (to REFL be-ashamed) (e) He (Agent1) does something that affects/embarrasses me/REFL (Patient) (e’) I (Agent2=former Patient) am ashamed by the behaviour of him (Theme=former Agent1)

With respect to the secondary proto-properties of subject and nonsubject, respectively, we can note that in the case of the nom/acc-REFL/gen verb sich schämen the subject referent (which is referentially identical with the accusative marked reflexive pronoun) constitutes the patient of (the first sub)event e und the agent of subevent e’, whereas the genitive object referent is agent of subevent e and theme of subevent e’. Such a distribution of proto-roles is not only observable with respect to schämen, but also with respect to erwehren (in two versions, with a +/-animate genitive object):

|| 64 As shown above, nom/dat interaction verbs also do not necessarily respect the ‘common aim property’. Nonetheless, even if this property does not characterise nom/dat interaction verbs very accurately, it seems to be a defining characteristic of the actants of nom/gen and nom/acc-REFL/gen verbs to have no common aim.

Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs | 91

(17) Ich konnte mich seiner erwehren (I could REFL his-GEN ward-off) sich erwehren (to REFL ward-off) (e) He (Agent1) affects (e.g. makes advances) me/REFL (Patient) (e’) I (Agent2=former Patient) am actively rebuffing the action of him (Theme=former Agent1)

(17’) Ich konnte mich der Angriffe erwehren (I could REFL the attacks-GEN ward-off) sich erwehren (to REFL ward-off) (e) Attacks (Theme1) affect (e.g. frighten) me/REFL (Patient) (e’) I (Agent2=former Patient) am actively rebuffing the attacks (Theme2=former Theme1)

Note the following: If the explanation that we have applied to dative topic drop data is correct (namely: a dative gap derived from a dative antecedent can be dropped even when the predicates in context and target differ, because the proto-properties of a dative object are fixed and can thus be read off from case features), the same kind of explanation should be applicable to genitive topic drop data in a situation where the genitive objects (and the respective actants) in context and target display the very same proto-properties (as it is the case with respect to sich schämen in (16) and sich erwehren in (17)). That such a dialogue should come out as grammatical is borne out by the following topic drop datum in (18): (18) A: B:

Ich schäme mich seiner Annäherungsversuche. I am-ashamed REFL his advances-GEN Ja, _ konnte ich mich aber trotzdem nicht erwehren. Yes, [GEN] could I REFL however not rebuff

Thus, (18) bears evidence for the supposition that it is not the general complexity of the event that makes identity between context and target predicate in genitive topic drop contexts necessary, but the unpredictability of the secondary proto-role of the genitive marked nonsubject argument (in contrast to the predictability of the secondary proto-role that is associated with the dative marked nonsubject argument (that is the proto-agent role)). As e.g. (19) shows, the nonsubject of nom/acc-REFL/gen verbs as sich enthalten (to abstain from something) or sich entledigen (to get rid of something) displays (no proto-agent but) unspecific theme properties within the first event, e:

92 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(19) Ich enthielt mich der Stimme (I abstained REFL-ACC from the vote-GEN) sich enthalten (to abstain) (e) I (Agent1) had a vote (Theme) / the possibility to be for or against something (e’) I (Agent2=Agent1) rejected this possibility for me/REFL (Patient=former Agent1=Agent2)

I.e., nom/acc-REFL/gen verbs behave irregularly (the genitive marked argument can be a proto-agent or a proto-patient/theme within the first subevent), similar to genuine two-place nom/gen verbs as e.g. gedenken (cf. the contrast between (13) and (13’)) where the genitive object can be an agent or a patient in subevent1, e. Thus, due to this variance, we should not treat nom/gen or nom/ acc-REFL/gen verbs in a particular way, because it is hardly possible to determine the proto-properties of the genitive object within the first subevent – which is here the supposed reason why genitive case is assigned on an idiosyncratic, unpredictable basis (whereas dative case is assigned on a (at least partly) predictable basis). Let me summarise what we have found above (with respect to nom/acc, nom/dat, nom/gen and nom/acc-REFL/gen verbs): (20) Summary: Event denotations of verbs with +/-multiple proto properties (a) nom/acc verbs We have observed that (typically) the nonsubject argument of nom/ acc predicates does not act as a secondary agent, i.e. there is no implicit event which is carried out by the accusative marked argument. Note that even if the accusative argument bears some agentive properties (i.e. when it actively participates within a (sub)event), it is mostly the case that the actions of the respective agents (the “primary subject agent” and the “secondary nonsubject agent”) participating in the relevant subevents are not connected to each other via a common aim, cf. the following nom/acc ʻinteraction verbsʼ whose nonsubject bears some proto-agent properties, as e.g. (Blume’s example) jagen (to hunt), hören (to hear), auslachen (to deride) etc. In most cases nom/acc verbs are used to describe an event in which the accusative marked argument does not have to be active in order that the whole event can take place, cf. “standard” cases as (perception) verbs as sehen (to see), riechen (to smell/scent) or lieben (to love), hassen (to hate), essen (to eat) etc. (b) nom/dat verbs In most cases, the nonsubject argument of a nom/dat verb is actively involved within a subevent, e (which often takes place prior to the main event e’ or overlaps with the main event e’, cf. Blume 1998:254).

Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs | 93

(c)

Moreover, (the referents of) the nominative (subject) and the dative (nonsubject) argument, respectively, often share a common aim. nom/gen, nom/acc-REFL/gen verbs Two-place nom/gen verbs and nom/acc-REFL/gen verbs involve two separate events (the first subevent is often completed when the second subevent starts). The two events are autonomous/independent of each other as they can be characterised by the circumstance that the actions of the actants of these two events (the nominative and the genitive argument) are not directed towards a common aim. Nevertheless, the two events denoted by nom/gen or nom/acc-REFL/gen verbs are often connected to each other via a cause-result relation – a further characteristics that differentiates them from nom/dat verbs. In contrast to dative case assignment, genitive case assignment is not associated with predictable proto-properties of the nonsubject actant within the (preceding) implicit subevent. Thus, the further “function” of genitive-marking is to indicate that an argument fulfils an additional role within a further (sub)event which departs from the ‘simple’ patient/theme role fulfilled by accusative arguments in nom/acc structures, as well as from the secondary agent-role role taken by dative arguments in nom/dat structures.

The findings outlined here for two-place predicates should be extendable to ditransitive verbs with regular (nom/dat/acc) and irregular patterns (nom/acc/ dat, nom/acc/gen, nom/acc/acc). This (naturally) follows from the assumptions (i) that case is linked to specific (combinations of) proto-roles, and (ii) that the specific (combinations of) proto-roles of a nonsubject argument are a reflex of the events that are denoted by the respective predicate. Thus, it is impossible for a nonsubject argument which displays no additional proto-properties to perform an additional action within a prior (sub)event. If an argument displays some additional/secondary proto-properties, these proto-properties should be marked by “extraordinary” case, in particular: by non-structural case. From this also follows that a nonsubject argument (of a two-place predicate) that bears only proto-patient properties receives structural (accusative) case simply because there is no reason to assign it a higher case65 (due to the fact that such a predicate does not denote an event which demands the nonsubject argument to act in an agentive way or to be passively involved in an additional subevent).

|| 65 By higher case I mean an oblique case (cf. the Keenan/Comrie case hierarchy).

94 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

On the other hand, dative or genitive marking the nonsubject of a two-place predicate signals that the referent of such an argument bears some additional proto-properties (due to its involvement within a complex event). Further evidence for the correlation between +complex meaning and +oblique case comes from “simple” nom/acc verbs that can be expanded by a preposition or “prefixed” so that the internal argument receives +oblique case: In this case, the respective verbs display a more complex meaning and/or the (then) non-structurally cased argument exhibits an additional agentive component. Consider e.g. hören (to hear, nom/acc) vs. zuhören (to listen to somebody/something, nom/dat): both predicates involve events that denote an activity of the nonsubject argument, however, in the case of hören subject and nonsubject do not share a common aim, whereas they do so in the case of zuhören: The action of the referent of the dative object of zuhören is (explicitly) directed towards the referent of the subject argument. A similar point can be made with respect to the differences of treffen (to meet, nom/acc) and sich treffen mit (to REFL meet with, nom/dat-PP): The meaning of the latter verb presupposes an active involvement of the person the subject referent is going to meet, whereas for treffen such an obligation does not hold to this extent (as the insertion of the adverb zufällig (coincidentally/by chance) indicates), cf. (21a) vs. (21b): (21) a.

Ich habe zufällig den Hans getroffen. I have coincidentally the Hans-ACC met b. * Ich habe mich zufällig mit dem Hans getroffen. I have REFL coincidentally with the Hans-DAT met

2.5.4 Investigation of (Events Denoted by) Three-Place Verbs I will go on by discussing ditransitive structures. As is well-known, ditransitive patterns have to be differentiated along various lines. Firstly, according to standard assumptions, the accusative is the regular case assignment option for direct objects of two-place and three-place predicates, whereas the dative is the regular case assignment option for indirect objects of three-place predicates. Secondly (modulo scrambling), it is assumed, cf. e.g. Haider (1993), Haider (2010), Fanselow (2000), Grewendorf (2002), Weiß (2004), and Woolford (2006), that the nom/dat/acc pattern is the unmarked word order of most ditransitive verbs in German (a view I share as well). This means that due to a limited case inventory, regular and irregular case marking

Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs | 95

variants can be kept apart from each other by different linearisation patterns/argument serialisations. In the following, I will examine whether accusative, dative and genitive case is connected to the same or to a different semantics when it is assigned to arguments in ditransitive structures as opposed to transitive structures. I will first check whether the dative in nom/dat/acc structures constitutes a structural/regular case assignment option, which can be tested under topic drop. I examine four scenarios, presented in (22)–(25), in order to find out in which ditransitive structure(s) the dative is assigned structurally. The respective predictions are in accordance with standard assumptions from the literature, cf. Haider (1993), (2000), Fanselow (2000), Woolford (2006) who claim that dative in nom/dat/acc verbs as schenken (to donate) or geben (to give) is assigned regularly (i.e. not idiosyncratically but instead: “inherently associated with θmarking”, cf. Woolford 2006:112), whereas dative assignment in nom/acc/dat verbs as unterziehen (to subject to) is idiosyncratic and hence not predictable. According to our previous finding we would expect that the first (inherent) dative type does not bear additional proto-properties whereas the second (lexical) dative type should bear additional proto-properties. However, the transfer of our hypothesis from two-place to three-place predicates is not without problems, because lexically assigned datives in threeplace nom/acc/dat predicates are most often not animate and hence it is difficult to ascribe them proto-agent properties. In three-place predicates the argument which is next to the verb is designated as +animate. Thus, the following examination has to be handled with care: (22) Scenario 1: predicted to be well-formed, cf. data as (22’) – the gapʼs antecedent receives dative case from a nom/dat verb – the gap receives dative case from a nom/dat/acc verb (22’) A: B:

Ich bin dem Hans ja so dankbar. I am the Hans-DAT PRT so grateful Ja, _ hab ich aus diesem Grund auch ein kleines Geschenk gekauft. Yes, [DAT] have I because-of this reason PRT a little present bought

(23) Scenario 2: predicted to be well-formed, cf. data as (23’) – the gapʼs antecedent receives accusative case from a nom/acc verb – the gap receives dative case from a nom/dat/acc verb

96 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(23’) A: B:

Du magst die Maria. You-sg like the Maria-ACC Ja, _ hab ich sogar ein Gedicht gewidmet. Yes, [DAT] have I even a poem dedicated

(24) Scenario 3: predicted to be (marginally) well-formed, cf. data as (24’) – the gapʼs antecedent receives dative case from a nom/dat verb – the gap receives dative case from a nom/acc/dat verb (24’) A:

Heute helfe ich der Gouvernante beim Kinderhüten. Today help I the nanny-DAT at child-caring B: ?* _ Solltest du den kleinen Hans nicht aussetzen. [DAT] should you the little Hans not expose-to

(25) Scenario 4: predicted to be not well-formed, cf. data as (25’) – the gapʼs antecedent receives accusative case from a nom/acc verb – the gap receives dative case from a nom/acc/dat verb (25’) A:

Morgen werde ich die Tests formulieren. Tomorrow will I the test-items-ACC formulate B: * _ Hab ich die Studenten schon gestern unterzogen. [DAT] have I the students already yesterday subjected-to

Thus, even if we should be cautious with respect to nom/acc/dat structures, cf. the scenarios described in (24)/(25) and the data in (24’)/(25’), we can state that the dative in a ditransitive nom/dat/acc structure patterns regularly, like a “structural” case, which means that it can pattern non-verbatim, cf. (22’) and (23’). Conversely, when dative case is assigned on an idiosyncratic/lexical basis, the respective topic drop dialogues are not well-formed/marginally wellformed, cf. (24’) and (25’) – i.e., we can assume that nom/dat/acc verbs do not denote a complex event, whereas for nom/acc/dat verbs this seems to be the case, modulo the fact that dative arguments of nom/acc/dat verbs are not designated as animate actants/proto-agents. I continue with a discussion of nom/acc/gen verbs. Note that the nom/acc/ gen order is the only possibility for a genitive argument to be realised in a ditransitive structure. As in the case of nom/dat/acc verbs, the obliquely cased argument can be optionally left out, e.g. with verbs connected to legal contexts as anklagen (to accuse), beschuldigen (to charge), berauben (to rob), cf. Lenz (1997:14):

Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs | 97

(26) Er beschuldigte sie (des Mordes). He-NOM charged her-ACC (the murder-GEN) When considering the verbs bezichtigen (to incriminate) and entbinden (to release/exonarate), two nom/acc/gen verbs, one can see that the genitive marked arguments (eines Vergehens (an offence); seiner Verantwortung (his responsibility)) constitute the theme within an implicit subevent e (that takes place prior to the main event, eʼ), cf. (27)/(28): (27) Ich bezichtigte den Hans eines Vergehens (I incriminated the Hans-ACC an offence-GEN) bezichtigen (to incriminate) (e) Hans (Agent1) has possibly committed an offence (Theme1) (e’) I (Agent2) incriminated Hans (Patient=former Agent1) to have committed an offence (Theme2=former Theme1)

(28) Ich entband den Hans seiner Verantwortung (I released the Hans-ACC from-his responsibility-GEN) entbinden (to release/to exonerate) (e) Hans (Agent1) was expected to be responsible for something (Theme1) (e’) I (Agent2) released Hans (Patient=Agent1) from this responsibility (Theme2=former Theme1)

However, note that sometimes it is highly unclear how a complex event denoted by a verb that assigns genitive case to its indirect argument might be decomposed, cf. (29) vs. (29’): In (29) the mother is a theme, and in (29’) the mother acts as a secondary agent within the first subevent e (?). (29) Ich entwöhnte das Kind der Mutter (I weaned the child-ACC the mother-GEN) entwöhnen (to wean) (e) Child (Agent1) is used to its mother (Theme1) (e’) I (Agent2) weaned the child (Patient) from the mother (Theme2=former Theme1)

(29’) Ich entwöhnte das Kind der Mutter (I weaned the child-ACC the mother-GEN) entwöhnen (to wean) (e) Mother (Agent1) cares for her child (Patient1) (e’) I (Agent2) weaned the child (Patient2=Patient1) from the mother (Theme=former Agent1)

98 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

From (29)/(29’) one can conclude that, like their counterparts in two-place verbs, genitive objects in three-place verbs do not have a fixed proto-role, i.e. they can have proto-agent or proto-patient properties (even in case of a single verb as entwöhnen). Let me shortly discuss the nom/acc/acc pattern as well. Fanselow (2000:181) claims that (passive data show that) only one (namely the higher) accusative marked argument is structurally case marked, whereas the lower accusative argument is not. Haider (2010), who goes in the same direction, notes the following (2010:252f): German has only very few verbs with two accusative objects (cf. ([30])). Passive (([31]) vs ([32])) shows that the two arguments are not equivalent. Only one of the two objects, namely the direct object, receives nominative: ([30])

Jemand fragt ihn die Vokabel[n] ab Someone checks him-ACC the vocabulary-ACC off ‘Someone tests him on the vocabulary’

([31])

Er wurde die Vokabel[n] abgefragt He-NOM was the vocabulary-ACC off-checked ‘He was tested on vocabulary’

([32]) * Die Vokabel[n] wurden ihn abgefragt The vocabulary-NOM was him-ACC off-checked

However, Haiderʼs example (32) does not sound so bad at all when the direct accusative object ihn (or some equivalent accusative marked NP) is moved to the prefield or to the top of the middlefield, cf. (33) or (34): (33)

Ihn wurden die Vokabeln abgefragt (und zwar von Maria). Him-ACC were the vocabulary-NOM checked (and namely by Maria).

(34)

Gestern bei der Prüfung wurde den Hans der ganze Cicero abgefragt. Yesterday at the test was the Hans-ACC the whole Cicero-NOM checked

I.e., grammaticality is improved when the arguments at the surface are ordered differently, namely acc > nom, cf. Haiderʼs (32) vs. (33) and (34): Thus, if the ungrammaticality of Haiderʼs (32) is only a problem of surface linearization, it is possible that the lower accusative is structurally assigned as well (otherwise the indirect accusative case could not by absorbed ‒ which is what Haider 2010 claims with respect to (32), but which is not borne out under a different surface pattern, cf. (33)/(34)). Given that there are speakers who accept patterns as (32)

Topic Drop and the Event Structure of Structural and Oblique Case Assigning Verbs | 99

and (33)/(34) where the lower accusative argument is passivised, it may be that generally accusative case is assigned structurally. As for now we can conclude that, as in the case of two-place verbs, oblique case marking in ditransitive structures serves to indicate a departure from a general/standard instantiation. However, due to the limited inventory of different cases, seemingly oblique (= dative) case marking is not sufficient to point out the presence of a complex event denotation. Therefore, irregular ditransitive patterns are (also) marked by different linearisation patterns. I.e., in three-place predicates not only “extraordinary” case marking but also exceptional/noncanonical argument linearisation indicates (syntactically/at the surface) that an argument bears a double proto-role (and hence participates in a complex event), cf. regular structural dative assignment in nom/dat/acc verbs as in (22’)/(23’) vs. irregular non-structural (and possibly idiosyncratic) dative assignment in nom/acc/dat verbs as in (24’)/(25’). (35) summarises what we have found in this and the former sections: (35) Predictability of case assignment: Structural vs. non-structural case assignment a. With respect to two-place predicates the accusative constitutes the structural case variant for nonsubject arguments, whereas the dative and the genitive constitute the non-structural (only partly predictable) case variant for nonsubject arguments. b. With respect to three-place predicates the dative (in “second” position, cf. the nom/dat/acc order) constitutes the structural/regular case variant for indirect objects, whereas all other orders constitute the non-structural/irregular case variant for indirect objects. I.e., not only the particular case marking plays a role but the underlying serialisation/linearisation of arguments as well. c. The non-structural/irregular case variants (mentioned in a. and b.) are in part predictable on a semantic basis. I.e., oblique case marking or extraordinary linearisation patterns, respectively, express an additional involvement of a nonsubject argument in an implicit subevent (which starts before the main event (or takes place simultaneously)), in analogy to the event/proto-role semantics of twoplace verbs. The above findings are illustrated in the following table, (36):

100 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(36)

(Un-)predictable case assignment

Kind of predicate

High predictability

Lower predictability

Low predictability

2-place predicates (dir. obj)

structural acc (nom/acc)

dat (nom/dat)

gen (nom/gen)

3-place predicates (indir. obj)

ʻstructuralʼ dat (nom/dat/acc)

ʻnon-structuralʼ dat gen (nom/acc/dat) (nom/acc/gen)

2.6 Topic Drop and the Representation of +/-Complex Events The next (and last) question that will be addressed in this part is how the event denotations that are linked (via an argument’s proto-properties) to +/-oblique case assignment might be represented syntactically. But before, I will shortly outline how the NVTD/VTD distinction might be connected to a predicate’s sselectional and c-selectional properties, respectively, and in which way they might possibly be arranged and ordered (within a predicate’s kernel).

2.6.1 The Relation between C-selection and S-selection in the Context of +/-Oblique Case Assignment The above discussion on topic drop can be summarised by stating that the more obliquely case marked a topic dropped element is, the more likely it is that it is licensed by the presence of an absolutely identical micro theta role (assigned to the gap and its antecedent, respectively). As topic drop is always fine as long as the gap bears structural case (cf. NVTD), we can presume that in nom/acc verbs s-selection and c-selection coexist but the respective properties are not intrinsically tied to each other. This, however, is probably the case with respect to dative and genitive gaps. As to dative gaps, case is meaningful and connected to s-(selectional) properties (cf. the secondary proto-agent properties of dative arguments in two-place predicates). Thus, dative drop can be seen as a semantically inflated version of NVTD. On the other hand, genitive case seems to be (just) an indicator for sproperties being absolutely idiosyncratic. Hence, a genitive gap is only licensed when the micro theta roles assigned by the context and target predicate are absolutely identical. As genitive case assigning verbs are hardly synonymous with other verbs, in most cases genitive drop is only licensed when context and

Topic Drop and the Representation of +/-Complex Events | 101

target predicate are identical, cf. ‘pure’ VTD (and a sociolinguistic register mismatch is not present as well). Thus, it suggests itself to map the different semantic and syntactic licensing strategies of nom/acc, nom/dat and nom(/acc)/gen verbs to the concept of s- and c-selection, respectively, cf. (1), repeated from chapter 2.4.1: (1) c-selection and s-selection, according to Bierwisch (2006:91): Categorial or c-selection determines the syntactic and morphological requirements that optional or obligatory complements of an expression must meet, while Semantic or sselection specifies the corresponding semantic constraints.

Moreover, let us also recapitulate what Bierwisch (2006:90) says about the architecture of theta roles: Theta-Roles must be able to reconcile the same semantic relation with alternative grammatical conditions and also the same morpho-syntactic properties with different semantic relations. To this effect, they must have access to semantic as well as syntactic and morphological information, participating in at least two levels of representation. In this sense, Theta-Roles are (part of) the interface mediating between formal or morpho-syntactic and semantic or conceptual aspects of linguistic expressions.

Summing up former sections we can state that (in two-place predicates) accusative case is assigned when a nonsubject argument is not designated to bear any additional proto-properties, whereas dative case indicates the presence of additional proto-agent properties of a nonsubject argument, and genitive case signals that a nonsubject argument bears additional proto-properties which are not a priori defined. I.e., only genitive case constitutes a “real” idiosyncratic case, indicating that an argument has non-predictable proto-properties. According to these observations/findings, the c- and s-selectional properties associated with nom/acc, nom/dat and nom/gen verbs, respectively, could be notionally described as bundles/fibres a theta role consists of, cf. (2)–(4). (2)

As to the “architecture” of accusative case, c- and s-selectional bundles/ fibres should not be tied to each other but coexist independently.

(3)

In accordance with the fact that case identity is sufficient to topic drop a dative NP, the label ‘dative’ should be able to guarantee that c-bundles contain a certain portion of s-fibres, which determine the proto-properties of a dative argument (= being a secondary agent within an implicit subevent).

102 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(4)

As to genitive case, it may be reasonable to think of c-fibres being fully captured/encapsulated by s-bundles; c-fibres just serving to signal that verbal meaning is idiosyncratic and not connected to a particular/ predictable proto-semantics.

2.6.2 The Syntax of Oblique Case Assignment – Some Introductory Notes In the following I will propose how an argument’s particular proto-properties could be represented syntactically. However, first, I will present some previously considered ways to account for (the difference between structural and) oblique case. One solution to the differentiation of structural vs. lexical cases is to assume that (in contrast to arguments which receive structural case) obliquely cased arguments are located in a distinctive syntactic position – e.g. an Applicative Phrase (ApplP), cf. McFadden (2006:56), referring to Pylkkänen (2002), Cuervo (2003), Hole (2005). In this case, the features that make up the dative (or genitive) are confined to a particular syntactic position. However, even if it may be the case that dative and genitive assignment finds its expression within designated syntactic positions, such assumptions must be evidenced independently (e.g. by semantic qualities). The suggestion of Bayer et al. (2001) goes in a slightly different direction, namely that obliquely cased arguments bear an additional layer above the DP – a K Phrase. Such an approach is located somewhere in-between a syntactic and a lexicalist approach to oblique case assignment. Bayer et al. (2001:465; 475f) note the following: In German, oblique Cases (dative and genitive) require morphological licensing while structural Cases (nominative and accusative) do not. This difference can be captured by assuming that in German, NPs bearing oblique Case have an extra structural layer Kase phrase (KP) which is missing in NPs bearing structural Case. […] We propose different structures for the two types: (i) Nominals which may project to NP and DP and (ii) Nominals which are in addition headed by the head K (K for Kase). Depending on external conditions, K is realized as oblique Case or as P. (p. 475) […]

Topic Drop and the Representation of +/-Complex Events | 103

([5])

KP ...

K’ DP ...

K D’

NP

D

With respect to German we make the assumptions in ([6]) about K. ([6]) a. b. c. d.

K stands for the unifying category KASE. K can be realized by a P which carries the feature [obl] or by the feature [obl] alone. The linear order between K and DP follows from principles of morphology. K must be specified. (p. 475)

In Ramchand (2008) events are represented in a linear syntactic order. The approach argued for by Ramchand becomes evident by the tree beneath which shows how the thematic properties of the subject are decomposed due to its multiple involvements within the respective subevents denoted by the verb (from Ramchand 2008:79): I will analyse a verb like arrive as containing a single DP argument which initiates its own transition to a final locational state – it is simultaneously the INITIATOR, UNDERGOER and RESULTEE. Moreover, because the verb identifies all three heads in this functional decomposition, the resulting predication is punctual ([7]).

104 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(7) initP

Michael init

procP

arrive < Michael > proc

resP

< arrive > < Michael > res

(XP)

< arrive > Ramchand (2008) represents the involvement of a particular argument within different (sub)events in a linear order, i.e., (i) the verb that denotes the relevant subevents and (ii) the argument(s) that is/are involved in the particular (sub)events are both multiply represented (by different occurrences) within a layered tree structure whereby each subevent node represents the event-related properties of the respective argument(s). In order to ensure that the features represented by the relevant heads can be interpreted, multiple occurrences of the verb and its argument(s) which are – due to their fixed positions within distinct functional projections – equipped with different features/properties have to be accumulated, e.g. via a process like Merge. In the following, I will neither adopt Ramchand’s (2008) approach nor the ones argued for by (inter alia) McFadden (2006) or Bayer et al. (2001). Instead, I will attribute the differences between differently cased nonsubject arguments (i.e. accusative, dative and genitive arguments) to their “in depth” representation and argue that different case assignment follows from a (verbal) predicate’s denotation of a +/-complex event. Obviously, such an approach must be able to account for the fact that verbal event semantics are crucial with respect to the determination of an argument’s proto-properties and, hence, the case features that it bears (or better: the case features it has to bear due to its involvement in a particular (+/-complex) event denoted by the verb). I.e., an adequate solution has to capture the empirical facts that (e.g.) a dative case marked nonsubject

Topic Drop and the Representation of +/-Complex Events | 105

argument (in a two-place predicate) fulfils two discrete actions, namely (i) being an agent in (sub)event1 (e) and being a patient in (sub)event2 (e’) – which constitutes the main event. As I assume that the events/event types (complex vs. simple) that verbs denote go hand in hand with certain proto-properties of their arguments, which are, in turn, spelled-out as case features, I will argue for an approach where the covertly manifest proto-properties and the overtly manifested morphological case of a nonsubject argument can be syntactically represented, in mutual dependence on each other.

2.6.3 +/-Oblique Case Assignment (and Topic Drop) First, in accordance with the above event decompositions of nom/acc, nom/dat and nom/gen verbs, respectively, repeated for convenience in (8), (9) and (10), I will assume that +/-oblique case marking is linked to verbal semantics as formulated in (11): (8)

Ich sehe den Hans (I see the Hans-ACC)

sehen (to see) (e) I (Agent) see Hans (Theme)

(9)

Ich helfe dem Hans (I help the Hans-DAT)

helfen (to help) (e) Hans (Agent1) does something (that possibly requires help wrt a Theme/Patient) (e’) I (Agent2) help Hans (Patient=former Agent1)

(10) Ich schäme mich seiner (I shame REFL his-GEN), (He=genitive object) sich schämen (to REFL be-ashamed) (e) He (Agent1) does something that affects/embarrasses me/REFL (Patient) (e’) I (Agent2=former Patient) am ashamed by the behaviour of him (Theme=former Agent1)

(11) Causal link between verbal event semantics and case marking A verb denotes a +/-complex event … (consequently) … the nonsubject argument displays +/-multiple proto-properties … (consequently) … it receives +/-oblique case Alternatively, (11) could also be formulated in terms of (12):

106 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(12) The “function” of oblique case marking Oblique case expresses a specific semantics, namely the involvement of a nonsubject argument within a complex event. From the event decompositions of verbs assigning +oblique case (dat, gen) to their nonsubject argument (cf. chapter 2.5) we can also conclude that (13) holds: (13) +/-Predictable oblique case marking a. The secondary proto-properties of a dative argument are predictable (= proto-agent), cf. Blume (1998, 2000). b. The secondary proto-properties of a genitive argument are not predictable (= proto-agent or proto-patient), cf. Lenz (1997). As already pointed out before, it is the unfixed character of the secondary protoproperties of a genitive object which makes genitive assignment so idiosyncratic and unpredictable. On the other hand, from dative assignment (in two-place predicates) one can infer the secondary proto-properties of the nonsubject argument (cf. proto-agent), hence topic drop is possible under case identity, whereas genitive topic drop requires identity between the context and target predicate (i.e. micro theta role identity), cf. the following data which show that in contrast to dative case assigning verbs (which do not have to be synonymous in order to license topic drop), genitive case assigning verbs can hardly be interchanged even when two predicates are perceived as “extremely” synonymous or synonymously related (i.e. possibly symmetric), cf. (14) vs. (15): (14) A: B:

Du hilfst der Maria ja, wo du nur kannst. You-sg help the Maria-DAT PRT, wherever you-sg just can Ja stimmt, _ folge ich sogar nach Paraguay! Yes right, [DAT] follow I even to Paraguay

(15) A:

Hans brüstete sich der wichtigsten Entdeckung Hans bragged-about REFL the most-important discovery-GEN der Welt. of-the world B : ? _ Rühmte sich Irene schon lange. [GEN] prided REFL Irene already for-long

Note that other topic drop combinations, e.g. dat/gen or gen/dat yield marginal outputs as well, cf. (16) and (17):

Topic Drop and the Representation of +/-Complex Events | 107

Dat-Gen (16) A: Ich werde Petra morgen beim Aufräumen helfen. I will Petra-DAT tomorrow with cleaning-up help B: ? _ Hab ich mich neulich auch mal erbarmt.66 [GEN] have I REFL recently also took-pity-on Gen-Dat (17) A: Ich habe mich neulich meiner Schüler erbarmt. I have REFL recently my pupils-GEN take-pity-on B: ? Ach, _ ist doch eh nicht mehr zu helfen. Ah, [DAT] is however anyway no longer to help The first question to be answered is why structurally cased NPs can occur in non-verbatim (different predicates in context and target) and verbatim environments (identical predicates in context and target) whereas obliquely cased NPs can only occur in verbatim environments. As already mentioned, I suggest that the relevant category by which topic dropped NPs (DPs) are represented is pro. Furthermore, as stated in (12), I assume that oblique case marking is meaningful because it expresses a particular semantics (namely, the additional involvement of the nonsubject argument within an implicit subevent), and, hence, it must be spelled out, cf. principle (18): (18) Spell out meaningful elements (features, properties).67 The line of argumentation I will pursue in order to explain why obliquely cased NPs demand a verbatim environment, whereas structurally cased NPs do not, is the following: Oblique case is meaningful (= associated with a particular semantics), cf. principle (12). Since pro is not represented at PF, topic drop of obliquely cased NPs is (in principle) impossible because the semantic content of oblique

|| 66 Note that erbarmen and helfen do have a similar meaning. Thus, one could suppose that they share a lot/at least some s-selectional properties. I.e., their respective micro theta roles are relatively similar (though not identical – as is evidenced by the fact that acceptability is degraded, cf. the survey and the respective judgements in 2.3). 67 Helmut Weiß (p.c.) proposed that possibly, the obligation Spell out meaningful elements (features, properties) may be inherent/included within Chomsky’s (1986) Principle of Full Interpretation “that requires that every element of PF and LF, taken to be the interface of syntax (in the broad sense) with systems of language use, must receive an appropriate interpretation.” (Chomsky 1986:98).

108 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

case must be spelled out overtly (otherwise principle (18) would be violated). However, one can evade the latter consequence by making sure that the protoproperties that cannot be spelled out by pro are nevertheless present. This can be done by establishing an identity relation with an element that overtly represents the properties that pro is lacking, namely the antecedent of the gap (cf. dative topic drop), or, when the proto-properties are not unambiguously determined by case (cf. genitive topic drop): by identity of micro theta roles of antecedent and gap (which can be assured by semantic identity of the context and target predicate). In the following I will present some independent evidence for the assumption that an obliquely case marked NP indicates/marks the presence of an implicit subevent, cf. (12). First, if it is true that oblique case indicates the presence of an implicit subevent, the overt realisation of such a subevent should have the effect that the additional semantics (= the presence of an implicit subevent) of oblique case is now spelled out overtly and, hence, an identity relation with an element that represents the semantics of an obliquely case marked NP is not a necessary condition anymore. Given such a situation, obliquely case marked NPs should be droppable in a non-verbatim environment as well – which is borne out by the following data, cf. (19)–(20) for dative drop and (21)–(22) for genitive drop. Note that in the a.-versions the event is not spelled out and hence, non-verbatim topic drop (NVTD) of obliquely cased NPs is marginal/ ungrammatical, whereas in the b.-versions the event is spelled out by an infinitival complement that bears an object controlled PRO (in the dative case) or, alternatively, by a CP that expresses the additional subevent68 (in the genitive case): (19) A: a.

Der Hans war gestern so unverschämt wie noch nie. The Hans-NOM was yesterday so impertinent as yet never B: ? _ Hab ich schon verziehen. [DAT] have I by-now forgiven

|| 68 Cf. Haider (2010:242) who notes that “arguments that require genitive for a DP object may be represented by a clause”. He gives the following examples: (i) Ich bin mir (dessen) bewusst, dass es so ist I am myself (it-GEN) conscious, that this so is ‘I am aware that it is so’ (ii) Jemand hat ihn (dessen) beschuldigt, dass er sich nicht um seine Kinder kümmere someone has him (it-GEN) accused that he himself not for his children cares ‘someone has accused him of not taking care of his children’

Topic Drop and the Representation of +/-Complex Events | 109

b.

B:

_ Hab ich schon verziehen, PRO die Party gesprengt zu haben. [DAT] have I by-now forgiven, the party braked to have

(20) A: a. b.

Ein aggressiver Hund hat gestern die Maria angefallen. An aggressive dog has yesterday the Maria-ACC attacked B: ? Ja, _ hab ich aber helfen können. Yes, [DAT] have I however help could B: Ja, _ hab ich aber helfen können, PRO dieses Monster abzuwehren. Yes, [DAT] have I however help could, this monster to fight-off

(21) A: a. b.

Dieser schmatzende Mann auf meiner Party war mir sehr peinlich. This munching man-NOM at my Party was (for) me very embarrassing B: * _ Hab ich mich auch ohne Ende geschämt. [GEN] have I REFL also endlessly shamed B: ? _ Hab ich mich auch ohne Ende geschämt, weil ich ihn doch [GEN] have I REFL also endlessly shamed, because I him PRT mitgebracht habe. with-me-brought have

(22) A: a. b.

Diesem Studenten sollte man eine bessere Note geben. This student-DAT should one a better mark give B: * Ne, _ hab ich mich schon bei der Bachelorarbeit erbarmt. No, [GEN] have I REFL already at the bachelor-thesis had-pity-with B: ? Ne, _ hab ich mich schon bei der Bachelorarbeit erbarmt, No, [GEN] have I REFL already at the bachelor thesis had-pity-with, weil er eine Schaffenskrise hatte. because he a creative-crisis had

Note that the marginality of the a-examples in (19)–(22) is not caused by the lack of additional context (as is given by the apodosis in the b-examples). This is shown by the following data, where a structurally cased gap is fine without an apodosis: (19’) A: a.

B:

Der Hans war gestern so unverschämt wie noch nie. The Hans-NOM was yesterday so impertinent as yet never _ Hab ich schon abserviert. [ACC] have I by-now ditched

110 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(20’) A: a.

B:

(21’) A: a.

B:

(22’) A: a.

B:

Ein aggressiver Hund hat gestern die Maria angefallen. An aggressive dog has yesterday the Maria-ACC attacked Ja, _ hab ich aber retten können. Yes, [ACC] have I however save could Dieser schmatzende Mann auf meiner Party war mir sehr peinlich. This munching man-NOM at my Party was (for) me very embarrassing _ Würde ich auch nicht nochmal einladen. [ACC] would I also not again invite Diesem Studenten sollte man eine bessere Note geben. This student-DAT should one a better mark give Ne, _ hab ich schon durch die Bachelorarbeit gehievt. No, [ACC] have I already through the bachelor-thesis heaved

Note further that in situations where the topic dropped element is a nominative/accusative NP (i.e., an instance of NVTD), the presence of an overtly realised subevent is irrelevant (as expected). Consequently, both variants (a. and b.) are equally well accepted, cf. (23) and (24): (23) A: a.

B:

b.

B:

(24) A: a.

B:

b.

B:

Mit dem Hans red ich nie wieder ein Wort. With the Hans-DAT speak I never ever a word Wieso? _ Hat sich doch entschuldigt. Why? [NOM] has REFL however said-sorry Wieso? _ Hat sich doch entschuldigt, PRO die Maria angemacht Why? [NOM] has REFL however said-sorry, the Maria made-a-pass-at zu haben. to have Kommt der Hans zu deinem Geburtstag? Comes the Hans-NOM to your birthday? Ja, _ hab ich ja extra d’rum gebeten. Yes, [ACC] have I purposely for-this-reason asked-for Ja, _ hab ich ja extra d’rum gebeten, PRO den Yes, [ACC] have I purposely for-this-reason asked-for, the Überraschungsgast zu spielen. surprise-guest to play

Topic Drop and the Representation of +/-Complex Events | 111

Besides the spell-out of an otherwise implicit subevent, there is further evidence for the supposition that oblique case marked elements must be realised at PF (because oblique case is meaningful, cf. principle (18)). This evidence comes from German dialects, particularly from relative clause data. Van Riemsdijk (1989) and Salzmann (2005) observe that in some dialects a resumptive pronoun must only be inserted when the relativised constituent is obliquely case marked, cf. (25a/b) vs. (25c): (25) a. b. c.

d=Frau, wo (*si) immer z=spaat chunt. the woman-NOM, which (she-NOM) always too late comes es Bild, wo niemert (*s) cha zale. this picture-NOM, which nobody (it-ACC) can pay (= afford) de Bueb, wo mer *(em) es Velo versproche händ. the boy-NOM, which we (him-DAT) the bike promised have

Note that a similar observation can be made with respect to relative complementisers in oblique function, cf. Fleischer (2006). See also Bayer (1984:221f) for the original observation and the datum in (26b): (26) a. b. c. d.

Der Mann, wo / den wo / den wir kennen. (nom-acc) The man-NOM, which / who-ACC which / who-ACC we know Der Mann, *wo / dem wo / dem ich das gegeben habe. (nom-dat) The man-NOM, which/whom-DAT which/whom-DAT I that given have Sie gem’s dem Mo, (dem) wo mir g’hoifa hom. (dat-dat) They give it the man-DAT, (whom-DAT) which we helped have i hans tǝm man ksaait, wu tɒ kaaɒtǝ kheeɒtǝ (dat-dat) I have it the man-DAT said, which-DAT the garden belonged (to)

With regard to (26b/c/d) Fleischer (2006:226) notes the following: It seems to be the case that […] [-case] strategies in the indirect object are only possible if the antecedent noun phrase displays dative case. Otherwise a (relative or resumptive personal) pronoun displaying dative case within the relative clause has to appear, i.e. if the head noun displays another case than dative, the [+case] strategy is obligatory.

With respect to (26d) Fleischer (2006:228) remarks that: In this variety, relative clauses are introduced by wo alone, and this seems to hold also for the indirect object cases, as illustrated by ([26d]). But this example displays dative case in the antecedent noun phrase […] Thus for the dialects of Zaisenhausen and Großrosseln, it is very likely that the case matching constraint for indirect object relative clauses does ex-

112 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

ist, but this cannot be stated with certainty since the respective grammatical descriptions give no explicit information on this phenomenon.

From the data in (25) and (26) we can conclude that a requirement to realise oblique case features overtly also holds for relative clauses in German dialects. Moreover, we can state that whenever dative marked elements are exempt from overt realisation (cf. (26c)/(26d)), this holds for all lower ranked cases on the Keenan & Comrie (1977) hierarchy as well.69 Moreover, the contrast between (26b) and (26c)/(26d) shows that a spell-out obligation of obliquely cased elements can only be circumvented by case matching with an overt element that is related to the gap via an identity condition70 – a situation which is analogously found with respect to topic drop of obliquely cased arguments.

2.6.4 The Syntactic Representation of +/-Oblique Case Assignment Before coming to the last point, let me shortly summarise the well-formedness conditions on topic drop (i.e., the dropping of spell-out): (27) Well-formedness conditions for topic-dropped NPs (DPs) a. Actants of verbs which participate in more than one subevent (that is denoted by the verb), display additional proto-properties and (therefore) receive semantically contentful (= oblique) case. b. Meaning (of elements/features) that cannot be inferred from syntactic (phrase) structure has to be spelled out at PF. c. If principle b. cannot be satisfied, the missing/not spelled out meaning of elements/features must be reconstructed/re-established via a (semantic) identity relation with elements/features that represent that particular meaning overtly. d. If c. cannot be satisfied, the resulting structure is ungrammatical. In the following, I will show how the implicit subevent denoted by an oblique case assigning verb could be represented. As an example I will consider a sentence as (28), whose event decomposition is given in the respective box: || 69 See also Trutkowski & Weiß (to be published) who mention relative clause data where resumptive pronouns occur with structural case. However, note that there are no (German) dialects with resumptive pronouns occurring only with structural case. 70 As nearly all German dialects have lacked the genitive object case for about 500 years, cf. Weiß (2012), we can test our hypotheses only with respect to dative case.

Topic Drop and the Representation of +/-Complex Events | 113

(28) Ich helfe dem Hans (I help the Hans-DAT) helfen (to help) (e) Hans (Agent1) does something (that possibly requires help wrt a Theme/Patient) (e’) I (Agent2) help Hans (Patient=former Agent1)

With respect to the syntactic representation of the events in (28) I assume... (i) that the lexical head of the verb helfen (to help) selects a silent ‘to do-VP’ as complement (ii) that the silent to do-VP represents the implicit subevent (e) and that it contains the agent (here: Hans) of this implicit subevent In (29a) subevent e is not yet spelled out, grey colour marks optional actants:71 (29a)

VP IchNOM

V’

VP Hansag

V° | V’ helfe

XPth/pat V° | to do

Here: The head of the lexical verb takes the silent to do-VP as a complement (note that due to its PF-deficiency the head of the latter can only assign a theta role, but no case)

In a further step, cf. (29b), the following happens: –

–

Agent Hans receives patient-properties und case when it is moved out of the silent to do-VP in order to participate within the second subevent, e’ – which constitutes the main event. As Hans is not newly inserted into the structure, but has already received proto-properties (in this case: proto-agent properties) within the silent to do-VP, Hans becomes obliquely case marked:

|| 71 Note that ‘ag’ means ‘agent’, ‘th’ means ‘theme’ and ‘pat’ means patient within the following tree(s). The case labels are the same as before, i.e. ‘NOM’ for nominative case etc.

114 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(29b)

VP IchNOM

V’

HansDAT VP Hansag

V’

Oblique case is assigned ...

V° | V’ helfe

... when an argument with proto-properties moves out of the silent to do-VP in order to participate within the main event, e’. An argument’s multiple proto-properties, as (e.g.) being an agent in one and a patient in another subevent, are made visible by oblique case marking.

XPth/pat V° | to do

I.e., dative-/genitive assignment is a means to overtly mark the presence of an abstract to do-VP; oblique case constituting a label which indicates that an argument has already participated in a preceding/simultaneous (sub)event. Note that the presence of the silent to do-VP is in principle the same as the overt spell-out of the subevent in the examples in 2.6.3 (the data in (19)–(22)). As for genitive case assignment in two-place predicates I assume an analogous structure and case assignment process as for dative case assignment, with the only difference that the genitive marked argument can also act as a theme or patient within the first (implicit) subevent. As to canonical dative assignment in three-place predicates consider the tree in (30): (30)

Ich gebe dem Hans ein Geschenk. I give the Hans-DAT a present-ACC VP IchNOM

V’

HansDAT

V’

GeschenkACC V° | gebe

Topic Drop and the Representation of +/-Complex Events | 115

As already mentioned, irregular dative assignment in three-place predicates is seldom found. Only few verbs, e.g. unterziehen (to subject to), aussetzen (to expose), zuführen (to feed/direct) show the nom/acc/dat pattern. Meinunger (2000) claims that the dative argument introduced by such verbs is a local argument of the verb. He suggests to analyse these dative objects in terms of “(the remnant of) a PP” (p. 58, see also Struckmeier 2014), which is supported by the fact that these verbs contain a prepositional morpheme, cf. unter- (below), aus(out/off) etc. Given the fact that topic drop of both, PPs and dative arguments of nom/acc/dat verbs, is very marginal (cf. the data in 2.5.4), we can explain this analogous behaviour quite consistently if we follow Meinunger (2000) and assume that such (pseudo) dative arguments can hardly be dropped because they have a PP head that is incorporated into the verb. (31a/b) and (32a/b) show how the multiple proto-properties of three-place nom/acc-REFL/gen predicates could be represented in a layered tree structure: (31a) Ich schäme mich meiner Katze. I am-ashamed myself-ACC (of) my cat-GEN VP IchNOM

V’

michACC VP meine Katzeag

V’ V° | V’ schäme (e.g. my cat (agent) has eaten a rat (patient))

Ratteth/pat V° | to do (e.g. to eat)

(31b) depicts the step of genitive case assignment:

116 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(31b) Ich schäme mich meiner Katze. I am-ashamed myself-ACC (of) my cat-GEN VP IchNOM

V’

michACC

V’

meiner KatzeGEN V’ VP meine Katzeag

V° | V’ schäme

Ratteth/pat V° | to do

(32a) Ich schäme mich des Kaffeeflecks. I am-ashamed myself-ACC (of) the coffee-stain-GEN VP IchNOM

V’

michACC

V’

VP

V° | Hansag V’ schäme (e.g. Hans (agent) has caused a coffee stain (theme)) Kaffeefleckth V° | to do (e.g. to cause something)

(32b) is the continuation of (32a):

Topic Drop and the Representation of +/-Complex Events | 117

(32b) Ich schäme mich des Kaffeeflecks. I am-ashamed myself-ACC (of) the coffee-stain-GEN VP IchNOM

V’

michACC

V’

KaffeeflecksGEN

V’

VP Hansag Kaffeefleckth

V° | V’ schäme V° | to do

In either variant, (31)/(32), the (later genitive case marked) argument moves out of the silent to do-VP in order to receive case – analogous to the dative example in (29b) the nonsubject argument receives oblique case in (31b)/(32b) because it already bears a proto-role. However, due to the fact that the (secondary) protoproperty of genitive arguments is not a priori defined, genitive case is assigned (signalling idiosyncratic meaning).72 As oblique arguments always participate in a (prior) subevent (and hence, they first receive proto-properties within the silent to do-VP), they are never (immediately) merged as direct complements of a lexical verb and, as a consequence, they can never receive structural case. As already mentioned, non-obliquely cased nonsubject arguments can also take part in complex events. However, the (for that circumstance) required additional proto-role is (obviously) not encoded via oblique case marking. As a consequence, the active involvement of the nonsubject argument within an additional (sub)event must be ensured without reference to a structure that would be designated for this purpose (which costs more, cf. Czypionka et al. who found that animate accusative objects are processed more slowly than animate

|| 72 For a detailed summary of the differences of nom/acc, nom/dat and nom/gen verbs see also 2.5.1, (20).

118 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

dative objects). I.e., exceptions are always possible: as a complex event can be denoted by nom/acc verbs, a non-complex event can also be denoted by nom/gen verbs, cf. bedürfen (to require). However, such instances should also be harder to process because an argument’s case marking (which indicates additional/multiple proto-properties) and a verb’s event denotation are not consistent (or, better: consistently marked). Of course, for the syntax proposed here some extra-assumptions have to be made, cf. (33): (33) Extra-assumptions with respect to oblique case assignment a. An event denoted by a verbal predicate can at most consist of two subevents. b. In a given tree structure at most one to do-VP is possible as a complement (this also holds for three-place predicates). (33) predicts that nom/dat/gen, nom/gen/dat, nom/dat/dat, nom/gen/gen patterns are ruled out. Moreover, it is expected that genitive case in three-place verbs indicates the presence of a complex event as well. With respect to dative marking, a nonsubject argument should not display additional proto-properties when it receives “structural” (= regular/inherent) dative from nom/dat/acc verbs as schenken (to donate), geben (to give) etc. As to “non-structural” dative case in nom/acc/dat verbs, I follow Meinunger (2000) who shows that a dative argument which is linearised in this way can be understood as the remnant of an incorporated PP. Summarising the above case assignment procedures, it can sloppily be said that – in essence – the technical case assignment procedure is the same for structurally and lexically cased NPs: Either way, nonsubject case is assigned by the head of a lexical verb – the only difference between structural and oblique case being that oblique arguments are not immediately merged as direct complements of a lexical verb, because they participate within a preceding subevent which is why they first acquire certain proto-properties (within the silent to doVP) before they move out of the silent to do-VP to take part within the main event, thereby receiving case overtly (in case position).73 The trigger for the later nonsubject to move out of the silent to do-VP is the visibility/spell-out condition in (18), which demands that semantic/meaningful content has to be present at surface structure/PF. || 73 As case cannot be assigned within the silent to do-VP, the (later obliquely case marked) actant must leave the to do-VP in order to not violate the Case Filter (cf. Helmut Weiß, p.c.).

Topic Drop and the Representation of +/-Complex Events | 119

Lexical/oblique case assignment can be seen as the result of an actant’s property acting as ‘the servant of two masters’, performing different actions within different subevents. As these hydra-like properties have to be ‘somehow’ spelled out within one single lexical item, the multiple proto-properties of such an argument (e.g. with respect to dative case: being an agent in event e and a patient in event e’) are combined, i.e. there are two instances of proto-role assignment: the nonsubject argument receives one proto-role from the to dopredicate and another proto-role from the lexical verb. These two steps of protorole assignment are amalgamated, resulting in oblique case realisation (at PF). Note that the assumed syntactic structure representations (cf. the above tree structures) reflect the fact that subevent1 (e) takes place prior to subevent2 (e’): I.e., the to do-VP (which contains a silent representation of the later dative/genitive object) is logically built up before its agent(/theme) or patient (/theme) (= the later obliquely marked nonsubject) moves up to a higher position where it is – structurally – designated as a patient (in accordance with Dowty 1991) which is why it receives object case (which is spelled out as oblique case due to the actant’s prior involvement within the additional (sub)event). Moreover, as already noted: If the semantic content representing a complement of a complex event denoting verb like helfen/schämen etc. is not realised at PF (as in the case of topic drop), the grammar module has to guarantee the identification of this element by other means, either (i) (ii)

via identity to an element that displays the very same semantic properties as the null element itself (= verbatim topic drop), or by overtly spelling out the missing subevent (via a controlled infinitive complement or a CP (that represents the missing event)).

Before closing this section, let me summarise the current chapter: In essence, I suppose that the fact that a particular actant (i.e. a verbal argument) plays different roles within an event denoted by an (oblique case assigning) verb is mirrored by its (multiple) proto-properties which are, in turn, manifested in particular case assignment. I.e., it is due to a verb’s internally complex (event) structure that an argument receives a complex case label, namely dative or genitive. As passivation shows, this internally complex label cannot be absorbed under syntactic transformations (due to the fact that oblique case assignment is semantically contentful, i.e. connected to an argument’s additional involvement in a prior (sub)event). Moreover, our findings on topic drop are coherent with other syntactic phenomena, cf. the dialectal relative clause data from Bayer (1984), Riemsdijk

120 | Topic Drop in German: (The Lack of) Identity between Antecedent and Gap

(1989), Salzmann (2005) and Fleischer (2006). I.e., our investigation of topic drop has shown (and confirmed) that an oblique case label must be overtly realised because it represents semantic content – if this semantic content is not overtly realised, specific identity requirements (with respect to dative case: case identity; with respect to genitive case: semantic predicate identity) have to be fulfilled in order to guarantee that the specific properties of an omitted obliquely cased argument can be interpreted – alternatively, the particular subevent, whose presence an obliquely cased element signifies, can be spelled out overtly. I.e., the syntax of different kinds of topic drop (VTD and NVTD) can be ultimately reduced to +/-semantic identity requirements.

2.7 Appendix: Topic Drop of Expletive Elements One last point, which has shortly been mentioned above, relates to the obligation that whenever there is an event, there must be some actant which represents this event or is somehow connected to it (a principle stated by Pustejovsky 1991:77). I assume that such an actant is also present in the context of verbs that select an expletive element as a (quasi-)argument, e.g. weather verbs (as well as other verbs that demand an expletive subject). Even if it is a standard assumption in the literature that an expletive subject is not referential and, hence, cannot be dropped, the following data suggest a different picture: Whenever the event is actually present, the expletive es can be omitted – however, this is not possible when the event is not present in the given discourse, cf. the following minimal pair (adapted from Trutkowski 2011:213):74 (1)

a.

_ Regnet grad’. [uttered while looking out of the window] [EXPL] Rains at-the-moment b. * _ Regnet bestimmt, wenn wir in Urlaub fahren. [EXPL] rains for-sure, when we for holidays drive (‘go on holidays’)

For further (non-German) data see Falk (1993:172) who notes that “[t]he notion ‘topic’ must be given a somewhat wider interpretation, including situations that are often expressed in impersonal expressions as a topic of discourse, like the weather.” Thus, as situations are referential and representable by overt elements (also by expletive pronouns), nothing speaks against the fact that these

|| 74 See also Haegeman & Ihsane (1999:121) for similar diary drop data.

Appendix: Topic Drop of Expletive Elements | 121

overt elements can be dropped. I.e., I assume that the (dropped) expletive pronoun represents the event that is the current situation. This is in accordance with Falk (1993) who argues that not a semantically empty element is dropped but the representative of an actual event.75 Interestingly, in some Northern varieties of German, the expletive in weather contexts can also be realised as a (referential) d-pronoun, cf. das regnet.76

|| 75 One could also argue that da (not in its function as a local adverb) represents an event or a proposition (to which da refers), cf. (i): (i) A: Hans kommt bald zurück. Hans comes surely soon back B: Da bin ich mir nicht so sicher. [There] am I me-DAT (= REFL) not so sure 76 Cf. a user discussion within a blog: http://www.wer-weiss-was.de/dialekte/hamburgisch: In Hamburg müssen wir oft sagen: „Das regnet!“ Jetzt hab ich von einer Besucherin aus Bayern gehört: „Das hört sich ja lustig an, es muß doch heißen ES regnet!“ Ja, nach der Grammatik stimmt das wohl, aber wird das wirklich nur im Norden so gesagt? Nein, auch in Bayern sagt man: „Des regnt heut vielleicht! Koan Hund jagt ma da ausse.“ Das ist emphatische Kritik zum Wetter. Während „Es regnt heut“ nur ein schlichte Mitteilung über die augenblickliche Witterung ist. Gruß! H. Re: Hamburgisch (Autor: Ο l g u, 19.5.2010 11:52 Uhr ) [Translation: In Hamburg, we often have to say Das regnet (That rains). Now I heard a visitor from Bavaria say: Das hört sich ja lustig an, es muß doch heißen ES regnet! (That sounds funny. The correct form should be IT rains) Yes, according to grammar this is correct. But is the first saying actually typical only for people from the North? No, in Bavaria they also say: Des regnt heut vielleicht! Koan Hund jagt ma da ausse. (That rains today indeed. No dog will be chased out) This is emphatic criticism of the weather, whereas Es regnet heut (It rains today) displays only a simple statement about the current weather situation. Greetings.]

3 Topic Drop in German: The Interpretation of the Gap 3.1 Introduction: (Im-)possible Interpretations of VTD and NVTD Gaps In the following I will discuss the interpretational possibilities of the gap in topic drop dialogues. Consider therefore (1), a dialogue involving a gap in a “perfect” verbatim environment (= identical predicates in context and target). (1)

A: B:

Der Hansi hat gestern seinei/k Mutter getroffen. The Hans has yesterday his mother-ACC met _i/k/m Hat der Ottom heute auch getroffen. [ACC] has the Otto today also met

In (1), the VTD (Verbatim Topic Drop) gap has three different interpretations, namely two strict ones, Otto has met Hans’ mother (= strict-1), Otto has met k’s mother (= strict-2), and a sloppy one (Otto has met his own mother).1 Note that it is irrelevant whether context and target predicate assign structural or lexical case to their nonsubject argument. In both examples, cf. (1) and (2), the gap receives two strict readings and one sloppy reading: (2) A: B:

Der Hansi hat gestern seineri/k Mutter geholfen. The Hans has yesterday his mother-DAT helped _i/k/m Hat der Ottom heute auch geholfen. [DAT] has the Otto today also helped

However, context and target predicate do not have to be identical to produce a sloppy reading. Predicate synonymy (i.e. semantic equivalence) is sufficient, cf. (3) and (4):

|| 1 In chapter 3.1 and 3.2 the antecedent of the gap is marked bold to prevent possible confusion regarding the question what the gap’s antecedent might be (especially in examples where an indexical pronoun is dropped). Moreover, for the reader’s convenience I often represent the gap (which I suppose to be of the category pro) as a crossed-out entity, especially when an indexical (1st/2nd person) pronoun is dropped and interpretational possibilities are hard to get.

Introduction: (Im-)possible Interpretations of VTD and NVTD Gaps | 123

(3) A: B:

(4) A: B:

Der Hansi ist gestern seinemi/k Prof begegnet. The Hans is yesterday his prof-DAT came-across _i/k/m Hat der Ottom heute auch getroffen. [ACC] has the Otto today also met Der Hansi hat gestern seineni/k Prof getroffen. The Hans has yesterday his prof-ACC met _i/k/m Ist der Ottom heute auch begegnet. [DAT] is the Otto today also came-across

The fact that sloppy readings also arise with synonymous predicates further confirms our approach to treating VTD with identical context/target predicates and synonymous context/target predicates on a par.2 Let me turn to NVTD (Non-Verbatim Topic Drop) examples: As (5) shows, a sloppy gap interpretation is neither possible in a non-verbatim environment (where context/target predicates are neither identical nor synonymous), nor in an environment where context and target predicate differ semantically but case features of antecedent and gap are the same,3 cf. (6): (5) A: B:

(6) A: B:

Der Hansi hat seineri/k Mutter gestern beim Umzug geholfen. The Hans has his mother-DAT yesterday at moving-house helped _i/k/*m Hat der Ottom heute im Supermarkt getroffen. [ACC] has the Otto today in-the supermarket met Der Hansi hat seineri/k Mutter gestern beim Umzug geholfen. The Hans has his mother-DAT yesterday at moving-house helped _i/k/*m Ist der Ottom heute im Supermarkt begegnet. [DAT] is the Otto today in-the supermarket came-across

The gaps in (5) and (6) cannot be bound by a respective binder in the target sentence, but can only be interpreted strictly, referring back to the contextually given antecedent. Hence, a NVTD gap can only have two readings, namely the

|| 2 Moreover, the fact that sloppy readings arise under synonymous predicates in context and target constitutes an additional (semantic) argument against an analysis of VTD in terms of stripping minus verb ellipsis (see also chapter 2.2.2). 3 I consider an ‘identical case + different predicates’ scenario as an instance of NVTD (but see chapter 2.3, chart (14) for details).

124 | Topic Drop in German: The Interpretation of the Gap

strict-1 and the strict-2 reading,4 lacking the sloppy interpretation. Therefore one might say that a NVTD gap is “discourse bound”/‘Dbound’ in the sense of Hardt (1993:13), i.e. “it may refer to some salient object in the current discourse context.” The property of non-verbatim environments that topic dropped elements can only be interpreted strictly therein (i.e. referentially dependent on an antecedent) correlates with the fact that non-referential NPs (e.g. QPs) can never be dropped in non-verbatim environments, cf. (7a) – whereas they can in verbatim environments, cf. (7b): (7) A: a. b.

Maria kennt fast jede Oper. Maria knows almost every opera-ACC B: * _ Kann der Hans sogar mitsingen. [ACC] can the Hans even sing-along B: _ Kennt der Hans auch. [ACC] knows the Hans as-well

Non-verbatim environment Verbatim environment

However, not all gaps dropped in verbatim environments display sloppy readings, because the sloppy reading depends on a binding configuration: If, as in (8), the dependent element (here: mein (my) in mein Prof) is not c-commanded by the coindexed element, binding is impossible (cf. Chomsky 1981, Reinhart 1983 and many others) and the gap only displays a strict reading: (8) A: B:

Meini/k Prof hat michi neulich kritisiert. My prof-NOM has me-ACC recently criticised _i/*m Hat michm neulich auch kritisiert. [NOM] has me-ACC recently also criticised

What is relevant in this respect (i.e. with respect to binding) is not the surface but the underlying order of the respective arguments in the middle field which

|| 4 Note with respect to the strict-2 reading (by which antecedent and the gap refer to the same “external” discourse referent) that the introduction of such an “external” discourse referent sweeps aside all other interpretations (in both, VTD/NVTD environments), cf. (i): (i) Was ist mit Ottok? (What about Otto?) A: Hansi hat gestern seinen?i/k Prof gesehen. Hans has yesterday his prof-ACC seen B: _?i/?m/k Hat Olafm heute auch gesehen/getroffen. [ACC] has Olaf today also seen/met

Introduction: (Im-)possible Interpretations of VTD and NVTD Gaps | 125

can be determined by existentially interpreted w-indefinites (which can hardly be moved from their base position, see e.g. Frey & Pittner 1998), cf. (9)/(10): (9) A: B:

Seini/k Auto liegt jedemi Studenten am Herzen. His car-NOM lies every student-DAT at-the heart _k/m Liegt auch jedemm Prof am Herzen. [NOM] lies also every prof-DAT at-the heart

(10) a.

Weil wem was am Herzen liegt Because someone-DAT something-NOM at-the heart lies b. * Weil was wem am Herzen liegt Because something-NOM someone-DAT at-the heart lies

In accordance with general conditions on binding, a binding relation in the target sentence of a topic drop dialogue can only be established when binder and bindee in the context are coindexed. Coindexication involves feature matching/feature sharing between a binder and a bindee or two coreferent elements, respectively. In German, the relevant (phi-)features are person, number and gender. If the contextually given elements are not coindexed, the gap in the target cannot be bound (i.e. a sloppy gap interpretation is impossible), cf. (11): (11) A: B:

Irenei mag seine*i/k Schwester. Irene likes his sister-ACC _k/*m Mag der Hansm auch. [ACC] likes the Hans as-well

Moreover, a sloppy gap reading is only possible when the bound element is a non-rigid designator so that its reference can vary across different contexts (here: across context and target). Thus, rigid designators in topic drop dialogues are mostly excluded from being interpreted sloppily. There is an apparent exception, namely definite expressions in ‘knight move’ binding contexts5 where free datives act as binders of a hidden possessive element, cf. (12): (12) A:

Heute ist miri deri Fernseher implodiert. Today is me-DAT the tv-set-NOM imploded

|| 5 For ‘knight move’ binding see e.g. Hole (2012).

126 | Topic Drop in German: The Interpretation of the Gap

B:

_*i/m Ist dem Hansm gestern auch implodiert. [NOM] is the Hans-DAT yesterday also imploded

(13) provides a short summary of the conditions under which sloppy gap readings are available in topic drop contexts: (13) Conditions on sloppy gap readings in topic drop dialogues (a) The antecedent of the gap must be a non-rigid designator such that its reference can vary across context and target (b) Syntactic conditions on binding must be met (cf. c-command) (c) Presence of a binding/coreference relation in the context (cf. two coindexed elements) (d) Identical/synonymous predicates in context and target (= verbatim environment) Before going on, I will shortly discuss the question by which category the null element could be represented. The following examples show that both, (i) the assumption of a deleted d-pronoun (cf. Balkenende 1995,6 Bayer et al. 2001) as well as (ii) the assumption of a PF-identical deleted antecedent copy cannot handle the data properly. A deleted d-pronoun cannot account for the sloppy reading of a topic drop gap, because a d-pronoun has to refer back (to some discourse given antecedent) in order to be interpreted. Thus, the assumption of a deleted d-pronoun is not compatible with a sloppy gap interpretation, cf. (14) and (15): (14) A: B:

(15) A: B:

Der Hansi hat gestern seinei/k Mutter getroffen. The Hans has yesterday his mother-ACC met Diei/k/*m hat der Ottom heute auch getroffen. [d-pronoun-ACC] has the Otto today also met Gestern hab ichi meineri Mutter geholfen. Yesterday have I my mother-DAT helped Deri/*m hab ichm heute auch geholfen. [d-pronoun-DAT] have I today also helped

|| 6 Balkenende (1995) is an M.A. Thesis (written in Dutch) that is referred to by Thrift (2003).

Introduction: (Im-)possible Interpretations of VTD and NVTD Gaps | 127

Furthermore, as (16) shows, deleted d-pronouns are restricted to the 3rd person and hence they cannot represent 1st/2nd person pronouns – neither in their strict (here: euch (you-pl-ACC)) nor in their sloppy (here: uns (us-ACC)) interpretation, cf. (16): Wiri mögen unsi. We like us/ourselves-ACC B: * Diei/m mögen wirm auch. [d-pronoun-ACC] like we as-well

(16) A:

Most authors who favour a copy-and-deletion approach to ellipses do not necessarily assume that antecedent and elided constituent have to be fully PFidentical. In fact, (one-to-one) PF-deleted antecedent copies cannot account for the strict readings of indexical pronouns, cf. (17), and of DPs which contain an indexical (possessive) pronoun, cf. (18): (17) A: B:

(18) A: B:

Ichi mag michi. I like me/myself-ACC Mich*i/m mag ichm auch. [Myself-ACC] like I as-well Ichi hab gestern meineri Mutter geholfen. I have yesterday my mother-DAT helped Meiner*i/m Mutter hab ichm heute auch geholfen. [My mother-ACC] have I today also helped

Thus, phonological deletion approaches, as e.g. Wilder (1996), propose that the syntactic structure of an elided and a non-elided constituent is more or less identical and that the only thing that the elliptic sentence is lacking is a phonological representation of the material in the prefield (cf. Wilder 1996:4): [...] so daß der elliptische Ausdruck und sein vollständiges Pendant im wesentlichen syntaktisch identisch sind. Was dem elliptischen Satz fehlt, ist nur ein Teil der entsprechenden phonologischen Repräsentation, der getilgt worden ist.

To prevent the application of a defective (non-V2) syntactic structure as (19a) from application, with respect to topic drop (and null subjects) Wilder (1996:4) argues for a representation as given in (19b):

128 | Topic Drop in German: The Interpretation of the Gap

(19) a. b.

[habe das schon beschrieben] [have it/that already described] [[ich] [habe das schon beschrieben]] [[I] [have it/that already described]

Thus, to account for the different readings of topic dropped elements neither the assumption of a deleted d-pronoun nor a PF-deleted (one-to-one) antecedent copy is a good choice. Besides approaches as Wilder’s (1996), a further alternative is the assumption that the topic drop gap is represented by a null pronoun pro, which is (also) sufficiently underspecified to account for data as (14)–(18). When we compare Wilder’s (1996) syntactic deletion approach with the assumption of a PF-flexible pro-category, the differences between the one and the other do not concern the nature of restrictions but rather the layer or level of their application: Whereas Wilder (1996) has to impose rules that constrain phonological deletion (e.g. the amount of deleted material or the position where deletion can take place), approaches arguing for the existence of a null pronoun must say something about its structure, its licensing and identification conditions and explain why pro is confined to the prefield. I will come back to that question later (cf. chapter 3.4), after having discussed the interpretational possibilities of topic dropped elements.

3.2 Two Different Kinds of Binding in Topic Drop Dialogues As already mentioned, one of the prerequisites for a sloppy interpretation of a topic drop gap is that binder and bindee in the context and in the target, respectively, are coindexed – which means that they must share relevant phi-features. As the gap’s phi-features cannot be read off from PF (= Phonetic/Phonological Form), I will look at the phi-features of the gap’s antecedent and see how the phi-feature sets of the antecedent’s bindee part (= context bindee) and those of the target binder are related to each other, i.e., whether binding is allowed despite phi-feature mismatches between the context bindee (which in most of the following examples is a possessive pronoun) and the target binder.7

|| 7 Note that with some relational nouns as mother, father etc. the sloppy reading is often preferred. This preference is due to pragmatic factors, cf. world knowledge (e.g. it is more reasonable/natural to assume that I met/love my own mother than the mother of someone else). In the following this fact is taken into account.

Two Different Kinds of Binding in Topic Drop Dialogues | 129

I will examine two kinds of binding relations which differ from each other with respect to the fact whether the element that binds the gap’s bindee part in the target sentence is a referential or a non-referential element. Consider (1) and (2) for a first differentiation: (1)

A: B:

(2) A: B:

Irenei mag ihrei/k Frisur. Irene likes her hair-cut-ACC _i/k/m Mag die Mariam auch. [ACC] likes the Maria as-well Irenei mag ihrei/k Frisur. Irene likes her hair-cut-ACC _i/k/m Mag fast jedem Frau. [ACC] likes almost every woman

Irene = Binder; ihre = Bindee ‘Referential binding’

Irene = Binder; ihre = Bindee ‘Non-referential binding’

Note that I will use the notions ‘binder’ and ‘bindee’ even if the relation between the coindexed elements is ambiguous between coreference and binding. Besides the sloppy reading, I will also discuss the strict readings (strict-1; strict-2) of topic drop gaps, which are in most examples indicated by the indices ‘i’ and ‘k’.

3.2.1 The Target Binder is a Referential Expression I will first examine the interpretational possibilities of a topic dropped element within (what I call) the ‘referential binding relation’, that is, a binding relation where the target binder is a specific and referential term. I start with a discussion of gender feature (mis)matches between the target binder and the context bindee (which is the antecedent of the gap’s bindee part). Consider first the minimal pair in (3): (3) A: a.

B:

b.

B:

Irenei mag ihrei/k Schwester. Irene likes her sister-ACC _i/k/m Mag die Annam auch. [ACC] likes the Anna as-well _i/k/*m Mag der Hansm auch. [ACC] likes the Hans as-well

Binding No binding

130 | Topic Drop in German: The Interpretation of the Gap

Note that the gap in (3a) displays a sloppy reading, whereas a sloppy reading of the gap is clearly blocked in (3b). By looking at the above example in more detail, we see that the only difference between (3a) and (3b) is that the target binder in (3a) is feminine and in (3b) it is masculine. Thus, referential binding seems to be blocked whenever the phi-features of the context bindee (here: ihre in ihre Schwester) are not compatible with the phi-features of the target binder (here: der Hans). In this case (e.g. due to a gender mismatch as in (3b)), binding cannot take place and the gap does not receive a sloppy reading. When we change the context binder and its bindee from feminine marked to masculine marked, the formerly bound gap in (3a) becomes unbound, cf. (4a), whereas the formerly unbound gap in (3b) becomes bound, cf. (4b), due to now different gender specifications of the coindexed elements in the context: (4) A: a.

B:

b.

B:

Ottoi mag seinei/k Schwester. Otto likes his sister-ACC _i/k/*m Mag die Annam auch. [ACC] likes the Anna as-well _i/k/m Mag der Hansm auch. [ACC] likes the Hans as-well

No binding Binding

An analogous phi-feature matching process between context bindee and target binder can be observed with respect to number feature (mis)matches, cf. (5a) vs. (5b): (5) A: a.

B:

b.

B:

Die Müllersi finden ihreni/k Hausarzt total kompetent. The Müllers find their family-doctor-ACC absolutely competent _i/k/m Finden die Mayersm auch kompetent. Binding [ACC] find the Mayers also competent _i/k/*m Findet der Hansm auch kompetent. No binding [ACC] finds the Hans also competent

Equivalent observations can be made with respect to person feature (mis)matches. However, with respect to person a departure from feature identity between context bindee and target binder seems to be marginally possible (at least for some speakers), cf. (6a) vs. (6b) and (7a) vs. (7b) as well as (8) and (9): (6) A:

Hansi mag seineni/k Prof. Hans likes his prof-ACC

Two Different Kinds of Binding in Topic Drop Dialogues | 131

a.

B:

b.

B:

(7) A: a.

B:

b.

B:

(8) A: B:

(9) A: B:

_i/k/?m Mag ichm auch. [ACC] like I as-well _i/k/m Mag der Olafm auch. [ACC] likes the Olaf as-well Ichi mag meineni Prof. I like my prof-ACC _i/m Mag ichm auch. [ACC] like I as-well _i/?m Mag der Olafm auch. [ACC] likes the Olaf as-well Die Erstklässleri mögen ihrei/k Lehrerin. The first-formers like their teacher-fem-ACC _i/k/?m Mögen wirm auch. [ACC] like we as-well

? Binding Binding

Binding ? Binding

? Binding

Ihri könnt eurei Lehrerin ja gar nicht leiden! You-pl can your-pl teacher-fem-ACC PRT absolutely not stand _i/?m Können die Zweitklässlerm auch nicht leiden. ? Binding [ACC] can the second-class-pupils also not stand

I.e., it seems that the context bindee must bear the same formal phi-feature specifications as the target binder (modulo person adjustment) such that a binding relation in the target can be established and the target gap can be interpreted sloppily. However, considering data as (10)–(12) we see that such a conclusion would be premature: Instead, it is relevant that the phi-features of context bindee and target binder are compatible at PF or, in other words, that the context bindee and the gap’s bindee part bear the same surface form (under the hypothetical assumption that the gap would be overtly realised). I.e., the phi-feature mismatches that are reported to disallow binding in (3)–(9) can be compensated under a syncretic form, as in (10b)/(11b)/(12a/b), where the (underlying) surface form of the possessive pronoun (involving a neuter possessum) of the 3rd person singular feminine and the 3rd person plural of all genders, respectively, is the same, (namely ihr). (10) A:

Irenei mag ihri/k Auto. Irene likes her car-ACC

132 | Topic Drop in German: The Interpretation of the Gap

a.

B:

b.

B:

(11) A: a.

B:

b.

B:

(12) A: a.

B:

b.

B:

_i/k/*m Mag der Hansm auch. No binding [ACC] likes the Hans as-well _i/k/m Mögen die Hansen-Brüderm auch. Binding (under syncretisms) [ACC] like the Hansen-brothers-masc as-well Die Hansen-Brüderi mögen ihri/k Auto. The Hansen-brothers-masc like their car-ACC _i/k/*m Mag der Hansm auch. No binding [ACC] likes the Hans as-well _i/k/m Mag die Irenem auch. Binding (under syncretisms) [ACC] likes the Irene as-well Die Hansen-Schwesterni mögen ihri/k Auto. The Hansen-sisters-fem like their car-ACC _i/k/m Mögen die Jacobsen-Brüderm auch. Binding (under syncretisms) [ACC] like the Jacobsen-brothers-masc as-well _i/k/m Mögen die Jacobsen-Kinderm auch. Binding (under syncretisms) [ACC] like the Jacobsen-children-neut as-well

The following table illustrates the distribution of syncretisms within the paradigm of possessive pronouns in German.8 In particular, (13a–c) show the distribution of syncretisms in the accusative case for all persons and numbers: (13a) ACC-possessum (= the object possessed by the NOM-possessor) is masculine

Person/Number

1st

2nd

3rd

Singular

meinen

deinen ihren (fem. possessor), seinen (masc./neut. possessor)

Plural

unseren euren

ihren

|| 8 Note that in German we have possessor-possessum agreement (which means that the stem of the possessive pronoun “agrees” with the possessor and its ending agrees with the possessum, cf. e.g. Irene mag ihren Bruder / ihre Schwester (Irene likes her brother-masc / her sisterfem); Hans mag seinen Bruder / seine Schwester (Hans likes his brother-masc / his sister-fem).

Two Different Kinds of Binding in Topic Drop Dialogues | 133

(13b) ACC-possessum (= the object possessed by the NOM-possessor) is female

Person/Number

1st

2nd

3rd

Singular

meine

deine

ihre (fem. possessor), seine (masc./neut. possessor)

Plural

unsere

eure

ihre

(13c) ACC-possessum (= the object possessed by the NOM-possessor) is neuter

Person/Number

1st

2nd

3rd

Singular

mein

dein

ihr (fem. possessor), sein (masc./neut. possessor)

Plural

unser

euer

ihr

From the data in (3)–(12) – and in particular the data (10)–(12) – we can draw the following conclusion, cf. (14): (14) Referential binding condition Phi-feature mismatches that are observable between a referential target binder and a context bindee are fine as long as the overtly realised form of the context bindee (= the antecedent of the gap’s bindee) is PF-identical with the covertly underlying form of the target bindee (= the gap’s bindee part). (15) illustrates how the relevant parts are involved in the referential binding relation: (15) ‘Referential Binding’ context binder … context bindee (= the gap’s bindee’s antecedent/ feature alignment = the antecedent of the bindee part of the gap) gap’s bindee part … target binder PF identity between context bindee and target bindee (contained within the gap) The single underlining (and the single line) relates to the PF identity obligation that has to hold between context bindee and target bindee (= the gap’s bindee part). The double underlining (and the double line) relates to ‘feature alignment’ between the target binder and the context bindee. Feature alignment can be understood as a PF-related checking mechanism regarding the phi-feature

134 | Topic Drop in German: The Interpretation of the Gap

identity requirements of the target binder and the antecedent of the bindee part of the gap observable at the surface – alternatively I will speak of (PF-related) phi-feature compatibility between target binder and context bindee. The PF identity obligation, on the other hand, is related to the context bindee and the gap’s bindee part and cannot be read off from PF because the gap’s bindee part is not realised at PF. Note that a sloppy reading of the target gap is independent of the particular referential properties of the context binder (which is a non-referential term in (16) and a referential term in (17)): (16) A: a.

B:

b.

B:

(17) A: a.

B:

b.

B:

Fast allei Studenten mögen ihreni/k Hund. Almost all students like their dog-ACC _k/m Mag die Mariam auch. Binding [ACC] likes the Maria as-well _k/*m Mag der Peterm auch. No binding [ACC] likes the Peter as-well Die Elterni mögen ihreni/k Hund. The parents like their dog-ACC _i/k/m Mag die Mariam auch. [ACC] likes the Maria as-well _i/k/*m Mag der Peterm auch. [ACC] likes the Peter as-well

Binding No binding

However, the referential properties of the context binder have an impact on the strict readings of the gap: In the presence of a (quantifier) binding relation in the context (i.e. when the context binder is a non-referential term) the gap in the target does not display a strict-1 reading, cf. the examples in (16a/b). By contrast, when the relation between the coindexed elements in the context is ambiguous between coreference and binding (i.e. when the context binder is a referential term), cf. the examples (17a/b), the gap displays a strict-1 reading. Thus, we can conclude that the lack of the strict-1 reading in (16) and its presence in (17) must be due to the referential properties of the respective context binder. The hypothesis expressed in (18) relates the properties of the context binder (or better: the relation between the coindexed elements in the context) to the possibility of a strict-1 reading of the gap:

Two Different Kinds of Binding in Topic Drop Dialogues | 135

(18) Conditions for the presence of a strict-1 reading of the gap a. The index that the context bindee has received from a coindexed element with which it stands in a possible coreference relation in the context can be transferred/copied from the context to the target (or, alternatively: inferred/taken from the context to the target). As a consequence the target gap displays a strict-1 reading. b. The index that the context bindee has received from a coindexed element with which it stands in a genuine binding relation cannot be transferred/copied from the context to the target (or, alternatively: not inferred/taken from the context to the target). As a consequence the target gap does not display a strict-1 reading. Intuitively, (18) must be correct because a non-referential term is not fixed or anchored within the discourse.9 Consequently, an index that is assigned from a non-referential binder to its bindee (in the context) cannot be transferred to the target gap – which (as I suppose) is due to the fact that it is assigned on a volatile/momentary “relational” basis and not on a fixed “referential” basis. I.e., an index that is assigned on the basis of a genuine binding relation is not fixed but variable, and hence it cannot be (further) passed on.10 By contrast, an index that is assigned on the basis of a coreference relation between two elements is not variable but fixed. I will refer to the respective indices as ‘binding index’ vs. ‘referential index’. The following data, (19)–(21), confirm the statements made in (18): (19) A: B:

(20) A:

Keini Student putzt regelmäßig seini/k Auto. No student cleans regularly his car-ACC _k/*m Putzt die Mariam auch nicht. [ACC] cleans the Maria also not Hansi putzt regelmäßig seini/k Auto. Hans cleans regularly his car-ACC

Non-referential binder No strict-1 reading/ No binding Referential binder

|| 9 Also, a non-referential element does not constitute a possible aboutness topic (see e.g. Reinhart 1981 or Frey 2004). 10 Consider also (i) where the antecedent of the gap is a non-referential element: (i) A: Annai mag jede Oper. Anna likes every opera-ACC B: _*m/*t Mag Irenem / Hanst auch. No referential anchoring [ACC] likes Irene/Hans as-well

136 | Topic Drop in German: The Interpretation of the Gap

B:

(21) A: B:

(22) A: B:

_i/k/*m Putzt die Mariam auch. [ACC] cleans the Maria as-well

Strict-1 reading/ No binding

Einei Frau kennt ihrei/k Schwächen. A woman knows her deficiencies-ACC _(i)/k/m Kennt die Mariam leider nicht. [ACC] knows the Maria unfortunately not

(Non-)referential binder

Annai kennt ihrei/k Schwächen. Anna knows her deficiencies-ACC _i/k/m Kennt die Mariam leider nicht. [ACC] knows the Maria unfortunately not

Referential binder

(No11) strict-1 reading/ Binding

Strict-1 reading/ Binding

(23) summarises the interpretational possibilities of a gap bound by a referential element: (23) Gap interpretations in the presence of a referential binder in the target a. A sloppy gap reading arises only when the phi-features of the context bindee are PF-compatible with the phi-features of the target binder (so that the surface form of the context bindee is PF-identical with the covertly underlying form of the target bindee). b. A strict-1 gap reading arises when the context binder is a referential element and the (coindexed) elements in the context stand to each other in a coreference relation. c. A strict-2 gap reading arises independently of the fact whether the context binder is a referential or a non-referential element and only when the context bindee is a 3rd person possessive pronoun. I.e., a strict-2 reading does not arise when the context bindee is an indexical pronoun or a 1st/2nd person possessive pronoun, cf. datum (7) above. Besides, the strict-2 reading of the gap arises independently of a coreference or binding relation in the context. Before turning to the next section, I would like to point out that formal phifeature mismatches in referential binding settings cannot be “salvaged” by natural/inherent feature matches, i.e. binding is not possible under identity of

|| 11 Note that this example is ambiguous. When the indefinite article is interpreted nonspecifically, the gap displays no strict-1 reading. However, when it is interpreted specifically, the gap displays a strict-1 reading.

Two Different Kinds of Binding in Topic Drop Dialogues | 137

natural gender features of the context bindee/the gap’s bindee part and the target binder as long as (the PF representation of) their formal features diverge(s). This is shown by the following example where the target binder das kleine Mädchen (the little girl) is formally neuter (although its natural gender/sex is feminine) and the context bindee/the bindee part of the gap ihre (her) is feminine (formally and naturally), cf. (24): (24) A: B:

Die Omai mag ihrei/k Nachbarin. The grandma likes her neighbour-fem-ACC _i/k/*m Mag das kleine Mädchenm auch. [ACC] likes the little girl-neut as-well

No binding

However, see Sauerland (2008), referring to Spathas (2007), who finds that in a German stripping construction pseudo-sloppy readings are allowed when natural phi-features are matched. Sauerland (2008:346) notes the following [F]or the availability of the pseudosloppy interpretation the grammatical gender of the subject of the elided IP is irrelevant, as long as the natural gender is FEM. This is confirmed by the availability of sloppy interpretation in both ([25a]) and ([25b]), where the grammatical gender of the subject of the elided IP is respectively MASC and NEUT.

(25) a.

b.

Das Mädchen hat ihre Zähne geputzt und der weibliche The-neut girl-neut has her-fem teeth cleaned and the-masc female Star auch. star-masc also (strict/sloppy) Das kleine Mädchen hat ihre Eltern angerufen und das The-neut little girl-neut has her-fem parents called and the-neut weibliche Opfer auch. female victim-neut also (strict/sloppy)

However, the fact that stripping allows for phi-feature mismatches does not mean that phi-feature mismatches are possible in topic drop dialogues as well: As shown in chapter 2, topic drop and stripping differ from each other essentially. This is e.g. proven by the fact that under stripping an idiomatic meaning is not lost, (26a), but it is lost under topic drop, (26b), cf. Fanselow (p.c.) who proposed this test to me: (26) A:

Gestern hab ich dem Hans Honig um den Mund geschmiert. Yesterday have I the Hans honey-ACC around the mouth spread (‘Yesterday I was buttering up Hans’)

138 | Topic Drop in German: The Interpretation of the Gap

a.

B:

b.

B:

[e] (Hab) ich dem Udo heute auch. [Honey around the mouth spread] (have) I the Udo today also _ Hab ich dem Udo heute auch um den Mund geschmiert.12 [Honey] have I the Udo today also around the mouth spread

Thus, I conclude that with respect to referential binding the only possible departure from feature identity between context bindee and target bindee concerns the feature person, as shown by (27) (= (8) from above):13 (27) A: B:

Die Erstklässleri mögen ihrei/k Lehrerin. The first-formers like their teacher-fem-ACC _i/k/?m Mögen wirm auch. [ACC] like we as-well

? Binding

3.2.2 The Target Binder is a Non-referential Expression In this section I will discuss what I call a ‘non-referential binding relation’, that is, a binding relation where the target binder is a non-specific, non-referential term. I will show that PF/phi-feature identity between context bindee and the gap’s bindee part is not a necessary condition when non-referential expressions act as possible target binders.14 I.e., topic dropped possessive elements can be bound despite formal or at PF noticeable phi-feature mismatches. The relevant data are presented below, in the respective a-examples (which are contrasted with the b-examples where the target binder is a referential element), cf. (28)– (30): Gender feature mismatch (28) A:

Irenei mag ihreni/k Vater. Irene likes her father-ACC

|| 12 For a further comparison of (quasi-)stripping and verbatim topic drop, see chapter 2.2.2. 13 As yet, I have no explanation why a departure from person features is (marginally) possible in referential binding contexts. However, it is worth noting that only referential terms can differ with respect to the feature ‘person’ because non-referential terms (as e.g. QPs) are always marked for the 3rd person (which some authors refer to as ‘non-person’, cf. Benveniste 1971). 14 As already mentioned above, instead of PF/phi-feature identity between context bindee and the gap’s bindee part one can also speak of PF/phi-feature compatibility between context bindee and target binder.

Two Different Kinds of Binding in Topic Drop Dialogues | 139

a.

B:

b.

B:

_i/k/m Mag auch fast jederm Junge. [ACC] likes also almost every boy _i/k/*m Mag der Hansm auch. [ACC] likes the Hans as-well

Binding (QP)15 No binding (DP/NP)

Person feature mismatch (29) A: a.

B:

b.

B:

Ichi mag meinei Frisur. I like my hair-cut-ACC _i/m Mag fast jedem Frau. [ACC] likes almost every woman _i/?m Mag der Hansm auch. [ACC] likes the Hans as well

Binding (QP) ? Binding (DP/NP)

Number feature mismatch (30) A: a.

B:

b.

B:

Hansi mag seineni/k Vater. Hans likes his father-ACC _i/k/m Mögen fast allem Jungs. [ACC] like almost all boys _i/k/*m Mögen die Hansen-Brüderm auch. [ACC] like the Hansen-brothers as-well

Binding (QP) No binding (DP/NP)

Clearly, the ‘referential binding condition’, cf. (14) from above, does not hold for non-referential target binders: (14) Referential binding condition Phi-feature mismatches that are observable between a referential target binder and a context bindee are fine as long as the overtly realised form of the context bindee (= the antecedent of the gap’s bindee) is PF-identical with the covertly underlying form of the target bindee (= the gap’s bindee part).

|| 15 When the target binder is a non-referential element, the insertion of auch (= as well, also, too) makes the strict reading a bit more prominent (as compared to Mag fast jeder Junge). However, when the target binder is a referential element, the insertion of auch is often obligatory, because the target constitutes a verbatim repetition of the context.

140 | Topic Drop in German: The Interpretation of the Gap

Instead, when the target binder is a non-referential element, we must assume something like (31), which negates the necessary condition expressed in (14): (31) Non-referential binding condition16 For a topic drop gap to be bound by a non-referential element it is not necessary that the phi-features of the context bindee are identical/PFcompatible with those of the target binder. Consider also the illustrations of the different binding instances/processes: first, when the target binder is a referential element, cf. (15) from above, and second, when the target binder is a non-referential element, cf. (32): (15) ‘Referential Binding’ context binder … context bindee (= the gap’s bindee’s antecedent/ feature alignment = the antecedent of the bindee part of the gap) gap’s bindee part … target binder PF identity between context bindee and target bindee (contained within the gap) is necessary (32) ‘Non-Referential Binding’ context binder … context bindee (= the gap’s bindee’s antecedent/ no feature alignment = the antecedent of the bindee part of the gap) gap’s bindee part … target binder PF identity between context bindee and target bindee (contained within the gap) is not necessary Note that in non-referential binding scenarios not only the phi-features/the PFform of context bindee and target bindee can be different, but distributed readings of the topic dropped element are possible, too, cf. (33) – in contrast to (34) where the target binder is a referential element: (33) A: B:

Die Nachbarskinderi mögen ihrei/k Mütter. The neighbour-children like their mothers-ACC _i/k/m Mag jederm Junge. (sloppy and distributed; [ACC] likes every boy cf. Seine Mütter17, Seine Mutter)18

|| 16 In fact, (31) is not a condition (on binding), but a negation of the application of a particular condition.

Two Different Kinds of Binding in Topic Drop Dialogues | 141

(34) A: a.

B:

b.

B:

Die Nachbarskinderi mögen ihrei/k Mütter. The neighbour-children like their mothers-ACC _i/k/*m Mag der Hansm auch. (not sloppy, not distributed; [ACC] likes the Hans as-well cf. *Seine Mütter, *Seine Mutter) _i/k/m Mag die Irenem auch. (sloppy but not distributed; [ACC] likes the Irene as-well cf. Ihre Mütter, *Ihre Mutter)

Of course, non-referential binding does not only apply to quantified expressions but to non-specific indefinite NPs as well, cf. (35): (35) A: B:

Einei Frau kennt ihrei/k Schwächen. A woman knows her deficiencies-ACC _(i)/k/(m) Kennt einm Mann leider nicht. [ACC] knows a man unfortunately not

(Non-)referential binder19

3.2.3 Interim Summary: Referential vs. Non-referential Binding The main findings with respect to the different interpretations of topic dropped elements which are derived from an antecedent that is a non-rigid designator can be summarised as follows, cf. (36) and (37): (36) Strict readings of a topic dropped element a. The presence of the strict-1 reading is dependent on the presence of a referential context binder – if the context binder is a non-referential element, the strict-1 reading disappears. Indefinite context binders are ambiguous, i.e. the gap receives a strict-1 reading only when the context binder is interpreted specifically. b. The strict-2 reading is present whenever the context bindee is a 3rd person possessive pronoun.

|| 17 Note that in principle this reading is semantically ruled out as a person cannot have more than one (biological) mother. However, a scenario where this interpretation is an option, is imaginable. 18 To account for the differently distributed sloppy reading(s), Ede Zimmermann (p.c.) suggested to include a distributive operator ‘Δ’ into the context, as in Δj Die Nachbarskinderi mögen ihrei/k/j Mütter. 19 The gap receives a sloppy reading when the target binder is interpreted non-specifically, and it receives a strict-1 interpretation when the context binder is interpreted specifically.

142 | Topic Drop in German: The Interpretation of the Gap

(37) Sloppy readings of a topic dropped element a. When the target binder is a referential element, binding the bindee part of the gap (= the target bindee) is only possible under (PFrelated) phi-feature compatibility of the gap’s bindee antecedent (= the context bindee) and the target binder. In this case context bindee and target bindee bear the same PF-form. b. When the target binder is a non-referential element it is not necessary that the phi-features of the gap’s bindee antecedent (= the context bindee) and the target binder are identical/PF-compatible. Note further that – as already mentioned above – a sloppy gap interpretation is (always) dependent on the preliminary conditions in (38) (= (13) from 3.1): (38) Conditions on sloppy gap readings in topic drop dialogues (a) The antecedent of the gap must be a non-rigid designator such that its reference can vary across context and target (b) Syntactic conditions on binding must be met (cf. c-command) (c) Presence of a binding/coreference relation in the context (cf. two coindexed elements) (d) Identical/synonymous predicates in context and target (= verbatim environment) In the next section I will shortly investigate and discuss topic drop of the reflexive pronoun sich. After that I will turn to topic drop of indexical pronouns (namely 1st/2nd person objects).

3.2.4 The Interpretation of the Reflexive Pronoun sich The German reflexive pronoun sich is an element whose reference cannot be established independently, without reference to a (binding) element (on which the interpretation of sich depends). The question is whether sich can act as antecedent of a topic dropped element and – if so – how it is interpreted. If we look at the data in (39) and examine the possible interpretations that emerge from a dropped reflexive pronoun, we see that there is nothing wrong with reflexives acting as antecedents of the gap,20 i.e. the gap can be interpreted || 20 Due to its grammatical function/its case features the context binder is not a possible antecedent for the gap, see chapter 2.1.2, data (26)–(29).

Two Different Kinds of Binding in Topic Drop Dialogues | 143

strictly. The sloppy reading is present whenever the conditions that apply with regard to referential or non-referential binding, respectively, are obeyed: (39) A: a.

B:

b.

B:

c.

B:

d.

B:

Der Hansi mag sichi. The Hans likes REFL-ACC (= himself) _i/m Mag der Olafm auch. [Him/Himself] likes the Olaf as-well _i/?m Mag ichm auch. [Him/Myself] like I as-well _i/m Mögen die Elternm auch. [Him/Themselves] like the parents as-well _i/m Mögen fast allem Studenten. [Him/Themselves] like almost all students

However, if we consider an example where the reflexive in the context is bound by a non-referential element, the gap bears no index which it could have inherited from the context (see also (16)–(18), chapter 3.2.1). As a consequence, it can at most display a sloppy reading (as it does in (40) – under phi-feature identity/compatibility between context bindee and target binder): (40) A: B:

Fast jederi Student mag sichi. Almost every student likes REFL-ACC (= himself) _m Mag der Olafm auch. [Himself] likes the Olaf as-well

I.e., the felicity of the sloppy interpretation of a gap derived from sich is independent of whether the context binder is a referential or non-referential element,21 because for the well-formedness of a gap’s sloppy reading its antecedent (i.e. sich) does not need to be anchored within the discourse. As already shown in chapter 2.2.2, sich can hardly be dropped when occurring with inherently reflexive verbs as schämen (to feel ashamed of oneself). Moreover, sloppy readings of topic dropped elements (not only with respect to sich) are “most well-formed” with verbs as mögen (to like), lieben (to love), hassen (to hate) etc. According to standard assumptions (see e.g. Bausewein 1990:92), the common property of these verbs is their (event-related) property that the nominative argument can be characterised as ‘Experiencer’ and the || 21 However, note that with respect to referential binding PF/phi-feature identity between target binder and context bindee is a necessary condition (as discussed in 3.2.1–3.2.3).

144 | Topic Drop in German: The Interpretation of the Gap

accusative argument as ‘Stimulus’. In reflexive contexts the stimulus is referentially identical with the experiencer. Note that regarding topic drop, “optimal” experiencer-stimulus verbs are those which do not allow adding an additional stimulus argument (as e.g. über/wegen/vor sich (about/because-of/of REFL)), cf. (41), that is, a PP whose reflexive part has the same reference as the subject (and the accusative marked reflexive pronoun). (41) a. * Der Hansi liebt / hört / mag / hasst sichi wegen sichi. The Hans loves/hears/likes/hates REFL-ACC because-of REFL b. * Der Hansi liebt / hört / mag / hasst sichi über sichi. The Hans loves/hears/likes/hates REFL-ACC about REFL With experiencer-stimulus verbs that allow the adding of an additional stimulus (as e.g. fürchten and ärgern), cf. (42), it is very hard to get the sloppy reading, cf. (43) – which is the reason why I use the verbs mögen, lieben, hassen (to like, to love, to hate) in most of my examples: (42) a. b.

(43) A: B:

Der Hansi fürchtet sichi vor sichi. The Hans fears REFL-ACC of REFL Der Hansi ärgert sichi über/wegen sichi. The Hans bothers REFL-ACC about/because-of REFL Hansi fürchtet / ärgert sichi. Hans fears/bothers REFL-ACC _i/?*m Fürchtet / Ärgert der Olafm auch. [Him/Himself] fears/bothers the Olaf as-well

In the next section I will discuss the drop of 1st/2nd person object pronouns. Note that due to the fact that German does not have genuine 1st/2nd person reflexive pronouns in its language inventory, 1st/2nd person object pronouns are used instead. I.e., in German 1st/2nd person object pronouns are ambiguous between an object reading and a reflexive pronoun reading.

Two Different Kinds of Binding in Topic Drop Dialogues | 145

3.2.5 The Interpretation of 1st and 2nd Person (Indexical) Pronouns The reason to examine 1st/2nd person object drop22 in more detail is not only because indexicals constitute a special instance of referential elements but also because the well-formedness of topic dropped 1st/2nd person objects is neglected throughout the (generative) literature,23 cf. Ross (1982), Rizzi (1994), Steinbach (2007), Costa, Lobo & Silva (2009).24 Cardinaletti (1990) stipulates a ‘featurebased’ ban on 1st/2nd person object gaps25 and Thrift (2003) claims that due to their indexical character 1st/2nd person objects cannot constitute proper antecedents – which cannot be upheld, as shown by data as e.g. (44): (44) A: B:

Wiri mögen unsi. We like us/ourselves-ACC Euchi / Unsm mögen wirm auch. [You-pl-ACC/Ourselves-ACC] like we as well

Before discussing topic drop and the possible interpretation(s) of indexical elements in detail let me shortly introduce some basic notions from Kaplan’s two papers published in 1989, ‘Demonstratives’ and ‘Afterthoughts’.

|| 22 1st/2nd person subject drop will be discussed separately in chapter 4. As will be shown, no antecedent is needed to drop a 1st/2nd person subject in German. Thus, 1st/2nd person null subjects are not (necessarily) an instance of topic drop. 23 However, Gisbert Fanselow (p.c.) pointed out to me that 1st/2nd person objects can indeed be dropped, but he claimed that this is only possible when no ‘vehicle change’ is involved (a view I do not share). His examples (plus acceptability judgements) were the following: (i) A: Was ist mit uns? What is about us-DAT? B: Uns hat er vergessen. [ACC] has he forgotten (ii) A: Was ist mit euch? What is about you-pl-DAT? B: * Uns hat er nicht eingeladen. [ACC] has he not invited 24 Costa, Lobo & Silva (2009) discuss topic drop in European Portuguese. 25 Cardinaletti (1990:59) gives the following (ungrammatical) example: (i) A: Habe ich dich gestört? Have I you-sg-ACC disturbed? B: * _ Hast du sehr gestört. [ACC] (= Me) have you very disturbed

146 | Topic Drop in German: The Interpretation of the Gap

In his seminal paper Kaplan (1989a:491) states that for ‘pure indexicals’ as ‘I’, ‘now’, ‘here’ “no associated demonstration is required […] The linguistic rules which govern their use fully determine the referent for each context.” I.e., in contrast to demonstratives as ‘that’ or ‘he’ no additional demonstration is needed to refer to a referent of a pure indexical (such as 1st/2nd person pronouns). Even when some emphasis, e.g. in form of a (pointing) gesture, is used for pure indexicals, so Kaplan (1989a:491), it is completely irrelevant.26 Kaplan (1989a:492) argues for two main principles cited in (45a/b): (45) Kaplan’s (1989a) Principles 1 and 2 (a) Principle 1: The referent of a pure indexical depends on the context, and the referent of a demonstrative depends on the associated demonstration. (b) Principle 2: Indexicals, pure and demonstrative alike, are directly referential. Kaplan (1989a:494) distinguishes possible occasions of use – which he calls ‘contexts’ – and says that “[a] directly referential term may designate different objects when used in different contexts”. Next to the ‘contexts of use’ he introduces the ‘circumstances of evaluation’. He says that “it is contents that are evaluated in circumstances of evaluation.” (Kaplan 1989a:501). In his Afterthoughts-paper where the notion ‘possible circumstance of evaluation’ is replaced by ‘particular possible world’ Kaplan (1989b:591) further notes that “[c]ontent is generated in a context, and each context is associated with a particular possible world. The agent, time, place are all drawn from that world”. Furthermore, Kaplan (1989a:500) states that the principles given in (45a/b) imply that sentences containing pure indexicals have two kinds of meaning, namely ‘content’ and ‘character’. One can sloppily say that the content of an indexical refers to the referent itself, cf. Kaplan (1989a:502) and that “[a]ll directly referential expressions (as well as rigid designators) have a fixed content” (ibid.).27 The term ‘character’, on the other hand, should be understood as follows (Kaplan 1989a:505): “The character of an expression is set by linguistic

|| 26 However, consider the following example (i) from Schlenker (2002:4) where a pure indexical’s reference stays unclear when no pointing gesture is involved: (i) Youi [pointing] are clever but youk [pointing again] are not. 27 In his Afterthoughts-paper (1989b:569) Kaplan adds: “The directly referential term goes directly to its referent, directly in the sense that it does not first pass through the proposition.”

Two Different Kinds of Binding in Topic Drop Dialogues | 147

conventions and, in turn, determines the content of the expression in every context.”, e.g. “‘I’ refers to the speaker or writer” (ibidem). With regard to the differentiation between indexical and non-indexical elements Kaplan (1989a:506f) states: Indexicals have a context-sensitive character. It is characteristic of an indexical that its content varies with context. Nonindexicals have a fixed character. The same content is invoked in all contexts […] If we were not only concerned with standardized representations (which certainly have their value for model-theoretic investigations) we might be inclined to say that the character of an indexical-free word or phrase just is its (constant) content.

In order to present the facts as clearly as possible, I indicate the different readings by substituting the gap by crossed out expressions (instead of an underline ‘_’). This, however, does not mean that these crossed out expressions (which may be associated with a phonological deletion approach) represent the gap. As is well-known, we have a special situation with (topic drop of) indexical pronouns in German. Due to the fact that German has no genuine (class of) reflexive pronouns for the 1st/2nd person (cf. in contrast to languages like English (myself, yourself etc.) or Polish (uniform reflexive się)), 1st/2nd person object pronouns are used. As a consequence, 1st/2nd person object pronouns have an ambiguous character: they are not only interpretable in dependence of an anaphoric relationship with some element they rely on, but constitute autonomously referring elements as well.28 In both functions they can serve as (referential) antecedents for topic dropped elements (independently of the reference of their context binder (as opposed to sich)). Note further that generally, dropped expressions that are or contain 1st/2nd person indexical pronouns only exhibit one strict reading (which is caused by the fact that they are directly referential and hence their reference in a particular context is not ambiguous). In the following, I will look at gaps whose antecedent is a 1st/2nd person object pronoun and see which interpretations are possible for these gaps. Note that I will only examine cases where the context binder and the gap’s antecedent are in a possible binding configuration (i.e. a configuration which is am-

|| 28 Consider (i) where the reading of the gap seems to be a sloppy one but in fact it is a strict one, because no binding is involved: (i) A: Hansi mag dichs. Hans likes you-sg-ACC B: Michs mag ichs auch. [Me] like I as-well

148 | Topic Drop in German: The Interpretation of the Gap

biguous between coreference and binding29). Thus, the only independent variable which can be changed/manipulated is the target binder, cf. (46): (46) Different scenarios with respect to 1st/2nd person object drop Constant/dependent variable: The gap’s antecedent is/contains an indexical pronoun Independent variables: (i) The target binder is a non-referential element, marked for 3rd person (ii) The target binder is a referential element, marked for 3rd person (iii) The target binder is a (directly) referential 1st/2nd person pronoun I start with (46i) which is illustrated by the examples in (47) and (48) beneath. Since indexical pronouns are directly referential elements, a gap derived from a 1st/2nd person object pronoun always receives a strict interpretation – as long as speaker B is able to perform the shifting operation. Furthermore (as we have already observed), a non-referential binder is able to establish a binding relation without additional identity requirements. However, a sloppy reading is only possible when the indexical is (a possessive pronoun) in determiner position (at LF), cf. (48), but impossible when the indexical is a ‘pure indexical’, cf. (47): (47) A: B:

(48) A: B:

Ichi mag michi. I like me/myself-ACC Dichi / Sich*m mag fast jederm. [You-sg-ACC/REFL-ACC] likes almost everyone Ichi mag meinei Oma. I like my grandma-ACC Deinei Oma / Seinem Oma mag fast jederm Student. [Your-sg grandma-ACC/His grandma-ACC] likes almost every student

|| 29 Instances which do not meet this requirement are not very interesting (cf. principle (13) in 3.1/(38) in 3.2.2) because in such cases the gap can only receive a strict reading, cf. (i): (i) A: Dui liebst michs nicht. You-sg love me-ACC not B: Na und, dichs liebt der Olafm doch auch nicht. So what, [you-sg-ACC] loves the Olaf yet as-well not

Two Different Kinds of Binding in Topic Drop Dialogues | 149

The second case I would like to discuss is (46ii), where the target binder is a 3rd person referential element. Such cases can be illustrated by examples as given in (49) and (50): (49) A: B:

(50) A: B:

Dui magst dichi. You-sg like you-sg/yourself-ACC Michi / Sich?m mag der Hansm auch. [Me-ACC/REFL-ACC] likes the Hans as-well Dui magst deinei Oma. You-sg like your grandma-ACC Meinei Oma / Seine?m Oma mag der Hansm auch. [My grandma-ACC/His grandma-ACC] likes the Hans as-well

As in (47)/(48), the strict reading is present in (49) and (50). However, a sloppy reading of these gaps is hard to get because (due to person mismatches) we have no PF/phi-feature identity between the context bindee and the bindee part of the gap – which is a necessary condition for a sloppy reading, given that the binder in the target is a referential element (here: der Hans). However, as already discussed in chapter 3.2.1 (cf. the examples (6)–(9)) in contrast to number or gender mismatches, person mismatches can be tolerated (at least by some speakers). The third case, (46iii), where the target binder is an indexical element, is illustrated by the examples (51) and (52): (51) A: a.

B:

b.

B:

(52) A: a.

B:

b.

B:

Ichi mag michi. I like me/myself-ACC Dichi / Uns*m mögen wirm auch. [You-sg-ACC/Ourselves-ACC] like we as-well Dichi / Michm mag ichm auch. [You-sg-ACC/Myself-ACC] like I as-well Ichi mag meinei Oma. I like my grandma-ACC Deinei Oma / Unsere*m Oma mögen wirm auch. [Your-sg grandma-ACC/Our grandma-ACC] like we as-well Deinei Oma / Meinem Oma mag ichm auch. [Your-sg grandma-ACC/My grandma-ACC] like I as-well

150 | Topic Drop in German: The Interpretation of the Gap

When PF/phi-feature identity conditions known from referential binding are obeyed, (elements containing) dropped 1st/2nd person objects can be interpreted sloppily and the relation that they thereby express is reflexive, cf. mich in (51b), or possessive, cf. meine Oma in (52b). When the respective identity constraints on referential binding are not obeyed, a gap derived from an indexical pronoun cannot be interpreted sloppily (here: due to number mismatches), but only strictly, cf. (51a) and (52a). A further example which can be added to the case/instance of (46iii) is datum (53) below – here the strict and the sloppy reading coincide, however, due to phi-feature mismatches (with respect to person) the (bound) sloppy reading is only marginally possible: (53) A: B:

Dui magst dichi. You-sg like you-sg/yourself-ACC Ja, michi / mich?m mag ichm. Yes, [me-ACC/myself-ACC] like I

A summary of the above section could read as follows: When a topic drop gap is strictly related to its antecedent, and this antecedent is (an expression that contains) a 1st or 2nd person indexical pronoun, antecedent and gap have to be represented by different characters – because the content of antecedent and gap is the same in context and target (note that I use character and content in Kaplanian sense, cf. Kaplan 1989a,b). Thus, strictly interpreted gaps that are derived from an antecedent which contains an/is an indexical pronoun are subject to ‘character conversion under content identity’. When a topic drop gap is sloppily related to its antecedent and this antecedent is (an expression that contains) a 1st or 2nd person indexical pronoun, antecedent and gap have to bear different contents. Here, two cases have to be differentiated, namely (i) ‘content conversion under character identity’ and (ii) ‘content conversion under character conversion’. Is the target binder a directly referential element, the form/phi-features of the target bindee must be the same as those of the context bindee. Thus, only ‘content conversion under character identity’ is possible. For ‘content conversion under character conversion’ target bindee and context bindee must have different forms/phi-feature specifications. Thus, according to the (form/phi-feature) restrictions on ‘referential binding’ ‘content conversion under character conversion’ is no option when the target binder is a referential element (modulo the marginal possibility of person conversion). However, ‘content conversion under character conversion’ is possible when a non-referential (i.e. a quantificational/indefinite) target binder binds an

Two Different Kinds of Binding in Topic Drop Dialogues | 151

indexical in possessive function/position, but not when the indexical is a pure indexical. The above observations/the possible readings of indexical pronouns in topic drop dialogues are summarised in (54): (54) Possible interpretations of topic dropped 1st/2nd person object pronouns a. Character conversion under content identity results in a strict reading of the gap (via referential identity). b. Content conversion under character identity results in a sloppy reading of the gap (via referential binding). c. Content conversion under character conversion results in a sloppy reading of the gap and is only possible when the dropped indexical is a possessive pronoun in determiner position and when it is bound by a quantificational/indefinite binder (via non-referential binding). Pure indexicals resist binding by a non-referential element. On the basis of the above data (contrasts) we can make the following statements with respect to the representation of different interpretations of topic drop gaps that are derived from indexical pronouns, cf. (55) and (56): (55) Strict readings of topic dropped 1st/2nd person objects – Syntactic representations On their strict reading, topic drop gaps that are derived from (expressions that contain) 1st/2nd person pronouns cannot be represented by deleted (one-to-one) antecedent copies. Furthermore, as a deleted d-pronoun cannot be the correct representation of a topic dropped pronoun with 1st or 2nd person features (because d-pronouns can only relate to 3rd person referents), I will suggest that in their strict interpretation topic dropped 1st/2nd person indexical pronouns and DPs that contain 1st/2nd person (possessive) pronouns are represented by pro. (56) Sloppy readings of topic dropped 1st/2nd person objects – Syntactic representations When the target binder is a referential element, on their sloppy readings topic drop gaps that are derived from (expressions that contain) 1st/2nd person pronouns can be represented by a PF-deleted (one-to-one) antecedent copy or, alternatively, by a (phi-feature-sensitive) pro category. Both assumptions can account for the occurrence of sloppy readings. When the target binder is a non-referential element, the PF-form/the phi-features

152 | Topic Drop in German: The Interpretation of the Gap

shared by antecedent and gap differ.30 Consequently, only a pro-category or the assumption of a syntactically intact structure which lacks a phonological matrix (cf. Wilder 1996) can account for such topic drop gaps. The question whether dropped indexical pronouns receive a sloppy interpretation in the target sentence or not, largely depends on the specifications of the target binder. With respect to ‘referential binding’, phi-feature identity/ compatibility (modulo person adjustment) between context bindee and target binder is obligatory.31 In case this obligation is not met, dropped elements (being or containing indexical pronouns) lack a bound reading. In the second case, cf. ‘non-referential binding’, dropped elements containing a 1st/2nd person pronoun receive their phi-features on a ‘volatile’, relational basis, in dependence on the (3rd person marked) non-referential target binder. However, a nonreferential target binder cannot “overwrite” the phi-feature specifications of a pure indexical. How the kind of binding relation decides about the (im)possible interpretation(s) of topic dropped elements is also shown by example (57): In this case the gap only has the strict-2 interpretation, because none of the respective conditions (for the strict-1 reading and the sloppy reading, respectively) is met: First, the context binder is a non-referential element – hence, the gap lacks a strict-1 reading. Second, the target binder is a (directly) referential element – thus, for the sloppy gap interpretation PF/phi-feature identity constraints must be respected. This, however, is not the case in (57) where (the surface form and) the phi-feature specifications of context bindee and target bindee depart from each other in number (and – possibly irrelevant – in person): (57) A: B:

Fast jederi Student mag seinei/k Oma. Almost every student likes his grandma-ACC Seinek Oma / Unsere*m Oma mögen wirm auch. [His (k’s) grandma/Our grandma] like we as-well

Consider also (58):

|| 30 As a matter of fact non-referential binders with 1st/2nd person features do not exist (in German). 31 Later I will assume that phi-feature identity between context binder and target binder is the relevant identity condition for referential binding and thus “explain away” the special case of ‘PF binding’ under syncretisms in the possessive pronoun paradigm (which makes recourse to the form of the context bindee necessary).

Survey on Bound Readings of the Gap: A (Better) Empirical Basis | 153

(58) A: B:

Fast jederi Student mag sichi. Almost every student likes REFL-ACC (= himself) Uns*m mögen wirm auch. [Ourselves] like we as-well

In (58) the gap does not have any interpretation, because none of the respective interpretational conditions is met: First, the context binder is a non-referential element binding a reflexive pronoun – hence, the gap lacks a strict reading. Second, the target binder is a (directly) referential element – thus for the sloppy gap interpretation form/phi-feature identity between context bindee and target bindee has to be respected. This, however, is – as in (57) – not fulfilled.

3.3 Survey on Bound Readings of the Gap: A (Better) Empirical Basis In the following I will present the results of a small online survey that I have conducted. In order to test what I have discussed in the above section(s), namely the availability of sloppy readings despite phi-feature mismatches, I asked 60 naive subjects (German native speakers, between 13 and 70 years old, mean age was 35.8, 42 persons were female and 17 were male, one person declared to be of gender ‘X’) for the availability of a sloppy interpretation of topic dropped elements. I introduced the topic to the subjects in the following way:32 (An analogous English translation is given below) In den folgenden Beispielen wird Ihnen ein Dialog zweier Sprecher (A, B) präsentiert, der – wie in der Umgangssprache üblich – ein unausgesprochenes Element enthält. Dieses nicht ausgesprochene Element ist durch einen Unterstrich _ gekennzeichnet. (1) A: B:

Gestern hat der Peter seinen Vermieter getroffen. _ hat der Otto heute auch getroffen.

|| 32 The items were unsystematically “distracted”. Besides the items presented here, I included three further topic drop dialogues as distracting elements.

154 | Topic Drop in German: The Interpretation of the Gap

Oft (aber nicht immer) hat ein Leerelement mehrere Interpretationen. Im Folgenden sehen Sie die potentiell möglichen Lesarten der Lücke in Satz (1B): (i) PETERS VERMIETER hat der Otto heute auch getroffen. (ii) SEINEN (=eigenen) VERMIETER hat der Otto heute auch getroffen. (iii) SEINEN (=von irgendeiner Person X) VERMIETER hat der Otto heute auch getroffen. Im Folgenden werde ich Sie zu den Interpretationsmöglichkeiten solcher (und ähnlicher) Lücken befragen. Die Bedeutung der Lücke _ wird dabei (wie im obigen Beispiel) durch GROßBUCHSTABEN repräsentiert. Lassen Sie sich nicht verwirren, wenn die Lücke mehrere Interpretationen erlaubt und kreuzen Sie nur an, ob die ANGEGEBENE Interpretation für Sie möglich (JA) oder nicht möglich ist (NEIN). Die ganze Befragung dauert ca. 5 Minuten. Da jede Sprecherin/jeder Sprecher eine eigene Intuition hat, bitte ich Sie, die Befragung ganz alleine, ohne die “Mithilfe” anderer durchzuführen. [Translation: The following examples present a conversation between two speakers, speaker A and speaker B. As it is customary in colloquial speech it includes a non-verbalized element. This element is marked by the underscore _. (1) A: B:

Gestern hat der Peter seinen Vermieter getroffen. Yesterday has the Peter his landlord met _ hat der Otto heute auch getroffen. _ has the Otto today also met

Regularly (but not always), the gap element can have several interpretations. In what follows you will be presented with possible readings of the gap in sentence (1B): (i) (ii)

PETERS VERMIETER hat der Otto heute auch getroffen. PETER’S LANDLORD has the Otto today also met SEINEN (=eigenen) VERMIETER hat der Otto heute auch getroffen. HIS (=own) LANDLORD has the Otto today also met

Survey on Bound Readings of the Gap: A (Better) Empirical Basis | 155

(iii) SEINEN (=von irgendeiner Person X) VERMIETER hat der Otto heute auch HIS (=of some person X) LANDLORD has the Otto today also getroffen. met In what follows I will ask you about the possible interpretations of these (or similar) gaps. The meaning of the gap _ will be represented (see example above) by CAPITAL LETTERS. Don’t be confused (don’t be puzzled) in case a gap allows for more than one interpretation. If so, only indicate whether the GIVEN interpretation is acceptable (YES) or not acceptable (NO). The questionnaire will take about 5 minutes. As every speaker has his or her own intuitions, please do the questionnaire without further assistance.] As mentioned above, the aim of this survey was to test whether a topic drop gap can be sloppily interpreted under the +/-presence of PF/phi-feature mismatches between antecedent and gap. The question which the informants were asked was always the same (i.e. I always asked for the sloppy interpretation and no other). The questions looked as the following one (as before, a translation is given below): A: B:

Der Peter mag seine Tante. _ mag die Maria auch.

Bitte kreuzen Sie an, ob Sie die obige Lücke folgendermaßen interpretieren können: IHRE (=eigene) TANTE mag die Maria auch.  JA  NEIN [Translation: A:

Der Peter mag seine Tante. The Peter likes his aunt-Acc

156 | Topic Drop in German: The Interpretation of the Gap

B:

_ mag die Maria auch. _ likes the Maria as-well

Please indicate whether you can interpret the above gap in the following way: IHRE (=eigene) TANTE mag die Maria auch. HER (=own) AUNT likes the Maria as-well  JA (Yes)  NEIN (No)] The results that I found were the following, cf. (2): (2)

Main results of the mini-survey on sloppy readings a. When the target binder was a referential element, about half of the speakers accepted the sloppy reading under PF/phi-feature mismatches between context bindee and target bindee. b. When the target binder was a non-referential element, nearly all speakers accepted the sloppy reading under PF/phi-feature mismatches between context bindee and target bindee.

In the following I (first) present those items (including the respective taskquestions) where phi-feature mismatches between the context bindee and the target binder were present: Gender mismatches, (3a) vs. (3b): (3a) A: B:

Der Peter mag seine Tante. The Peter likes his aunt-ACC _ Mag die Maria auch. [ACC] likes the Maria as-well

Interpretation that was questioned: IHRE (=eigene) TANTE mag die Maria auch. HER (=own) AUNT likes the Maria as-well Referential binding relation Sloppy interpretation: YES – 28 | NO – 32

Survey on Bound Readings of the Gap: A (Better) Empirical Basis | 157

60 50 40

Yes

30

No

20 10 0 Sloppy Gap Interpretation (3a)

(3b) A: B:

Jeder Mann pflegt seine Bekanntschaften. Every man cherishes his acquaintances-ACC _ Pflegt jede Frau auch. [ACC] cherishes every woman as-well

Interpretation that was questioned: IHRE (=eigenen) BEKANNTSCHAFTEN pflegt jede Frau auch. HER (=own) ACQUAINTANCES cherishes every woman as-well Non-referential binding relation Sloppy interpretation: YES – 51 | NO – 9 60 50 40

Yes

30

No

20 10 0 Sloppy Gap Interpretation (3b)

Person mismatches, (4a) vs. (4b): (4a) A: B:

Ich mag meinen Professor. I like my professor-ACC _ Mag der Hans auch. [ACC] likes the Hans as-well

158 | Topic Drop in German: The Interpretation of the Gap

Interpretation that was questioned: SEINEN (=eigenen) PROFESSOR mag der Hans auch. HIS (=own) PROFESSOR likes the Hans as-well Referential binding relation Sloppy interpretation: YES – 32 | NO – 28 60 50 40

Yes

30

No

20 10 0 Sloppy Gap Interpretation (4a)

(4b) A: B:

Ich bewundere meinen Professor. I admire my professor-ACC _ Bewundert doch jeder Doktorand. [ACC] admires PRT every PhD-student

Interpretation that was questioned: SEINEN (=eigenen) PROFESSOR bewundert doch jeder Doktorand. HIS (=own) PROFESSOR admires PRT every PhD-student Non-referential binding relation Sloppy interpretation: YES – 50 | NO – 10 60 50 40 30 20 10 0 Sloppy Gap Interpretation (4b)

Yes No

Survey on Bound Readings of the Gap: A (Better) Empirical Basis | 159

Number mismatches, (5a) vs. (5b): (5a) A: B:

Die Uni-Professoren mögen ihre Sekretärinnen. The Uni(versity) professors like their secretaries-ACC _ Mag der Hans auch. [ACC] likes the Hans as-well

Interpretation that was questioned: SEINE (=eigene) SEKRETÄRIN mag der Hans auch.33 HIS (=own) SECRETARY likes the Hans as-well Referential binding relation Sloppy interpretation: YES – 32 | NO – 28 60 50 40

Yes

30

No

20 10 0 Sloppy Gap Interpretation (5a)

(5b) A: B:

Peter mag seine Nase. Peter likes his nose-ACC _ Mögen fast alle Männer. [ACC] like almost all men

Interpretation that was questioned: IHRE (=eigene) NASE mögen fast alle Männer. THEIR (=own) NOSE like almost all men

|| 33 Note that in this case the dropped element is indicated to bear singular number even if its antecedent is a plural expression. I.e., the gap’s interpretation is distributive.

160 | Topic Drop in German: The Interpretation of the Gap

Non-referential binding relation Sloppy interpretation: YES – 47 | NO – 13 60 50 40 Yes

30

No

20 10 0 Sloppy Gap Interpretation (5b)

Thus, the above results confirm the hypothesis that (underlying/supposed34) phi-feature mismatches between context bindee and target bindee/target binder “do not matter” when the target binder is a non-referential element, however (by contrast) they play a big/greater role when the target binder is a referential element. Note further that in non-referential binding contexts the subjects detected a sloppy reading independently of phi-feature mismatches between context bindee and target binder. I.e., when the target binder was a non-referential element, the presence/lack of mismatches did not play any role (which means that mismatched items, cf. (3b), (4b), (5b), were “as good as” matched items, cf. (6)): No mismatches (at PF), (6) vs. (7) vs. (8): (6) A: B:

Der Hans mag seine Kommilitonen. The Hans likes his fellow-students-ACC _ Mag fast jeder Student. [ACC] likes almost every student

|| 34 The terms ‘underlying’ and ‘supposed’ refer to the fact that the gap’s bindee part is not represented at PF and possible phi-feature mismatches between context bindee and target bindee can only be read off from the features of the context bindee and target binder.

Survey on Bound Readings of the Gap: A (Better) Empirical Basis | 161

Interpretation that was questioned: SEINE (=eigenen) KOMMILITONEN mag fast jeder Student. HIS (=own) FELLOW-STUDENTS likes almost every student Non-referential binding relation (no mismatch) Sloppy interpretation: YES – 51 | NO – 9 60 50 40 Yes

30

No

20 10 0 Sloppy Gap Interpretation (6)

In a situation where the target binder was a referential element and the strict reading constituted a probable option, the sloppy gap interpretation was slightly dispreferred – even when the context bindee and the target binder were not mismatched with respect to their phi-features, cf. (7): (7) A: B:

Der Peter hat seine Schwester geschlagen. The Peter has his sister-ACC beaten _ Hat der Otto auch geschlagen. [ACC] has the Otto also beaten

Interpretation that was questioned: SEINE (=eigene) SCHWESTER hat der Otto auch geschlagen. HIS (=own) SISTER has the Otto also beaten Referential binding relation (no mismatch) Sloppy interpretation: YES – 24 | NO – 36

162 | Topic Drop in German: The Interpretation of the Gap

60 50 40

Yes

30

No

20 10 0 Sloppy Gap Interpretation (7)

In a situation where the target binder was a referential element and the strict reading was possible, but, in contrast to (7), very unlikely, the great majority of the subjects could interpret the gap sloppily, cf. (8): (8) A: B:

Die Schüler haben ihre Handschuhe vergessen. The pupils have their gloves-ACC forgotten _ Hat die Lehrerin auch vergessen. [ACC] has the teacher-fem also forgotten

Interpretation that was questioned: IHRE (=eigenen) HANDSCHUHE hat die Lehrerin auch vergessen. HER (=own) GLOVES has the teacher-fem also forgotten Referential binding relation (Number/Gender mismatch, but “accidental” (PFrelated) phi-feature match) Sloppy interpretation: YES – 56 | NO – 4 60 50 40 30 20 10 0 Sloppy Gap Interpretation (8)

Yes No

Survey on Bound Readings of the Gap: A (Better) Empirical Basis | 163

Obliquely cased gaps, (9) and (10): Further items I asked for, cf. (9) and (10), were two examples where the topic dropped element was obliquely case marked (and the binder was a referential expression): Here about half of the 60 speakers accepted the sloppy reading: (9) A: B:

Der Hans hat heute seinem Doktorvater geholfen. The Hans has today his supervisor-DAT helped _ Hat der Peter gestern auch geholfen. [DAT] has the Peter yesterday also helped

Interpretation that was questioned: SEINEM (=eigenen) DOKTORVATER hat der Peter gestern auch geholfen. HIS (=own) SUPERVISOR has the Peter yesterday also helped Referential binding relation (dative antecedent and gap, no mismatch) Sloppy interpretation: YES – 37 | NO – 23 60 50 40 Yes

30

No

20 10 0 Sloppy Gap Interpretation (9)

(10) A: B:

Monika hat sich ihrer Schüler erbarmt. Monika has REFL her pupils-GEN take-pity-on _ Hat sich Maria auch erbarmt. [GEN] has REFL Maria also take-pity-on

Interpretation that was questioned: IHRER (=eigenen) SCHÜLER hat sich Maria auch erbarmt. HER (=own) PUPILS has REFL Maria also take-pity-on Referential binding relation (genitive antecedent and gap, no mismatch) Sloppy interpretation: YES – 32 | NO – 28

164 | Topic Drop in German: The Interpretation of the Gap

60 50 40 30 20

Yes No

10 0 Sloppy Gap Interpretation (10)

With respect to the difference between the data (7), (9) and (10) it can be noted that, interestingly, more speakers rejected the sloppy interpretation when the gap was accusative marked, cf. datum (7), than when it was obliquely case marked, cf. the data (9) and (10). This could hint at the supposition that, generally, structurally cased gaps tend to be interpreted strictly, whereas for obliquely cased gaps the sloppy interpretation is favoured: This becomes especially evident when one considers the dative example in (9) where about half of the speakers could interpret the gap sloppily, cf. (37:23). Moreover, astonishingly, more speakers (32:28) could interpret the genitive gap in (11) sloppily than the accusative gap in (7), cf. (24:36). So, with respect to (9) and (10) one could suppose that the approval of the sloppy interpretation is much higher for obliquely cased NPs because they can only be dropped in verbatim environments (and a verbatim environment constitutes a prerequisite on a gap’s sloppy interpretation), whereas, on the other hand, topic drop of structurally cased NPs is per se independent of a verbatim environment. I.e., for the latter (structurally cased gaps), the sloppy interpretation does not come “for free” whereas it could be argued that for dative and genitive cased gaps this is the case. Thus, the main hypothesis formulated within the theoretical part (binding in topic drop dialogues has to be differentiated with respect to the referential properties of the target binder) could be confirmed. Unfortunately, my supposition that – given a referential binder in the target – the sloppy reading becomes severely degraded when phi-features of antecedent and gap are mismatched (at PF) could not be confirmed. Possibly a different (experimental) design could lead to more reliable evidence for this hypothesis.

The Representation of the Gap | 165

3.4 The Representation of the Gap In this part I describe the circumstances under which a topic drop gap can receive particular interpretations and examine by which empty category the different readings of a topic drop gap might be represented. I will focus on bound readings and start by presenting a summary of the empirical facts (discussed throughout chapter 3), followed by some theoretical considerations.

3.4.1 Binding in Topic Drop Dialogues: Empirical Findings As to binding in topic drop dialogues consider first the following (necessary) conditions that have to hold such that a topic drop gap can be interpreted sloppily, cf. (38) from 3.2.4, repeated as (1): (1)

Conditions on sloppy gap readings in topic drop dialogues (a) The antecedent of the gap must be a non-rigid designator such that its reference can vary across context and target (b) Syntactic conditions on binding must be met (cf. c-command) (c) Presence of a binding/coreference relation in the context (cf. two coindexed elements) (d) Identical/synonymous predicates in context and target (= verbatim environment)

According to (1), a sloppy reading of the gap is only possible when context and target are semantically parallel (i.e. when a verbatim environment is present). Furthermore, in case the bindee part of the gap gets bound it must be derived from an antecedent which has also entered (i) a binding relation or (ii) a relation that is ambiguous between coreference and binding with some element in the context. In (2a/b) this is not the case and, as a consequence, the gap cannot be interpreted sloppily: (2) A: a.

B:

b.

B:

Hansi mag ihrenk Prof. Hans likes her prof-ACC Ihrenk/*m Prof mag die Irenem auch. No binding/no coreference [ACC] likes the Irene as-well Ihrenk/*r Prof mag jeder Doktorandin. No binding [ACC] likes every PhD-student-fem

166 | Topic Drop in German: The Interpretation of the Gap

(2a/b) show that index/feature sharing of “binder” and “bindee”35 is a necessary condition for binding.36 When it is not fulfilled, a possible target binder (in the above examples this is Irene, jede Doktorandin, respectively) lacks any “binding force” and, as a consequence, the gap cannot be interpreted sloppily – note that this is so even if the (silent PF) form of the target bindee and the respective target binder bear the same features (which is a prerequisite for a coreferential or bound reading), cf. (2a) Ihren Prof mag die Irene auch / (2b) Ihren Prof mag jede Doktorandin. In cases as (2a/b) antecedent and gapped constituent refer to the same entity, in this case an “external” antecedent named ‘k’.37 As before, I will refer to this reading as strict-2 reading. Another instantiation of a strict reading is the strict-1 reading, indexed by ‘i’ in (3). As the contrast between (3) and (4) shows, the strict-1 reading is dependent on a coreference relation in the context. (3) A: B:

(4) A: B:

Hansi mag seineni/k Prof. Hans likes his prof-ACC Seineni/k/m Prof mag der Olafm auch. [ACC] likes the Olaf as-well. Jederi Student mag seineni/k Prof. Every student-masc likes his prof-ACC Seinenk/m Prof mag der Olafm auch. [ACC] likes the Olaf as-well.

Expressions that are/contain (directly referential) 1st/2nd person pronouns, only have one strict reading, cf. (5): (5) A: B:

Ichi mag meineni Prof. I like my prof-ACC Deineni Prof / Seinen?m Prof mag der Hansm auch. [ACC] likes the Hans as-well

|| 35 Note that the terms ‘binder’ and ‘bindee’ are also used when the relation is ambiguous between coreference and binding. 36 Two sentential arguments that bear the same referential or binding index have to share particular phi-features as well (in German: Person, Number, Gender). 37 Some authors, e.g. Thornton & Wexler (1999:114), call the strict-2 reading a deictic interpretation. On the other hand, they refer to the strict-1 reading as strict coreference interpretation.

The Representation of the Gap | 167

Besides its strict interpretations, a topic dropped element can also receive a sloppy interpretation. Depending on the referential properties of the target binder I will speak of ‘referential binding’ vs. ‘non-referential binding’ (that is, ‘bound by a referential term’ vs. ‘bound by a non-referential term’). Referential binding can only take place when the phi-features of the target binder bear the same specifications as those of the context binder (see chapter 3.2.1, data (3)–(9) for details). An example where this necessary condition is fulfilled and the sloppy reading is generated, is given in (6). I will refer to this kind of sloppy reading as sloppy-1. (6) A: B:

Irenei mag ihrei/k Lehrerin. Irene likes her teacher-fem-ACC Ihrei/k/m Lehrerin mag die Mariam auch. [ACC] likes the Maria as-well

In (7), on the contrary, target binder and context binder bear different phifeature specifications: (7) A: B:

Irenei mag ihrei/k Lehrerin. Irene likes her teacher-fem-ACC Ihrei/k Lehrerin / Seine*m Lehrerin mag der Hansm auch. [ACC] likes the Hans as-well

As a consequence, the gap in (7) cannot be interpreted sloppily, but can only receive a strict-1 and a strict-2 interpretation.38 However, there is a way to obtain a sloppy gap interpretation despite the fact that phi-features of context binder and referential target binder are mis-

|| 38 Note that in (quasi-)stripping contexts (where the finite verb is elided too) the gap lacks a sloppy reading (too) whenever context binder and target binder have different phi-feature specifications (here: different gender features), cf. (i): (i) A: Irenei mag ihrei/k Frisur. Irene likes her haircut-ACC B: ei/k/*m Hansm auch. [Her/His haircut likes] Hans as-well Thus, phi-feature identity between context binder and target binder seems to play a role not only in topic drop but also in stripping dialogues. Obviously, the conditions on the interpretation of stripped elements are in part comparable with those that hold for topic dropped elements. However, further investigation and research is needed to state the exact differences; see also 2.2.2 for a comparison of (quasi-)stripping and topic drop.

168 | Topic Drop in German: The Interpretation of the Gap

matched: this is possible when syncretisms in the possessive pronouns paradigm permit that the (underlying) PF-forms of the context bindee and the target bindee are identical, i.e., when the phi-feature demands of the context binder and the target binder, respectively, are met at the same time.39 This kind of “accidental” sloppy reading will be referred to as sloppy-2, cf. (8): (8) A: B:

Irenei mag ihrei/k Lehrerin. Irene likes her teacher-fem-ACC Ihrei/k/m Lehrerin mögen die Hansen-Brüderm auch.40 [ACC] like the Hansen-brothers-masc as-well

Besides the sloppy-1 and the sloppy-2 cases, a sloppy interpretation of the gap also arises in circumstances where the binder in the target is a non-referential element (see also chapter 3.2.2). However, in this case neither phi-feature identity between context binder and target binder as in (6), nor accidental PFidentity of context bindee and target bindee as in (8) is a necessary condition. I.e., binding takes place without a fulfilment of these requirements. This sloppy reading will be referred to as sloppy-3, cf. (9): (9) A: B:

Irenei mag ihrei/k Lehrerin. Irene likes her teacher-fem-ACC Ihrei/k Lehrerin / Seiner Lehrerin mag jederr Doktorand. [ACC] likes every PhD-student-masc

With respect to the above examples and our previous findings, we can postulate the existence of three different interpretations (strict-1, strict-2, sloppy) and five different ways to receive them, cf. (10): (10) Possible interpretations of a topic dropped element (i) Strict interpretation via referential identity (strict-1) ‘coreferential’ (ii) Strict interpretation via referential identity (strict-2) ‘deictic’ (iii) Referential binding via phi-feature identity (sloppy-1) (iv) Referential binding via PF identity (sloppy-2)

|| 39 See (13) in 3.2.1 for a short overview of the possessive pronouns paradigm in German. 40 Note that an indexation of the gap as in (8) constitutes a very lax (and in fact incorrect) notation as it suggests that the gap bears both phi-features of the feminine singular and the masculine plural. Later I will adjust the indexation and the gap’s representation, cf. (30)–(33).

The Representation of the Gap | 169

(v)

Non-referential binding via quantified or non-specific indefinite elements (sloppy-3)

Applying a characterisation as in Fox (2000) to topic drop dialogues, we can state that the strict-1 and the strict-2 reading are instances of a Referential Parallelism and that all sloppy readings are instances of a Structural Parallelism, cf. (11): (11) NP Parallelism (according to Fox 2000:117) NPs in the antecedent and elided VPs must either a. have the same referential value (Referential Parallelism) or b. be linked by identical dependencies (Structural Parallelism). The general conditions on binding a topic dropped element mentioned in (1) can now be formulated more precisely, in dependence of whether the target binder is a non-referential or a referential element (and whether phi-feature identity between context binder and target binder is obeyed or not). Besides the conditions mentioned in (1), we can state that the sloppy-1 reading, illustrated by example (6), is only well-formed when (12) holds: (12) Condition on the gap’s sloppy-1 interpretation in topic drop dialogues When the target binder is a referential element, context binder and target binder must have identical phi-feature specifications. Otherwise the bindee part of the gap cannot be bound. As to the question how the quasi-accidental sloppy-2 interpretation, cf. example (8) above, is generated, two possibilities come to mind: First, one could assume that the respective binders in context and target are “aware” of their choice which possessive they might select, thereby supposing that the language system has some internal knowledge about the occurrence of syncretisms (within the possessive pronouns paradigm). According to this solution, binding would be expected from the very beginning. Another way of explaining this unexpected sloppy reading would be to assume that the sloppy reading is the result of a post-syntactic operation, caused by the ambiguity of the (underlying) PFrepresentation of the possessive pronoun(s) (in (8) the form ihre is ambiguous between feminine singular and gender-underspecified plural). Thus, the relevant condition on the well-formedness of the sloppy-2 interpretation could read as follows:

170 | Topic Drop in German: The Interpretation of the Gap

(13) Condition on the gap’s sloppy-2 interpretation in topic drop dialogues ((i) and (i’) constituting variants) (i) The target binder is endowed with information about the PF-form of the possessive pronoun it might choose and binds the empty pronoun (despite phi-feature mismatches between target binder and context binder). (i’) The overtly realised form of the context bindee (= the antecedent of the gap’s bindee) must be PF-identical with the covertly underlying form of the target bindee (= the gap’s bindee part), or, put another way: the phifeature specifications of the target binder and the context bindee must be PF-compatible. If this is the case, binding applies post-syntactically. In my opinion, (13i) is an improbable option. However, no matter what line of reasoning one prefers it can be stated that for the elements which are involved in this particular kind of binding (underlying) surface identity plays a crucial role. As pointed out above, no PF/phi-feature identity requirements obtain when the binder is a non-referential element. I.e., context binder and target binder need not have identical phi-feature specifications neither do the target binder and the context bindee have to have a compatible PF-form. Rather, context bindee and the bindee part of the gap can have different PF/phi-feature specifications. Thus, there is no additional condition to be added to (1), which is repeated here for convenience: (1)

Conditions on sloppy gap readings in topic drop dialogues (a) The antecedent of the gap must be a non-rigid designator such that its reference can vary across context and target (b) Syntactic conditions on binding must be met (cf. c-command) (c) Presence of a binding/coreference relation in the context (cf. two coindexed elements) (d) Identical/synonymous predicates in context and target (= verbatim environment)

3.4.2 Binding in Topic Drop Dialogues: Some Theoretical Considerations In the following I will try to link the empirical findings on binding in topic drop dialogues to a theoretical background. First, I will examine by what kind of empty category the different readings of a topic drop gap could be represented. So far (see chapter 3.1) we have dis-

The Representation of the Gap | 171

cussed four possibilities; I will shortly summarise their pros and cons: First, a direct (one-to-one) antecedent copy can account for sloppy readings in referential binding contexts and for the strict readings of 3rd person gaps that do not contain a 1st/2nd person possessive pronoun (but not for 3rd person gaps that contain a 1st/2nd person possessive pronoun and not for dropped indexicals). A deleted d-pronoun is also determined as it can solely represent the strict readings of 3rd person gaps. As pointed out in the course of this and the preceding chapter, the category which is able to represent any interpretation is pro. The same holds for the forth option, argued for by Wilder (1996), according to which the elided constituent is present with its full-fledged syntactic structure but lacks a phonological matrix. Possibly, the choice between pro and Wilder’s variant is just a matter of taste (instead of constraining the syntax of a null pronoun pro, Wilder has to formulate constraints on phonological deletion of full structure). The relevant question (for me who favours a pro approach) is whether all topic drop instances are represented by pro or not. By considering data as (14)/(15) we see that a deleted (one-to-one) antecedent copy could account for sloppy gap readings in referential binding scenarios in a quite elegant way, without stipulating the identity constraints which have to hold between antecedent and gap in these settings: (14) A: a.

B:

b.

B:

(15) A: B:

Gestern hat Mariai ihreni/k Lehrer getroffen. Yesterday has Maria her teacher-masc-ACC met Ihreni/k/m Lehrer hat die Annam heute auch getroffen. [Her teacher-masc-ACC] has the Anna today also met Ihreni/k/m Lehrer haben die Erstklässlerm heute auch getroffen. [Their teacher-masc-ACC] have the first-formers today also met Wiri mögen unsi. We like us/ourselves-ACC Euchi / Unsm mögen wirm auch. [You-pl-ACC /Ourselves-ACC] like we as well

As we know that the occurrence of the sloppy reading in referential binding scenarios is constrained by phi-feature identity between context binder and target binder, it suggests itself to refer to a feature checking mechanism rather than to a phonological copy device under PF-identity. Moreover, as departure from phi-feature identity is marginally possible with respect to ‘person’ but not with respect to gender and number (see datum (5) and chapter 3.2.1, (3)–(9)),

172 | Topic Drop in German: The Interpretation of the Gap

the assumption of a phi-feature sensitive pro-category makes more sense than the assumption of a PF-deleted (one-to-one) antecedent copy. Besides, the postulation of a phi-feature checking mechanism is in line with what we know from chapter 2 where we learned that in topic drop dialogues the case features/protoroles of the gap and its antecedent are checked/compared as well. The (desired) consequence of making the licensing of pro in referential binding contexts dependent on phi-feature identity requirements is that we cannot derive the sloppy reading in cases as (14b) or (16), where context binder and target binder bear different phi-feature specifications (feminine singular and gender-underspecified plural, respectively). (16) A: B:

Irene mag ihre Lehrerin. Irene likes her teacher-fem-ACC [“Ihre” Lehrerin] mögen die Hansen-Brüder auch. [Her (= Their) teacher-fem-ACC] like the Hansen-brothers as-well

As the sloppy reading of the gap in (16) is caused by an ambiguous form (within the paradigm of possessive pronouns), this special case should be treated differently from the other readings, which will all be represented by pro. I argue that the sloppily interpreted gap in (16) cannot be represented by pro, because in cases where the target binder is a referential element, pro is only licensed under phi-feature identity between context binder and target binder. In particular, I suppose that the PF-dependent sloppy reading (which I have termed sloppy-2 reading) operates on the underlying (supposed) surface form of the strict-1/strict-2 reading (which is “ihre” Lehrerin in (16)). It emerges due to a PFrelated reinterpretation of the strict reading(s) and it is instantiated post hoc (post-syntactically) and as such it is not represented as pro in the syntactic structure of (16). As to the structure of pro, I will suggest that the correct representation of topic dropped elements is an internally structured pro, i.e., an empty category that is of the category pro and that contains the category pro. This representation ensures that the gap is sufficiently underspecified and compatible with both, referential binding relations as well as non-referential binding relations. Furthermore, nothing speaks against the possibility that pro can be bound by a c-commanding binder and receive a sloppy interpretation, cf. Lobeck (1995).

The Representation of the Gap | 173

In (17) two options are given to represent pro: The first, in (17a), assumes that the dropped nominal is of the category pro as well, whereas the second, in (17b), assumes that the dropped nominal is a deleted antecedent copy.41 (17) a. b.

pro = [pro [pro]] pro = [pro [ N ]]

As the PF-flexible category pro can account for character transformations (as e.g. from meine Mutter (my mother) to deine Mutter (your-sg mother)) we might think that (17b) is underspecified enough. However, the (17b)-option may be incompatible with data where syntactic identity between antecedent and gap is not given, cf. (18). (18) A: B:

Wie war denn die Griechenlandreise der Abiklasse? How was PRT the Greece-trip-NOM of-the A-level class-GEN? _ Sind ganz begeistert zurückgekommen.42 [NOM] are very impressed came-back ‘How was the Greece trip of the A-level class? They (= the participants) came back quite impressed.’

Moreover, the (17b)-option cannot account for distributive sloppy readings under a non-referential target binder (cf. (33) from chapter 3.2.2), repeated as (19): (19) A: B:

Die Nachbarskinderi mögen ihrei/k Mütter. The neighbour-children like their mothers-ACC Ihrei/k Mütter / Seiner Mütter | Seiner Mutter mag jederr Junge. [ACC] likes every boy

Thus, the internally structured pro-category in (17a) fits the data better. Before discussing the licensing/identification conditions on pro and the question why topic drop is only possible in the prefield, I would like to shortly introduce the work of Takahashi & Fox (2005) as well as of Hartman (2011) on

|| 41 An internally structured pro-category is also assumed for indefinite argument drop in Greek by Merchant & Giannakidou (1997:8). 42 As already pointed out in 2.1.2, in this example the referent which is represented by the gap and to which the B-speaker refers, is neither the Greece trip nor the A-level class, but the people participating in the trip (the pupils/teacher(s) of the A-level class).

174 | Topic Drop in German: The Interpretation of the Gap

re-binding.43 By means of this approach I will try to derive and explain the Structural Parallelism (constraint) that is necessary to account for the sloppy readings in verbatim topic drop dialogues. According to Hartman (2011:369) re-binding refers to a scenario where “the elided constituent contains a variable whose binder lies outside the elided constituent” – a situation that we also find in topic drop dialogues. In order to capture the intuition that re-bound sentences display a sloppy reading, Takahashi & Fox (2005), following Rooth (2002), introduce the notion of a Parallelism Domain (PD), on p. 229: ([20]) For ellipsis of EC [elided constituent] to be licensed, there must exist a constituent, which reflexively dominates EC, and satisfies the parallelism condition in ([10]). We will call such a constituent a Parallelism Domain or a PD: ([21]) Parallelism PD satisfies the parallelism condition if PD is semantically identical to another constituent AC, modulo focus-marked constituents. PD is semantically identical to AC modulo focus marked constituents, if there is a focus alternative to PD, PDAlt, such that for every assignment function, g, [[PDAlt]]g = [[AC]]g. PDAlt is an alternative to PD if PDAlt can be derived from PD by replacing focus marked constituents with their alternatives.

With respect to Takahashi & Fox’ (2005) principles given in (20) and (21), Hartman (2011:369) notes that

|| 43 The re-binding data go back to a generalisation of Sag and Williams (cf. e.g. Sag 1976) who claim that a sloppy reading is not possible when the lambda-operator is not internal to the elided constituent (as in (ia)), but outside, cf. (ib): (i) a. John said Mary hit him, and BILL also did . b. * John said Mary hit him, and BILL also λx. x said she did . (cited after Takahashi & Fox 2005:225, following Sag 1976:131) Takahashi & Fox (2005:228) note: “As mentioned above, the Sag-Williams parallelism condition always applies to EC and AC and requires that they be semantically identical. In Rebinding, which is represented again in ([i]), the presence of free variables within these constituents ensures that the condition will not be satisfied.” ([i]) Antecedent Clause: [ZP XPx ... [ ... [AC ... x ...]]] |_____________| Ellipsis Clause: [WP YPy ... [ ... < EC ... y ... >]] |_____________|

The Representation of the Gap | 175

[...] the definition in ([20]) leaves the size of the PD unfixed; in principle, the PD may be the elided constituent itself, or larger. The condition in ([21]), however, entails that there is one particular scenario in which the PD must be larger than the elided constituent. This scenario occurs when the elided constituent contains a variable whose binder lies outside the elided constituent. In this configuration, which Takahashi and Fox term rebinding, the semantic condition in ([21]) will require the PD to be large enough to include the binder. If the PD did not include the binder, the PD would contain a free variable, rendering it semantically nonidentical to its antecedent.

I assume that the verbatim environment (i.e., identical predicates in context and target, cf. chapter 2.2.1) provides the relevant Parallelism Domain in topic drop dialogues which involve binding of the gap in the target.44 Moreover, the inclusion of the binder into the PD is a prerequisite for the sloppy reading in topic drop dialogues. If the target binder was not part of the PD, its features could not be checked or compared with the features of the context binder in the context/ AC. Note that this also holds in case of non-referential binding, where we might be tempted to assume that the binder is not part of the PD, referring to what Hartman (2011:369) notes (see above), namely: “If the PD did not include the binder, the PD would contain a free variable, rendering it semantically nonidentical to its antecedent.” It seems that this is exactly what is required for nonreferential binding (where phi-features of context binder and target binder can differ (and thus, the form of the antecedent and gap can differ, too)). However, if we assume that in non-referential binding scenarios the inclusion of the binder in the PD is not a necessary condition, we would expect binding despite a lack of parallelism. Moreover, any other element located outside the PD could act as a potential binder too.

3.4.3 The Identification and Licensing of pro I suppose that the reason why pro is restricted to the prefield is that only in this position optimal access to the (salient) antecedent is ensured. According to Rizzi (1994, 2002), only the prefield (‘the spec of the root’) guarantees optimal (free)

|| 44 Takahasi & Fox (2005) as well as Hartman (2011) discuss the size of a PD mainly with respect to the principle of MaxElide which says: “Elide the biggest deletable constituent reflexively dominated by PD” (Takahashi & Fox 2005:229, Hartmann 2011:369). However, as shown in chapter 2.2.2, verbatim topic drop is not an instance of (quasi-)stripping without MaxElide application, because the non-elided material is responsible for interpretational differences between topic drop and stripping. To prevent the application of MaxElide to verbatim topic drop one could refer to the Gricean Maxim of Quantity.

176 | Topic Drop in German: The Interpretation of the Gap

access to the discourse, all other (sentence-internal) positions may be dominated by some other element (e.g. an interfering antecedent). Thus, I (also) assume that only in sentence-initial position identity between the gap and its antecedent can be checked. As we know from our examination of topic drop (here and in chapter 2), the following features/properties are involved in a checking process that concerns the (identity) relation between antecedent and gap: (i) phi-features, (ii) protoroles/case and (iii) reference. To a certain extent, antecedent and gap can differ with regard to the quality of these features/properties. This, I suppose, is why optimal access to the discourse is so important for topic dropped elements (in contrast to other kinds of ellipses where departure from identity between antecedent and the gap is much more constrained). Thus, we can formulate the well-formedness conditions on antecedentdependent pro (= topic drop) as follows, cf. (22): (22) Identification and licensing of antecedent-dependent pro a. (Referential) Identification (modelled after Rizzi 199445): Only in the spec of the root-position an empty element is available for the identification by a discourse antecedent. b. (Parallel) Licensing: For pro to be licensed it must be parallel to another (equivalent) constituent. The interpretation of pro depends (i) on how big the Parallelism Domain (PD) is, (ii) on the referential properties of the binder of the gap (as well as on the referential properties of the binder of the gap’s antecedent (that is, the context binder), see chapter 3.2.1) and (iia), in case, the target binder is a referential element: on its phi-feature specifications. The respective conditions (and the consequences of their application) are summarised in (23), which is related to (22b) and which explicates that it depends on the size of the Parallelism Domain which properties and features are checked for identity (the bigger the PD, the more is checked), cf. (23i–iii):

|| 45 Rizzi (1994:162) says: “The specifier of the root [then] is the only position in which an empty element can fail to have a clause internal identification, and is available for discourse identification. Under this interpretation of ([i]), an unbound null constant can survive in the specifier of the root in structures like ([ii]), and receive its referential value in discourse.” ([i]) = Rizzi’s (23): “ECP (identification) Empty categories must be chain connected to an antecedent” (... if they can = (Rizzi’s (31)). ([ii]) = Rizzi’s (30): [nc habe [t es schon gesehen]] (null constant have it already seen)

The Representation of the Gap | 177

(23) Identity checking in dependence of the extension of the PD (i) If the PD only contains the gap, antecedent and gap are only checked for identical reference. As context and target predicate are semantically different, identity of proto-roles/case cannot be checked. Hence the gap can only bear structural case. (ii) If the PD contains the gap and the predicate (but further syntactic material in context and target differs), antecedent and gap are checked for identical reference and for identity of proto-roles/case. As a consequence, the gap can also bear oblique case. (iii) If the PD is semantically identical to the context sentence (modulo focus marked elements), antecedent and gap are checked for identical reference and for identity of proto-roles/case. Moreover, if the relevant requirements with respect to binding are met, cf. (1), one has to differentiate two cases, (a) vs. (b). (a) When the target binder is a referential element phi-feature specifications of context binder and target binder must be identical. (b) When the target binder is a non-referential element no PF or phifeature identity checking with the context binder takes place.46 When parallelism between context A and target B is satisfied, an unindexed pro-category is generated and bound (by the target binder). Scenario (23i) relates to ‘classical’ instances of topic drop which I named ‘nonverbatim topic drop’ (NVTD) in chapter 2. In a non-verbatim environment context and target predicate are semantically different. As a consequence, only structurally cased elements can be topic dropped (e.g. Bayer et al. 2001, Steinbach 2007, Haider 2010), regardless of the case of its antecedent, cf. (24): (24) A: a.

B:

Dem Hans ist heute nicht gut. The Hans-DAT is today not well (‘Hans is sick today.’) Komisch, _ hab ich gestern Abend noch Fußballspielen sehen. Strange, [ACC] have I yesterday evening still football-playing seen

|| 46 I.e., the features of the gap’s antecedent can be replaced by the features of a non-referential target binder – however, there is one exception, namely ‘pure’ indexicals. As shown in chapter 3.2.5, 1st/2nd person personal pronouns cannot be bound by non-referential binders: In this case the connection between character and content (cf. Kaplan 1989a,b) cannot be absorbed, but only altered when the one or the other stays the same, i.e. character conversion under content identity or content conversion under character identity is possible, but content conversion under character conversion is not.

178 | Topic Drop in German: The Interpretation of the Gap

b.

B: * _ Hab ich deswegen sogar gedacht. [GEN] have I because-of-this even commemorated

Scenario (24ii) refers to instances I referred to by ‘verbatim topic drop’ (VTD). As largely discussed in chapter 2, VTD is characterised by the fact that antecedent and gap bear the same (finely granulated) micro theta role, which involves semantic identity of the context and target predicate (and hence, identity of the antecedent’s and gap’s proto-properties), case identity playing a minor role (and if, then only with respect to dative drop, see chapter 2.5). In contrast to NVTD environments, in VTD environments gaps with any case can be dropped (i.e., elements bearing oblique case as well): (25) A: B:

Maria braucht eine Gehhilfe. Maria needs a walker-ACC _ Bedarf Martin auch. [GEN] requires Martin as-well

(26) A:

Maria bedarf einer Gehhilfe. Maria requires a walker-GEN B: * _ Schämt sich Martin ganz schön. [GEN] ashamed REFL Martin quite nicely (= badly)

Scenario (23iii) constitutes a special instance of a verbatim environment which allows for binding in that it demands that (the AC in the) context and (the PD in the) target are identical despite focus marked elements (and lambda-bound variable indices, cf. Hardt 2006:3). That this is a necessary condition, is evidenced by the contrast in (27) where, in case (the VP in) the target contains material that is non-focused (cf. Rooth 2005, Takahashi & Fox 2005) and nonidentical to the context material (cf. eine Orange vs. eine Banane), the sloppy reading is not generated, cf. (27a). (27) A: a.

B:

b.

B:

Peteri hat seinemi/k Prof eine Banane geschenkt. Peter has his prof-DAT a banana given Seinemi/k/*m Prof hat der OTTOm eine Orange geschenkt. [DAT] has the Otto an orange given Seinemi/k/m Prof hat der OTTOm auch eine Banane geschenkt. [DAT] has the Otto also a banana given

The Representation of the Gap | 179

By means (of the different extensions) of the Parallelism Domain, cf. (23i–23iii), we are able to describe the “effects” of ‘non-verbatim environments’ (= different predicates in context and target) and ‘verbatim environments’ (= semantically identical predicates in context and target) more accurately (as we now see that the size/extension of the PD correlates with the option/necessity to check specific features and properties for identity (in context and target)). In the following I will represent the different readings of topic drop gaps more accurately, cf. (28)–(37). As to the sloppy-1 reading of topic dropped elements I assume that (given general conditions on binding, cf. (1), are fulfilled) pro is licensed when the PD extends to the whole sentence and the binders in context and target bear the same phi-feature specifications, so that the phi-feature specifications of the target binder (and its bindee) in the PD are identical with the phi-feature specifications of the context binder (and its bindee) within the AC. I.e., the PD extends to the whole sentence modulo the denotation of the focus marked target binder: (28) A: (i) B: (ii) B:

(29) A: B:

Hansi mag [seinei Nase]s. Hans likes his nose-ACC [proi [pro]]s mag DER EGONt auch. [prot [pro]]s mag DER EGONt auch. [ACC] likes the Egon as-well

s = si

Hansi mag [seinek Nase]s. [prok [pro]]s mag DER EGONt auch.

s = sk (strict-2, s = sk)

(strict-1, s = si) (sloppy-1, s = st)

In case a gap is interpreted strictly, the reference of antecedent and gap stays the same. Thus, one could assume that in the examples above the referential indices ‘i’ and ‘k’ (at pro) are transferred from context to target (or, alternatively: inherited from the context by the target).47 In terms of Hardt (1993) we could say

|| 47 I assume that (in contrast to referential indices) binding indices cannot be transferred from the context to the target, because the referents that bear them are not fixed in the discourse (cf. the impossibility of the strict-1 reading when the antecedent of the gap is bound by a nonreferential element, see (16)–(18) in chapter 3.2.1). However, Ede Zimmermann (p.c.) suggested to copy the gap’s antecedent with all its indices from the context to the target. The elegant outcome of the latter idea is (i) that referential indices cannot/need not be inherited/ transferred from the context to the target and (ii) that binding indices can (and must) be copied from the context to the target. Although E. Zimmermann’s suggestion is a way out of a theoretical/representational problem (namely, where do all the indices in the target come from?), it

180 | Topic Drop in German: The Interpretation of the Gap

that in its strict interpretation the gap is Dbound (“discourse-bound”) by its antecedent; in Fox’ (2000) terms we can speak of a Referential Parallelism. In (30)/(31) bound pro is not licensed since the referential binders in context and target (Anna, die Hansen-Brüder) bear different phi-feature specifications: (30) A: B:

(31) A: B:

Annai mag [ihrei Nase]s. Anna likes her nose-ACC [proi [pro]]s mögen die Hansen-Brüdert auch. [ACC] like the Hansen-brothers as-well

s = si

Annai mag [ihrek Nase]s. [prok [pro]]s mögen die Hansen-Brüdert auch.

s = sk (strict-2, s = sk)

(strict-1, s = si)

However, as mentioned before, the sloppy reading arises due to the fact that the possessive pronoun ihre is ambiguous between ‘her’ and ‘their’. I.e., a bound reading becomes possible by (accidental) PF identity of context bindee and the bindee part of the gap / by PF-related phi-feature compatibility of context bindee and target binder, which is illustrated by (32): (32) B:

[“Ihre” [pro]]s mögen die Hansen-Brüdert auch.

As I assume that the sloppy-2 reading emerges post-syntactically, it is (in contrast to the gap’s strict readings) not represented in the syntax. The sentence’s (underlying) PF-representation (including indexation), however, could look as follows, cf. (33): (33) B:

[Ihret [Nase]]s mögen die Hansen-Brüdert auch.

(sloppy-2, s = st)

As to non-referential binding (i.e. binding by a quantified/indefinite element), there are no obligations on PF/phi-feature identity between the target binder and the context binder. I.e., the PD is semantically identical to the AC modulo the focus marked non-referential target binder (and its bindee). Thus, the nonreferential target binder’s denotation as well as its phi-feature specifications can

|| sweeps away the structural differences between “real” binding relations and relations that are ambiguous between binding and coreference as it treats the two relations (technically – and hence, structurally) alike as it says that a coreference relation that can be interpreted in terms of binding should be interpreted as (variable) binding.

The Representation of the Gap | 181

differ from those of the context binder (and, accordingly, the denotation and phi-feature specifications of the (bindee part of the) gap can differ from those of its antecedent/the context bindee: (34) A: (i) B: (ii) B:

(35) A: B:

[Peteri]r mag [seinei/r Nase]s. Peter likes his nose-ACC [proi [pro]]s mögen FAST ALLEr MÄNNER. [pror [pro]]s mögen FAST ALLEr MÄNNER. [ACC] like almost all men

s ∈ {si, sr}48

Peteri mag [seinek Nase]s. [prok [pro]]s mögen FAST ALLEr MÄNNER.

s = sk (strict-2, s = sk)

(strict-1, s = si) (sloppy-3, s = sr)

The only “obstacle” for non-referential binders are pure indexicals, namely directly referential 1st/2nd person pronouns as ich (I), mich (me) ihr (you-pl) etc. In this case, a non-referential binder cannot impose its own features on the dropped indexical. In contrast to other examples where this is possible, e.g. in (34) or in Johnny did his homework, but I didn’t from Heim (2005:45), a topic dropped pure indexical cannot be interpreted as if it were featureless, cf. (36): (36) A: (i) B: (ii) B:

Ichi mag michi. I like me/myself-ACC proi mögen FAST ALLEr MÄNNER. (strict-1) pro*r mögen FAST ALLEr MÄNNER. (*sloppy-3) [ACC] like almost all men (okyou-sg/*themselves)

However, the problem is the position and not being a pure indexical. This is shown by (non-referentially) bound instances of dropped indexicals in determiner/possessive position which can be interpreted as if they were featureless, cf. (37):49

|| 48 Note that here (and in the following) the index ‘r’ operates as a binding index (and not as a referential index, see e.g. Heim 1998 (1993) or Sternefeld 1998 for this differentiation). Note further that indexing the context binder (and its bindee) with a binding index (as in (34)) might be problematic as it suggests that speaker A has intended binding before Speaker B formulates his/her reply to A’s utterance. 49 Consider therefore (i) where the particle selbst (self) is a functor that enables the indexical pronoun to be in determiner position (at LF): (i) A: [Ichi]r mag [michi/r selbst]s. s ∈ {si, sr} I like me/myself-ACC self

182 | Topic Drop in German: The Interpretation of the Gap

(37) A: (i) B: (ii) B:

[Ichi]r mag [meinei/r Nase]s. s ∈ {si, sr} I like my nose-ACC [proi [pro]]s mögen FAST ALLEr MÄNNER. (strict-1, s = si) [pror [pro]]s mögen FAST ALLEr MÄNNER. (sloppy-3, s = sr) [ACC] like almost all men (okyour-sg nose/oktheir nose)

The summary in (38) resumes the different interpretations (and the respective mechanisms) which we can find in topic drop dialogues: (38) Kinds of gap interpretations in topic drop dialogues (i) Strict interpretations via referential identity (strict-1, strict-2) (d(iscourse)-binding, cf. Hardt 1993) (ii) Referential binding via phi-feature identity (sloppy-1) (morpho-syntactic binding / phi-feature based coreference) (iii) Referential binding via PF identity (sloppy-2) (post-syntactic binding / PF-based coreference) (iv) Non-referential binding via quantified or non-specific indefinite elements (sloppy-3) (semantic binding) As to the internal structure of pro I suggest the following representations (for the different readings of topic dropped elements), cf. (39): (39) The morpho-syntax of pro (in topic drop dialogues) Referential Identity (Referential Parallelism) (i) Contextually indexed pro [proindex [pro]] (pure indexicals: proindex) D-binding, strict-1 and strict-2 interpretation Referential Binding (Structural Parallelism) (ii) Contextually unindexed but morphologically constrained pro [prophi [pro]] (pure indexicals: prophi) Morpho-syntactic binding, sloppy-1 interpretation

|| (a) B: (b) B:

[Dichi selbst] mag fast jederr. (strict-1, s = si) [Sichr selbst] mag fast jederr. (sloppy-3, s = sr) [You-sg-ACC self/REFL-ACC self] likes almost everyone Thanks to Ede Zimmermann for discussing this point with me.

The Representation of the Gap | 183

(iii) Contextually unindexed but phonologically constrained “pro” [“context bindee” [pro]] Post-syntactic binding, sloppy-2 interpretation Non-Referential Binding (Structural Parallelism) (iv) Contextually unindexed pro [pro [pro]] (pure indexicals: *pro) Semantic binding, sloppy-3 interpretation (The differences between) referential binding and non-referential binding can be depicted as follows, cf. (40) (40) Binding in topic drop dialogues (i) Elements in context and target are considered as equivalent/parallel (verbatim environment) (ii) A Parallelism Domain is built up (iii) The potential target binder is checked for its referential properties Referential binder

Non-referential binder

Context binder and target binder are (successfully) checked for phi-feature identity (iv) The target binder is focus marked (v) An unindexed pro-copy is generated (vi) Binding takes place According to what has been elaborated above, the (referential) relation which is ambiguous between binding and coreference can be distinguished from genuine (quantifier) binding. In topic drop dialogues only the latter applies freely, without reference to PF/phi-feature identity conditions. Disregarding the “accidental” PF sloppy reading, we can state the following: If the binder in the target sentence of a topic drop dialogue is a referential element, binding can only take place when it bears the same phi-feature specifications as the context binder. The effect of this ‘phi-feature identity condition’ is that context and target are semantically identical, modulo the denotation of the focus marked target binder – which means that in principle the Parallelism Domain also contains a phifeature skeleton of the focus marked target binder. Is the binder in the target sentence of a topic drop dialogue a non-referential element, binding can take

184 | Topic Drop in German: The Interpretation of the Gap

place without matching the phi-feature specifications of the context binder. In this case, context and target are (also) identical modulo focus marked elements/ the focus marked target binder (however, phi-feature identity between context binder and target binder plays no role). Thus, with one exception (namely, pure indexicals) non-referential binding involves a kind of ‘tabula rasa principle’ with respect to the specifications of the context binder. The structural constraint applying to referential binding but not to non-referential binding allows us to define/refine the distinction between (i) relations which are ambiguous between coreference and binding (‘referential binding’) and (ii) relations that cannot be reduced to coreference phenomena (‘non-referential binding’) through +/-necessary phi-feature identity. Due to their structural (morpho-syntactic) determination the former should be kept apart from the latter. I.e., relations that can be reduced to coreference should not be equated with / analysed as (variable) binding but rather considered as coreference relations with morphosyntactically determined binding options.50 The possibility of phi-feature mismatches in non-referential binding relations and their absence in referential binding relations hints at the conclusion that the topic drop construction allows for a systematic distinction of binding and coreference in natural language.

|| 50 This stands in contrast to what e.g. Heim (1998:205) notes, namely that “[o]ther semantic relations, in particular such as may obtain between two referring NPs, are not even represented on any syntactic level and can therefore not be directly licensed or ruled out by structural conditions. They are regulated instead by an extragrammatical principle which says, very roughly, that coreference is unavailable whenever the same meaning can be conveyed by means of variable binding.” Consider also Jäger (2006:81): “Given that these conceptually similar phenomena – binding and coreference – are not overtly distinguished in natural language, it seems somewhat artificial to separate them in linguistic theory.”

4 Null Subjects at the Syntax-Pragmatics Interface 4.1 Introduction: (Antecedent-Independent) Null Subject Omission in German This chapter is mainly about referential, definite null subjects in the German prefield of main (1a) and embedded (1b) verb-second (V2) clauses.1 (1)

a. b.

Ø Bin mal wieder zu spät dran. [I] am once again too late PRT Ich glaube, Ø kommen mal wieder zu spät.2 I think, [we] come-1(/3)pl once again too late

The subject omissions I focus on are only possible in the German prefield3 and ungrammatical in the middle field (cf. Ross 1982 and subsequent literature): (2) a.

* Heute bin Ø mal wieder zu spät dran. Today am [I] once again too late PRT b. * Ich glaube, heute kommen Ø mal wieder zu spät. I think, today come-1(/3)pl [we] once again too late

To a minor extent, I will also discuss occurrences where the null subject is not in sentence initial position, cf. (3). However, it is questionable whether subject

|| 1 Some of the ideas within this chapter were previously published as Trutkowski (2011). 2 Note that though the verb form in (1b) is syncretic between a 1st and a 3rd person plural, the null subject unambiguously refers to a 1st person plural. In the following I will indicate the reading the particular gap has in the glosses and put impossible interpretations into brackets (as above). When the form of verbal inflectional endings plays no role for the discussion and the respective referent(s) can be easily inferred from the glosses’ context, I will not indicate person/number specifications, as e.g. in ich glaube (I think) above. Note that in contrast to topic drop gaps which are represented by an underline, [ _ ], (so called ‘out of the blue’) null subjects are represented by the symbol [ Ø ]. 3 According to standard assumptions the German prefield can only host one constituent (however, see Müller 2002 for a more differentiated view). Consequently, only one null argument per sentence is present in the prefield (cf. e.g. Huang 1984, Cardinaletti 1990, Wilder 1996).

186 | Null Subjects at the Syntax-Pragmatics Interface

omissions as in (1) and (3) should be grouped together. In chapter 4.3.4 I will come back to omissions as illustrated in (3). (3) a.

Was würdest Ø mir empfehlen? What would-2sg [you-sg] me recommend? b. (?) Was würdet Ø mir empfehlen? 4 What would-2pl [you-pl] me recommend?

Following Trutkowski (2011) I will argue that occurrences of null subjects as in (1a/b) are neither instances of topic drop, diary drop nor of ‘classical’ pro-drop or radical/discourse pro-drop, but instead a peculiar phenomenon of German syntax. At first, I will show (on an empirical basis) which factors have an influence on the occurrence of null subjects in German, i.e. I will determine the exact (morpho-)syntactic and pragmatic / information structural conditions under which they are well-formed. After that I will shortly examine how German fits into a general typology of (non-)null subject languages. Note that the kind of null subjects that I discuss in the following occurs (like topic drop) mostly in spoken and/or informal registers which are characterised by an economic language use (e.g. telegrams, emails, text messages, chats). They do not occur in formal language. Nevertheless, my examination of their identification and licensing conditions will come to the conclusion that the conditions under which they are well-formed are sufficiently robust to impose a language change (whenever). In the (generative) literature prefield null subjects in German are often treated as a side issue within the discussion on topic drop.5 Topic drop (cf. Ross 1982, Huang 1984, Fries 1988, Sigurðsson 1993, Rizzi 1994, Huang 2000, Rizzi 2002, Ackema & Neeleman 2007, Sigurðsson 20116) allows nearly all kinds of constituents to be dropped as long as there is some (non-)linguistic discourse antecedent which the gap in the prefield can refer to, cf. (4) vs. (5):

|| 4 Note that such (non-prefield) occurrences of 2nd person plural null subjects are not very well accepted by German native speakers. However, as they occur relatively regularly, I include them in the discussion. 5 However, see Cardinaletti (1990), who differentiates null object and null subject cases in German. She notes that in the subject case a null pronoun, pro, is involved: “pro in (25) [(25) = pro habe es gestern gekauft, ET] differs from the null subject found in Italian in that the recovery of its feature content does not depend on the agreement specification of the verb, but on linguistic or pragmatic context.” (cf. Cardinaletti 1990:81). 6 Sigurðsson (2011:280) notes that in non-null subject languages as in Swedish or German “agreement affects the identification of null subjects” (italics are mine).

Introduction: (Antecedent-Independent) Null Subject Omission in German | 187

(4) A:

a.

B:

b.

B:

Kennst du den Hans? Know you the Hans-ACC? ‘Do you know Hans?’ Na klar, _ is’ mein Nachbar. Of course, [NOM] is my neighbour ‘Of course, he is my neighbour.’ Ne, _ kenn’ ich nicht. No, [ACC] know I not ‘No, I don’t know him.’

(5) a. * _ Is’ mein Nachbar. b. * _ Kenn’ ich nicht. In contrast to the data in (4) and (5), datum (6) suggests that there are null elements where this obligation is not met, i.e. drop is possible without a contextually given, salient discourse antecedent. However, as (6) shows, this seems to hold only for 1st and 2nd person null subjects but not for 3rd person null subjects and objects gaps (as already shown by (5a/b), see also Trutkowski 2011): (6) a. b.

Ø Komme/Kommst/*Kommt leider immer zu spät. 1sg/2sg/*3sg [I/You-sg/He, she, it] come(s) unfortunately always too late Ø Kommen/Kommt/*Kommen leider immer zu spät. 1pl/2pl/*3pl [We/You-pl/They] come unfortunately always too late

Thus, according to the data examples in (4)–(6) we can differentiate two cases of subject omission in German, summarised in (7a/b): (7) Different means of identification and licensing subject gaps in German (a) The gap is identified/licensed in dependence on a salient discourse antecedent = Instances of Topic Drop (TD) – cf. 3rd person null subject/null object occurrences as in (4) (b) The gap is identified/licensed independently of a salient discourse antecedent = Instances of ‘Out of the Blue’ Drop (OBD) – cf. 1st/2nd person null subject occurrences as in (6) In the course of this chapter I will try to find an empirically and theoretically adequate account for instances of (7b), i.e. of antecedent-independent subject omissions in the German prefield to which I will refer by OBD (‘Out of the Blue Drop’ or ‘Out of the Blue Dropped’ null subjects).

188 | Null Subjects at the Syntax-Pragmatics Interface

However, before presenting more (corpus) data, I will first give a definition of what I understand as an out of the blue context. I will define an out of the blue context on pragmatic grounds, as follows, cf. (8): (8) Out of the blue context An out of the blue context is a null context, i.e. a context which provides no salient discourse antecedent(s) at all. Discourse environments in which a discourse antecedent could be inherently given should, of course, be excluded by this definition. Although it is never clear whether all nuisance factors have been eliminated, I will assume that out of the blue contexts indeed exist. Thus, I take titles of books, novels, poems etc., headings of (newspaper) articles as well as song titles, beginnings of novels, paragraphs and conversations to be good examples of out of the blue contexts. The following data, (9)–(12), are (corpus) examples of subject omissions in the prefield for all out of the blue omittable person/number combinations (i.e. all persons/numbers except the 3rd person singular/plural) that are observable in German. In the following examples the context either provides no discourse antecedent at all or an incompatible one.7 Note that out of the blue uttered null subjects are not only a characteristic of Modern German, especially of electronic or more colloquial registers, cf. (9a), but are also found in older stages and/or non-colloquial registers of German, cf. (9b) and (9c). 1st person singular (9) a.

Ø Würde gerne mal auf einem richtig großen Gut in Frankreich [I] would-1(/3)sg like PRT on a really big manor in France Urlaub machen. Ø Finde zwar was im Internet, holidays make. [I] find-1sg indeed something on-the internet, Ø möchte es mir aber lieber empfehlen lassen. Wichtig sind [I] will-1(/3)sg it me however rather to-recommend let. Important are mir stilvolles Wohnen im Landhausambiente, tägliches Ausreiten for-me stylish lodging in-a country-style, daily riding-out und natürlich landestypische Küche. Ø Freu mich and of-course country-characteristic food. [I] look-forward-1sg REFL

|| 7 As will be seen, this is questionable for some of the examples given.

Introduction: (Antecedent-Independent) Null Subject Omission in German | 189

auf jede Menge Tipps. Danke ! 8 for a-lot hints. Thanks! b.

Ø Bin in Essen glücklich angekommen! Besten Gruß 9 [I] am in Essen luckily arrived! Best greetings

Figure 1: Postcard, dated 1918, cf. example (9b)

c.

Ø Ging heut Morgen übers Feld 10 [I] went-1(/3)sg today morning across-the field

2nd person singular (10) a.

b.

Ø Bist mir so nah 11 [You-sg] are-2sg me so close Ø Wirst mir fehlen 12 [You-sg] Will-2sg me be-missing ‘I will be missing you.’

|| 8 http://andivaswelt.wordpress.com/2011/06/ (checked: 22.05.2012) 9 Author unknown, original picture was decolourised; http://www.ebay.de/itm/Essen-56070gut-angekommen-/220947543152?pt=Ansichtskarte_Zubeh%C3%B6r&hash=item33717ff470 (checked: 22.05.2012) 10 Title/first line of a song of Gustav Mahler, 1884/1885 11 Song title of the pop singer Sandra Weiss, http://www.kristallklang-musik.at/shop/maxicd/sandra-weiss---bist-mir-so-nah.php (checked: 17.05.2014) 12 Article heading, http://blog.zeit.de/politiker-wg/?p=51 (checked: 26.04.2009)

190 | Null Subjects at the Syntax-Pragmatics Interface

c.

Dann: „Die lässt sich die Haare trotzdem so schneiden. Then: “She gets REFL the hair nevertheless in this way cut. Ø Kennst sie doch.” 13 [You-sg] know-2sg her PRT” ‘Anyway, she will get her hair cut like that. You know her.’

1st person plural (11) a.

b.

Wolfram Müller – Ø Nutzen jetzt unsere Stärken 14 Wolfram Müller – [We] exploit-1(/3)pl now our strengths ‘Wolfram Müller – We are now exploiting our strengths’ Ø Sind mal kurz drüben. Wandern in Nordböhmen. [We] are-1(/3)pl PRT shortly over-there. Hiking in Northern Bohemia.15

Figure 2: Book title, cf. example (11b)

|| 13 DWDS Corpus, Berliner Zeitung, 22.01.2005, p. 3. 14 News heading, http://www.deutscher-leichtathletik-verband.de/index.php?NavID=1& SiteID=28&NewsID=21594 (checked: 17.05.2014) 15 Book title, K. & A. Micklitza, “Sind mal kurz drüben” (Lusatia Verlag, Bautzen 2010).

Introduction: (Antecedent-Independent) Null Subject Omission in German | 191

2nd person plural (12) a.

[Ø Suche guten Fernseher , Ø hoffe] Ø könnt mir helfen 16 [I] search-1sg a good TV set, [I] hope-1sg [you-pl] can-2pl me help

b.

Bitte einmal abstimmen. Ø Würdet mir sehr mit eurer Meinung Please PRT vote. [You-pl] would-2pl me very with your-pl opinion helfen! 17 help

Most of the data from (9)–(12) fit the definition in (8), i.e. the respective null subjects are uttered in an out of the blue context. However, not all examples are clear instances of OBD:18 I will shortly discuss those which are possibly problematic – these are the examples in (9b), (10c), (11a) and (11b). Consider first (9b): I will not count data like (9b) as instances of OBD, because the figure on the postcard (who tripped over the stairs) possibly constitutes a non-linguistic antecedent for the 1st person singular null subject. The same holds for (10c): In this dialogue the hearer (that is addressed by the speaker) is provided by the discourse situation. Examples as (10c) are well known from English diary drop registers which I will shortly discuss in the following, thereby pointing out some differences between OBD and diary drop. Consider next (11a): In (11a) the gap is preceded by an incompatible antecedent: the reason why the null subject in (11a) cannot be understood as referring back to the antecedent Wolfram Müller is contingent on the fact that in German the verb has to agree with the (omitted) subject. If the gap was identified by the antecedent Wolfram Müller, subject-verb agreement would be violated (as the verb displays plural morphology and the only antecedent, Wolfram Müller, which is provided by the context, bears 3rd person singular features).

|| 16 http://gutefrage.net/frage/suche-guten-fernseher--hoffe-koennt-mir-helfen (checked: 17.05.2014) 17 http://www.rund-ums-baby.de/forenarchiv/vornamen/Bitte-einmal-abstimmen-Wuerdetmir-sehr-mit-eurer-Meinung-helfen_24878.htm (checked: 17.05.2014) 18 Note that OBD null subjects cannot be (hidden) imperatives: Whereas imperative forms cannot occur with modal verbs, (i), OBD null subjects can, (ii): (i) * (Du) soll! (You-sg) should-imp-sg (ii) Ø Solltest dich benehmen. [You-sg] should-2sg REFL behave

192 | Null Subjects at the Syntax-Pragmatics Interface

In a similar vein, what holds for (11a) can also be maintained for (11b): In (11b) the book cover only shows a single person, but the null subject must refer to at least two persons.19 Thus, I will count (11a) and (11b) as instances of OBD, because the features of the antecedents which are provided by the contexts (respectively) are incompatible with the features of the respective null subjects.20 Consequently, these null subject occurrences are not dependent on the presence of a contextually given antecedent. Hence, the only cases which are possibly no instances of OBD are (9b) and (10c). Note in particular that the null subjects in (9c), (11a) and (11b) are licensed and well-formed despite the lack of an (appropriate) discourse antecedent and despite the fact that verb forms are syncretic: in (9c) ging is ambiguous between a 1st and 3rd person singular interpretation, and nutzen in (11a) as well as sind in (11b) are ambiguous between a 1st and 3rd person plural interpretation. This observation will be discussed later in great detail. As to the differences between OBD and diary drop (cf. Haegeman 1997, 2007, Haegeman & Ihsane 1999, 2001, Weir 2008) I will show that in diary drop scenarios null subjects can only be omitted when the discourse provides a (default) referent or an antecedent/addressee (which is particularly addressed in the given context, often accompanied by a specific intonation), cf. (13): (13) a. b. c.

Ø Feel so real 21 He lives up the road a couple of miles, Pebwater Farm, Ø can’t mistake it.22 Ø Shouldn’t have took more than you gave 23

Although the omissions in (13) look as if the respective null subjects therein were omitted out of the blue, I will argue that they are constrained by the discourse context. The null subject in (13a) refers (default-like) to the speaker/singer, in (13b) it addresses the person(s) that has/have asked for the

|| 19 Here and in the following I will suppose that 1st/2nd person plural pronouns refer to at least two persons, i.e. I will depart from an analysis of 1st/2nd person plural pronouns as Pluralis Maiestatis or Pluralis Modestiae, referring to a single person with a distinguished social status. Anyway, the use of the Pluralis Maiestatis or Pluralis Modestiae is not coherent with the register-specific, informal use of topic drop and OBD. 20 However, note that the antecedents Wolfram Müller in (11a) or the person on the book cover in (11b), respectively, could be regarded as partial/split antecedents for the gap. 21 Song title of Steve Arrington. 22 Symons, J., The Progress of a Crime, from Haegeman (2007), her example (5c). 23 Song title of the band ‘Traffic’.

Introduction: (Antecedent-Independent) Null Subject Omission in German | 193

way, and in (13c) its reference is highly ambiguous even if the preferred interpretation is one in which it refers to a 2nd person addressee (being coreferent with the overt pronoun you). To point out the difference to prefield OBD null subjects in German, I will compare (13c) with its German equivalents and show that in German it is not necessarily the pragmatic and syntactic context, but the discreteness of verbal inflectional endings which restricts the interpretation of null subjects. As a matter of fact, there is not only one German equivalent to (13c), but six, cf. (14a–f): (14) a. b. c. d. e. f.

Ø Sollte nicht mehr nehmen, als du gabst. [I] should-1(/3)sg not more take than you-sg gave-2sg Ø Sollte nicht mehr nehmen, als ihr gabt. [I] should-1(/3)sg not more take than you-sg gave-2pl Ø Solltest nicht mehr nehmen, als du gabst. [You-sg] should-2sg not more take than you-sg gave-2sg Ø Solltest nicht mehr nehmen, als ihr gabt. [You-sg] should-2sg not more take than you-sg gave-2pl Ø Solltet nicht mehr nehmen, als du gabst. [You-pl] should-2pl not more take than you-sg gave-2sg Ø Solltet nicht mehr nehmen, als ihr gabt. [You-pl] should-2pl not more take than you-sg gave-2pl

Note first that even if the inflectional endings of the 1st and the 3rd person coincide, the null subject in (14a–b) is unambiguously interpreted as referring to a 1st person. For the moment I will leave this point aside. However, as will be shown later, in German the 1st person does not need to be identified/licensed by a (default) discourse antecedent, nor does its identification/licensing take place by speaker/hearer features. Looking at (14c–f), we see that the respective verb forms determine the choice of a null subject ultimately. This however, cannot be said with respect to (13c): In the English version the null subject can be interpreted as referring to a 1st/2nd(/3rd)24 person singular/plural, whereas in German distinct inflectional endings allow a non-ambiguous interpretation. Moreover, in German the number (singular vs. plural) of the second pronoun (which corresponds to you in (13c)) can differ from the number of the null subject (under the

|| 24 As will be shown later, an out of the blue interpretation of a null subject referring to a (specific) 3rd person referent is generally impossible as long as the discourse does not provide an identifying antecedent. This also holds for (e.g.) pro-drop languages, cf. Samek-Lodovici (1996), Frascarelli (2007).

194 | Null Subjects at the Syntax-Pragmatics Interface

assumption that the null subject is interpreted as referring to a 2nd person), cf. (14c) vs. (14d) and (14e) vs. (14f) – this, however, is hardly an option in (13c): here a 2nd person null subject and the pronoun you must be interpreted in coreference to each other. I.e., next to an impersonal reading, only a singularsingular or a plural-plural interpretation is possible. The reason for this difference is – as before – the fact that German has richer inflectional paradigms than English (in both, the verbal and the (pro)nominal domain). Thus, we can hypothesise that due to the latter the well-formedness of null subjects in German is independent of the presence of a surrounding context – whereas the licensing and identification of null subjects in English depends on a given context (which provides a default antecedent, a coreferring pronoun or a discourse antecedent in (13)).25 Summarising the above data, we can state that, firstly, OBD is possible with st 1 and 2nd person singular and plural subjects, and that, secondly, syncretisms do not seem to matter, i.e., null subjects in the German prefield (= the specifier position of CP) of main (and embedded26) V2 clauses are licensed out of the blue and receive an unambiguous interpretation despite syncretisms within the verbal inflectional paradigms. Thus, the null subject occurrences in (9)–(12) cannot be considered as instances of (a German version of) diary drop. As shown by (13), (English) diary drop is only possible if the discourse provides an accessible (default) antecedent. Moreover, in case a verb form is syncretic (as e.g. in (13c)), English diary dropped elements remain ambiguous (e.g. between a singular vs. plural reading). Conversely, this is not the case for German OBD instances, cf. (14). Thus, the interpretation of German OBD is not context-dependent / not determined by the presence of a topic-drop like identification/licensing mechanism (which would make a (default) antecedent necessary). Moreover, as will be shown later, OBD is not licensed by speaker/hearer features. According to standard views (non-dialectal) German does not count as a pro-drop language. Nonetheless, I will suppose (as a working hypothesis) that OBD null subjects in German are (like topic dropped elements, see chapter 2 and 3) represented by pro (cf. Cardinaletti 1990), which I assume to be a genuine

|| 25 Under the conditions that have to hold for null subject licensing/identification in English, null subjects in German are well-formed, too. However these conditions do not constitute necessary licensing/identification conditions on German OBD null subjects. 26 I will discuss this occurrence of OBD later.

Well-Formedness Conditions on Antecedent-Independent Null Subjects | 195

pronoun whose special syntactic distribution and semantic properties are determined by the lack of a phonological matrix.27

4.2 Well-Formedness Conditions on Antecedent-Independent Null Subjects With respect to the well-formedness conditions on null subjects it is sometimes difficult to differentiate between licensing and identification conditions. One can generally say that the identificational component provides a null pronoun with (semantic) content/reference whereas “[f]ormal licensing restricts the occurrence of pro to a particular syntactic position, or particular positions, in a language”, cf. Ackema, Brandt et al. (2006:4). However, a comparison between Italian, German and Polish shows that licensing cannot be equated/narrowed down to the +/-presence of a particular syntactic configuration. Instead, the information structural status of a null subject antecedent plays a role as well: Samek-Lodovici (1996:31ff) shows that a 3rd person null subject in Italian (and other languages) is not well-formed when introduced by a by-phrase which constitutes a non-topical discourse antecedent,28 cf. (1), but well-formed when the discourse antecedent is a topic,29 cf. (2). In German and Polish, on the other hand, 3rd person null subject antecedents can be non-topical PPs, cf. (3)/(4): (1)

Questa mattina, la mostra é stata visitata da Giannii. This morning, the exhibition was visited by John. Piú tardi, *Øi / eglii / luii ha visitato l'universitá. More late, [he] / he / he has visited the university.

|| 27 Note that this assumption possibly conflicts with (older) standard views, e.g. Haegeman (1997:235) a “positive setting of the pro-drop parameter is associated with a cluster of syntactic properties”, e.g. ‘the definite subject can identify a postverbal position’, ‘quasi-argument subjects of weather verbs are necessarily nonovert’ etc. However, this cluster of properties associated with the pro-drop parameter is problematic anyway for German and German dialects, cf. Haider (1994). See Roberts & Holmberg (2010) and Camacho (2013) for a current discussion of the pro-drop parameter. 28 Samek-Lodovici (1996) attributes this topic test to Strawson (1964). The grammaticality judgements of the Italian data are from Samek-Lodovici (1996). 29 See also Frascarelli (2007).

196 | Null Subjects at the Syntax-Pragmatics Interface

(2)

Questa mattina, Giannii ha visitato la mostra. This morning, John has visited the exhibition. Piú tardi, Øi / ?eglii / ??lui ha visitato l'universitá. More late, [he] / he / he has visited the university.

(3)

Diese Ausstellung wurde auch von Hansi besucht. This exhibition was also by Hans visited Ja stimmt, Øi / eri war danach sogar noch an der Uni. Yes right, [he] / he was later even at the uni(versity)

(4)

Ten samochód został zaprojektowany przez Jankai. This car was designed by Janek No, Øi / oni nawet dostał nagrodę za ten wspaniały pomysł. Indeed, [he] / he even received a prize for that fabulous idea

I assume that whenever the identification component is satisfied, languages should only differ with respect to formal licensing conditions. If a null element cannot be realised – although it can be identified – this must be due to the fact that licensing conditions are not properly fulfilled in a given language, or alternatively, that they cannot be fulfilled at all, as e.g. the lack of topic drop in prodrop languages shows (European Portuguese, cf. Raposo 1986, may be an exception30). Thus, considering the data (1)–(4) we can suppose that the +/-topical status of an antecedent is also part of the licensing conditions on (3rd person) null subjects. Of course, identification and licensing are mutually dependent on each other (as there can neither be a licensed but unidentified nor an identified but unlicensed null element). In this study I focus on the well-formedness conditions for null subjects in general and I hope to make clear which particular identification and licensing conditions must be met for 1st and 2nd (vs. 3rd) person null subjects to be well-formed in the sentence initial position (the prefield) of German V2 clauses. Given the lack of an appropriate antecedent (cf. out of the blue-drop), two mechanisms remain as possible identifiers/licensers of German OBD null subjects:

|| 30 However, as pointed out by Esther Rinke (p.c.), the occurrence of null objects in European Portuguese is hardly comparable to (German) topic drop, cf. also Hernani Da Costa who kindly discussed some EP data with me.

Well-Formedness Conditions on Antecedent-Independent Null Subjects | 197

(5) Possible sources for the well-formedness of (1st/2nd person) null subjects (a) Discrete verbal inflectional endings (for each person/number combination) (b) Speaker/hearer features (for speaker and hearer, respectively)31 With respect to (5a) the question arises how distinct verbal inflectional endings must be for null subjects in (partial) pro-drop languages to be licensed. See e.g. Jaeggli & Safir (1989) for a very early discussion on that topic, Vainikka & Levi (1999) on the distinctness of verbal inflectional endings in partial pro-drop languages (cf. Finnish and Hebrew) and Roberts & Holmberg (2010) for a recent overview on the topic. With respect to (5b) we can ask how the presence of syntactically manifested speaker/hearer features (or, alternatively: speech participant features) can be verified. Giorgi (2010) claims that speaker-related features are present in the left-most position of the C-layer, providing evidence from Italian. Sigurðsson (2011:269; 281) suggests the following: Universal Grammar does not contain any null-subject parameter, licensing of null arguments instead following from general factors (in the spirit of Chomsky 2005). Second, any definite argument, overt or silent, positively matches at least one C/edge linker in its local C-domain, where C/edge linkers include Top(ic) features and speech participant features (“speaker,” “hearer”). [...] referred to as the logophoric agent (ΛA) and the logophoric patient (ΛP).

Notably, when speaker/hearer features exist, they should be universally present and (syntactic) effects of their existence should show up in all languages. However, even if we can detect such effects, it does not automatically mean that speaker/hearer features are active null subject licensers/identifiers. In the following I will try to empirically determine which mechanism, cf. (5a) or (5b), allows null subjects in the German prefield to be well-formed. Besides, I will show how the occurrence of null subjects is constrained across different languages/language types.

|| 31 One could suggest that speaker/hearer features are present in each number.

198 | Null Subjects at the Syntax-Pragmatics Interface

4.2.1 Empirical Investigations of German and Other (Non-)Null Subject Languages I have already shown above that 1st and 2nd person null subjects in the German prefield can be uttered out of the blue. I.e., OBD is not (necessarily) subject to an antecedent-dependent identification/licensing condition and hence not an instance of topic drop/topic drop-like omissions. Thus, the further investigation will be focused on the question(s) whether OBD is licensed/identified by discrete verbal inflectional endings (similar to what is assumed for pro-drop languages as Polish, Italian, Spanish etc.) or, independently of discrete verbal inflectional endings, by speaker/hearer features (as suggested (not only) for radical pro-drop languages like Chinese by e.g. Sigurðsson 2011). I will first examine if it makes a difference whether a 1st/2nd or a 3rd person subject is omitted. Siewierska (2004:5) notes: It is generally acknowledged that “there is a fundamental, and ineradicable, difference between the first and second person, on the one hand, and the third person on the other” (Lyons 1977:638). One manifestation of this difference is that whereas the first and second persons are regularly referred to essentially only by person markers, reference to the third person can be achieved by any lexical expression.

In fact, the prevailing majority of researchers assume that the most relevant split of person features concerns the 1st/2nd vs. the 3rd person, cf. also Ritter & Harley (2002:486), Déchaine & Wiltschko (2002:419).32 Note that the split between the 1st/2nd and the 3rd person also concerns the semantics of (pro)nouns, cf. Kaplan (1989b:569), who states that “[t]he directly referential term goes directly to its referent, directly in the sense that it does not first pass through the proposition.” In order to test whether expressions that are related to 1st/2nd vs. 3rd person referents behave differently, I will apply the (what I refer to as) ‘Coordinated Antecedents Test’ (CAT). The idea for this kind of ‘test’ goes back to/is inspired by Cole (2009:567– 571). Cole observes that in some pro-drop languages (which license/identify null subjects via discrete inflectional endings) some inflectional paradigms display syncretisms, e.g. the Spanish Imperfect paradigm (between the 1st and the 3rd person singular) or the Italian Subjunctive paradigm (between the 1st and 2nd (and 3rd) person singular). Cole (2009) notes that despite syncretisms within inflectional paradigms, a null subject can be identified when there is a matching antecedent, as e.g. Juan in the Spanish example in (6): || 32 However, see de Schepper (2012) for an alternative view.

Well-Formedness Conditions on Antecedent-Independent Null Subjects | 199

(6)

Juan llegaba a casa. Ø Tenía las llaves. Juan arrived at home. [He] had-1(/3)sg the keys ‘Juan was arriving home. He had the keys.’

However, when the preceding discourse provides an incompatible antecedent (a coordinated expression, cf. Juan y yo), the null subject cannot be identified, cf. (7): (7) Juan y yo llegábamos a casa. *Ø/El tenía las llaves. Juan and I arrived-1pl to home. [I/He] had-1/3sg / He had-(1/)3sg the keys33 ‘Juan and I were arriving home. I/he had the keys.’ Cole’s conjecture (2009:570f) reads as follows: [O]ne can see that [..] null thematic subjects may be recovered by reference to unique agreement and, in its absence, by reference to a suitable antecedent in context. In the absence of both, it may have a preferred interpretation. If the preferred interpretation is not the one required, an overt pronoun must be used.

In the following, I will examine null subject licensing under a CAT. In addition to German I will look at a couple of other languages (e.g. Polish, Spanish and Chinese). Besides, in contrast/in addition to Cole (2009) I will not only look at +/-null subject licensing under syncretic verb forms, but examine whether there is a difference between 1st/2nd vs. 3rd person null subject licensing which is independent of a syncretic verb form. If we find syncretism-independent differences between 1st/2nd vs. 3rd person null subject licensing we can possibly make a stronger statement with respect to the salience of 1st/2nd vs. 3rd person referents and possibly answer the question whether speaker/hearer features exist or not. Consider first (German and Polish) examples in which the verb form is nonsyncretic: (8) [[Hans] und [ich]] haben den Film schon gesehen. Hans and I have-1(/3)pl the movie already seen a. Ø Bleibe deshalb lieber zuhause. 1st person singular [I] stay-1sg therefore rather at-home

|| 33 Note that in contrast to example (6), in (7) the syncretic verb form in (7) is crucial with respect to the (here impossible) interpretation of the null subject, whereas in (6) the 1st/3rd person ambiguity can be resolved by an identifying antecedent in the preceding context.

200 | Null Subjects at the Syntax-Pragmatics Interface

b. * Ø Bleibt deshalb lieber zuhause. [He] stay-3sg therefore rather at-home (9) [[Janek] i [ja]] byliśmy w kinie. Janek and I were-1pl at cinema a. Niestety Ø już znałam ten film. Unfortunately [I] already knew-1sg this film b. * Niestety Ø już znał ten film. Unfortunately [he] already knew-3sg this film

3rd person singular

1st person singular 3rd person singular

(8) and (9) show that (in contrast to a 1st person null subject) a 3rd person null subject is not licensed when it is derived from an antecedent that is part of a coordinated expression (cf. the ‘Coordinate Structure Constraint’, Ross 1967). This holds independently of the kind of (non-)null subject language, cf. German and Polish. As it can hardly be assumed that the coordinated XPs in (8)/(9) differ with regard to their syntactic position and function, it is reasonable to assume that ‘syntactically’/from a syntactic point of view the coordinated XPs constitute equally salient34/non-salient antecedents, i.e. the “extrinsic” salience of the 1st singular pronoun(s) and the 3rd person singular XP(s) in (8)/(9) should be the same,35 and hence their syntactically determined information structural properties as well.36 This is borne out by the data in (10)–(13):37

|| 34 I will define salience along the line of Chiarcos et al. (2011:2): “Salience defines the degree of relative prominence of a unit of information, at a specific point in time, in comparison to the other units of information.” 35 A reversal of the coordinated arguments, cf. [Ich und Hans], [Ja i Janek] does not change the interpretational options of the respective gap(s). 36 I.e. ‘extrinsic’ salience refers to the syntactic/information structural properties that an element displays independently of its intrinsic, structural properties (as e.g. being a pronoun or not, a quantified vs. non-quantified element etc.). 37 Note that – as expected – coordinated 1st and a 3rd person singular pronouns make the referent of a 1st person plural pronoun given and designate it as an aboutness topic in the sense of Reinhart (1981): (i) A: Ich erzähl dir mal was über den Hans und mich: I tell you-sg PRT something about the Hans and me B: Ø / Wir / Hans und ich waren gestern im Kino. [We] / We / Hans and I were-1(/3)pl yesterday at-the cinema (ii) A: Opowiem ci no coś o Janku i mnie: (I) tell-1sg you-sg PRT something about Janek and me B: Ø / My / Janek i ja byliśmy wczoraj w kinie. [We] / We / Janek and I were-1pl yesterday at-the cinema

Well-Formedness Conditions on Antecedent-Independent Null Subjects | 201

(10) A: a. b.

Ich erzähl dir mal was über den Hans: I tell you-sg PRT something about the Hans B: # Hans und ich waren gestern im Kino. Hans and I were yesterday at-the cinema B: Hans war gestern im Kino. Hans was yesterday at-the cinema

(11) A: a. b.

Ich erzähl dir mal was über mich: I tell you-sg PRT something about me B: # Hans und ich waren gestern im Kino. Hans and I were yesterday at-the cinema B: Ich war gestern im Kino. I was yesterday at-the cinema

(12) A: a. b.

Opowiem ci no coś o Janku: (I) tell-1sg you-sg PRT something about Janek B: # Janek i ja byliśmy wczoraj w kinie. Janek and I were yesterday at-the cinema B: Janek był wczoraj w kinie. Janek was yesterday at-the cinema

(13) A: a. b.

Opowiem ci no coś o mnie/o sobie: (I) tell-1sg you-sg PRT something about me/about self B: # Janek i ja byliśmy wczoraj w kinie. Janek and I were yesterday at-the cinema B: (Ja) byłam wczoraj w kinie. (I) was yesterday at-the cinema

(10)–(13) show that a 1st person pronoun is not a better topic/“more topic-like” than a 3rd person expression when it occurs in a coordination. Hence we can conclude the following: As the CAT does not provide an antecedent that is compatible with the feature specifications displayed by the inflectional endings of the subsequent target verb, a null subject that is nevertheless licensed must be

|| Note that throughout the CAT examples the antecedent-dependent/topic drop-like omission of the expression which refers back to the two coordinated antecedents is not (further) discussed.

202 | Null Subjects at the Syntax-Pragmatics Interface

(inherently) given / inferrable.38 As (8) and (9) show, this holds for 1st person null subjects but not for 3rd person null subjects, because, in contrast to 1st person null subjects, 3rd person null subjects always need an identifying (discourse) antecedent to be well-formed – even if inflectional verbal endings are distinct for all person/number combinations (as e.g. in Polish or Italian, cf. SamekLodovici 1996). This is not only proven by the minimal pairs in (8) and (9) but also by out of the blue null subject occurrences as in (14) and (15): (14) a.

Ø Werde glücklich sein. German [I] will-1sg lucky be b. Ø Wirst glücklich sein. [You-sg] will-2sg lucky be c. * Ø Wird glücklich sein. [She/He/It] will-3sg lucky be

(15) a.

Ø Będę szczęśliwa. Polish [I] will-be-1sg lucky-fem b. Ø Będziesz szczęśliwa. [You-sg] will-be-2sg lucky-fem c. * Ø Będzie szczęśliwa. [She] will-be-3sg lucky-fem

From the above data and discussion we can conclude that 1st/2nd vs. 3rd persons have to be differentiated with respect to their “intrinsic” salience, that is, their inherent morpho-syntactic and semantic properties (which are observable independently of the presence of syncretisms within verbal inflectional paradigms, “contra” Cole 2009), 1st/2nd person referents being easier inferrable than 3rd person referents. I.e., there may be some evidence that speaker/hearer features indeed exist and that their presence shows up in the fact that – in contrast to 3rd person null subjects – 1st/2nd person null subjects are exempted from an antece-

|| 38 Note that besides formal licensing and identification conditions a null subject can only be licensed when context and target constitute a pragmatically/syntactically coherent discourse, cf. (i): (i) A: Wo ist denn der Hans? Where is the Hans? B: * Ø Bist nicht da. Are-2sg not there

Well-Formedness Conditions on Antecedent-Independent Null Subjects | 203

dent-dependent identification.39 However, even if speaker/hearer features exist, we still do not know how to determine their locus (syntax, semantics, discourse pragmatics (?)). Moreover, as already mentioned, their existence (possibly across all languages) does not automatically entail that they act as null subject licensers (in a given language). Let me now turn to languages that display +/-systematic syncretisms within their verbal inflectional paradigms. I will start with Chinese, a language in which inflectional endings are systematically syncretic (i.e. null subject licensing takes place independently of discrete verbal inflectional endings). By applying the CAT to Chinese, we see that 1st/2nd person singular null subjects are wellformed, whereas 3rd person singular null subjects are not. I.e., the supposition that (in contrast to 3rd person null subjects) 1st/2nd person null subjects are exempted from an antecedent-dependent identification can be proven to be right for radical pro-drop languages as well:40 (16)

Wo3 he2 Zhang1san1 zuo2tian1 qu4 dian4ying3yuan4, hui2jia1 Me and Zhangsan yesterday go cinema, back-home yi3jing1 hen3 wan3 le. already very late PFV ‘Me and Zhangsan went to the cinema yesterday. I/We/*Zhangsan was/were back home very late.’ (We were back home very late: entails that we were back at the same home)

(17)

Hansi he2 ni3 (, ni3men) ye3 ke3yi3 lai2. Hans and you-sg (, you-pl) also can come Ken3ding4 hui4 chuan1 de hen3 piao4liang. Sure FU dress PRT very pretty ‘Hans and you can also come. I am sure, you/*Hans will be dressed nicely’

Due to the fact that Chinese does not display grammaticalised overt means (e.g. inflectional endings) to resolve the reference of the above 1st/2nd person null subjects (respectively), it may be reasonable to assume that in Chinese null subject resolution/interpretation is pragmatically determined (Huang 1984) or

|| 39 This holds for out of the blue occurrences of full pronouns as well, cf. (i) *He comes vs. (ii) I come. 40 Thanks to Janna Lipenkova and Jin Cui for the Chinese data.

204 | Null Subjects at the Syntax-Pragmatics Interface

that it takes place via inherently given speaker/hearer features (e.g. Sigurðsson 2011). As we have already seen in the Spanish example in (7) (subsequently repeated as (18)), in pro-drop languages, which license null subjects via discrete verbal inflectional morphology, a null subject is not well-formed when the respective verb displays a syncretic form and the context does not provide a matching antecedent: (18)

Juan y yo llegábamos a casa. *Ø Tenía las llaves. Juan and I arrived-1pl to home. [I/He] had-1/3sg the keys. ‘Juan and I were arriving home. I/he had the keys.’

What holds for Spanish can also be observed in Polish: Generally, Polish displays distinct forms for each person/number combination. However, in text messages (‘SMS’) Polish writers often “simplify” certain letters because they involve diacritic signs (e.g. ę, ś, ń) which costs the text message user more. As a consequence, in a language which marks person/number (and gender) combinations in an unambiguous way, verbal forms become ambiguous, cf. the comparison between Standard Polish (SP) and Text Message Polish (TMP) in (19): (19)

1 sg

Verbal inflectional paradigms in Standard Polish and Text Message Polish: indicative present active of odpisać / odpisac (to write-back) and the simple future of wrócić / wrocic (to return); syncretisms are marked bold

odpisać (standard)

odpisac (text message)

wrócić (standard)

wrocic (text message)

odpiszę

odpisze

wrócę

wroce

2 sg

odpiszesz

odpiszesz

wrócisz

wrocisz

3 sg

odpisze

odpisze

wróci

wroci

1 pl

odpiszemy

odpiszemy

wrócimy

wrocimy

2 pl

odpiszecie

odpiszecie

wrócicie

wrocicie

3 pl

odpiszą

odpisza

wrócą

wroca

As (20) shows, the presence of syncretisms in Text Message Polish has an impact on the licensing and identification of null subjects: (20)

* Gdy Tomek i ja wrocimy z Czestochowy, Ø zaraz odpisze. (TMP) When T. and I return-1pl from Cz., [I/he] at once write-back-1/3sg

Well-Formedness Conditions on Antecedent-Independent Null Subjects | 205

In (20), a 3rd person null subject is not interpretable/not well-formed because the only available discourse antecedent Tomek (which acts as clause-external identifier for the null subject) is not accessible/not inferrable from the gdyclause as it is packed within a coordination. The fact that the (otherwise discrete) verbal inflectional ending (which serves as clause-internal licenser for the null subject) is ambiguous between a 1st and a 3rd person is not responsible for the ungrammaticality of (20): This is shown by example (21) where the orthography complies with the standards of Polish, and where the difference between the 1st person singular odpiszę and 3rd person singular odpisze is wittingly realised (nonetheless the null subject in (21) remains uninterpretable):41 (21)

* Gdy Tomek i ja wrócimy z Częstochowy, Ø zaraz odpisze. (SP) When T. and I return-1pl from Cz., [he] at once writes-back-3sg

In analogy to the 3rd person case, a 1st person interpretation of the null subject in (20) is impossible too, because the syncretic verb form cannot identify the 1st person null subject as long as a salient 1st person singular antecedent (which could act as a clause-external identifier) is not provided by the discourse. However, when verbal inflectional endings are discrete, the presence of a clauseexternal identifier (that is, a discourse antecedent) is not a necessary condition on 1st person null subjects (in contrast to 3rd person null subjects) – this is shown by (the contrast between (21) and) (22): (22)

Gdy Tomek i ja wrócimy z Częstochowy, Ø zaraz odpiszę. (SP) When T. and I return-1pl from Częstochowa, [I] at once write-back-1sg

Thus, the contrast between (21) and (22) reveals why the null subject in (20) neither receives a 1st nor a 3rd person interpretation and shows that contrary to 3rd person null subjects, 1st (and 2nd)42 person null subjects are exempted from a clause-external (= antecedent-dependent) identification condition. Note that in a situation where we do have an identifying (possibly default) antecedent, this antecedent can “recover” the meaning of a null subject under a

|| 41 See also (9b) above, which is parallel to (21). 42 Cf. the following example which involves a coordination of a 2nd and 3rd person singular pronoun and which is parallel to (22), cf. (i): (i) Gdy Tomek i ty wrócicie z Częstochowy, Ø zaraz odpiszesz. When Tomek and you-sg return-2pl from Częstochowa, [you-sg] soon write-back-2sg

206 | Null Subjects at the Syntax-Pragmatics Interface

syncretic (1st/3rd person) verb form, cf. the Spanish examples43 in (23), (23a) being from Cole (2009): (23) a.

b.

Juan llegaba a casa. Ø Tenía las llaves. Juan arrived-3sg to home. [He] had-(1/)3sg the keys ‘Juan was arriving home. He had the keys.’ (Yo/pro) llegaba a casa. Ø Tenía las llaves.44 I arrived-1sg to home. [I] had-1(/3)sg the keys ‘I was arriving home. I had the keys.’

It seems that data as (23) can be covered by Cole’s (2009) “Maximality Principle”, cf. (24): (24) Maximality Principle, cf. Cole (2009:559): [E]very language has its point of morphological maximality […] [R]ecovery of such [= null, ET] subjects is achieved first by reference to discrete agreement and, if this fails, by reference to an antecedent in context.

However, further data show that Cole’s principle must be extended with respect to one particular detail because languages can also make use of an obviation strategy in order to assign an unambiguous interpretation to a null subject, cf. the Italian data in (25)–(27) where fossi is ambiguous between a 1st and 2nd person singular: (25) Pensavo Ø fossi malato. Thought-1sg [you-sg] were-(1/)2sg sick ‘I thought you were sick.’ (≠ ‘I thought I were sick.’) (26) Pensavi Ø fossi malato. Thought-2sg [I] were-1(/2)sg sick ‘You thought I were sick.’ (≠‘You thought you were sick.’)

|| 43 Thanks to Núria Bertomeu Castelló for providing me with additional Spanish examples and discussing the data with me. 44 Note that given a matching antecedent Text Message Polish behaves like Spanish: (i) Tomek wraca pozno do domu. Ø Duzo pracuje. Tomek return-3sg late to home. [He] a-lot work-(1/)3sg (ii) (Ja) wracam pozno do domu. Ø Duzo pracuje. (I) return-1sg late to home. [I] a-lot work-1(/3)sg

Well-Formedness Conditions on Antecedent-Independent Null Subjects | 207

In (25) and (26), the subject obviation strategy prohibits that the null subject in the embedded clause is interpreted in coreference with the matrix subject. Only when the matrix subject has 3rd person features, cf. (27), fossi is ambiguous (because the obviation strategy cannot apply) and the reference of the null subject must be determined by (external) contextual factors (e.g. a topical antecedent).45 (27) Pensava Ø fossi malato. Thought-3sg [I/you-sg] were-1/2sg sick ‘S/he thought I/you were sick.’ Let us now look at German in more detail. Although German displays a relatively rich inflectional paradigm, verbal forms display a lot of syncretisms, cf. the following table (syncretisms between the 1st and 3rd person singular/plural are marked bold, and syncretisms between the 3rd person singular and the 2nd person plural are marked by bold italics): (28)

1 sg

German verbal inflectional paradigms, cf. the indicative present/preterite active of the verbs können (can); sagen (to say); kommen (to come); tragen (to wear); sein (to be)

Praeteritopraesentia

Weak conjugation

Strong conjugation

Strongumlauting conjugation46

Suppletive conjugation

kann/konnte

sage/sagte

komme/kam

trage/trug

bin/war

kommst/kamst

trägst/trugst

bist/warst

kommt/kam

trägt/trug

ist/war

2 sg

kannst/konntest sagst/sagtest

3 sg

kann/konnte

1 pl

können/konnten sagen/sagten

2 pl

könnt/konntet

3 pl

können/konnten sagen/sagten

sagt/sagte sagt/sagtet

kommen/kamen tragen/trugen

sind/waren

kommt/kamt

seid/wart

tragt/trugt

kommen/kamen tragen/trugen

sind/waren

From (28) we can infer that (according to the presence of syncretisms) three conjugation classes have to be (systematically47) distinguished, cf. table (29): || 45 Thanks to Irene Caloi and Cecilia Poletto for the data and some discussion on them. 46 Note that the strong umlauting conjugation also includes verbs without orthographical Umlaut (as ä, ö or ü), e.g. sehen, to see (cf. er sieht (he sees), ihr seht (you-pl see). 47 See Müller (2005a,b) for a discussion whether the mixed number syncretism (between the 3rd person singular and the 2nd person plural) is systematic or not. However, the OBD licensing principle is independent of the fact whether the 3sg/2pl syncretism is a systematic syncretism or not. As shown above, e.g. by the Polish and Spanish minimal pairs (and as will be shown

208 | Null Subjects at the Syntax-Pragmatics Interface

(i) praeterito-praesentia (ii) the strong/weak conjugation (iii) the strong-umlauting conjugation and suppletive conjugation of auxiliary verbs (29)

Syncretisms within German inflectional paradigms (suppletive sein (to be) is left out)

Syncretism

Number

Conjugation class

Tense

1 sg / 3 sg

singular

all conjugation classes

preterite

praeterito-praesentia

present

1 pl / 3 pl

plural

all conjugation classes

all tenses

3 sg / 2 pl

mixed

weak/strong conjugation classes

present

Now consider an application of the ‘Coordinated Antecedents Test’ in German where the (underlined) verb forms are syncretic (i.e. parallel to the Spanish example in (18) and the (Text Message) Polish example in (20)) and note that in contrast to these two pro-drop languages in German a 1st(/2nd)48 person null subject is licensed, cf. (30a): (30) [[Hans] und [ich]] haben den Film schon gesehen. Hans and I have-1pl the movie already seen a. Ø Will deshalb lieber zuhause bleiben. 1st person singular [I] want-1(/3)sg therefore rather at-home stay b. * Ø Will deshalb lieber zuhause bleiben. 3rd person singular [He] want-(1/)3sg therefore rather at-home stay

|| subsequently), null subjects are “equally well” blocked by systematic as well as by nonsystematic (occasional) syncretisms. Thus, I assume that null subject licensing is not sensitive to how syncretisms arise/how they are “composed”. 48 Note that this also holds for 2nd person singular null subjects, cf. the following equivalent example that contains a form (lässt) which is syncretic between a 3rd person singular and a 2nd person singular: (i) [[Hans] und [du]] (ihr) scheint ja das Risiko zu lieben. Hans and you-sg (you-pl) seem-2pl PRT the risk to love a. Ø Lässt ständig die Balkontüre offen. 2nd person singular [You-sg] leave-2sg(/3sg) always the balcony-door open b. * Ø Lässt ständig die Balkontüre offen. 3rd person singular [He] leave-(2sg/)3sg always the balcony-door open

Well-Formedness Conditions on Antecedent-Independent Null Subjects | 209

I.e., the CAT yields different outputs for German and pro-drop languages. Whereas null subjects in pro-drop languages are not licensed when the verbal form is syncretic (between 1st and 3rd person singular, cf. (18) and (20)), in German this constitutes no problem: the null subject in (30) can be interpreted as referring to the 1st person singular49 (for convenience (18) and (20) are repeated below): (18) Juan y yo llegábamos a casa. *Ø Tenía las llaves. Juan and I arrived-1pl to home. [I/He] had-1/3sg the keys. ‘Juan and I were arriving home. I/he had the keys.’ (20) * Gdy Tomek i ja wrocimy z Czestochowy, Ø zaraz odpisze. When T. and I return-1pl from Cz., [I/he] at once write-back-1/3sg In coherence with the data contrasts (18)/(20) vs. (30) I will attribute the null subject licensing differences between German and pro-drop languages to the presence of topic drop. Whereas German has topic drop, pro-drop languages do not have topic drop, cf. Polish (31), or only a very restricted version of (verbatim) topic drop – maybe in terms of (32), which is from Kowaluk (1999). Moreover, it is not clear in which syntactic positions such NP/DP omissions take place (not necessarily in sentence initial position): (31) A:

Tam stoi Janek. There stands Janek-NOM B: * Nie widziałam _ od lat. Not saw-1sg-fem [ACC] for years

(32) A: B:

Czy podlałaś moją palmę? If water-2sg-fem my palm-ACC (= ‘Have you watered my palm tree?’) Podlałam _. Water-1sg-fem [ACC] (= ‘I have watered your palm tree.’)

I.e., I hypothesise that null subject distribution differences50 between German and pro-drop languages can be attributed to the fact that in German 1st/2nd person null subjects are identified and licensed via discrete verbal inflectional

|| 49 The 3rd person null subject is, as expected, not licensed, cf. the discussion of the German datum in (8b). 50 Here I disregard the positional restriction of OBD (cf. the restriction to the prefield).

210 | Null Subjects at the Syntax-Pragmatics Interface

endings, whereas 3rd person null subjects in German are identified and licensed via an antecedent (i.e. via topic drop). On the contrary, pro-drop languages only need a (topical) antecedent for the identification but not for the licensing of 3rd person null subjects; 1st/2nd person null subjects are identified and licensed via discrete verbal inflectional endings – as in German. However, in contrast to German, in pro-drop languages not only 1st/2nd person null subjects but also 3rd person null subjects are licensed by discrete verbal inflectional endings. I.e., in pro-drop languages 3rd person null subjects are not exempted from an inflection-dependent licensing mechanism – which, however, is the case in German (cf. topic drop) where, as a consequence, syncretisms between the 1st/2nd and 3rd persons do not matter. The different licensing/identification mechanisms of German and pro-drop languages are illustrated/summarised in the following tables, (33/34): (33)

Null subject licensing and identification in German (OBD and topic drop)

Number

Person

Discourse Domain/Licensing Mechanism

Construction Type

SINGULAR

1/2 sg

Antecedent-independent / out of the blue, by non-syncretic inflection

OBD

3 sg

Antecedent-dependent

Topic Drop

1/2 pl

Antecedent-independent / out of the blue, by non-syncretic inflection

OBD

3 pl

Antecedent-dependent

Topic Drop

PLURAL

(34)

Null subject licensing and identification in pro-drop languages (pro-drop + antecedentdependent identification of 3rd person null subjects)

Number

Person

Discourse Domain/Licensing Mechanism

Construction Type

SINGULAR

1/2 sg

Antecedent-independent / out of the blue, by non-syncretic inflection

3 sg

Non-syncretic inflection + antecedent-dependent identification

1/2 pl

Antecedent-independent / out of the blue, by non-syncretic inflection

pro-drop (for all person and number combinations) . . . . .

PLURAL

Well-Formedness Conditions on Antecedent-Independent Null Subjects | 211

Number

Person

Discourse Domain/Licensing Mechanism

3 pl

Non-syncretic inflection + antecedent-dependent identification

Construction Type . . .

Note that when compared to a “simple” out of the blue context, the CAT provides a further diagnostic for the involved identification and licensing mechanisms of null subjects (not only) in German and pro-drop languages, respectively, which cannot be provided by an out of the blue context. Consider therefore the minimal pair in (35)/(36) where it is shown that in ‘pure’ out of the blue contexts 1st person singular null subjects accompanied by syncretic verbal inflections are licensed in Spanish51 as well as in German: (35)

Ø Tenía mucho que hacer. [I] had-1(/3)sg lot to do

(36)

Ø Hatte viel zu tun. [I] had-1(/3)sg lot to do

I suppose that in an out of the blue context Spanish compensates the syncretic verb form plus the lack of an antecedent by choosing a default (1st person) interpretation, cf. (35) – such a mechanism could also be assumed to be active in German, cf. (36). However, a comparison of German and Spanish CAT data that contain a syncretic verb form shows that this cannot be the relevant null subject licensing mechanism for German, because – in contrast to the parallel Spanish example in (37) – the 1st person null subject in German is licensed, despite the lack of a default antecedent, cf. (38): (37)

Juan y yo llegamos tarde. *Ø Tenía mucho que hacer. Juan and I came-1pl late. [I/he] had-1/3sg lot to do

(38)

Hans und ich kamen spät. Ø Hatte viel zu tun. Hans and I came-1pl late. [I] had-1(/3)sg a-lot to do

|| 51 Text Message Polish behaves analogously, cf. (i): (i) Ø Pracuje caly dzien. [I] work-1(/3)sg whole day

212 | Null Subjects at the Syntax-Pragmatics Interface

Note that the non-presence of a default antecedent can only be manifested under the CAT, because the CAT provides a possible however incompatible antecedent and thus makes a default antecedent choice impossible – which is exactly the reason why the Spanish null subject (37) remains uninterpretable,52 whereas the Spanish null subject in (35) can be interpreted. We can thus state that 1st/2nd person null subjects in German are not licensed by a default antecedent but probably by discrete inflectional endings. To exclude that speaker/hearer features are the accurate null subject licensers I will provide further independent evidence. But before, I will suggest that the following hypotheses, cf. (39a) and (39’a), (still) constitute variants: (39) Null subject licensing/identification hypothesis for German (variant 1) (a) 1st/2nd person null subjects are well-formed out of the blue because they can be identified/licensed by discrete verbal inflectional endings. (b) 3rd person null subjects are identified and licensed in dependence of an external antecedent (topic drop) – as a consequence it does not matter whether 3rd person verbal inflectional endings are syncretic with 1st/2nd person verbal inflectional endings. (39’) Null subject licensing/identification hypothesis for German (variant 2) (a) 1st/2nd person null subjects are well-formed out of the blue because they can be identified/licensed by speaker/hearer features. (b) 3rd person null subjects are identified and licensed in dependence of an external antecedent (topic drop) – as a consequence it does not matter whether 3rd person verbal inflectional endings are syncretic with 1st/2nd person verbal inflectional endings. In the following I will shortly point out why (39) has to be preferred over (39’), i.e., why OBD is not licensed by speaker/hearer features but by distinct verbal inflectional endings. The data set that proves (39) to be preferred over (39’) are /s/-stem verbs as rasen (to race), küssen (to kiss), hassen (to hate), motzen (to

|| 52 Null subjects remain uninterpretable, when they cannot be licensed/identified by the means mentioned in (i)–(iii), cf. Cole’s (2009) Maximality Principle: (i) discrete verbal inflectional endings | speaker/hearer features (ii) a compatible antecedent (iii) a default (1st person) antecedent The CAT ensures that only (i) can apply.

Well-Formedness Conditions on Antecedent-Independent Null Subjects | 213

grumble) etc. The table in (40) illustrates the additional syncretisms of /s/-stem verbs (in contrast to a non-/s/-stem ending verb as machen) by bold italics: (40)

Inflectional paradigms, indicative present/preterite active, German machen (to make); motzen (to grumble); lassen (to let); heißen (to be called)

Weak conjugation, no /s/-stem

Weak conjugation, /s/-stem [+syncretismpres]

Strong umlauting conjugation, /s/-stem [+syncretismpret]

Strong conjugation, /s/-stem [+syncretismpres+pret]

mache/machte

motze/motzte

lasse/ließ

heiße/hieß

2 sg machst/machtest

motzt/motztest

lässt/ließ(es)t

heißt/hieß(es)t

3 sg macht/machte

1 sg

motzt/motzte

lässt/ließ

heißt/hieß

1 pl

machen/machten

motzen/motzten

lassen/ließen

heißen/hießen

2 pl

macht/machtet

motzt/motztet

lasst/ließ(e)t

heißt/hieß(e)t

3 pl

machen/machten

motzen/motzten

lassen/ließen

heißen/hießen

Now consider the minimal pairs in (41)–(43) and note that whenever a 2nd person singular verbal ending coincides with a 2nd person plural verbal ending, a non-ambiguous out of the blue interpretation of the respective null subject(s) is impossible: (41) a.

* Ø Kotzt mich total an! [You-sg/You-pl] vomit-2sg(/3sg)/2pl me totally on (‘... can’t stand’) b. Ø Machst mich total an! [You-sg] turn-2sg me totally on

(42) a.

* Ø Hießt mich noch gestern einen [You-sg]/[You-pl] dubbed-2sg/2pl me still yesterday a Kostverächter. non-bon-vivant b. Ø Hießest / Hießet mich noch gestern einen [You-sg] dubbed-2sg/[You-pl] dubbed-2pl me still yesterday a Kostverächter. 53 non-bon-vivant

|| 53 Note that archaic forms as hießest and hießet belong to more formal (= non-OBD) registers. Nevertheless, the contrast between the a. and b. version is clearly evident.

214 | Null Subjects at the Syntax-Pragmatics Interface

(43) a.

* Ø Löst mal wieder keine der [You-sg/You-pl] solve-2sg(/3sg)/2pl PRT again none of-the Algebra-Aufgaben. algebra-exercises b. Ø Löse / Lösen mal wieder keine der Algebra-Aufgaben. [I] solve-1sg/[We] solve-1(/3)pl PRT again none of the a.-exercises

The reason why the null subjects in (41a)/(42a)/(43a) are uninterpretable is that 2nd person singular and 2nd person plural belong to the same discourse domain, namely the antecedent-independent/out of the blue null subject licensing domain.54 I.e., we have a situation partly analogous to pro-drop languages displaying syncretisms in particular paradigms with the same effect, namely that null subjects that are “usually” licensed/identified by discrete verbal inflections can neither be licensed nor identified because they are affected by (accidental) syncretisms (cf. the parallel CAT data for Spanish (18) and Polish (20)). As “regular” null subject licensing is suspended by the presence of additional syncretisms, cf. (41a)/(42a)/(43a), we can conclude that it is the discreteness of verbal inflectional endings that constrains null subject licensing in German. The assumption that speaker/hearer features license null subjects in German could not account for the ungrammaticality/“non-interpretability” of the null subjects in (41a)/(42a)/(43a). Rather, under a speaker/hearer feature licensing the null subjects in (41a)/(42a)/(43a) would be expected to be wellformed/interpretable. With respect to /s/-stem verbs not only the out of the blue data in (41a)/(42a)/(43a) but also the (now following) application of the CAT shows that the well-formedness of OBD null subjects is constrained by discrete verbal inflections (and does not depend on the presence of grammaticalised speaker/ hearer features), cf. (44): (44) [[Du Psychologe] und [ihr Linguisten]] habt die You-sg psychologist and you-pl linguists have-2pl the Versuchspersonen echt ausgenutzt. experimental-subjects really exploited a. (*) Ø Ließt sie sogar am [You-sg]/[You-pl]/[You-sg+you-pl] make-2sg/2pl them even on

|| 54 Note that in German the forms of the 1st person singular and the 1st person plural never coincide. Thus, the following argument cannot be independently checked (e.g. with 1st person singular vs. 1st person plural data).

Well-Formedness Conditions on Antecedent-Independent Null Subjects | 215

b. c.

Sonntag kommen. (/s/-stem verb) Sunday come Ø Hast sie sogar am Sonntag kommen lassen. (no /s/-stem verb) [You-sg] have-2sg them even on Sunday come make Ø Habt sie sogar am Sonntag kommen lassen. (no /s/-stem verb) [You-pl]/[You-sg+you-pl] have-2pl them even on Sunday come make

As expected, the null subject in (44a) receives no interpretation, because the verb form is syncretic between a 2nd person singular/plural. However, with respect to (44a) it has to be noted that, of course, an interpretation where the null subject refers to the whole plurality, i.e. to Du Psychologe und ihr Linguisten is possible55 (hence, I have put the star ‘*’ in brackets – note that such an interpretation could also arise from topic drop). Only when the forms of the 2nd person singular and 2nd person plural do not coincide, the null subject is interpreted unambiguously, cf. (44b) and (44c) – (44c) modulo the different groups denoted by the 2nd person plural form.56 The interpretative options of null subjects under /s/-stem verb forms (displaying form identity between the 2nd person singular and the 2nd person plural) can now be contrasted with the interpretative options of null subjects under verb forms that are syncretic between the 1st and the 3rd person singular, cf. the German CAT data in (30) and (38) above, repeated here for convenience: (30) [[Hans] und [ich]] haben den Film schon gesehen. Hans and I have-1pl the movie already seen a. Ø Will deshalb lieber zuhause bleiben. 1st person singular [I] want-1(/3)sg therefore rather at-home stay b. * Ø Will deshalb lieber zuhause bleiben. 3rd person singular [He] want-(1/)3sg therefore rather at-home stay (38) Hans und ich kamen spät. Ø Hatte viel zu tun. Hans and I came-1pl late. [I] had-1(/3)sg a-lot to do

|| 55 Cf. Cole’s (2009) maximality principle that predicts reference to an external antecedent when inflection is not discrete enough. 56 Note that this is exactly the reason why it makes no sense to apply the CAT to coordinated items marked for the 1st person plural and the 3rd person plural, because one cannot exclude that the (for these persons) always syncretic verb form refers to the plurality of the 1st person plural and the 3rd person plural, parallel to (44c) where it cannot be determined whether the null subject refers to Ihr Linguisten or to Du Psychologe und ihr Linguisten.

216 | Null Subjects at the Syntax-Pragmatics Interface

In (30) and (38) verbal syncretisms between the 1st and the 3rd person (singular) do not block the presence of 1st person singular null subjects. By contrast, null subjects are blocked when syncretisms occur between the 2nd person singular and 2nd person plural, cf. (41a)/(42a)/(43a) and (44a), because 2nd person singular and 2nd person plural null subjects belong to the same null subject licensing domain (cf. out of the blue). Consider therefore also example (45) where a null subject is well-formed when the verb form (here: lässt) is syncretic between a 2nd person singular and a 3rd person singular, but not well-formed when the verb form (here: motzt) is syncretic between a 2nd person singular, a 3rd person singular and a 2nd person plural: (45) [[Hans] und [du]] (ihr) scheint ja das Risiko zu lieben. Hans and you-sg (you-pl) seem-2pl PRT the risk to love a. Ø Lässt ständig die Balkontüre offen. (dissoluble syncretism) [You-sg] leave-2sg(/3sg) always the balcony-door open b. * Ø Motzt ständig rum wegen Langeweile. (indissoluble syncretism) [ ] grumble-2sg(/3sg)/2pl PRT because-of boredom Thus, the null subject blocking in /s/-stem verbs is in parts analogous to the null subject blocking in certain verbal inflectional paradigms of pro-drop languages. With respect to the differences between German and pro-drop languages we can state that 1st/2nd person OBD null subjects in German are well-formed out of the blue, independently of an antecedent, whereas 3rd person null subjects in German are licensed and identified in dependence from an antecedent – hence, syncretisms that occur between (antecedent-independent) 1st/2nd and (antecedent-dependent) 3rd person forms do not matter. However, syncretisms between the 2nd person singular and the 2nd person plural do play a role, because these persons both belong to the antecedent-independent (null subject) licensing (and identification) domain. On the contrary, in pro-drop languages syncretisms between the 1st/2nd and 3rd person forms matter. Although 3rd person null subjects are exempted from an inflection-dependent (clause-internal) identification (as they are identified via a (clause-external) antecedent), they are (as 1st/2nd person null subjects) not exempted from an inflection-dependent (clauseinternal) licensing. As a consequence of the fact that in pro-drop languages 1st, 2nd and 3rd persons are subject to the same null subject licensing mechanism (via discrete verbal inflectional endings), syncretisms that occur between 1st/2nd person and 3rd person forms are crucial, cf. (18), (37), (20), as long as there is no

Well-Formedness Conditions on Antecedent-Independent Null Subjects | 217

accessible discourse antecedent which could resolve that ambiguity (see Cole’s Maximality Principle in (24) above). (18) Juan y yo llegábamos a casa. *Ø Tenía las llaves. John and I arrived-1pl to home. [I/He] had-1/3sg the keys. ‘John and I were arriving home. I/he had the keys.’ (37) Juan y yo llegamos tarde. *Ø Tenía mucho que hacer. Juan and I came-1pl late. [I/he] had-1/3sg lot to do (20) * Gdy Tomek i ja wrocimy z Czestochowy, Ø zaraz odpisze. When T. and I return-1pl from Cz., [I/he] at once write-back-1/3sg Given the above contrasts and /s/-stem verbs as additional and independent evidence for preferring (39) over (39’), we can reformulate (39) in a more abstract way,57 cf. (46): (46) Licensing/identification of 1st/2nd vs. 3rd person null subjects (in German) The fact that null subjects are well-formed despite the fact that particular inflectional endings within verbal paradigms are syncretic with other ones does not matter as long as the respective person/number combinations bearing these forms are subject to different licensing conditions / ‘licensing domains/paradigms’, namely (i) antecedent-independent (inflectionbased) subject omission and (ii) antecedent-dependent subject omission. Note that hypothesis (46) is also able to capture ungrammatical/uninterpretable topic drop data: when two elements belong to the same licensing domain (here: antecedent-dependent), they are (consequently) subject to the same null subject/object licensing condition and hence not licensed in situations as the following ones, cf. (47)–(48): (47) A:

Der Hans ist eingeladen und der Otto auch. The Hans-NOM is invited and the Otto-NOM as-well B: * _ Hat aber keine Zeit. [NOM] has however no time

|| 57 For a null subject licensing principle that is also related to discourse domains (differentiating between anaphoric and non-anaphoric content licensing, but with different implications) see Cabredo Hofherr (2006).

218 | Null Subjects at the Syntax-Pragmatics Interface

(48) A:

Die Eltern haben den Hans und den Otto eingeladen. The parents-NOM have the Hans-ACC and the Otto-ACC invited B: * _ Hat aber keine Zeit. [NOM] has however no time

4.3 More on the Identification/Licensing of Null Subjects (in German) In the following I will present more empirical evidence for the supposed licensing/identification condition(s) on null subjects. After that I will outline why OBD is restricted to the prefield and discuss non-prefield occurrences of null subjects in German (dialects) as well. I will close this chapter by pointing out that Colloquial (High) German may count as a partial pro-drop language.

4.3.1 Independent Evidence for the Assumed Identification/Licensing Conditions on Null Subjects In this section I will give additional evidence for the fact that the hypotheses (39) and (46) (numeration from the preceding chapter) make the correct predictions with respect to null subject licensing in other languages/German dialects as well. Instead of referring to (39) and (46) I will refer to (1) (which equals (39)), and to (2) (which is a more abstract formulation of (46)): (1)

Null subject licensing/identification hypothesis for German (a) 1st/2nd null subjects are well-formed out of the blue because they can be identified/licensed by discrete verbal inflectional endings. (b) 3rd person null subjects are identified and licensed in dependence of an external antecedent (cf. topic drop) – as a consequence it does not matter whether 3rd person verbal inflectional endings are syncretic with 1st/2nd person verbal inflectional endings.

(2)

Licensing/identification of 1st/2nd vs. 3rd person null subjects (in general) The fact that null subjects are well-formed despite the fact that particular inflectional endings within verbal paradigms are syncretic with other ones does not matter as long as the respective person/number combinations bearing these forms are subject to different licensing conditions / ‘licensing domains/paradigms’, namely (i) antecedent-independent (inflection-

More on the Identification/Licensing of Null Subjects (in German) | 219

based or speaker/hearer feature based) subject omission and (ii) antecedent-dependent subject omission. In order to find out whether the principle in (2) can capture further null subject occurrences, consider (e.g.) Modern Greek: In Modern Greek impersonal and referential 3rd person null pronouns exist simultaneously: with respect to this fact Lavidas & Papangeli (2007:11) note that “transitive [and intransitive58] verbs may be ambiguous between arbitrary and non-arbitrary interpretation of the subject […] It is only contextual considerations, pragmatic or extra-linguistic context that may decide between the two readings”. They provide minimal pairs similar to those in (3) and (4): (3)

Εδώ χορεύουν συχνά (Οι φοιτητές) Edo xorevun sixna (i fitites) Here dance-3pl often (the students) ‘Here, one / people dance often’ (‘Here, students often dance’)59

(4)

Εδώ τρώνε πολλά μήλα (Οι φοιτητές) Edo trone pola mila (i fitites) Here eat-3pl many apples (the students) ‘Here, one / people eat many apples’ (‘Here, students eat many apples’)

As (3) and (4) show, Modern Greek (a pro-drop language) licenses referential and impersonal null pronouns by the same verb form, however, the occurrence of each of them, referential vs. impersonal null subjects, can be determined by the assumption of different discourse domains as null subject licensing areas, as formulated in (2). I.e., a Modern Greek 3rd person plural null subject receives an impersonal interpretation when the discourse does not provide an antecedent – then the null subject is licensed independently of an antecedent (out of the blue). On the other hand, it receives a definite (referential/thematic) inter-

|| 58 Note that Lavidas & Papangeli’s claim made for transitive verbs also holds for intransitive verbs. Thanks to Vasiliki Koukoulioti for pointing that out and discussing the data with me. 59 The presence of the bracketed NP the students serves to indicate that the agreeing verb form is not specialised for some impersonal/generic pronoun use. Moreover, it indicates that whenever the discourse provides the students as an antecedent, the null subjects in (3)/(4) refer to the students.

220 | Null Subjects at the Syntax-Pragmatics Interface

pretation when the discourse provides an antecedent – then the null subject is licensed by reference to its antecedent. Thus, principle (2) predicts that whenever two identical forms co-occur (due to syncretisms in the verbal inflectional paradigm), only one of these occurrences can act as an out of the blue null subject (whereas the other one has to be licensed via an external antecedent). Further evidence for the hypotheses in (1) and (2) comes from German dialects, cf. Swabian and Zurich-German. As the following table shows, in these (Southern German) dialects only singular verb forms are distinctly marked. Plural forms are identical for all persons: (5)

1 sg

Inflectional paradigms, indicative present/preterite60 active of different Swabian verbs könne (can)

sage (to say)

komme (to come)

trage (to wear)

koa/konnt

sei (to be)

sag/sagte

komm/kam

trag/trug

ben/war

2 sg koasch/ konntesch

sagsch/ sagtesch

kommsch/ kamsch

trägsch/ trugsch

bisch/ warsch

3 sg koa/konnt

sagt/sagte

kommt/kam

trägt/trug

isch/war

1 pl

könnet/konntets

saget/sagtet

kommet/kamet

traget/truget

sen/waret

2 pl

könnet/konntets

saget/sagtet

kommet/kamet

traget/truget

sen/waret

3 pl

könnet/konntets

saget/sagtet

kommet/kamet

traget/truget

sen/waret

According to hypothesis (1) we would predict that due to syncretisms within all conjugation classes in the plural number, OBD in Swabian should only be possible when related to a 1st and 2nd person singular. This is exactly what is borne out by the data in (6): (6) a.

Passet uff, Ø komm glei nüba! Take care, [I] come-1sg soon along

1st sg

|| 60 It is generally assumed in the literature that Oberdeutsch dialects do not have a preterite form, cf. Abraham & Conradi (2001); see also Frey (1975) who notes that a preterite form only occurs with the form war (I/he was). However, in modern uses of Swabian the preterite form does occur, cf. the following datum in (i): (i) oah hello pogo DU sagtesch doch ich soll ins solarium‼‼ oah hello pogo you-sg said PRT I should (go) to-the solarium (http://www.bym.de/forum/kosmetik/400358-solaaaaaarium.html; checked 11.02.2016)

More on the Identification/Licensing of Null Subjects (in German) | 221

b.

Ø Bisch wirklich a Liebe. [You-sg] are-2sg really a lovely-one-fem c. * Ø Kommt net. [He] comes-3sg not d. * Ø Gebet nix. [We] give-1/2/3pl nothing e. * Ø Kommet wohl immer z’ spät. [You-pl] come-1/2/3pl surely always too late f. * Ø Sen wieder net g’komme. [They] are-1/2/3pl again not come

2nd sg 3rd sg 1st pl 2nd pl 3rd pl

The same distribution of null subjects correlated with the same occurrence of syncretisms within inflectional paradigms can be observed in Zurich-German, which is an Alemannic dialect as well (thanks for the data to Martin Salzmann (and Peter Gallmann)), see the table in (7) and the data in (8): (7)

Inflectional paradigms, indicative present active; Alemannic choo (can); sage (to say); komme (to come); träge (to wear); sii (to be) choo (can)

sage (to say)

komme (to come)

1 sg

cha

säge

2 sg

chasch

säisch

träge (to wear)

sii (to be)

chume

träge

bin

chunnsch

träisch

bisch isch

3 sg

cha

säit

chunnt

träit

1 pl

chönd

säged

chömed

träged

sind

2 pl

chönd

säged

chömed

träged

sind

3 pl

chönd

säged

chömed

träged

sind

(8) a.

Passed uuf, Ø chume glii übere. Take care, [I] come-1sg soon along b. Ø Bisch würkli e Liebi. [You-sg] are-2sg really a lovely-one-fem c. * Ø Chunnt nöd. [He] comes-3sg not d. * Ø Gänd nüüt. [We] give-1/2/3pl nothing e. * Ø Chömed immer z spaat. [You-pl] come-1/2/3pl always too late

1st sg 2nd sg 3rd sg 1st pl 2nd pl

222 | Null Subjects at the Syntax-Pragmatics Interface

f.

* Ø Sind wider nöd choo. [They] are-1/2/3pl again not come

3rd pl

Thus, the data in (6)/(8) show that prefield null subjects are dependent on the presence of discrete verbal inflections not only in (High) German but also in (two loosely related) German dialects. Note, however, that all occurrences of OBD which are blocked due to syncretisms can be topic dropped, cf. (9): (9) A: B:

Warum isch d’r Hans denn so sauer auf ons? Why is the Hans so angry on (= with) us-ACC? Ha, _ kommet halt immer z’spät. INTERJ,61 [you-pl-NOM] come-(1/)2(/3)pl PRT always too late

I will conclude by stating that the principles formulated in (1) and (2) make the correct predictions with regard to the well-formedness of out of the blue 1st/2nd person null subjects (as well as with regard to not out of the blue licensed 3rd person null subjects) in the German prefield. Moreover, I hope to have pointed out the circumstances under which syncretisms between particular person/number combinations become (ir-)relevant – not only in German, but in a couple of (unrelated) languages. We can summarise this particular finding by saying that syncretic forms can license null subjects when occurring within different (complementary) discourse domains (antecedent-dependent vs. antecedent-independent (= out of the blue)).

4.3.2 Positional Licensing of ‘Out of the Blue Drop’ (OBD) and the Empty Category Associated with OBD In the following I will try to indicate why German OBD is restricted to the prefield and with which empty category it is associated. After that I will show how OBD in embedded V2 clauses differs from OBD in V2 main clauses. With respect to the first question, I will assume that German OBD null subjects are confined to the prefield, because they can only be licensed under a Spec-Head configuration (see also Cardinaletti 1990,62 Speas 2006), cf. (10):

|| 61 INTERJ = interjection 62 Cardinaletti (1990:81) also assumes that null subjects in the German prefield are represented by pro. However, she notes the following: “[ ] pro in (25) [= pro habe es gestern gekauft (I have it yesterday bought), ET] differs from the null subject found in Italian in that the recovery

More on the Identification/Licensing of Null Subjects (in German) | 223

(10) OBD ‘positional’ licensing OBD is licensed ‘locally’, via a Spec-Head configuration, i.e. the finite verb in C° carries the person/number specifications of the null subject (located in Spec-CP) and stands in an Agree relation with it. ‘Local’ licensing of OBD shows up in OBD’s dependency on distinct verbal inflectional endings realised within individual conjugation classes/paradigms: the morpho-syntactic OBD licensing rule does not extend systematically to the whole paradigm/all conjugation classes, cf. /s/-stem verbs, dialects. Let me also shortly point out the relevant reasons (ex negativo) why I assume that OBD is represented by ‘little’ pro, cf. (11): (11) The empty category resulting from OBD cannot be represented… a. by PRO, as the well-formedness of OBD is not dependent on a controlling argument (except one assumes finite PRO (cf. Landau 2004) for OBD in embedded V2 clauses63), b. by a trace, t, as we would have to assume an empty category which ccommands that trace (however, there is no evidence for such an empty category), c. by a bound operator, as the well-formedness of OBD is contextually independent. Thus, on the basis of (10) and (11), I will claim that (12) holds: (12) Kind of category OBD should be associated with (OBD) null subjects in the German prefield are represented by pro, which constitutes a silent but genuine pronoun.

4.3.3 OBD in Embedded V2 Clauses Following Jaensch (2005:98), contra Cardinaletti (1990) und Rizzi (1994), topic drop is not restricted to V2 main clauses but can also occur in Spec-CP of embedded V2 clauses, (13):64

|| of its feature content does not depend on the agreement specification of the verb, but on the linguistic or pragmatic context.” 63 Cases of embedded OBD will be discussed in the following section.

224 | Null Subjects at the Syntax-Pragmatics Interface

(13) A: B:

Will Peter wirklich das neue Auto kaufen? Wants Peter really the new car-ACC buy? Ich glaube, _ hat er schon gekauft. I think, [ACC] has he already bought

The occurrence of OBD in embedded V2 clauses can also be proven by attested (corpus) examples, as presented in (14): (14) a. b. c.

[Ø Suche guten Fernseher , Ø hoffe] Ø könnt mir helfen 65 [I am] looking for a good TV set, [I] hope-1sg [you-pl] can-2pl me help Ø Glaub, Ø war sogar erst in der 3.! 66 [Schwangerschaftswoche] [I] Think, [I] was-1(/3) even only in the third [week of pregnancy] o gott, Ich glaub Ø bin verknallt.67 Oh God, I think [I] am in-love

However, the data in (15) show that OBD in V2 main clauses has to be kept apart from OBD in embedded V2 clauses: Whereas verbal syncretisms between a 1st person and a 3rd person have no influence on null subject interpretation in V2 main clauses, the existence of 1st/3rd person syncretisms within the verbal inflectional paradigm makes null subject interpretation in embedded V2 clauses dependent on the form of the matrix subject, cf. (15a) vs. (15b/c), and (15d) vs. (15e/f): (15) a. b.

Ø War gestern zu lange auf. (1st sg / no 3rd sg) [I] was-1(/3)sg yesterday too long awake Ich glaub, Ø war gestern zu lange auf. (1st sg / no 3rd sg) I think, [I] was-1(/3)sg yesterday too long awake

|| 64 Note that topic drop can also apply to arguments moved across complementisers (contra Cardinaletti 1990:77), cf. Werner Frey and Andreas Pankau (p.c.): (i) A: Was ist mit Hans? What is with Hans-DAT (= ‘What about Hans?’) B: _ Glaub ich nicht, dass ich [ ] nochmal treffen möchte. [ACC] think I not, that I [ ] once-again to-meet want. 65 http://gutefrage.net/frage/suche-guten-fernseher--hoffe-koennt-mir-helfen (checked: 17.05.2014) 66 http://rund-ums-baby.de/schwanger-wernoch/Glaub-war-sogar-erst-in-der-3-g-Gesehenhat-man-nix-o-T_238622.htm (checked: 17.05.2014) 67 Personally (received) text message, 4.11.2010.

More on the Identification/Licensing of Null Subjects (in German) | 225

c. d. e. f.

Hans glaubt, Ø war gestern zu lange auf. (3rd sg / no 1st sg) Hans thinks, [he] was-(1/)3sg yesterday too long awake Ø Waren gestern zu lange auf. (1st pl / no 3rd pl) [We] were-1(/3)pl yesterday too long awake Wir glauben, Ø waren gestern zu lange auf. (1st pl / no 3rd pl) We think, [we] were-1(/3)pl yesterday too long awake Die Eltern glauben, Ø waren gestern zu lange auf. (3rd pl / no 1st pl) The parents think, [they] were-(1/)3pl yesterday too long awake

Note that coreference between matrix subject and embedded null subject is obligatory when agreement features of the embedded verb match the person/number specifications of the matrix subject, cf. (15b/c/e/f).68 The only way to circumvent a coreferential interpretation is by imposing a matching context, e.g. with respect to (15b): Was macht der Hans? (What makes the Hans?) – then the embedded null subject is interpreted as referring to Hans.69 Similarly, under a context where the speaker is designated as an aboutness topic, the null subjects in (15c) and (15f) can be interpreted with respect to the speaker ((15f) with respect to the speaker and the group she/he belongs to). Only 2nd person singular null subjects can always be interpreted independently of the matrix subject as 2nd person singular inflectional endings are never syncretic within all conjugation classes, cf. (16a) – except for verbs whose stem ends with an /s/, cf. (16b). In the latter case, the reference of an OBD null subject can be resolved via a matching antecedent (if there is one): (16) a. b.

Hans glaubt, Ø kommst immer zu spät. Hans thinks, [you-sg] come-2sg always too late Hans glaubt, Ø löst jede Aufgabe Hans thinks, [you-sg/he/you-pl] solve-(2sg/)3sg(/2pl) every task im Handumdrehen. in-the blink-of-an-eye

Of course, OBD null subjects in embedded clauses remain uninterpretable, when the matrix subject cannot resolve the syncretisms that occur within certain verbal paradigms, cf. löst in (17a), which is parallel to OBD in matrix

|| 68 As mentioned above, one could account for such instances of OBD by Landau’s (2004) finite PRO. 69 Such a case of null subject identification is parallel to topic drop cases where an external antecedent sweeps away all alternative gap interpretations, see footnote 4 in chapter 3.1.

226 | Null Subjects at the Syntax-Pragmatics Interface

clauses, cf. the data (41a/42a/43a) in 4.2.1. However, as in (16b), the otherwise crucial syncretisms can be resolved when a matching antecedent is able to identify the null subject reference, cf. (17b/c/d): (17) a. * Ich glaube, Ø löst jede Aufgabe. I think, [you-sg/he/you-pl] solve-2sg/3sg/2pl every task b. Du glaubst, Ø löst jede Aufgabe. You-sg think, [you-sg] solve-2sg(/3sg/2pl) every task c. Hans glaubt, Ø löst jede Aufgabe. Hans thinks, [he] solve-(2sg/)3sg(/2pl) every task d. Ihr glaubt, Ø löst jede Aufgabe. You-pl think, [you-pl] solve-(2sg/3sg/)2pl every task Summarising the behaviour of main clause OBD and embedded clause OBD so far, we can state (18): (18) V2-OBD licensing/identification succeeds according to: (i) The presence of discrete verbal inflections at the finite verb, as e.g. in (16a). (ii) The presence of a matrix subject which matches the verbal inflectional endings of the embedded verb, being able to resolve (indissoluble) syncretisms within (specific) conjugation classes via a coreferential interpretation, as e.g. in (17b/c/d). (iii) The presence of an external and matching (topic drop) antecedent (which is not the matrix subject), as suggested with respect to example (15b). As in main clauses, the well-formedness of OBD null subjects in embedded V2 clauses depends on the presence of (i) a Spec-Head configuration and (ii) distinct inflectional morphology displayed at the finite verb that is located in C°. Note that some occasional syncretisms, namely those in the context of /s/stem verbs, (which could not be resolved in main clauses) can now be resolved by an identifying (coreferent) antecedent in the matrix clause. I.e., in contrast to OBD null subjects in main clauses, OBD null subjects in embedded clauses are interpreted in dependence of a matrix subject whenever this is possible: On the one hand this widens the interpretational options of embedded V2 null subjects as e.g. in (17b/c/d) because now null subjects can be identified notwithstanding the lack of discrete verbal inflectional endings. On the other hand, some OBD instances get lost when OBD takes place in an embedded clause, because the

More on the Identification/Licensing of Null Subjects (in German) | 227

presence of a matching matrix subject antecedent prevents ‘pure’ out of the blue-driven instances cf. (15c/f). Thus, the above data confirm the hypotheses formulated in (1) and (2) (repeated here for convenience) which state that generally OBD null subjects in German are licensed via discrete verbal inflectional endings. (1)

Null subject licensing/identification hypothesis for German (a) 1st/2nd null subjects are well-formed out of the blue because they can be identified/licensed by discrete verbal inflectional endings. (b) 3rd person null subjects are identified and licensed in dependence of an external antecedent (cf. topic drop) – as a consequence it does not matter whether 3rd person verbal inflectional endings are syncretic with 1st/2nd person verbal inflectional endings.

(2)

Licensing/identification of 1st/2nd vs. 3rd person null subjects (in German) The fact that null subjects are well-formed despite the fact that particular inflectional endings within verbal paradigms are syncretic with other ones does not matter as long as the respective person/number combinations bearing these forms are subject to different licensing conditions / ‘licensing domains/paradigms’, namely (i) antecedent-independent (inflectionbased) 1st and 2nd person subject omission and (ii) antecedent-dependent 3rd person subject omission.

Besides, we can also state that in a situation where verbal inflectional endings are not discrete (enough) and reference to a discourse antecedent is possible, OBD null subjects are identified by reference to an antecedent70 – i.e., in this case Cole’s (2009) Maximality Principle applies, cf. (19): (19) Maximality Principle, cf. Cole (2009:559) [E]very language has its point of morphological maximality […] [R]ecovery of such [= null, ET] subjects is achieved first by reference to discrete agreement and, if this fails, by reference to an antecedent in context.

|| 70 I.e., insufficient licensing/identification via inflectional endings can be “rescued” by the presence of a matching discourse antecedent, as e.g. in (16b) which is parallel to pro-drop instances in Spanish inflectional paradigms that contain syncretic forms, cf. the Spanish data in (23) in 4.2.1.

228 | Null Subjects at the Syntax-Pragmatics Interface

4.3.4 Non-Prefield Null Subjects in German In this section I will discuss data as (20) and (21), already mentioned at the very beginning of this chapter: (20)

Was würdest Ø mir empfehlen? What would-2sg [you-sg] me recommend?

(21) (?) Was würdet Ø mir empfehlen? What would-2pl [you-pl] me recommend? According to the literature (since Ross 1982), it is generally assumed that in German null subjects (and objects) are only well-formed in the prefield. When no reduced (clitisised) pronoun is attached to the verb, occurrences in other positions are ungrammatical, cf. (22) and (23):71 (22)

Was *würde Ø / würdich dir empfehlen? What would-1(/3)sg [I] / would-1(/3)sg+cl1sg you-sg-DAT recommend?

(23)

Was *würden Ø / würdenwɐ dir empfehlen? What would-1(/3)pl [we] / would-1(/3)pl+cl1pl you-sg-DAT recommend?

A standard objection against the occurrence of a 2nd person singular null subject in (20) is the assumption that the verb possibly contains a clitic-like/(silent) incorporated d-pronoun that represents a shortened version of du (which is the 2nd person singular (nominative) subject pronoun) whose +/-presence in data as (20) is hard to detect because the 2nd person singular verb ending in German is – st, and so, the verbal ending and a 2nd person singular clitic-like element with the form –t(e) cannot be dissociated from each other on phonetic grounds (i.e. würdest could be considered as a shortened version of würdeste).72 Thus, if one assumes that (Colloquial) German also has clitics of the form –t(e) in its inventory, datum (20) constitutes ambiguous evidence for the presence of null subjects in other positions than the prefield.

|| 71 Note that würde and würden (would) are both syncretic between the 1st and the 3rd person, singular and plural, respectively. 72 However, Helmut Weiß (p.c.) claims that in German (dialects) only 2nd person singular verb forms that contain a clitic end with an –e (cf. würdeste); forms without an –e (cf. würdest) do not contain a clitic.

More on the Identification/Licensing of Null Subjects (in German) | 229

A way to exclude that 2nd person singular null subject occurrences as in (20) may involve a clitic is to look at 2nd person plural data as in (21) and to draw a conclusion by analogy (under the assumption that, syntactically, 2nd person singular and 2nd person plural data behave alike). Although data as (21) are very marginal, we can definitely be sure that (in contrast to 2nd person singular occurrences) no phonological clitisation process is involved. Regrettably, there are no DWDS findings of 2nd person plural null subjects that do not occur in the prefield. However, by conducting a google search73 I found a lot of occurrences of non-prefield 2nd person plural null subjects, so that the existence of the following examples cannot be ascribed to chance (in the sense of being accidental). Moreover, the contexts in which the data in (24) were uttered did not indicate that the respective speakers were low performers of German. Google search: (24) a. “Was würdet Ø mir empfehlen?” What would-2pl [you-pl] me recommend? b. “Was würdet Ø mir raten?” What would-2pl [you-pl] me advise? c. “Wie würdet Ø euch verhalten?” How would-2pl [you-pl] yourself-pl behave? d. “Was würdet Ø an meiner Stelle tun?” What would-2pl [you-pl] in my position do? e. “Was könnt Ø empfehlen?” What would-2pl [you-pl] recommend? f. “Was könntet Ø mir empfehlen?” What would-2pl [you-pl] me recommend? g. “Was habt Ø gemacht?” What have-2pl [you-pl] made? h. “Was habt Ø gesehen?” What have-2pl [you-pl] seen? i. “Wo habt Ø das her?” Where have-2pl [you-pl] that from? j. “Wann wart Ø das letzte Mal [...] ?” When were-2pl [you-pl] the last time? k. “Wann seid Ø losgefahren?” When are-2pl [you-pl] moved-off?

|| 73 I did the google search in July 2013/September 2015.

37 independent matches 22 independent matches 4 independent matches 2 independent matches 27 independent matches 15 independent matches 11 independent matches 5 independent matches 3 independent matches 2 independent matches 1 match

230 | Null Subjects at the Syntax-Pragmatics Interface

l.

“Das solltet Ø nicht tun.” That should-2pl [you-pl] not do

1 match

Within my google search I only found 2nd person plural null subjects occurring with modal and auxiliary verbs but none with lexical verbs (as e.g. Wann kommt vorbei? (When come-2pl [you-pl] along?)). Thus, we have to ask, what makes 2nd person plural modals and auxiliaries special (as to this particular person/number specification and in contrast to lexical verbs). Because of its possibly phonetically amalgamated form the 2nd person singular will (first) be left out of the discussion. As we already know from the previous sections, the well-formedness of null subjects occurring in the German prefield depends on the +/-presence of syncretisms within the verbal paradigm. Hence it suggests itself to investigate the pattern of syncretisms of modal and auxiliary verbs (in contrast to those of lexical verbs) when investigating null subjects which do not occur in the prefield. (25)

Inflectional paradigms, indicative present/preterite active, German; können (can), sagen (to say); kommen (to come); tragen (to wear); sein (to be)

Modals (Praeteritopraesentia)

Weak conjugation

Strong conjugation

Strong-umlauting conjugation

Suppletive conjugation

1 sg kann/konnte

sage/sagte

komme/kam

trage/trug

bin/war

2 sg kannst/konntest

sagst/sagtest

kommst/kamst

trägst/trugst

bist/warst

3 sg kann/konnte

sagt/sagte

kommt/kam

trägt/trug

ist/war

1 pl

können/konnten

sagen/sagten kommen/kamen tragen/trugen

sind/waren

2 pl

könnt/konntet

sagt/sagtet

seid/wart

3 pl

können/konnten

sagen/sagten kommen/kamen tragen/trugen

kommt/kamt

tragt/trugt

sind/waren

The empirical observation that can be inferred from the above table is that 2nd person plural (and singular) modal and auxiliary verbs are non-syncretic throughout all inflectional paradigms in which they occur, i.e., verbs of these paradigms lack the 3rd person singular/2nd person plural syncretism (note e.g. that in the past tense of verbs of the weak and strong conjugation this kind of syncretism does not occur as well, but it occurs in the present tense). Thus, a possible hypothesis could read as follows:

More on the Identification/Licensing of Null Subjects (in German) | 231

(26) Licensing of non-prefield null subjects in German Verb forms which are discrete throughout all tenses of their inflectional paradigms license null subjects not only in the prefield (i.e. independently of a Spec-Head configuration). However, according to (25)/(26) not only modal verbs and auxiliaries should license 2nd person plural null subjects in the middle field, but also verbs of the strong-umlauting conjugation (which do not display syncretisms between the 3rd person singular and the 2nd person plural throughout their whole conjugational paradigms as well) – Note that this is exactly what is borne out by the data in (27)–(30) that involve the verbs raten (to recommend) and vorschlagen (to suggest):74 (27) Guten Tag, Ich brauche Hilfe bei der Auswahl meines Bikes. Ich habe ein Budget von ca. 1400 € und hoffe dafür ein gutes gebrauchtes Bike zu bekommen. Zu was ratet Ø mir?75 | To what recommend-2pl [you-pl] me-DAT? [Will mir keinen Fehlgriff leisten und bräuchte ein Bike das schon etwas aushält, bin keine völliger Anfänger und habe mich mit meinem Hardtail schon ein bisschen experimentiert, konnte natürlich nicht allzuweit gehen, da es nicht dafür ausgelegt war, hat auch nur 500 € gekostet und dementsprechend war es mir zu gefährlich mehr zu riskieren. Ich hoffe ich bekomme einige hilfreiche Tipps] (28) wo ich arbeite gibt es eine hübsche Blondine, diemich wohl mag. Sie ist sehr schüchtern. Habe sie angesproche, ob wir nicht einmal ein Kaffee trinken können. Antwort mal sehen , jetzt noch nicht. Ich hab die Hoffnung schon aufgegeben. Soll ich um sie kämpfen. Was ratet Ø mir?76 | What recommend-2pl [you-pl] me-DAT? (29) Ich lese unheimlich gern Bücher und möchte mir auch die einzelnen Wörter, die ich nicht verstehe und im Duden nachschlage, - um auch einige Beispiele durchzulesen und es mir auch zu merken -, am besten für immer im Kopf be-

|| 74 As the surrounding context indicates, the language use of the speakers in (27), (29) and (30) can be classified as ‘rather distinguished’. 75 http://www.mtb-news.de/forum/t/dh-freeridebike-fuer-anfaenger.653304/ (checked: 17.05. 2014) 76 http://forum.gofeminin.de/forum/couple1/__f45840_couple1-Sie-will-oder-sie-will-auchnicht.html (checked: 17.05.2014)

232 | Null Subjects at the Syntax-Pragmatics Interface

halten. Doch geht das eurer Meinung nach? Sollte man sich diese Wörter + Beispiel & Bedeutung in einem Blatt Papier aufschreiben und auswendig lernen? So hab’ ich es zumindest damals gemacht und war halt ein wenig aufwendig.Was ist eure Methode? Was schlagt Ø vor? 77 | What suggest-2pl [you-pl] PRT? (30) “Nun,...” setzt Vabranar etwas zögerlich an, um nach einer kurzen Pause entschlossener fortzufahren: “Vermutlich habt ihr recht und wir sollten wirklich nicht tatenlos herumsitzen. Aber was schlagt Ø vor zu tun? 78 | But what suggest-2pl [you-pl] PRT to do? [Wenn wir uns jetzt auf den Weg machen und ihre Spuren verfolgen, kann es durchaus passieren, dass wir ihnen schließlich im Dunkeln und noch dazu außerhalb des Hofes gegenüberstehen. Vielleicht sollten wir Vorkehrungen gegen einen erneuten Angriff treffen.” Er wendet sich Dylan zu: “Könnt ihr nicht Fallen aufstellen oder zumindest in der Nähe des Tores eine Fallgrube errichten?”] Of course, the examples (27)–(30) only constitute positive evidence. The only “real”, i.e. independent evidence for the claim in (26) comes from dialects, namely Swabian. In Swabian, no clitic-like pronoun can be assumed for the 2nd person singular. Due to the fact that the 2nd person singular form is nonsyncretic throughout the whole paradigm, 2nd person singular null subjects are (as expected) licensed, cf. (31a). However, 2nd person plural subjects remain unlicensed, cf. (31b) – which is no wonder, because the Swabian plural paradigm is uniform (see the table in 4.3.1): (31) a.

Wann bisch Ø losg’fahre? When are-2sg [you-sg] moved-off? b. * Wann sen Ø losg’fahre? When are-pl [you-pl] moved-off?

Swabian Swabian

Note that the examples in (31) are absolutely parallel to the Colloquial High German data in (32), with the only exception(s) that, firstly, the Swabian 2nd person singular form does not end with a (possibly) amalgamated form, cf. (31a), and – as a consequence – we can be sure that no clitic is present. Sec-

|| 77 http://www.wer-weiss-was.de/deutsch/fremdwoerter-wie-am-besten-merken (checked: 17.05.2014) 78 http://www.rpgboard.org/thread.php4?id=47014&page=9&lvis=1006961887 (checked: 17.05.2014)

More on the Identification/Licensing of Null Subjects (in German) | 233

ondly, in contrast to Swabian, cf. (31b), the 2nd person plural null subject in Colloquial High German is licensed, cf. (32b) – which, according to (26), is so because the form of the 2nd person plural (seid) is non-syncretic throughout the whole paradigm of sein (to be). (32) a. b.

Wann bist Ø losgefahren? When are-2sg [you-sg] moved-off? Wann seid Ø losgefahren? When are-2pl [you-pl] moved-off?

(Colloquial) High German (Colloquial) High German

Being aware of the fact that the 2nd person plural null subjects in non-prefield position are not well accepted by most native speakers, I will nevertheless maintain that German has the force to license non-prefield null subjects under forms that are non-syncretic throughout their whole inflectional paradigm (i.e. across all tenses), cf. the principle in (26), which predicts the occurrence of 2nd person singular and 2nd person plural null subjects in a non-prefield position (in the presence of certain verbs). In particular, I will suppose that non-prefield null subjects in German stand in Wackernagel position (WP), which is the same slot that is occupied by null subjects in Bavarian and in older stages of German (which count as partial null subject languages, see e.g. Bayer 1984, Weiß 1998, Axel & Weiß 2011), cf. (33): (33) a. b.

(I glaub,) moang bisd Ø wieda gsund Bavarian I think, tomorrow are-2sg [you-sg] again healthy wennsd Ø moang wieda gsund bisd Bavarian If-2sg [you-sg] tomorrow again healhy are-2sg

As to the formal/positional licensing of WP null subjects, I will adapt the licensing rule in (34) from Axel & Weiß (2011:37) which makes the correct predictions with respect to German, German dialects (cf. Bavarian, but also Swabian) and older stages of German:79

|| 79 With respect to the occurrence of 2nd person null subjects in later stages Axel & Weiß (2011:35, fn 15) note the following: “The special role of 2nd person can also be observed in Early Modern High German. As Held (1903: 129) shows, the number of examples with omitted 2nd person subject pronouns is high both in the poetry and in the prose. For example, the texts by Hans Sachs witness many examples:

234 | Null Subjects at the Syntax-Pragmatics Interface

(34) pro must be c-commanded by pronominal / discrete agreement.80 As shown above, (34) can be fulfilled by different means: (i) (ii)

Movement of the finite verb to C° (as in (31a), (32) and (33a)) The presence of inflected complementisers (as in (33b))

In Swabian and Colloquial German we do not have inflected complementisers. Thus, the only way to ensure (34) is via verb movement. Note that in contrast to Wackernagel null subjects, OBD null subjects have different properties. First, they are (obviously) only possible in the prefield, but not in WP. Second, for the licensing of OBD null subjects the licensing rule in (34) does not apply. Instead OBD null subjects are licensed by a Spec-Head configuration in C, with the finite verb in C° and the null subject in Spec-position. I will finish this chapter by pointing out some general remarks with respect to the type of (non-)null subject language German may belong to. Evidently, the particular distribution of null subjects makes it difficult to put German in any of the “traditional” null subject classes.

4.3.5 German – a Partial Null Subject Language (?) As German licenses null subjects by discrete inflections it patterns along the lines of pro-drop and partial pro-drop languages. On the other hand, as OBD

|| (i) was fragst? what ask-2sg ‘why are you asking?’ (Hans Sachs, Fasnachtsspiele 4, 34, 30; from Held 1903: 134) (ii) wie bist erplichen? how are-2sg turned-pale ‘how have you turned pale?’ (Hans Sachs, Fabeln 77, 17; from Held 1903: 134) In the other persons, examples are much rarer and occur largely in poetical texts, where metrical requirements might often have triggered the omission of the subject pronoun (Held 1903: 112). Moreover, in letters 1st person subjects are often omitted (ibid.: 127). However, this type of subject omission is probably not an instance of pro-drop but rather of so-called ‘diary drop’, a kind of pragmatically licensed subject omission restricted to certain genres that is widespread even in non-pro-drop languages.” 80 In fact, the rule in (34) is extended: Axel & Weiß (2011) speak of pronominal agreement only.

More on the Identification/Licensing of Null Subjects (in German) | 235

null subjects can only be realised in a particular syntactic position (the prefield) and WP null subjects are confined to the 2nd person, it can hardly be maintained that German is a typical null subject language. Moreover, the occurrence of null subjects in German is wide-spread and regular but (sociolinguistically) confined to colloquial uses. Besides, German lacks generic pro which is a typical property of partial null subject languages but an untypical property of genuine pro-drop languages (cf. Holmberg & Sheehan 2010). On the other hand, in German and partial pro-drop languages (but unlike in pro-drop languages), the interpretation of a null pronoun is not (necessarily) different from that of a full pronoun picking up the matrix subject, cf. Finnish (Holmberg et al. 2009:65), German and (on the other hand) Italian and Greek:81 (35)

Juhanii kertoi että (häni) oli ostanut talon. Juhani said that he had bought house ‘Juhani said that he had bought a house.’

Finnish

(36)

Hansi sagte, Øi / eri/k hätte dieses Haus sofort gekauft. Hans said, [he]/he had that house immediately bought

German

(37)

a.

Italian

b.

Il professorei ha parlato dopo che (lui*i/j) è arrivato. The professor has spoken after that (he) is arrived ‘The professor spoke after he arrived.’ I Mariai jelase afou (afti*i/j) idhe ton Yianni. The Mary laughed after (she) saw the Yiannis ‘Mary laughed after she saw Yiannis.’

Greek

In addition, (38) shows that split antecedents for null subjects are (marginally) possible in German – as in most partial null subject languages (e.g. Marathi, to a lesser extent in Finnish, see Holmberg & Sheehan 201082): (38)

Hansi erzählte Mariam, _i+m seien heute Abend bei den Hans-NOM told Maria-DAT, [they-NOM] were-(1/)3pl today evening at the

|| 81 With respect to the data in (37) Roberts & Holmberg (2010:7) note: “Here the overt pronoun in the adverbial clause does not allow the interpretation in which it corresponds to the subject of the main clause (see Vanelli, Renzi & Beninca 1986, Samek-Lodovici 1996, and Frascarelli 2007 for discussion).” 82 See Holmberg & Sheehan (2010:132) for an overview of the characteristics of partial null subject languages.

236 | Null Subjects at the Syntax-Pragmatics Interface

Müllers eingeladen. Müllers invited Thus, I will conclude that German bears some intermediate status, between a non-null subject language and a null subject language. However, the rich occurrence of null subjects in different positions, connected to specific licensing and identification conditions makes it reasonable to put (Colloquial) German into the class of partial null subject languages. Even if German lacks some (defining) characteristics of partial null subject languages (e.g. a generic null pronoun, non-thematic null subjects), German differs significantly from English, Swedish and other consistent non-null subject languages. I.e., considering the fact that in Colloquial German null subjects are used quite frequently (and under conditions which are predictable), we can suppose that German is (on the way to become) a partial pro-drop language (by assumption of a cyclical development as argued for by van Gelderen 2011 or Trutkowski 2012).

References Abraham, W. & C. J. Conradie (2001): Präteritumschwund und Diskursgrammatik, Amsterdam: John Benjamins. Ackema, P., P. Brandt et al. (2006): “The role of agreement in the expression of arguments”, in: Ackema, P., P. Brandt, M. Schoorlemmer & F. Weerman (eds.), Arguments and Agreement, Oxford: Oxford University Press, 1–32. Ackema, P. & A. Neeleman (2007): “Restricted Pro Drop in Early Modern Dutch”, The Journal of Comparative Germanic Linguistics 10(2), 81–107. Ackerman, F. & J. Moore (2001): Proto-properties and grammatical encoding: a correspondence theory of argument selection, Stanford, California: CSLI Publications. Axel, K. & H. Weiß (2011): “Pro-drop in the History of German. From Old High German to the modern dialects”, in: P. Gallmann & M. Wratil (eds.), Null Pronouns (Studies in Generative Grammar [SGG]), Berlin, New York: De Gruyter, 21–51. Balkenende, P. (1995): Top ik Drop je: Over Topic-drop in het Nederlands, M.A. Thesis, Utrecht University. Bausewein, K. (1990): Akkusativobjekt, Akkusativobjektsätze und Objektsprädikate im Deutschen. Untersuchungen zu ihrer Syntax und Semantik, Tübingen, Niemeyer (Linguistische Arbeiten 251). Bayer, J. (1984): “Comp in Bavarian Syntax”, Linguistic Review 3, 209–274. Bayer, J., M. Bader & M. Meng (2001): “Morphological underspecification meets oblique case: Syntactic and processing effects in German”, Lingua 111, 465–514. Benveniste, É. (1971): Problems in General Linguistics, Translated by Mary Elisabeth Meek. Cora Gables, FA. University of Miami Papers. Bierwisch, M. (2006): “Thematic Roles – Universal, Particular, and Idiosyncratic Aspects”, in Bornkessel, I., M. Schlesewsky et al., 89–126. Biberauer, T., A. Holmberg et al. (2010): Parametric Variation. Null Subjects in Minimalist Theory, Cambridge: Cambridge University Press. Blume, K. (1998): “A contrastive analysis of interaction verbs with dative complements”, Linguistics 36, 253–280. Blume, K. (2000): Markierte Valenzen im Sprachvergleich: Lizenzierungs- und Linkingbedingungen, Tübingen: Niemeyer. Bornkessel, I., M. Schlesewsky et al. (2006): Semantic role universals and argument linking: Theoretical, typological and psycholinguistic perspectives, Berlin: De Gruyter Mouton. Cabredo Hofherr, P. (2006): “‘Arbitrary’ pro and the theory of pro-drop”, in: Ackema, P., P. Brandt et al. (eds.), Arguments and Agreement, Oxford: Oxford University Press, 230–257. Camacho, J. A. (2013): Null Subjects, Cambridge: Cambridge University Press. Cardinaletti, A. (1990): “Subject/object asymmetries in German null-topic constructions and the status of specCP”, in: Mascaró, J. & M. Nespor (eds.), Grammar in Progress, Dordrecht: Foris, 75–84. Chiarcos, C., C. Berry & M. Grabski (2011): “Salience in linguistics and beyond”, in: Salience. Multidisciplinary Perspectives on its Function in Discourse (Trends in Linguistics, Studies and Monographs [TiLSM] 227), Berlin, New York: De Gruyter Mouton, 1–28. Chomsky, N. (1981): Lectures on government and binding. The Pisa lectures, Studies in generative grammar 9, Dordrecht: Foris. Chomsky, N. (1986): Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger.

238 | References

Chomsky, N. (2005): “Three Factors in Language Design”, Linguistic Inquiry 36, 1–22. Cole, M. (2009): “Null subjects: a reanalysis of the data”, Linguistics 47(3), 559–587. Costa, J., M. Lobo & C. Silva (2009): “Null objects and early pragmatics in the acquisition of European Portuguese”, Probus 21, 143–162. Cuervo, M. (2003): Datives at large, Doctoral Dissertation, MIT. Czypionka, A. (2013): Resümee der Dissertation The interplay of object animacy and verb class in representation building, Humboldt Universität zu Berlin, 8.11.2013. Czypionka, A., K. Spalek & I. Wartenburger (2012): Object animacy effects in more or less transitive sentences, Poster abstract for the 25th Annual CUNY Conference on Human Sentence Processing, CUNY 2012. Déchaine, R.-M. & M. Wiltschko (2002): “Decomposing Pronouns”, Linguistic Inquiry 33, 409– 442. Dowty, D. (1991): “Thematic Proto-Roles and Argument Selection”, Language 67(3), 547–619. Engelberg, S. (1995): “Event Structure and the Meaning of Verbs”, in: Bærentzen, P. (ed.), Aspekte der Sprachbeschreibung. Akten des 29. Linguistischen Kolloquiums, Aarhus 1994, Tübingen: Niemeyer, 37–41. Engelberg, S. (2006): “A Theory of Lexical Event Structures and its Cognitive Motivation”, in: Wunderlich, D. (ed.), Advances in the Theory of the Lexicon. Berlin, New York: De Gruyter Mouton, 235–285. Erteschik-Shir, N. (2007): Information structure. The syntax-discourse interface, Oxford: Oxford University Press. Falk, C. (1993): “Non-referential subjects and agreement in the history of Swedish”, Lingua 89(2–3), 143–180. Fanselow, G. (2000): “Optimal Exceptions”, in: Stiebels, B. & D. Wunderlich (eds.), Lexicon in focus (Studia grammatica 45), Berlin: Akademie Verlag, 173–209. Fox, D. (2000): Economy and Semantic Interpretation, Massachusetts: MIT Press. Fleischer, J. (2006): “Dative and indirect object in German dialects: evidence from relative clauses”, in: Hole, D., A. Meinunger & W. Abraham (eds.), Datives and Other Cases: between Argument Structure and Event Structure (Studies in Language Companion Series 75), Amsterdam: Benjamins, 213–238. Frey, E. (1975): Stuttgarter Schwäbisch. Marburg: Elwert. Frey, W. (2004): “A medial topic position for German”, Linguistische Berichte 198, 153–190. Frey, W. & K. Pittner (1998): “Zur Positionierung der Adverbiale im deutschen Mittelfeld”, Linguistische Berichte 176, 489–534. Frascarelli, M. (2007): “Subjects, topics and the interpretation of referential pro. An interface approach to the linking of (null) pronouns”, Natural Language and Linguistic Theory 25(4), 691–734. Fries, N. (1988): “Über das Nulltopik im Deutschen”, Sprache und Pragmatik 3, 19–49. Fuß, E. (2011): “Historical pathways to null subjects: Implications for the theory of pro-drop”, in: Gallmann, P. & M. Wratil (eds.), Null Pronouns, Berlin: De Gruyter Mouton, 53–98. van Gelderen, E. (2011): Linguistic cycle: Language Change and the Language Faculty. Oxford: Oxford University Press. Giorgi, A. (2010): About the speaker, Oxford: Oxford University Press. Grewendorf, G. (2002): Minimalistische Syntax, Tübingen: Francke. Grimshaw, J. (1990): Argument Structure, Cambridge, Mass.: MIT Press. Haegeman, L. (1997): “Register variation, truncation, and subject omission in English and in French”, English Language and Linguistics 1(2), 233–270.

References | 239

Haegeman, L. (2007): “Subject omission in present-day written English: On the theoretical relevance of peripheral data”, Rivista di Grammatica Generativa [RGG] 32, 91–124. Haegeman, L. & T. Ihsane (1999): “Subject ellipsis in embedded clauses in English”, English Language and Linguistics 3(1), 117–145. Haegeman, L. & T. Ihsane (2001): “Adult Null Subjects in the non-pro-drop Languages: Two Diary Dialects”, Language Acquisition 9(4), 329–346. Haider, H. (1993): Deutsche Syntax – generativ. Vorstudien zur Theorie einer projektiven Grammatik, Tübingen: Narr. Haider, H. (1994): “(Un)heimliche Subjekte”, Linguistische Berichte 153, 372–385. Haider, H. (1997): “Precedence among predicates”, The Journal of Comparative Germanic Linguistics 1, 3–41. Haider, H. (2010): The Syntax of German, Cambridge: Cambridge University Press. Hardt, D. (1993): Verb Phrase Ellipsis: Form, Meaning, and Processing, Ph.D. Thesis, University of Pennsylvania. Hardt, D. (2006): Re-binding and the Derivation of Parallelism Domains, Ms., Copenhagen Business School. Harley, H. & E. Ritter (2002): “A feature-geometric analysis of person and number”, Language 78(3), 482–526. Hartman, J. (2011): “The Semantic Uniformity of Traces: Evidence from Ellipsis Parallelism”, Linguistic Inquiry 42(3), 367–388. Heim (1998): “Anaphora and Semantic Interpretation: a Reinterpretation of Reinhart’s Approach”, in: Sauerland, U. & O. Percus (eds.), The Interpretive Tract. MIT Working Papers in Linguistics 25, 205–246. Heim, I. (2008): “Features on Bound Pronouns”, in: Adger, D., S. Bejar & D. Harbour (eds.), Phi Theory: Phi Features across Interfaces and Modules, Oxford: Oxford University Press, 35– 56. Held, K. (1903): DasVerbum ohne pronominales Subjekt in der älteren deutschen Sprache, Berlin: Mayer & Müller. Hole, D. (2005): “Reconciling ‘possessor’ datives and ‘beneficiary’ datives – Towards a unified voice account of dative binding in German”, in: Maienborn, C. & A. Wöllstein (eds.), Event Arguments: Foundations and Applications, Tübingen: Niemeyer, 213–242. Hole, D. (2012): “German free datives and Knight Move Binding”, in: Alexiadou, A., T. Kiss & G. Müller (eds.), Local Modelling of Non-Local Dependencies in Syntax, Tübingen: Niemeyer, 213–246. Holmberg, A., A. Nayudu & M. Sheehan (2009): “Three partial null-subject languages: a comparison of Brazilian Portuguese, Finnish, and Marathi”, Studia Linguistica 63(1), 59–97. Holmberg, A. & M. Sheehan (2010): “Control into finite clauses in partial null-subject languages”, in: Biberauer, T., A. Holmberg et al., Parametric Variation. Null Subjects in Minimalist Theory, Cambridge: Cambridge University Press, 125–152. Huang, C.-T. J. (1984): “On the Distribution and Reference of Empty Categories”, Linguistic Inquiry 15(4), 531–574. Huang Y. (2000): Anaphora. A cross-linguistic study, Oxford: Oxford University Press. Jackendoff, R. (1990): Semantic Structures, Cambridge, Mass.: MIT Press. Jaeggli, O. & K. Safir (1989): “The Null Subject Parameter and Parametric Theory”, in: Jaeggli, O. & K. Safir (eds.), The Null Subject Parameter. Studies in Natural Language & Linguistic Theory 15, Dordrecht: Kluwer, 1–44.

240 | References

Jaensch, C. (2005): “UG and Poverty of the Stimulus: Topic Deletion Evidence from L2 learners of German”, Essex Graduate Student Papers in Language and Linguistics 7, 97–117. Jäger, G. (2006): Anaphora and Type Logical Grammar. Springer: Dordrecht. Kaplan, D. (1989a): “Demonstratives. An essay on the semantics, logic, metaphysics and epistemology of demonstratives and other indexicals”, in: Almog, J., J. Perry & H. Wettstein (eds.), Themes from Kaplan, Oxford: Oxford University Press, 481–563. Kaplan, D. (1989b): “Afterthoughts”, in: Almog, J., J. Perry & H. Wettstein (eds.), Themes from Kaplan, Oxford: Oxford University Press, 565–614. Keenan, E. & Comrie, B. (1977): “Noun phrase accessibility and universal grammar”, Linguistic Inquiry 8, 63–99. Kowaluk, A. (1999): “Null objects in Polish: Pronouns and determiners in Second Language Aquisition”, Working Papers of the Research Centre for English and Applied Linguistics, Volume 6, University of Cambridge, 135–152. Landau, I. (2004): “The Scale of Finiteness and the Calculus of Control”, Natural Language & Linguistic Theory 22, 811–877. Lavidas, N. & D. Papangeli (2007): “Impersonals in the diachrony of Greek”, in: Proceedings of the 7th International Conference of Greek Linguistics, University of York. Lenz, B. (1997): Genitiv-Verben und Objektvariation. Arbeiten des Sonderforschungsbereichs 282, Theorie des Lexikons. Wuppertal. Lobeck, A. (1995): Ellipsis. Functional Heads, Licensing, and Identification, Oxford: Oxford University Press. Lyons, J. (1977): Semantics (vols. I and II). Cambridge: Cambridge University Press. Maling, J. (2001): “German inherent datives and argument structure”, Lingua 111, 419–464. Mater, E. (1971): Deutsche Verben. Band 6: Rektionsarten. Leipzig: Bibliographisches Institut. McFadden (2006): “Dative: the heterogeneity of the mapping among morphological case, grammatical functions, and thematic roles”, in: Hole, D., A. Meinunger & W. Abraham (eds.), Datives and Other Cases: between Argument Structure and Event Structure (Studies in Language Companion Series 75), Amsterdam: Benjamins, 49–77. Meinunger, A. (2000): Syntactic aspects of topic and comment (Linguistik Aktuell 38), Amsterdam: John Benjamins. Meinunger, A. (2007): “Der Dativ im Deutschen - Eine Verständnishilfe für das Phänomen der gespaltenen Ergativität”. Linguistische Berichte 209, 3–33. Merchant, J. (2008): “Variable island repair under ellipsis”, in: Johnson, K. (ed.), Topics in ellipsis, Cambridge University Press: Cambridge, 132–153. Merchant, J. (2013): Ellipsis: A survey of analytical approaches. Ms., University of Chicago. Merchant, J. & A. Giannakidou (1997): “On the interpretation of null indefinite objects in Greek”, Studies in Greek Linguistics 18, 141–154. Müller, S. (2002): Complex Predicates: Verbal Complexes, Resultative Constructions, and Particle Verbs in German (Studies in Constraint-Based Lexicalism 13), Stanford: CSLI Publications. Müller, G. (2005a): Subanalyse verbaler Flexionsmarker, Ms. Universität Leipzig, 03. May 2005. Müller, G. (2005b): Pro-Drop and Impoverishment, Ms. Universität Leipzig, 06. October 2005. Primus, B. (2012): Semantische Rollen, Heidelberg: Carl Winter. Pustejovsky, J. (1991): “The syntax of event structure”, Cognition 41, 47–81. Pylkkänen, L. (2002): Introducing Arguments, Ph.D. Dissertation, MIT. Ramchand, G. (2008): Verb Meaning and the Lexicon. A first phase syntax, Cambridge: Cambridge University Press.

References | 241

Raposo, E. (1986): “On the Null Object in European Portuguese”, in: Jaeggli, O. & C. SilvaCorvalán (eds.), Studies in Romance Linguistics, 373–390. Reinhart, T. (1981): “Pragmatics and Linguistics: An Analysis of Sentence Topics”, Philosophica 27, 53–94. Reinhart, T. (1983): “Coreference and Bound Anaphora: A Restatement of the Anaphora Questions”, Linguistics and Philosophy 6(1), 47–88. van Riemsdijk, H. (1989): “Swiss Relatives. Sentential Complementation and the Lexicon”, in: D. Jaspers, W. Klooster et al. (eds.), Studies in Honour of Wim de Geest, Dordrecht: Foris, 343–354. Rizzi, L. (1994): “Early null Subjects and Root Null Subjects”, in: Hoekstra, T. & B. D. Schwartz (eds.), Language acquisition studies in generative grammar, Amsterdam: John Benjamins, 151–176. Rizzi, L. (2002): On the Grammatical Basis of Language Development: A Case Study, Ms., Università di Siena. Roberts, I. & A. Holmberg (2010): “Introduction: parameters in minimalist theory”, in: Biberauer, T., A. Holmberg et al. (2010) Parametric Variation. Null Subjects in Minimalist Theory. Cambridge: Cambridge University Press, 1–57. Rooth, M. (2002): “Ellipsis Redundancy and Reduction Redundancy”, in: Berman, S. & A. Hestvik (eds.) Proceedings of the Stuttgart Ellipsis Workshop, Report Series, Sprachtheoretische Grundlagen für die Computerlinguistik 29, SFB 340, IBM Heidelberg, 1–26. Rooth, M. (2005): “Scope Disambiguation by Ellipsis and Focus without Scope Economy”, in: Dekker, P. & M. Franke (eds.) Proceedings of the Fifteenth Amsterdam Colloquium, ILLC/Department of Philosophy, University of Amsterdam. Ross, J.R. (1967): Constraints on Variables in Syntax, Doctoral Dissertation, MIT. Ross, H. (1982): Pronoun-Deleting Processes in German, handout, LSA Annual Meeting, San Diego. Sag, I. (1976): Deletion and Logical Form, Doctoral Dissertation, MIT, Cambridge, Mass. Salzmann, M. (2005): Resumptive Pronouns and Matching Effects in Zurich German Relative Clauses as Distributed Deletion, Ms., Leiden Center for Linguistics (ULCL). Samek-Lodovici, V. (1996): Constraints on Subjects: An Optimality Theoretic Analysis, Ph.D. Thesis, Rutgers University, New Brunswick, NJ. Sauerland, U. (2008): “Pseudo-Sloppy Readings in Flat Binding”, in: Bonami, O. & P. Cabredo Hofherr (eds.), Empirical Issues in Syntax and Semantics 7, 331–349. de Schepper, C. (2012): You and me against the world? First, second and third person in the world’s languages, Dissertation (Proefschrift), University Nijmegen, Utrecht: LOT. Schlenker, P. (2002): Indexicality, Logophoricity, and Plural Pronouns, Ms. UCLA & Institut Jean-Nicod. Siewierska, A. (2004): Person, Cambridge: Cambridge University Press. Sigurðsson, H. Á. (2011): “Conditions on argument drop”, Linguistic Inquiry 42(2), 267–304. Sigurðsson, H. Á. & V. Egerland (2009): “Impersonal null-subjects in Icelandic and elsewhere”, Studia Linguistica 63(1), 158–185. Smith, M. B. (1987): The Semantics of Dative and Accusative in German: An Investigation in Cognitive Grammar, Ph.D. Dissertation, UC San Diego. Spathas, G. (2007): On the interpretation of gender on nouns and pronouns, Draft Ms. University of Utrecht. UiL-OTS.

242 | References

Speas, M. (2006): “Economy, Agreement and the Representation of Null Arguments”, in: Ackema, P., P. Brandt et al. (eds.), Arguments and Agreement, Oxford: Oxford University Press, 35–75. Steinbach, M. (2007): “Integrated Parentheticals and Assertional Complements”, in: Dehé, N. & Y. Kavalova (eds.), Parentheticals, Amsterdam: Benjamins, 53–87. Sternefeld, W. (1998): “Reciprocity and Cumulative Predication”, Natural Language Semantics 6(3), 303–337. Strawson, P. F. (1964): “Identifying Reference and Truth Values”, in: Strawson, P. F. (ed.), Logico-Linguistic Papers (1971), London: Methuen & Co. Strecker, B., L. Hoffmann & G. Zifonun (1997): Grammatik der Deutschen Sprache, Berlin: De Gruyter. Struckmeier, V. (2014): Scrambling ohne Informationsstruktur: Zur komplexen Architektur grammatischer Beschreibungen (Studia grammatica 77), Berlin: DeGruyter. Takahashi, S. & D. Fox (2005): “MaxElide and the re-binding problem”, Ms. MIT [Appeared in Georgala, E. & J. Howell (eds.), Proceedings of SALT 15, Ithaca, N.Y.: CLC Publications, Cornell University]. Thornton, R. & K. Wexler (1999): Principle B, VP ellipsis, and interpretation. Cambridge, MA: MIT Press. Thrift, K. E. (2003): Object Drop in the L1 Acquisition of Dutch, University of Utrecht. UiL-OTS. Trutkowski, E. (2011): “Referential null subjects in German”, in: Cummins, C., C.-H. Elder, T. Godard et al. (eds.), Proceedings of the Sixth Cambridge Postgraduate Conference in Language Research, Cambridge: Cambridge Institute of Language Research, 206–217. Trutkowski, E. (2012): The Loss and Rise of Null Subjects within the Subject Cycle, talk at the DGfS Jahrestagung, AG 4, “Language change at the syntax-semantics interface”, Frankfurt am Main, 7.–9. March 2012. Trutkowski, E. & H. Weiß (to be published): “When Personal Pronouns Compete with Relative Pronouns”, in: Grosz, P. & P. Patel-Grosz (eds.), The Impact of Pronominal Form on Interpretation (Studies in Generative Grammar [SGG] 125), Berlin, New York: De Gruyter Mouton. Vainikka, A. & Y. Levi (1999): “Empty Subjects in Finnish and Hebrew”, Natural Language and Linguistic Theory 17(3), 613–671. Vanelli, L., L. Renzi & P. Benincà (1985/2007): “Typologie des pronoms sujets dans les langues romanes”. Actes du XIIe Congrès de Linguistique et Philologie Romanes. Aix-en-Provence. Published in English translation as Vanelli, L., L. Renzi & P. Benincà. 2007. “A typology of Romance subject pronouns”, in I. Roberts (ed.), Comparative grammar, Volume II: the null subject parameter. London: Routledge, 234–45 (cited by Roberts & Holmberg 2010 as Vanelli, Renzi & Benincà 1986). Volodina, A. & H. Weiß (to be published): “Diachronic development of null-subjects in German”. In: S. Featherston & Y.Versley (eds.), Firm Foundations: Quantitative Approaches to Grammar and Grammatical Change (Trends in Linguistics: Studies and Monographs 290), Berlin, New York: De Gruyter. Wegener, H. (1985): Der Dativ im heutigen Deutsch, Tübingen: Narr. Wegener, H. (1991): “Der Dativ - ein struktureller Kasus?”, in: Fanselow, G. & S. W. Felix (eds.), Strukturen und Merkmale syntaktischer Kategorien, Tübingen: Narr 1991, 70–103. Weiß, H. (1998): Syntax des Bairischen. Studien zur Grammatik einer natürlichen Sprache (Linguistische Arbeiten 391), Tübingen: Niemeyer.

References | 243

Weiß, H. (2002): “A Quantifier Approach to Negation in Natural Languages. Or Why Negative Concord is Necessary”, Nordic Journal of Linguistics 25(2), 125–153. Weiß, H. (2004): „Information structure meets Minimalist syntax. On argument order and case morphology in Bavarian“, in: Alice ter Meulen, Werner Abraham (eds.), The Composition of Meaning: From Lexeme to Discourse (Current Issues in Linguistic Theory 255), Amsterdam, Philadelphia: Benjamins, 139–165. Weiß, H. (2005): “The double competence hypothesis. On diachronic evidence”, in: Kepser, S. & M. Reis (eds.), Linguistic Evidence: Empirical, Theoretical, and Computational Perspectives, Berlin, New York: De Gruyter Mouton, 557–575. Weiß, H. (2012): “The Rise of DP-Internal Possessors”, in: G. de Vogelaer & G. Seiler (eds.), The dialect laboratory: dialects as testing ground for theories of language change, Amsterdam: John Benjamins, 271–293. Weiß, H. (2015): “When the subject follows the object. On a curiosity in the syntax of personal pronouns in some German dialects”, Journal of Comparative Germanic Linguistics 18(1), 65–112. Weir, A. (2008): Subject pronoun drop in informal English, M.A. Dissertation, University of Edinburgh. Wilder, C. (1996): “V2-Effekte: Wortstellung und Ellipse”, in: Lang, E. & G. Zifonun (eds.), Deutsch – Typologisch, IdS-Jahrbuch 1995, Berlin: De Gruyter, 142–180. Wilder, C. (1997): “Some Properties of Ellipsis in Coordination”, in: Alexiadou, A. & T. Hall (eds.), Studies in Universal Grammar and Typological Variation (Linguistik Aktuell 13), Amsterdam: John Benjamins, 59–107. Winhart, H. (2002)/(2005): Funktionsverbgefüge im Deutschen. Zur Verbindung von Verben und Nominalisierungen, Doctoral Dissertation, Universität Tübingen. Woolford, E. (2006): “Lexical Case, Inherent Case, and Argument Structure”, Linguistic Inquiry 37(1).

Index (aboutness) topic 27, 135, 200, 225 Abraham 220 Ackema 15, 20, 186, 195 Ackerman 56, 60, 62 Alemannic 221 animacy/animateness 63, 64, 69–73, 78, 79 antecedent-dependent licensing / identification 1, 10, 15, 176, 198, 201, 203, 205, 210, 211, 216, 217, 219, 222, 227 antecedent-independent (= ‘out of the blue’) licensing / identification 10, 11, 185, 187, 195, 210, 214, 216–218, 222, 227 Applicative Phrase 102 Argument Selection Principle (Dowty) 57 Axel 233, 234 Balkenende 126 Bausewein 143 Bavarian 233 Bayer 26, 49, 102, 104, 111, 119, 126, 177, 233 Beninca 235 Benveniste 138 Biberauer 13 Bierwisch 53–55, 59, 101 binding (vs. referential) index 135, 147, 166, 181 Blume 2, 5, 43, 58, 60, 62, 63, 65–68, 76, 78–87, 92, 106 Bornkessel 31 bound operator (Rizzi) 223 Brandt 195 Cabredo Hofherr 217 Camacho 195 Cardinaletti 8, 15, 18, 23, 145, 185, 186, 194, 222–224 CAT (Coordinated Antecedents Test) 12, 13, 115, 116, 198, 199, 201, 203, 209, 211, 212, 214, 215 C/edge linker (Sigurðsson) 197 c-selection (categorial selection) 32, 53–55, 100, 101, 107

character (Kaplan) 8, 9, 106, 145–147, 150, 151, 173, 177 Chiarcos 19, 200 Chinese 12, 13, 198, 199, 203 Chomsky 32, 107, 124, 197 circumstances of evaluation (Kaplan) 146 clause-external licensing/identification 205, 216 clause-internal licensing/identification 205, 216 clitic 228, 229, 232 Cole 12, 198, 199, 202, 206, 212, 227 Comrie 5, 58, 93, 112 Conradi 220 content (Kaplan) 8, 9, 146, 147, 150, 151, 177 contexts of use (Kaplan) 146 Coordinated Antecedents Test → CAT Costa 145 Cuervo 102 Czypionka 78, 117 d(iscourse) binding 180, 182 default antecedent/interpretation 12, 192– 194, 205, 211, 212 deleted antecedent copy 126, 173 (deleted) d-pronoun 26, 121, 126–128, 151, 171, 228 depictive 65 diacritic signs 204 diary drop 10, 13, 120, 186, 191, 192, 194, 234 discrete/non-syncretic verbal inflectional endings 10, 11, 12, 197, 198, 203–206, 209, 210, 212, 214–216, 218, 222, 226, 231, 234 distributed interpretation 140, 141, 173 ditransitive verbs 43, 88, 93–96, 99, Dowty 5, 55–57, 62, 119 d-pronoun → (deleted) d-pronoun embedded V2 clauses 222–224, 226 Engelberg 5, 79, 80 English 13, 15, 21, 147, 191, 193, 194, 236 Erteschik-Shir 15

246 | Index European Portuguese 145, 196 experiencer (-stimulus) verbs 68, 143, 144 expletive 27, 120, 121 Falk 27, 120, 121 Fanselow 37, 43, 94, 95, 98, 137, 145 feature alignment 133, 140 (phi-)feature identity 7, 8, 130, 138, 143, 149, 150, 152, 153, 167–172, 180, 182, 183 (phi-)feature mismatch 20, 130, 138, 139, 161, 162 Finnish 59, 197, 235 Fleischer 111, 120 Fox 9, 33, 169, 173–175, 178, 180 Funktionsverben 61, 74 Frascarelli 193, 195, 235 Frey, E. 220 Frey, W. 19, 125, 135, 224 Fries 2, 3, 15, 23, 25, 28, 186 van Gelderen 236 gender feature mismatches 129, 130, 137, 138, 156, 162, 167, 171 generic null pronoun 236 genitive 2, 4, 21, 25, 26, 31, 43, 50, 52, 53, 56, 59, 61, 73–75, 80, 82, 88–102, 104–106, 108, 112, 114, 115, 117–120, 163, 164 Giorgi 197 (Modern) Greek 13, 21, 173, 219, 235 Grewendorf 94 Grice (Gricean Maxim of Quantity) 33, 175 Grimshaw 53, 80 Haegeman 120, 192, 195 Haider 2, 5, 44, 50, 52, 56, 58, 59, 65, 94, 95, 98, 108, 177, 195 Hardt 124, 178, 179, 182 Harley 198 Hartman 33, 173–175 Hebrew 197 Held 233, 234 Hole 102, 125 Holmberg 195, 197, 235 Huang, C.T.J. 15, 19, 185, 186, 203 Huang, Y. 15, 186

idiosyncratic 43, 44, 57–59, 61, 75, 78, 79, 82, 89, 92, 95, 96, 99, 100–102, 106, 117 Ihsane 120, 192 impersonal pronouns 120, 194, 219 incomplete VP deletion 18 indefinite 21, 136, 141, 150, 151, 169, 173, 180, 182 pure indexical 9, 146, 148, 151, 152, 181 interaction verb 58, 62, 63, 66, 76, 77, 80, 81, 83, 85, 87, 90, 92 information structure 20, 23, 186, 195, 200 Italian 11–13, 186, 195, 197, 198, 202, 206, 222, 235 Jackendoff 55 Jaeggli 197 Jaensch 223 Kaplan 8, 145–147, 150, 177, 198 Kase Phrase (KP) 102 Kiezdeutsch 28 knight move binding 125 Kowaluk 209 Landau 223, 225 Lavidas 219 Lenz 5, 79–81, 96, 106 Levi 197 Lobeck 172 Lobo 145 Lyons 198 Maling 65 Mater 85 MaxElide 33, 175 Maximality Principle (Cole) 206, 212, 215, 217, 227 McFadden 56, 59, 65, 102, 104 Meinunger 28, 58, 85, 115, 118 Merchant 21, 33, 173 micro theta role 3–5, 31–33, 50, 53–55, 60, 69, 100, 106–108, 178 modal (and auxiliary) verbs 14, 191, 208, 230, 231 Moore 56, 60, 62 morpho-syntactic binding 182

Index | 247

Müller, G. 207 Müller, S. 185 Neeleman 15, 20, 186, 195 non-referential binding 2, 8, 129, 138, 140, 141, 143, 151, 152, 157, 158, 160, 161, 167, 169, 172, 175, 180, 182–184 non-rigid designator 8, 125, 126, 141, 142, 165, 170 non-thematic null subjects 236 non-verbatim topic drop (NVTD) 3, 4, 6, 27– 29, 32, 33, 37, 39, 41, 44, 45, 48, 49, 51– 53, 75, 100, 108, 110, 120, 122–124, 177, 178 non-verbatim environment 39, 43, 44, 49, 108, 123, 124, 177 number feature (mis)matches 130, 139 OBD (out of the blue drop) 10, 11, 13, 14, 187, 191, 192–196, 198, 207, 209, 212–2014, 216, 218, 220, 222–227, 234 OBD licensing 207, 223, 226 obviation 206, 207 out of the blue context 12, 13, 188, 191, 211 Papangeli 219 Parallelism Domain (PD) 9, 174–180, 183 particular possible world 146 partial pro-drop 13, 197, 218, 233–236 passive 62, 65, 67, 71, 74, 77, 78, 98 person feature (mis)matches 130, 139 PF-compatibility 6, 134, 138, 142, 180 phi-feature identity/mismatch 2, 7, 8, 128, 131, 133, 137–139, 143, 149, 150, 152, 153, 155, 156, 160, 167–172, 180, 182– 184 Pluralis Maiestatis 192 Pluralis Modestiae 192 Polish 13, 147, 195, 198–200, 202, 204–209, 211, 214 possessive pronoun 2, 127, 128, 131, 132, 136, 141, 148, 151, 152, 168–172, 180 post-syntactic (PF) binding 169, 170, 172, 182 PP (Prepositional Phrase) 23, 26–31, 39, 4143, 45–49, 51–54, 56, 58, 68, 94, 115, 118, 144

predicate identity/synonymy 6, 30, 31, 33, 39, 44, 75, 81, 82, 88, 120, 122 Primus 31 pro-drop languages 11–13, 193, 194, 197, 198, 204, 208–211, 214, 216, 219, 235 Pronoun Zap 28 proto-role 57, 58, 60, 61, 79, 81, 82, 88-91, 93, 98, 99, 117, 119, 177 proto-properties 5, 55, 56, 58, 59, 61, 75, 79, 81, 82, 83, 85, 87–95, 100–102, 104– 106, 108, 112–115, 117–119, 178 Pustejovsky 83, 120 Pylkkänen 102 radical/discourse pro-drop 13, 186, 198, 203 Ramchand 60, 103, 104 re-binding/rebinding 9, 174, 175 referential binding 2, 7, 8, 129, 130, 133, 136, 138–141, 143, 150-152, 156–163, 167, 169, 171, 172, 175, 180, 182–184 referential (vs. binding) index 135, 147, 181, Referential Parallelism 169, 180, 182 reflexive pronoun sich 34, 35, 54, 90, 142– 144, 147, 153 register specific use 14, 26, 28, 43, 50, 52, 101, 186, 188, 191, 192, 213 relative clause 5, 111, 112, 119 Reinhart 27, 124, 135, 200 Renzi 235 resumptive pronoun 111, 112 Riemsdijk 111, 119 rigid designator 125, 146 Ritter 198 Rizzi 8, 15, 19, 23, 145, 175, 176, 186 Roberts 195, 197, 235 Rooth 174, 178 Ross 5, 8, 12, 15, 23, 145, 185, 186, 200, 228 s-selection (semantic selection) 32, 53–55, 100, 101, 107 /s/-stem verb 14, 212–217, 223, 226 Safir 197 Sag 174 salience 12, 19, 20, 23, 62, 124, 175, 187, 188, 199, 200, 202, 205 Salzmann 111, 120, 221

248 | Index de Schepper 198 Samek-Lodovici 11, 193, 195, 202, 235 Sauerland 137 Schlenker 146 semantic binding 182 Sheehan 235 Siewierska 22, 198 Sigurðsson 15, 186, 197, 198, 204 silent to do-VP 113–119 Silva 145 Smith 62, 63, 66, 68, 76, 79, 82 Spathas 137 Spanish 11–13, 198, 199, 204, 206–208, 211, 212, 214, 227 speaker/hearer features 12, 193, 194, 197– 199, 202–204, 212, 214, 219 Spec-Head configuration 13, 14, 222, 223, 226, 231, 234 split antecedent 192, 235 Steinbach 8, 19, 23, 49, 145, 177 Sternefeld 181 Strecker 29 stripping 33–37, 51, 123, 137, 138, 167, 175 Struckmeier 115 Structural Parallelism 169, 174, 182 Swabian 12, 220, 232–234 Swedish 186, 236 syncretisms/syncretic 2, 10–12, 14, 131, 132, 152, 168, 169, 185, 192, 194, 198, 199, 202–218, 220–228, 230–233, 236 Takahashi 9, 33, 174, 175, 178 thematic (theta) roles 30, 31, 55, 57–60, 76, 80–82, 103 (see also ‘micro theta role’) Thornton 166

three-place predicates 2, 56, 58, 79, 88, 94, 95, 98, 99, 114, 115, 118 Thrift 8, 126, 145 topological field model 1, 18 topic/topical 20, 27, 195, 196, 207, 210 Trutkowski 10, 112, 120, 185–187, 236 two-place predicates 43, 44, 56, 65, 79, 80, 83, 85, 88, 92–95, 98–101, 105, 106, 114 unindexed pro 177, 182, 183 Vainikka 197 Vanelli 235 verbatim topic drop (VTD) 3, 4, 26, 28–37, 39–41, 43, 44, 51–53, 100, 101, 120, 122–124, 127, 178 verbatim environment 4, 7, 8, 39, 43, 44, 49, 54, 107, 108, 122–124, 126, 142, 164, 165, 170, 175, 177, 178, 183 Wackernagel position (WP) 13, 14, 233, 234 Wegener 43, 62, 68 Weir 192 Weiß 20, 21, 64, 94, 107, 112, 118, 228, 233, 234 Wexler 166 Wilder 17, 18, 127, 128, 152, 171, 185 Williams 174 Wiltschko 198 Winhart 61, 74, 86 Woolford 43, 94, 95 Zurich-German 220, 221