Null Pronouns 9783110238716, 9783110238709

Table of contents :
Introduction
Pro-drop in the history of German – From Old High German to the modern dialects
Historical pathways to null subjects: Implications for the theory of pro-drop
Uncovered pro – On the development and identification of null subjects
Silent resumptives in Zurich German possessor relativization
Anti-agreement with subjects and possessors from a typological perspective: A case for null pronouns or for economy?
Language index
Subject index

Citation preview

Null Pronouns

Studies in Generative Grammar 106

Editors

Henk van Riemsdijk Harry van der Hulst Jan Koster

De Gruyter Mouton

Null Pronouns

Edited by

Melani Wratil Peter Gallmann

De Gruyter Mouton

The series Studies in Generative Grammar was formerly published by Foris Publications Holland.

ISBN 978-3-11-023870-9 e-ISBN 978-3-11-023871-6 ISSN 0167-4331 Library of Congress Cataloging-in-Publication Data Null pronouns / edited by Melani Wratil, Peter Gallmann. p. cm. ⫺ (Studies in generative grammar; 106) Includes bibliographical references and index. ISBN 978-3-11-023870-9 (alk. paper) 1. Grammar, Comparative and general ⫺ Pronoun. 2. Generative grammar. I. Wratil, Melani. II. Gallmann, Peter. P279N85 2011 4151.55⫺dc23 2011036670

Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.d-nb.de. ” 2011 Walter de Gruyter GmbH & Co. KG, Berlin/Boston Typesetting: RoyalStandard, Hong Kong Printing: Hubert & Co. GmbH & Co. KG, Göttingen ⬁ Printed on acid-free paper Printed in Germany www.degruyter.com

Table of contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Melani Wratil and Peter Gallmann

1

Pro-drop in the history of German – From Old High German to the modern dialects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Katrin Axel and Helmut Weiß

21

Historical pathways to null subjects: Implications for the theory of pro-drop. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Eric Fuß

53

Uncovered pro – On the development and identification of null subjects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Melani Wratil

99

Silent resumptives in Zurich German possessor relativization . . . . . Martin Salzmann

141

Anti-agreement with subjects and possessors from a typological perspective: A case for null pronouns or for economy? . . . . . . . . . . Albert Ortmann

223

Language index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Subject index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

265 267

Introduction Melani Wratil and Peter Gallmann 1. Null pronouns in the classical theory of Principles and Parameters In the middle of the last century many grammarians noted that utterances sometimes contain functional elements that lack a phonetic shape but nevertheless refer to particular entities of the extra-linguistic world. Under the influence of Chomsky’s (1965) Aspects of the Theory of Syntax, such elements were regarded as being inserted on the deep structure level of the clause and subsequently deleted on its surface structure level. Rosenbaum (1967), Postal (1969) and Perlmutter (1971) point out that the subjects of English infinitival clauses belong to those non-overt functional elements. In contrast to Rosenbaum (1967), who analysed the subjects of infinitives as gaps caused by a special deletion procedure, Postal (1969) classified them as genuine null pronouns. In accordance with Postal (1969), Chomsky (1973) finally called the pronominal subjects of non-finite clauses (big) PRO. Together with Lasnik (cf. Chomsky & Lasnik 1977) he showed that PRO represents the selected external argument of subordinated nonfinite verbs and that it combines pronominal with anaphoric properties. So he specified PRO as [þanaphoric; þpronominal]. Displaying this special feature combination, PRO is subject to two contradictory conditions of the Binding Theory. While anaphors have to be bound in their governing category according to Principle A, pronouns must be free in this domain according to Principle B. Hence, PRO is licensed only if it is not located within any governing category and consequently has neither a governor nor a case (PRO theorem). Since, according to Chomsky (1981), non-finite verbs are not capable of governing their subject, PRO only occurs in the SpecI-position of infinitival clauses – thus, in a position that overt DPs are excluded from (cf. (1)). (1) a. Sue decided [PRO / *she to kiss the frog]. b. [PRO / *Sue to kiss the frog] would be a bit strange. Bouchard (1984) and Franks & Hornstein (1992) ascribed this complementary distribution to the fact that only the caseless PRO is allowed to occupy a position no case is assigned to. Chomsky & Lasnik (1993) put

2

Melani Wratil and Peter Gallmann

forward the objection that PRO as antecedent of an A-chain has to be case-marked according to the Visibility Condition. They concluded that non-finite verbs assign null case to their subject DP and that PRO is the only element that can bear it. In the 1970s Perlmutter (1971) and Taraldsen (1978) observed that in a number of languages, as for example in many Romance languages, even the referential subject pronouns of finite clauses are able to obtain a null representation. Together with Jaeggli (1981), Chomsky (1981) initially argued that the languages in question are distinguished by their verbal inflectional morphemes moving to their stems in the deeper syntax (Rule R). In his view, the functional head INFL of those languages does therefore not govern the subject position at the S-structure and consequently allows PRO to fill this position. The often observed fact that the null subjects of finite clauses nevertheless still appear in head-governed positions at the S-structure as for example in the canonical post-verbal subject positions of interrogative clauses finally led him to introduce another pronominal null element called (small) pro (cf. (2a)). In contrast to PRO, pro is defined as an exclusively definite and pronominal ([anaphor; þpronominal]) category. As a consequence pro is thought to have the same syntactic distribution as definite overt subject pronouns and lexical DPs (cf. (2a, b)). (2) Spanish a. Quiere pro venir conmigo? want-3sg come with me ‘‘Does he want to come with me?’’ ?

?

b.

Quiere Pedro venir conmigo? want-3sg Pedro come with me ‘‘Does Pedro want to come with me?’’

Following Neeleman & Szendro¨i (2007), Fuß (this volume) contests the view that the null subject of finite clauses has to be defined as a special empty category pro. He assumes that, just like overt weak pronouns, null pronouns are simply D heads, which, in contrast to the former, are not associated with phonological features at the point of Vocabulary Insertion. 2. The pro-drop parameter According to Chomsky’s (1981) traditional Theory of Principles and Parameters, the knowledge of language is a mental state that is determined

Introduction

3

by a limited system of abstract principles and rules. Several principles of this so-called Universal Grammar (UG) are represented in a parametric fashion therefore permitting grammatical variations between the particular languages. The corresponding parameters are binary in their nature. That is, they allow to select one of two opposing grammatical features or structural properties. Language learners fixate these parameters on the basis of the specific input they are exposed to during the acquisition of their target language. The capability of omitting the subjects of finite clauses or to represent them as null pronouns, respectively, is presumed to be the result of such a parameter setting. The speakers of the concerned speech communities have fixated the null subject (or pro-drop) parameter positively. So they distinguish themselves from speakers of other speech communities who have fixated this parameter negatively and therefore do not allow the subjects of finite clauses to remain phonologically null. Since parameter fixations influence various domains of grammar, they coincide with other specific parameter settings. Chomsky (1981) infers that parameters are arranged in bundles. Thus, an independent grammatical characteristic is always deducible from another independent grammatical property in the same language. Chomsky (1981) and Rizzi (1982) point out that the null subject property is typically connected with the absence of *[that-t]-e¤ects, with the possibility of free subject inversion and with the possibility of extracting a subject long distance over a lexically filled complementizer. Fuß (this volume) argues against the binary concept of the pro-drop parameter, which implies that referential pro-drop is either generally available in all persons and numbers or completely absent. He elucidates on the basis of Bavarian, colloquial French and Finnish data that the pro-drop phenomenon is often confined to certain slots of the pronominal paradigm yielding partial pro-drop. Wratil (this volume) goes one step further. She not only refuses the basic assumption that the language learner simply makes a definite decision in favour of or against [þpro-drop]. She even doubts that the occurrence of subject null pronouns in finite clauses is subject to the specific setting of an underlying null subject parameter. She argues that the classical concept of this parameter is incompatible with the vast majority of partial null subject languages, with the small number of non-null subject languages and with the di¤erent grammatical properties that the various null pronouns exhibit. Recent studies in language acquisition support Wratil’s (this volume) reservations.

4

Melani Wratil and Peter Gallmann

3. Null subjects and verbal inflection It is commonly assumed in the literature on null subject properties that it is the morphological make-up of verbal inflection in the target language that determines how the pro-drop parameter is finally fixated by the language learner. In the late 1970s, Taraldsen (1978) found out that null subjects predominantly occur in those languages that have distinctive morphological agreement specifications on verbs. Chomsky (1981), Jaeggli (1981), Safir (1985) and Rizzi (1986), among others, observed that Taraldsen’s (1978) discovery is verified in many Indo-European language areas. For that reason, it was generally agreed that there is a strong correlation between the morphological distinctiveness of verbal inflection and the positive setting of the pro-drop parameter. The general requirement that pro, just like all other empty categories, has to be identified in its clause or immediate context supports this view. More precisely, only if the agreement marking of the finite verb distinguishes all available phi-feature specifications, the corresponding pronominal null subject can be associated with exactly one individual a‰x. In this way it is, like for example in Spanish (cf. (3)), fully recoverable and therefore licensable by the verbal morphology (Jaeggli 1981). (3) Spanish pro bail-o (1sg)

‘‘I dance’’

pro bail-as (2sg)

‘‘you dance’’

pro bail-a (3sg)

‘‘he/she/it dances’’

pro bail-amos (1pl) ‘‘we dance’’ pro bail-a´is (2pl) pro bail-an (3pl)

‘‘you dance’’ ‘‘they dance’’

Rizzi (1986) therefore considers rich verbal inflection as the fundamental precondition for the insertion and identification of null subjects. But he remarks that it is nevertheless an insu‰cient prerequisite for their licensing. In his opinion, pro is only licensed if it is governed by a head whose features it is co-indexed with and which belongs to a special designated class. The rich agreement specification of inflectional a‰xes is not per se connected with the required designation of the governing verbal heads. The inflectional paradigm of the German verb, for example, is morpho-

Introduction

5

logically rich with regard to its verbal agreement marking (cf. (4)). However, since German does not belong to the full pro-drop languages because of its underlying parametric definition, it does not allow referential subject pronouns to be null in finite clauses (cf. (5)).1 (4) German ich fluch-e (1sg)

‘‘I curse’’

du fluch-st (2sg)

‘‘you curse’’

er/sie/es fluch-t (3sg)

‘‘he/she/it curses’’

wir fluch-en (1pl)

‘‘we curse’’

ihr fluch-t (2pl)

‘‘you curse’’

sie fluch-en (3pl)

‘‘they curse’’

(5) German Sie / *pro findet, dass er / *pro dummes Zeug redet. she think-3sg that he stupid things say-3sg ‘‘She thinks that he says stupid things.’’ Huang’s (1984) studies of Mandarin Chinese obviously contradict Rizzi’s (1986) pro-drop theory. They reveal that referential null subjects are licensed even if finite verbs generally lack morphological agreement markers. According to Jaeggli & Safir (1989), the full null subject status of Mandarin Chinese is attributed to the morphological make-up of the corresponding conjugational system. In their opinion, null subjects typically occur in languages with morphologically uniform inflectional paradigms. They define a paradigm as morphologically uniform if it has either only underived inflectional forms like the Chinese verbal paradigm (cf. (6)) or only derived inflectional forms like the Spanish verbal paradigm (cf. (3)). Accordingly, if a paradigm combines underived inflectional forms with derived inflectional forms like the verbal paradigm of the full non-null subject language English (cf. (7)), it is not regarded as morphologically uniform.

1. In contrast to Standard Dutch and English among others, German is classified as semi pro-drop language because it allows non-argumental, i.e. expletive subjects to be null (see for example Cardinaletti 1990; Abraham 1993; Hulk & v. Kemenade 1993). Haider (1990) and Hoeing (1991) argue against such a categorisation.

6

Melani Wratil and Peter Gallmann

(6) Chinese pro shuo (1sg)

‘‘I say’’

pro shuo (2sg)

‘‘you say’’

pro shuo (3sg)

‘‘he/she/it says’’

pro shuo (1pl)

‘‘we say’’

pro shuo (2pl)

‘‘you say’’

pro shuo (3pl)

‘‘they say’’

(7) English I smoke (1sg) you smoke (2sg) he/she/it smoke-s (3sg) we smoke (1pl) you smoke (2pl) they smoke (3pl) Rohrbacher (1994), Speas (1994), Alexiadou & Anagnostopoulou (1998), Manzini & Savoia (2002) and Platzack (2004) among others dispense with the null pronoun pro entirely. Rohrbacher (1994) and Speas (1994) are convinced that languages allowing pronominal subjects of finite clauses to be omitted di¤er from languages that always require the insertion of overt subjects in finite clauses in that the former are endowed with a strong verbal agreement morphology whereas the latter display a weak verbal agreement morphology. They define the morphological agreement marking on the verb as strong as soon as the corresponding inflectional paradigm exhibits a morphological distinction between the 1st and the 2nd person form in at least one number category and, moreover, if it features a clear morphological di¤erence between the singular and the plural form in at least one person category. In all languages whose finite verbs are specified by such a strong morphological agreement marking, agreement a‰xes are said to be base generated in AgrS , rendering SpecAgrS unnecessary in accordance with Speas’ (1994) principle of Projection Economy. As a result, the concerned languages do not demand any subject DP in finite clauses. In languages whose finite verbs exhibit a weak verbal agreement marking, however, inflectional a‰xes are attached to their verbal stems in the underlying lexicon. In these languages the presence of any subject DP in finite clauses is required for projecting the

Introduction

7

functional AgrSP. Rohrbacher (1994) and Speas (1994) finally argue that languages that entirely lack verbal agreement markers, like Chinese, do not merge the functional category AgrS. As a consequence, they do not rely on the insertion of syntactic subjects in finite clauses. Alexiadou & Anagnostopoulou (1998), Manzini & Savoia (2002) and Platzack (2004), however, assume that the verbal agreement morphology of null subject languages is itself referential and therefore fully interpretable. With that it does not need to enter into an Agree relation with any interpretable subject pronoun in order to get specified. According to Alexiadou & Anagnostopoulou (1998), Manzini & Savoia (2002) and Platzack (2004), it has to be defined as an a‰xed definite pronoun that functions as the syntactic subject by itself receiving the corresponding y-role. Of course, the verification of such a hypothesis presupposes the presence of overt and moreover distinctive subject agreement morphemes – hence, of rich agreement – in all null subject languages. It excludes the insertion of referential null subjects in languages that do not exhibit any agreement inflection. Furthermore, it prompts questions about the representation of overt referential and non-referential subjects in consistent and partial null subject languages (cf. De Crousaz & Shlonsky 2003; Holmberg 2005). Some non-null subject languages, as for example the Germanic languages Icelandic and German, pose di‰culties for Rohrbacher’s (1994) and Speas’ (1994) analysis and, at first glance, also for Jaeggli & Safir’s (1989) account. The verbal paradigms of the concerned languages contain only derived inflectional forms and show a clear morphological distinction between the singular and the plural marking of the distinctive 1st and 2nd person forms. Nevertheless, they are not able to license referential null subjects in finite clauses (cf. Sigurðsson 1989). Jaeggli & Safir (1989) agree with Gilligan (1987) that the V2 phenomenon is responsible for the nonexistence of null subjects in Icelandic as well as in German. They point out that AGR is able to identify small pro only if it is a morphological specification of the relevant case marking inflectional head. Together with Bennis & Haegeman (1984), Vikner (1989) and Wexler (1994) they presume that the functional head C  assigns structural case to the subject DP located in SpecI in V2 languages because the C-projection of these languages is host to essential tense features. Hence, due to V2, AGR, which is included in the functional domain of INFL, is structurally separated from the case marking head, therefore being unable to identify pro (cf. (8)).

8

Melani Wratil and Peter Gallmann

(8) Icelandic [ IP e´g t i það ekki ]]. a. [ CP Nu´na þekki i now recognize-1sg I that not ‘‘I do not recognize that now.’’ b. *[ CP Nu´na þekki i [ IP pro t i það ekki ]]. now recognize-1sg that not Beside Rizzi (1986), Rohrbacher (1994) and Speas (1994), Mu¨ller (2006) holds morphological richness or strength, respectively, responsible for the omission of subject pronouns. More precisely, he argues that pro is excluded from all languages whose verb forms exhibit a poor inflectional morphology. On the theoretical basis of Distributed Morphology, he describes the inflectional marking of verbs as morphologically poor if it is subject to an impoverishment procedure that leads to a neutralization of phi-features. Such impoverishment operations are reflected in systematic syncretisms that extend over the paradigms of all inflectional classes in a language. In this way, the inflectional morphology of the Icelandic verb has to be defined as rich (or strong) in the sense of Taraldsen’s (1978), Rizzi’s (1986), Jaeggli & Safir’s (1989), Rohrbacher’s (1994) and Speas’ (1994) accounts but as poor according to Mu¨ller’s (2006) considerations. Mu¨ller (2006) observes that the inflectional markings of the 1st person singular and the 3rd person singular in the past tense on the one hand and the inflectional markings of the 2nd person singular and the 3rd person singular in the present tense on the other hand are homonymous in Icelandic. They form syncretisms resulting from an impoverishment rule which neutralises person di¤erentiations. In Mu¨ller’s (2006) view, the corresponding impoverishment operations block the licensing of pro. As a consequence, the present-day Icelandic finite verbs do not co-occur with referential null subjects. Axel & Weiß (this volume), Fuß (this volume) and Wratil (this volume) contest Jaeggli’s & Safir’s (1989) statements about V2 languages. They illustrate that the alleged incompatibility of the V2 phenomenon with null subject properties is not borne out. Whereas Wratil (this volume) addresses the V2 null subject language Old Icelandic, Axel & Weiß (this volume) investigate the appearance of referential null subjects in the German language area. They show that, under certain conditions, many German varieties and even earlier language periods of Standard High German – all of them exhibiting strict V2 patterns – allow subject pronouns to obtain a null realization. Axel & Weiß (this volume) and Fuß (this volume)

Introduction

9

are sure that the morphological make-up of verbal agreement markers is the decisive factor in the licensing of null arguments. However, Fuß (this volume) dissociates himself from the traditional approaches that unconditionally highlight the notions of morphological richness and poverty. On the basis of Bavarian and colloquial French data, he elucidates that null subjects emerge in exactly those slots of pronominal paradigms where the corresponding overt clitic pronoun has turned into a subject agreement morpheme in order to eliminate a systematic syncretism within the verbal inflectional paradigm. Hence, the replacement of overt weak pronouns by null subjects is restricted to gaps caused by reanalysis, that is to say, to those contexts where new verbal agreement markers arose. It is not related to properties of the verbal agreement paradigm as a whole. This implies that full null subject languages evolve step by step, passing through various partial pro-drop stages. The study of Axel & Weiß (this volume) supports this claim. It reveals that the null subjects of the modern German dialects, including Bavarian, only appear in contexts where they are c-commanded by an agreement marker that immediately descends from a cliticized pronoun. Wratil (this volume) does not take the view that the null subject properties of a language are always causally related to its verbal inflectional morphology. She argues that weak pronouns are able to exhibit a null representation as soon as they are identifiable in some way. According to her observations, the identification by the inflectional agreement marking on the verbal head is only one possible strategy to recover a subject null pronoun in a tensed clause. But there are, as she goes on to say, some other, at least equally e‰cient strategies, for instance the identification by anaphoric binding, which can be observed in some Celtic languages, as well as the identification by the association with a highly prominent discourse topic, which can be observed in some creole languages.

4. Null pronouns and language change The generative works on various language periods of French provided the foundation for the diachronic investigation of pro-drop phenomena in the late 1980s and in the 1990s. Adams (1987) belongs to the cutting-edge linguists who scrutinized the null subject properties in Old French and their entire loss in modern French. Also Vance (1988), Dupuis (1988), Roberts (1992) and Hirschbu¨hler (1995) tried to find out why null subjects could not survive in the present-day standard variety. They came to the

10

Melani Wratil and Peter Gallmann

conclusion that their loss is due to the erosion of verbal subject agreement markers and to the prevalence of SVO word order patterns in finite root clauses. Referring to the theoretical accounts of Ashby (1977), Lehmann (1988) and Roberge (1990), amongst others, Fuß (this volume) and Wratil (this volume) point out that the reintroduction of null subjects in modern colloquial French has to be traced back to the rise of new distinctive agreement markers. Brandi & Cordin (1989), Renzi (1992) and Poletto (1995) examined certain other varieties of the Romance languages under this aspect. They showed that several subject clitics of some Northern Italian dialects have been re-analysed as distinct functional agreement markers that are accompanied by new, fully identifiable null pronouns. As illustrated by Axel & Weiß (this volume) and by Fuß (2005, this volume), such a developmental process is traceable in the German language area as well. Axel & Weiß (this volume) explore the occurrence and distribution of referential null subjects in Old High German and give an explanation for their gradual loss. According to their observations, the decline of null pronouns in late Old High German is due to a change of the restrictions on null subject licensing caused by the re-analysis of an asymmetric verbal agreement morpheme as pronominal marker and to the subsequent erosion of this marker. Axel & Weiß (this volume) illustrate that the new restricted licensing conditions are still valid in the modern German varieties. Since some pronominal subject clitics have turned into C-oriented agreement markers being attached to the tensed verbs of root clauses or to complementizers, the prerequisites for null subject licensing are nowadays fulfilled in several German dialects. Hence, null subjects are still wide spread in the German language area, though not in modern Standard German. Fuß (this volume) regards the historical development of agreementrelated null subjects as an interrelation between morphological Blocking and morphological Deblocking. The principle of morphological Blocking predicts that a morpheme which exhibits several appropriate phonological realizations in the Primary Linguistic Data always obtains the phonological form that matches the greatest subset of its morphosyntactic features, whereas it blocks all other forms. As soon as a Blocking procedure causes a gap in a paradigm in order to establish a more distinct form in another paradigm, a previously blocked realization is deblocked. Fuß (this volume) shows that the procedures of morphological Blocking and Deblocking were responsible for the rise of partial pro-drop in Bavarian and in present-day Non-Standard French. In both varieties, Blocking forced certain subject clitics to turn into verbal agreement markers, thereby

Introduction

11

enabling null representations to be deblocked in the corresponding slots of the pronominal paradigm. Fuß (this volume) moreover argues that languages in which verbal agreement markers have not yet come into being are able to become full null subject languages through close contact with null subject languages that license their null arguments by a special verbal morphology. Wratil (this volume) takes the view that weak subject pronouns, like many other functional elements, are involved in a diachronic cycle of grammaticization. She argues that the competing forces of the morphological and phonological economy, on the one hand, and of the semantic and pragmatic economy, on the other hand, precipitate this cycle. These active forces not only give rise to a permanent alteration of weak subject pronouns with respect to their syntactic representation, but they also provoke the constant change of the null subject status that a language displays. The full null subject stage and the full non-null subject stage merely represent the antipodal points of this cycle. The graduations between those points manifest themselves in the numerous partial null subject languages. Wratil (this volume) illustrates on the basis of French, Welsh, Mauritian Creole and many other data that the null pronouns in the various intermediate stages of this diachronic cycle usually do not unconditionally fit into the traditional scheme of the empty category pro.

5. Null possessors In head-marking languages, the semantic relationship between a possessor and a possessed item is usually indicated by a special agreement morpheme representing the phi-feature content of the corresponding possessor argument as agreement marker on the possessum noun. Such NP-external agreement relations can be observed in Abkhaz (cf. (9)), Maltese, Turkish and Hungarian, amongst others. (9) Abkhaz a. sara` se -y 0ne pron.1sg poss.1sg-house ‘‘my house’’ b.

a` -cˇ ’k 0’n-c 0a r -y 0n -k 0a` det-boy -pl poss.3pl-house-pl ‘‘the houses of the boys’’

12

Melani Wratil and Peter Gallmann

Languages that exhibit possessor agreement markers do not require their pronominal possessor arguments to have a phonological representation (cf. (9b)). Indeed, overt possessor pronouns are quite rare in the possessive constructions of these languages. Hence, it seems to be at least questionable whether possessors of agreeing possessed items are really always structurally realized. Abney (1987), Szabolcsi (1990, 1994) and den Dikken (1999) are sure of the existence of pronominal null possessors. They argue that a‰xes that mark an agreement relation on the possessum have to be licensed by a congruent possessor phrase. Ortmann (this volume) investigates the morphological marking of possessor agreement and the distribution of possessor phrases in Hungarian and Turkish. In contrast to den Dikken (1999), whose null possessor hypothesis is chiefly based on Hungarian data, he denies the existence of any possessor null pronouns within Hungarian and Turkish possessive constructions. As can be observed, the possessum nouns of Hungarian always display morphological agreement with respect to the category of person. They also display a morphological number agreement marking in cases where their possessor is covert or an overt pronominal expression. Unlike the overt 3rd person pronominal subjects of Hungarian finite clauses, the overt 3rd person pronominal possessors of Hungarian possessive constructions never exhibit any plural morpheme. Den Dikken (1999) accounts for these facts by assuming a special movement procedure. In accordance with Rouveret (1991), he defines full lexical noun phrases as DPs but pronouns as NumPs. In contrast to the number features of full lexical DPs, the number features of pronominal NumPs are licensed by head movement from Num in SpecF to Agr located in the possession phrase. Since Agr is the functional head responsible for the representation of agreement relations, only pronouns or null pronouns (hence covert possessors), respectively, are able to trigger any morphological number agreement marking on the possessum. He furthermore claims that Num-to-Agr movement severs the stems of 3rd person plural possessor pronouns from their agglutinative number morpheme. The latter a‰xes to Agr and subsequently moves downwards to the possessum noun by A‰x Hopping. Ortmann (this volume) contests den Dikken’s (1999) approach whose analysis he shows to be empirically as well as conceptually problematical or even untenable. According to Ortmann (this volume), the special morphological representation of possession in Hungarian can be explained by an economy-based analysis that dispenses with null possessors entirely but takes into account the essential role of saliency. On the typological

Introduction

13

basis of well-documented agreement asymmetries, which are determined by dimensions such as animacy, topic-worthiness and definiteness, he illustrates that the morphological agreement marking on Hungarian possessum nouns reflects the individual saliency value of the corresponding possessor entities. Furthermore, he convincingly argues against the background of the modern Optimality Theory that the morphological number marking on Hungarian possessor pronouns guarantees the most economical and therefore optimal representation of possessive relationships. The relevance of economy can also be observed in the possessive constructions of Turkish, even though their possessum nouns – in contrast to the possessum nouns of Hungarian – are always marked by a singular agreement morpheme if the corresponding possessor phrase is overt. In Ortmann’s (this volume) view the restriction of plural agreement to possessive constructions whose possessor is dropped guarantees the most economical morphosyntactic representation of possession in Turkish. Salzmann (this volume) investigates Standard and Non-Standard German possessive constructions that involve a possessive pronoun. In Standard German, the relationship between a possessum and a corresponding possessor does not manifest itself in an agreement morpheme attached to the possessum noun but rather in a special case marking on the possessor phrase (cf. (10a)) or in a specifying possessive pronoun, which precedes the possessum (cf. (10b)). In colloquial German and in many Non-Standard German varieties, as for example in the Alemannic dialect Zurich German, however, 3rd person possessive pronouns usually combine with case marked possessor phrases yielding so-called possessor doubling constructions (cf. Haegeman 2004) (cf. (10c)). (10) German a. die Fliege des Kellners det bow tie det waiter-gen ‘‘the bow tie of the waiter’’

b.

seine Fliege his bow tie ‘‘his bow tie’’

c. dem Kellner seine Fliege det(dat) waiter his bow tie ‘‘the bow tie of the waiter’’ Salzmann (this volume) describes possessor doubling constructions as DPs whose specifier is represented by the possessor phrase and whose head is represented by the possessive pronoun, which takes the possessum phrase as its complement. According to his observations, such DPs have to be

14

Melani Wratil and Peter Gallmann

regarded as the basic structural realization of possessor relationships in the German language area. In cases where the possessor phrase is covert, like in simple possessive constructions involving a 1st or 2nd person possessive pronoun (or even a 3rd person possessive pronoun in Standard German), the SpecD-position is filled by the empty pronoun pro. Salzmann (this volume) shows that the possessor pro is not only able to control embedded PRO-subjects but, if necessary, also functions as correlative pronoun in relative clauses. The latter can be proved on the basis of Zurich German relative clauses. In contrast to Standard German relative clauses, they are introduced by an invariant complementizer. Their correlative element, which is coreferential with the modified head noun located in the matrix clause, is represented by a gap if it stands for the syntactic subject or for the direct accusative object. Otherwise it is represented by a resumptive pronoun. Salzmann (this volume) argues that resumption is a last resort strategy. It applies only if operator movement to SpecC is not feasible or an oblique case has to be realized. Since resumption does not create any variables, relativizing is insensitive to locality constraints in Zurich German as soon as a resumptive pronoun represents the correlative. Consequently, relativizing on a possessor whose phi-features are identified by a possessive pronoun is always practicable in Zurich German. Just like all other correlative resumptives, the possessor resumptive pro is bound by a non-overt relative operator base generated in the embedded SpecC position.

6. Conclusion The short survey of the common classical and current approaches to the null subjects of finite clauses and to the null possessors of possessive constructions presented in the preceding sections has illustrated that even the well known basic assumptions concerning the licensing of null arguments are not unconditionally verifiable. Much of the data that is discussed in the following papers and many of the theoretical considerations put forward reveal that a large part of these assumptions is at least debatable. So, it seems to be legitimate to ask some fundamental questions once again. For example, do all phonologically empty subject pronouns that occur in the finite clauses of full and partial null subject languages belong to a special group of empty categories called pro that combine the features [þpronominal], [anaphoric] and [þdefinite]? Can we be sure that the innate mental abilities of language learners provide a specific parameter,

Introduction

15

which is binary in its nature, and whose individual fixation during language acquisition determines whether subject null arguments are allowed or forbidden in the target language? Is the richness of the verbal inflectional morphology always the decisive factor in the occurrence and licensing of subject null arguments? Does the rise and loss of null subjects in a language therefore always depend on the rise and loss of its inflectional markers? Are there any morphological or syntactic properties of natural languages that generally block the insertion of null arguments? Do all head marking languages, whose possessum nouns are endowed with possessor agreement markers, really license null possessors? And do null possessors only occur in head marking languages? According to the papers compiled in this book, every single one of these questions can be answered with a ‘‘no’’. Fuß (this volume) argues that the null subjects of full and partial null subject languages cannot be defined as empty categories in the sense of the early Principles and Parameter Theory. In his view, they are weak pronouns that are not associated with phonological features at the point of Vocabulary Insertion. Wratil (this volume) analyses the occurrence of various indefinite null subjects and moreover shows that the finite verbs of a number of languages license null subjects with anaphoric properties. Both of them dispute the binary nature of the so called pro-drop parameter. Beside languages that do not exhibit any agreement morphology on the finite verb but nevertheless belong to the null subject languages, Wratil (this volume) discusses a language that in fact exhibits strong agreement morphology but nevertheless uses discourse strategies in order to license and identify null subjects. Axel & Weiß (this volume) investigate the subject null arguments of languages that are traditionally defined as non-null subject languages because of their syntactic properties. The diachronic analysis of these languages shows that the loss of subject null arguments is not always concomitant with any erosion of verbal markers. Like Fuß (this volume) they moreover describe processes in language change that lead to the emergence of null subjects in specific slots of the pronominal paradigm yielding partial null subject languages. Ortmann (this volume) illustrates that at least in a number of head marking languages the insertion of any null possessors would violate some basic principles of economy. In his view, the concepts of salience, topic-worthiness and definiteness play the decisive role in the morphosyntactic realization or non-realization of possessor arguments. Salzmann (this volume) convincingly argues that there are null possessors even in languages that cannot be classified as consistently head marking and whose possessum nouns are not endowed

16

Melani Wratil and Peter Gallmann

with any possessor agreement markers. He shows that such null possessors are merged in the specifier position of possessive DPs. Since they may function as possessor correlatives in relative constructions, they allow the relativizing on possessors whose phi-features are specified by an overt possessive pronoun. Thus, the authors of the following papers do not only prove that the basic claims, which have been formulated in connection with pro-drop, are problematical or simply not universally applying but they also propose some alternative approaches to this phenomenon. These contributions are an important step in the research of null arguments and we hope that this volume will be a valuable addition to the literature on the structure and development of pronominal systems.

References Abney, S. 1987 Abraham, W. 1993 Adams, M. 1987

The English Noun Phrase in its Sentential Aspect. Ph.D. Diss.: MIT Cambridge, Mass. ‘‘Null Subjects in the History of German: From IP to CP’’. Lingua 89, 117–142.

‘‘From Old French to the Theory of Pro-drop’’. Natural Language and Linguistic Theory 5, 1–32. Alexiadou, A. & E. Anagnostopoulou 1998 ‘‘Parametrizing AGR: Word order, V-movement, and EPPchecking’’. Natural Language and Linguistic Theory 16, 375– 416. Ashby, W. 1977 Clitic Inflection in French. An Historical Perspective. Amsterdam: Rodopi. Bennis, H. & L. Haegeman 1984 ‘‘On the Status of Agreement and Relative Clauses in West Flemish’’. In: W. de Geest & Y. Putseys (eds.), Sentential Complementation. Dordrecht: Foris, 33–53. Bouchard, D. 1984 On the Content of Empty Categories. Dordrecht: Foris. Brandi, L. & P. Cordin 1989 ‘‘Two Italian Dialects and the Null Subject Parameter’’. In: O. Jaeggli & K.J. Safir (eds.), The Null Subject Parameter. Dordrecht: Kluwer, 111–142.

Introduction

17

Bresnan, J. & S.A. Mchombo 1987 ‘‘Topic, Pronoun, and Agreement in Chichewa’’. Language 63, 741–782. Cardinaletti, A. 1990 ‘‘Es, pro and Sentential Arguments in German’’. Linguistische Berichte 126, 135–163. Chao, W. 1980 ‘‘PRO-drop Languages and Nonobligatory Control’’. University of Massachusettes Occasional Papers in Linguistics 7, 46–74. Chomsky, N. 1965 Aspects of the Theory of Syntax. Cambridge: MIT Press. Chomsky, N. 1973 ‘‘Conditions on Transformations’’. In: S. R. Anderson & P. Kiparsky (eds.), A Festschrift for Morris Halle. New York: Holt, Rinehart & Winston, 232–286. Chomsky, N. 1981 Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. 1982 Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, Mass.: MIT Press. Chomsky, N. & H. Lasnik 1977 ‘‘Filters and Control’’. Linguistic Inquiry 8, 425–504. Chomsky, N. & H. Lasnik 1993 ‘‘Principles and Parameters Theory’’. In: J. Jakobs, A. von Stechow, W. Sternefeld & T. Vennemann (eds.), Syntax: An International Handbook of Contemporary Research. Berlin: de Gruyter, 506–569. De Crousaz, I. & Shlonsky, U. 2003 ‘‘The Distribution of a Subject Clitic Pronoun in a FrancoProvencal Dialect and The Licensing of Pro’’. Linguistic Inquiry 34, 413–442. den Dikken, M. 1999 ‘‘On the Structural Representation of Possession and Agreement’’. In: I. Kenesi (eds.), Crossing Boundaries. Advances in the Theory of Central Eastern European Languages. Amsterdam: Benjamins, 137–178. Dupuis, F. 1988 ‘‘Pro-drop dans les subordonne´es en ancien franc¸ais’’. Revue que´be´coise de linguistique the´orique et applique´e 7, 47–62. Franks, S. & N. Hornstein 1992 ‘‘Secondary Predication in Russian and Proper Government of PRO’’. In: R.K. Larson, S. Iatridou, U. Lahiri & J. Higginbotham (eds.), Control and Grammar. Dordrecht: Kluwer, 1–50.

18

Melani Wratil and Peter Gallmann

Fuß, E. 2005 Gilligan, G.M. 1987 Haegeman, L. 2004

Haider, H. 1990

The Rise of Agreement. A formal approach to the syntax and grammaticalization of verbal inflection. Amsterdam: Benjamins. A Cross-Linguistic Approach to the pro-drop Parameter. Ph.D. Diss., University of Southern California. ‘‘DP-Periphery and Clausal Periphery: Possessor Doubling in West Flemish’’. In: D. Adger, C. de Cat & G. Tsoulas (eds.), Peripheries. Syntactic Edges and their E¤ects. Dordrecht: Kluwer, 211–240. ‘‘Null Subjects and Expletives in Romance and Germanic Languages’’. In: W. Abraham, W. Kosmeijer & E. Reuland (eds.), Issues in Germanic Syntax. Berlin: de Gruyter, 49–66.

Hirschbu¨hler, P. 1995 ‘‘Null Subjects in Verb-First Embedded Clauses in Philippe de Vigneulles’’. In: Battye & I. Roberts (eds.), Clause Structure and Language Change. Oxford: Oxford University Press, 257–291. Hoeing, R.G. 1991 ‘‘Out of Control: Control Theory and its Implications for Empty Categories, Expletives and Missing Subjects in German’’. In: E.H. Antonsen & H.H. Hock (eds.), Stæfcræft. Studies in Germanic Linguistics. Amsterdam: Benjamins, 91–102. Holmberg, A. 2005 ‘‘Is there a little pro? Evidence from Finnish’’. Linguistic Inquiry 36, 533–564. Huang, J.C.-T. 1984 ‘‘On the Distribution and Reference of Empty Pronouns’’. Linguistic Inquiry 15, 531–574. Huang, J.C.-T. 1989 ‘‘Pro-drop in Chinese: A Generalized Control Theory’’. In: O. Jaeggli & K.J. Safir (eds.), The Null Subject Parameter. Dordrecht: Kluwer, 185–214. Hulk, A. & A. van Kemenade 1993 ‘‘Subjects, Nominative Case, Agreement and Functional Heads’’. In: Lingua 89, 181–215. Jaeggli, O. 1981 Topics in Romance Syntax. Dordrecht: Foris. Jaeggli, O. & K.J. Safir 1989 ‘‘The Null Subject Parameter and Parametric Theory’’. In: O. Jaeggli & K.J. Safir (eds.), The Null Subject Parameter. Dordrecht: Kluwer, 1–44.

Introduction Lehmann, C. 1988

19

‘‘On the Function of Agreement’’. In: M. Barlow & C.A. Ferguson (eds.), Agreement in Natural Language. Stanford: CSLI, 55–65. Manzini, M.R. & Savoia, L. 2002 ‘‘Parameters of Subject Inflection in Italian Dialects’’. In: Svenonius, P. (ed.), Subjects, Expletives, and the EPP. Oxford: Oxford University Press, 157–200. Mu¨ller, G. 2006 ‘‘Pro-drop and Impoverishment’’. In: P. Brandt, E. Fuß (ed.), Form, Structure, and Grammar. A Festschrift Presented to Gu¨nther Grewendorf on Occasion of His 60th Birthday. Berlin: Akademie Verlag, 183–203. Neeleman, A. & K. Szendro¨i 2007 ‘‘Radical pro-drop and the Morphology of Pronouns’’. Linguistic Inquiry 38, 671–714. Perlmutter, D. 1971 Deep and Surface Structure Constraints in Syntax. New York: Holt, Rinehart and Winston. Platzack, C. 2004 ‘‘Agreement and the Person Phrase Hypothesis’’. Working Papers in Scandinavian Syntax 73, 83–112. Poletto, P. 1995 ‘‘The Diachronic Development of Subject Clitics in North Eastern Italian Dialects’’. In: A. Battye & I. Roberts (eds.), Clause Structure and Language Change. Oxford: Oxford University Press, 295–324. Postal, P.M. 1969 ‘‘Anaphoric Islands’’. In: R.I. Binnick, A. Davison, G.M. Green & J.L. Morgan (eds.), Papers from the Fifth Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago University Press, 205–239. Renzi, L. 1992 ‘‘I pronomi soggetto di due varieta` substandard: fiorentino e francais avance´’’. Zeitschrift fu¨r romanische Philologie 108, 72– 98. Rizzi, L. 1982 Issues in Italian Syntax. Dordrecht: Foris. Rizzi, L. 1986 ‘‘Null Objects in Italian and the Theory of pro’’. Linguistic Inquiry 17, 501–558. Roberge, Y. 1990 The Syntactic Recoverability of Null Arguments. Kingston / Montreal: McGill-Queen’s University Press.

20

Melani Wratil and Peter Gallmann

Roberts, I. 1992

Verbs and Diachronic Syntax. A Comparative History of English and French. Dordrecht: Kluwer.

Rohrbacher, B. 1994 The Germanic VO Languages and the Full Paradigm: A Theory of V to I Raising. Ph.D. Dissertation. University of Massachusetts. Amherst. Rosenbaum, P. 1967 The Grammar of English Predicate Complement Constructions. Cambridge, Mass.: MIT Press. Rouveret, A. 1991 ‘‘Functional Categories and Agreement’’. The Linguistic Review 8, 353–387. Safir, K. 1985 Syntactic Chains. Cambridge: Cambridge University Press. Siewierska, A. 1999 ‘‘From Anaphoric Pronoun to Grammatical Agreement Marker: Why objects don’t make it’’. Folia Linguistica 33, 2, 225–251. Sigurðsson, H.A. 1989 Verbal Syntax and Case in Icelandic. Doctoral dissertation, University of Lund. Speas, M. 1994 ‘‘Null Arguments in a Theory of Economy of Projections’’. University of Massachusetts Occasional Papers 17, 179–208. Szabolcsi, A. 1990 ‘‘Is DP Analogous to IP or CP?’’. In: J. Payne (ed.), Proceedings of Manchester Colloquium on Noun Phrases. Den Haag: Mouton. Szabolcsi, A. 1994 ‘‘The Noun Phrase’’. Syntax and Semantics 27, 179–274. Taraldsen, K.T. 1978 On the Nominative Island Condition, Vacuous Application, and the That–Trace Filter. Blomington: Indiana University Linguistics Club. Vance, B. 1988 Null Subjects and Syntactic Change in Medieval French. PhD Diss.: Cornell University. Vikner, S. 1989 ‘‘Object Shift and Double Objects in Danish’’. Working Papers in Scandinavian Syntax, Vol. 44, 141–156. Wexler, K. 1994 ‘‘Optional Infinitives, Head Movement and the Economy of Derivations’’. In: D. Lightfoot & N. Hornstein (eds.), Verb Movement. Cambridge: Cambridge University Press, 305–350.

Pro-drop in the history of German – From Old High German to the modern dialects Katrin Axel and Helmut Weiß 1. Introduction1 Modern Standard German does not license (referential) pro-drop even though verbal inflection is relatively rich. It has therefore been claimed in the generative literature that in verb-second languages a syntactic prerequisite for the licensing of pro is not fulfilled (e.g. Jaeggli & Safir 1989; Rohrbacher 1999). What has often been disregarded, however, is that at the dialectal level the situation is entirely di¤erent. In German dialects (as, in fact, in all the major dialect groups within the Continental West-Germania) pro is licensed under certain conditions. Furthermore, referential pro-drop was possible in historical times, notably in earlier Old High German. In this paper we will compare the null-subject properties of Old High German to that of the modern German dialects. We shall argue that the seemingly di¤erent syntactic distribution of pro is only a surface phenomenon: The syntactic licensing conditions for pro are the same both in the historical and the current dialects. What has changed, however, is that the type of Agr necessary for the identification of pro has become more specific.

2. Old High German 2.1. Null-subject constructions Old High German is the oldest attested period of the German language. More precisely, the term refers to a group of dialects of the West Germanic

1. We are very grateful to the audience of the Kolloquium Nullpronomen (Jena, September 2006) and to an anonymous reviewer for their comments on this paper. Thanks also to the editors, Melani Wratil and Peter Gallman, for their help.

22

Katrin Axel and Helmut Weiß

branch of the Germanic family. In the earlier texts, referential subject pronouns are often not overtly realised:2 (1) a.

b.

In dhemu druhtines nemin archennemes . . . fater in the Lord’s name recognise-1.pl father ‘in the name of the Lord we recognise . . . the Father’3 In persona enim domini patrem accipimus (I 279f.) nu uuizumes thaz thu allu uueist now know-1.pl that you all know ‘now we know that you know all things’ nunc scimus quia scis omnia

(T 289, 25)

Referential null subjects only occur in the 8th and 9th century texts. In late Old High German (11th century) they are hardly found anymore. What is a drawback for syntactic research is that the written attestation of Old High German comprises hardly any autochthonous prose texts. Our data here have been gained from the Old High German Isidor translation (ca. 790/800, South-Rhine-Franconian), the Monsee Fragments (ca. 800, Bavarian) and from the Tatian (ca. 830, East Franconian). These texts are all translations from Latin sources. As can be seen in the examples in (1) and in the following examples, the Latin source sentences usually do not contain an overt subject pronoun either since Latin was a prodrop language. However, as will be argued below, the fact that null subjects have a special syntactic distribution shows that they are not merely a syntactic loan. 2.2. Syntactic distribution of Old High German null subjects The syntactic distribution of null subjects in the earlier Old High German prose texts is characterised by an asymmetry between main and subordinate clauses (cf. also Axel 2005, 2007: Chapter 6; Axel and Weiß 2010 for a more comprehensive discussion). It should be noted that in Old High German, main and subordinate clauses are already distinguished 2. Furthermore, quasi-arguments and arbitrary subject pronouns could be omitted (cf. Axel 2007: Chapter 6). See also Abraham (1993) on non-referential null subjects in Old High German. 3. The English translations of the Isidor examples have been taken from Robinson (1997).

Pro-drop in the history of German

23

structurally by the position of the finite verb. As in present-day German, verb placement in Old High German is ‘asymmetric’: In independent clauses (without complementisers) there is V-to-C-movement, whereas in subordinate clauses introduced by a complementiser the verb remains in its base position at the end of the clause (see also Axel 2007: Chapter 2). Interestingly, referential null subjects are almost exclusively attested in root sentences, more precisely in sentences with verb movement. The examples in (2) to (4) illustrate this for independent declarative sentences with verb-second order, for wh-interrogatives and for yes/no-interrogatives. (2) a.

b.

[ CP Sume i [ C’ [ C hahetj ] [t i in cruci t j ]]] some-acc hang-2.pl to cross ‘some of them you will crucify’ et ex illis . . . crucifigetis (MF XVIII, 17f; Mt 23: 34) [CP [In dhemu druhtines nemin] i [C’[ C archennemes j ] [. . . t i fater t j ]]] in the Lord’s name recognise-1.pl father ‘in the name of the Lord we recognise . . . the Father’ In persona enim domini patrem accipimus (I 279f.)

(3) [ CP uuannei [ C’ [ C gisahumes j ] [thih t i t j ]]] when saw-1.pl you-acc ‘when did we see you?’ quando te uidimus (4) [ CP [ C’ [ C quidis i ] [zi uns thesa parabola t i ]]] say-2.sg to us this parable ‘are you telling this parable to us?’ ad nos dicis hanc parabolam

(T 267, 11)

(T 259, 2)

Unlike in canonical null-subject languages, where the subject pronoun can only be overtly realised in case of an emphatic interpretation, in Old High German sentences with V-to-C-movement the subject pronoun could also be realised overtly without evoking a special semantic-pragmatic e¤ect. The overt and the null variant systematically co-occur even within individual texts. As can be seen in Table 1, in Isidor and Tatian, roughly 40 per cent and in the Monsee Fragments about two thirds of main clauses with pronominal subjects exhibit subject omission (cf. Axel 2005, 2007: Chapter 6; Axel and Weiß 2010).

24

Katrin Axel and Helmut Weiß

Table 1. Overt/null-subject pronoun use in main vs. subordinate clauses in three 8th and 9th century prose texts4 Isidor

Monsee Fragments

Tatian

pronoun subject

pronoun subject

pronoun subject

overt

null

overt

null

overt

null

main

61 (56%)

48 (44%)

48 (36%)

84 (64%)

1434 (60%)

960 (40%)

subordinate

85 (91%)

8 (9%)

73 (85%)

13 (15%)

1180 (92%)

95 (8%)

clause type

In subordinate clauses null subjects are much rarer. Here the rate of subject omission is low, which is between 9 and 15 per cent in all three texts (cf. Table 1). In the complex sentences in (5), for example, subject pronouns have been inserted into the middle field of the verb-final subordinate clauses (in square brackets), whereas they have been left out in the respective main clauses. In each case, neither the subordinate clause nor the main clause in the Latin source contains an overt subject pronoun. (5) a.

[Dhar ir auh quhad >. . .< ,] dhar meinida leohtsamo zi where he also said there meant easily to archennenne dhen heilegan geist recognise the Holy Spirit ‘where he also said >. . .< , he clearly meant the Holy Spirit’ Item dicendo >. . .< sanctum spiritum euidenter aperuit (I 274¤.)

b.

Enti [so aer  danan fuor]  quuam in iro  dhinchu#s . . . and when he thence went came to their synagogue ‘when he had departed from there, he went into their synagogue’ Et cum inde transisset, uenit in synagogam eorum . . . (MF IV, 19; Mt 12: 9)

The main/subordinate asymmetry shows that Old High German null subjects depend on verb fronting. Apparently, they are only licensed in a configuration where they are c-commanded by the agreement features of

4. The figures have been calculated on the basis of Eggenberger (1961). The Bavarian version of the Isidor, which is part of the Monsee Fragments, has not been included.

Pro-drop in the history of German

25

the finite verb.5 In Old High German, the only way to obtain the configuration in (6) is via verb movement to C  . (6) [V þ Agr] i pro t i The situation in Old High German is certainly not exceptional. Barbosa (1995), for example, presents cross-linguistic evidence from modern Italian and from the historical stages of various Romance6 and Germanic languages, where null subjects only occur in post-finite environments. She even proposes (p. 80) that there is a general connection between XP–V–Sorder and the possibility of subject drop. It should be noted that many scholars, for example Grimm ([1898] 1967: 235) and Paul ([1919] 1968: 22), have claimed that referential null subjects had largely died out before the earliest attested stages of Old High German, even though there is a large consensus that they were present in Proto-Germanic and in Proto-Indo-European.7,8 The phenomenon of null subjects in Old High German has often been dismissed as a phenomenon of loan syntax (see Eggenberger 1961; Hopper 1975: 80f.; Schrodt 2004: 73 § S68). Proponents of the loan-syntax hypothesis have argued that there are virtually no cases in the Old High German translations where an Old High German null subject does not correspond to a Latin empty subject in the first place (cf. Meillet 1908/1909: 89; Hopper 1975: 81). However, this is not really surprising if one takes into account that in Latin, a full pro-drop language, the non-realization of the subject pronoun is the rule. So the chances of detecting a pattern of subject omission 5. It could be objected that this post-finiteness condition is not absolute since the rate of null subjects in the three texts is between 9 and 15 per cent in subordinate clauses (cf. Table 1) according to Eggenberger (1961). Note, however, that Eggenberger’s figures for subordinate clauses also include special types of subordinate clauses (e.g. causal clauses introduced by huuanta ‘because, for, since’) which show clear indications of verb-fronting. Also, in Old High German the amount of ambiguous word orders was higher than, for example, in modern German. There are thus quite a few cases that Eggenberger has classified as subordinate clauses which could receive an alternative analysis as main (or subordinate) clauses with V-to-C movement. See also Axel (2007: Chapter 6) for further comments on Eggenberger’s classifications. 6. See also Adams (1987) and Vance (1997) on pro-drop in Medieval French. 7. Behaghel (1928: 443) notes that 1st and 2nd person subject pronouns have partly been omitted in Isidor, Tatian, and in Otfrid. He considers these cases as exceptional and is not sure whether they should be regarded as relics of an archaic stage or as a more recent development. 8. See also Schulze (1924) and Ferraresi (2005) on null subjects in Gothic.

26

Katrin Axel and Helmut Weiß

in the Old High German translations which is independent of the Latin are very low from the start. While it is true that almost all cases of omitted subject pronouns in Old High German correspond to pro-drop in the Latin, the main-subordinate asymmetry shows that the distribution of referential null subjects in Old High German does not simply mimic the situation in the Latin where pro-drop is also the standard in subordinate clauses. Further counterarguments against the loan-syntax hypothesis can be found in Axel (2005, 2007: Chapter 6). 2.3. Morphological distribution of null subjects Since Taraldsen (1978), it has been argued in the generative literature that there is a connection between the ‘richness’ of a morphological paradigm and pro-drop. Moreover, the introduction of overt subject pronouns in early Old High German has traditionally been linked to the weakening of inflectional endings as a consequence of the introduction of word-initial accent (see Ko¨gel 1882: 127; Held 1903: XIII.; Behaghel 1928: 442). As is objected by Grimm ([1898] 1967: 235) and Eggenberger (1961), these phonological and subsequent morphological changes did not lead to a substantial levelling of inflectional distinctions in the verbal paradigm during the Old High German period. It should also be noted that from a cross-linguistic perspective even the inflectional system of present-day German would still be su‰ciently ‘strong’ for referential pro-drop to be possible according to standard assumptions. However, there is one interesting piece of evidence which might corroborate the hypothesis that there is a connection between rich inflection and pro-drop (cf. Axel & Weiß 2010). In earlier Old High German the 1st person plural has two alternative endings: a short ending -m (or -n) and a long ending (-me#s). Pronouns occurring with the short ending are virtually always overt. With the long ending, the subject pronoun is often omitted, but only if it occurs in post-finite position. Note that the long and the short endings are found within individual texts (e.g. in Tatian or Isidor). (7) a.

Endi dhes selben christes, . . . , chichundemes auh nu and of-that same Christ demonstrate-1.pl also now dhes e˛dhili endi odhil of-that lineage and home ‘We now demonstrate the lineage and homeland of that same Christ’ et cuius . . . demonstretur et genus et patria (I 520¤.)

Pro-drop in the history of German

b.

Chiuuisso chio¤anodom uuir nu hear dhazs . . . surely revealed-1.pl we now here that ‘surely we have now revealed here that . . .’ Probauimus . . .

27

(I 484f.)

Harbert (1999: 263) speculates that the long ending might be su‰ciently strong to license pro-drop in contrast to the shorter one since it can be argued that it separately encodes person and number. He recounts the proposal by Shields (1996) that -me#s consists of both a number-neutral 1st person marker -m and a plural desinence -e#s from the perspective of its Indo-European etymology. What speaks against this hypothesis is that if the separate encoding of person and number was really a necessary condition for the licensing of pro-drop in Old High German, then we would expect it to be ruled out in most other persons/numbers, which is not the case. It should also be noted that during the Old High German period, the short ending -m was replaced by -n, a development which led to some syncretism between the 1st and 3rd person plural in the present and preterite conjunctive of both strong and weak verbs. This might explain why prodrop is ruled out after the short ending -n. However, pro-drop is also already absent in the context of the older ending -m, which does not give rise to such syncretism. In Isidor, for example, a late 8th century text, the short ending is still -m, as can be seen in (7)-b. Thus, there is no syncretism between the 1st and 3rd person plural in the present and preterite conjunctive. Nevertheless, in sentences with the short ending -m subject pronouns are always overt in this text. As was already observed by Paul (1877: 421¤.), after the long ending (-me#s) the subject pronoun is (virtually) always omitted. The overt pronoun (¼uuir) only occurs after the short ending -m/n or in pre-finite position. Paul thus suggests that -me#s is not an inflectional su‰x, but rather a kind of clitic subject pronoun. This would explain why -me#s could not be followed by a further subject pronoun (¼uuir). However, this explanation is hardly convincing since the etymology of -me#s is entirely unclear. As Roberts (1935: 224) objects ‘‘the pronominal hypothesis, as formulated by Kuhn and Ko¨gel [and Paul, Axel & Weiß], is not valid. No form has existed in Germanic which could provide a pronominal foundation for Old High German -me#s. There is, to be sure, excellent reason to believe that all personal endings were originally pronouns; but they must have been appended to the verbal roots in primordial IE times, doubtless many thousands of years before the separation of the several languages from the mother speech’’. In section 4.2 we

28

Katrin Axel and Helmut Weiß

will nevertheless put forward the hypothesis that the Old High German speakers might have reanalysed -me#s as being of pronominal nature. As will be argued below, pronominal agreement is indeed a prerequisite for pro-drop in contemporary dialects. For example, in modern Bavarian, pro-drop is possible in the 2nd person singular, which ends in /st/. /t/ is a pronominal element. Thus it can argued that the presence of pronominal agreement makes it possible for pro to be identified in the 2nd person singular in contrast to the 1st person singular, which does not exhibit pronominal agreement (see below). In earlier Old High German, the 2nd person singular still frequently showed the ending -s and thus pronominal agreement had not yet fully evolved. According to standard assumptions, the ending of the 2nd person singular (¼-st), which originally was only -s, originated from cases where the verb was followed by the clitic pronoun thu, du. In this configuration, the word boundary was then reanalysed in such a way that the dental plosive became part of the verbal ending. Already in Tatian, expressions such as gisihist tu ‘‘see you-2.sg’’ are attested. In Christus und die Samariterin, forms such as ke¨ro#st thu ‘‘choose/desire you’’ can be found alongside forms such as thu wissı #s ‘‘you-2.sg know’’  (Braune 2004: 261; § 306, n 5). It could be speculated that pro could not be identified by the original -s ending, which was still frequent at the time. However, this prediction is not fulfilled. As examples such as (8) suggest, see also (4) above, pro-drop was possible with the original, ‘non-pronominal’ ending as well. Such examples occur in all major prose texts from the 8th and 9th centuries. (8) Ni ghibis dhinemu heileghin zi chisehanne unuuillun neg give your holy-one to see corruption You will not give to your holy one to see corruption Nec dabis sanctum tuum uidere corruptionem

(I 721f.)

Furthermore, the 1st and 3rd person singular do not show any pronominal agreement at all; nevertheless, subject drop is also attested in these cases. Notably 3rd person subjects pronouns are frequently omitted in the 8th and 9th century texts. This might be due to the fact that they can be used anaphorically. SigurÐsson (1993) shows that in Old Icelandic, subject and object pro are licensed under free coindexing with a DP in the preceding discourse. Note, however, that the surface distribution of subject omission in Old Icelandic is almost reversed as compared to Old High German. As Sigurðsson’s material shows, in Old Icelandic, the condition of free coindexing is often satisfied in complex clauses with postposed subordinate

Pro-drop in the history of German

29

clauses since in this configuration the subject pronoun can often be coindexed with a DP in the matrix clause. In Old High German, by contrast, null subjects are largely absent from subordinate clauses as we have demonstrated above. Also, as can be seen in (9)-a, the (overt) subject pronoun of the subordinate clause (in square brackets) and the null subject of the main clause are not always coreferential. Nor is it the case that the main clause containing the null subject always follows the subordinate clause with the overt subject pronoun in complex sentences: In (9)-b, for example, it is the subordinate clause which is postposed. (9) a.

. . . zuuene man stu´ontun nah In. In scı´nentemo giuu´ate. two men stood-3.pl to them in shining garment [mittiu sio tho forohtun Inti helditun iro annuzi as they then feared-3.pl and bowed-3.pl their faces In erda] quadun ziı´n . . . to earth said-3.pl to-them ‘two men stood by them in shining garments. As they (¼the women) were afraid, and bowed down their faces to the earth, they (¼the men) said unto them’ . . . duo uiri st&erunt secus illas In ueste fulgenti Cum timerent autem & declinarent uultum In terram dixerunt ad illas. (T 324, 23¤.)

b.

niquam zi thiu [thaz ih giladoti rehte neg-came-1.sg to that that I called-conj righteous ouh suntige zi riuuu.,] but sinners to repentance ‘for I have not come to call the righteous, but sinners to repentance’ Non enim ueni uocare iustos sed peccatores in poenitentiam; (T 91, 24f.)

It should also be noted that in the Old High German texts, null subjects do not always have an antecedent. For example, they also occur discourseinitially and they do not necessarily require narrative discourse topicality. However, there seems to be at least a clear tendency to use an overt subject pronoun when a change of subject reference occurs (see Lippert 1974: 35). As has been shown, subject omission is very frequent in 8th and 9th century texts. In the late Old High German period subject pronouns are

30

Katrin Axel and Helmut Weiß

almost always realised overtly: Notker’s texts from the 11th century only contain a very small number of omitted subject pronouns. The same is true for Williram von Ebersberg’s paraphrase of the Song of Song, a late eleventh century text. According to the secondary literature, pro-drop is found only sporadically in Middle High German texts. Interestingly, nullsubject usage was quite frequent again in the Early Modern High German period.9 However, the surface distribution of omitted subject pronouns at the time was somewhat di¤erent from that of the Old High German period (see fn. 15). Instead, null-subject usage then already showed a similar distribution as in the modern dialects. The syntax and morphology of pro-drop in the dialects are discussed in the following section.

3. Modern dialects At first sight, the modern dialects of German show an entirely di¤erent pattern than the Old High German dialects: Pro-drop occurs predominantly in non-root clauses. The development of this pattern is the topic of the following sections. 3.1. Complementiser agreement One of the most curious features of modern Continental West-Germanic (and hence German) dialects is complementiser agreement (cf. Weiß 2005; Zwart 2005, 2006).10 As the sentences in (10) demonstrate, in addition to the verb, the complementiser agrees with the subject: (10) a. b.

wennst willst if-2.sg want-2.sg

(Bavarian, Weise 1907: 204)

waaln mer graad besamn senn because-1.pl we at-the-moment together are-1.pl (East Franconian, Weise 1907: 204)

9. Held (1903: 111) argues that the di¤erence between Middle High German and Early New High German is partly due to the fact that in the latter case, text genres had developed that were closer to oral communication. Middle High German court language, by contrast, was very remote from popular speech and strongly shaped by literary conventions. 10. According to Zwart (2006: 69), ‘‘complementizer agreement of the type observed in Continental West-Germanic is typologically extremely rare, possibly unique (Ken Hale, p.c.)’’.

Pro-drop in the history of German

31

Due to the lack of systematic investigations it is hard to give the precise areal extension of complementiser agreement. Judging from the data and remarks given in traditional dialect grammars (cf. Weiß 2005), it seems to occur (or to have occurred) in all major dialect groups within the Continental West-Germania, though there are some dialects which appear to not show it.11 Variation can be observed with regard to the paradigmatic extension of complementiser agreement. While many dialects seem to exhibit it only in the 2nd person singular as for example West Frisian and Low German dialects, cf. (11), there are others, such as Upper Saxon or East Franconian, which additionally show morphologically visible complementiser agreement in the entire plural; cf. (12).12 (11) a. b.

dat-st juˆn komst that-2.sg tonight come-2.sg vanste if/when-2.sg-you

c. datste that-2.sg-you

(Frisian, Zwart 1993: 253) (Soest, Holthausen 1886: 89)

(Mu¨hlheim at the Ruhr, Maurmann 1898: 68)

(12) a.

2.sg: 1.pl: 2.pl: 3.pl:

vu:sde o¨penmer venter vi:nze;

‘where-2.sg-you’ ‘whether-1.pl-we’ ‘if-2.pl-you’ ‘how-3pl-they’ (Upper Saxon, Richter 1979)

b.

2.pl: 1.pl: 2.pl: 3.pl:

wailst-da` wailn me¨ wailt-e¨ wailn-sa

‘because-2.sg-you’ ‘because-1.pl we’ ‘because-2.pl-you’ ‘because-3.pl-they’ (East Franconian, Rowley 1994)

11. An uncertain case is Alemannic, which probably does not show complementiser agreement, as the reviewer objects (cf. Weiß 2005 for an alternative analysis of the relvant Alemannic data). Note however, that dialects with pro-drop in root sentences, but without complementiser agreement (as Alemannic is according to the reviewer) pose no problem whatsoever for our approach. 12. Further information is given in Weiß (2005).

32

Katrin Axel and Helmut Weiß

3.2. Double agreement Two further issues that are related to complementiser agreement are of relevance for our purposes. The first one is double agreement (cf. Zwart 1993), which is, as we shall argue below, a necessary requirement for the development of complementiser agreement (besides V-to-C raising). Double agreement means that there are two possibilities to inflect the verb and that it depends on the position of the verb which inflectional marker appears: When the verb occurs in final position it bears another inflection than when it occurs in second position. This is illustrated in (13) and (14) with examples from Central and North Bavarian. (13) a. b. c.

dasma mia aaf Minga fahrn/*ma that-1.pl we to Munich go mia fahrma/*n aaf Minga fahrma/*n mia aaf Minga?

(14) a.

wents neat tiets a# tepa# if-2.pl not you too with b. weits iwet pruk when-2.pl over-the bridge

(Central Bavarian)

wa#e t were khumt, seatses wı #etshaus come see-2.pl-the pub (Sangerberg, Pfalz 1918: 18)

Interestingly, in these dialects the inflection on the verb in second position and on the complementiser is identical, whereas the verb in final position bears a di¤erent inflectional marker.13 There is a second type of doubleagreement dialect where the inflection depends on the position of the pronominal subject, that is there is a di¤erent inflection when the pronominal subjects follows the verb in second position. Again, the important point is that this special inflection is identical with complementiser inflection (cf. Goeman 2000); cf. (15): (15) a.

datte wij speult/*e that-1.pl we play b. Wy speult/*e c. Waar speule/*t wij? Where play we

(East Netherlandic)

13. We will use the term complementiser agreement for inflection on complementisers and on verbs in C , in contrast to the traditional verbal agreement which is sometimes called subject agreement, cf. van Koppen (2005).

Pro-drop in the history of German

33

Regarding the emergence of double agreement there is synchronic variation in Bavarian which reveals one possible way as to how the alternative set of inflectional markers might have developed (cf. Weiß 2005). The relevant case is the 1st person plural in Central Bavarian, where we have the two inflectional markers -(a)n and -ma. The former occurs in verb-final sentences and the latter in verb-second sentences and on complementisers, that is in C  : (16) a. b. c.

dasma mia aaf Minga fahrn/*ma that-1.pl we to Munich go/drive mia fahrma/*n aaf Minga fahrma/*n mia aaf Minga?

(Central Bavarian)

The source of the inflectional marker -ma is the pronominal subject mia in its clitic form ma which cliticised onto the verb in second position and was eventually reanalysed as verbal inflection. Phonological processes like assimilation may have made the process of reanalysis possible. Supposedly there was a development similar to fahrn’ma > fahrm’ma > fahrma, in the course of which alveolar /n/ first assimilated to bilabial /m/, and then vanished completely. In modern Central Bavarian, -ma is obligatory even when C  is filled with a complementiser instead of a verb, and it co-occurs with the full pronoun mia; cf. (16)-a above. These properties demonstrate that the complementiser agreement morpheme -ma is an inflectional su‰x (rather than a subject clitic). As mentioned before, in the special case of the 1st person plural, we can observe intra-dialectal variation in Bavarian. In the North Bavarian variant spoken in the northern part of the Upper Palatinate (Bachmann and Scheuerer 1995, Bachmann 2000), -ma is only a subject clitic which cliticises onto the complementiser or onto the verb in inversion structures, as shown in (17)-a, -b. It can be replaced by the full form, cf. (17)-c, but the two forms are never allowed to appear together in the same sentence; cf. (17)-d.14

14. (17)-a, b are taken from (Bachmann 2000, 99, 114), (17)-c, d are formed according to his remark that it is not possible to double a full pronoun by a clitic, with the exception of the 2nd person singular/plural (cf. Bachmann 2000: 100).

34

Katrin Axel and Helmut Weiß

(17) a. b. c. d.

wuo ’me han where- we are priÐme bring-we wuo mie han where we are *mie priÐme we bring

In the southern part of the Upper Palatinate, however, there is evidence that the subject clitic is being reanalysed as an inflectional su‰x, since it can be doubled by the full form in main clauses (cf. Bachmann and Scheuerer 1995, Rowley 1994), though not yet in embedded clauses; cf. (18): (18) a.

mia liÐma we lie-we/lie-1.pl b. *wa’ma mia Hunt hand because-we we dogs are

In the next step, -ma is no longer a subject clitic, but an inflectional element that cannot be absent even when the full pronoun is present, cf. (19), which means that complementiser agreement is necessary. Note that it is still restricted to the comp position (from where it can license pro-drop). This state of a¤airs is attained in Central Bavarian. (19) a.

wama (mia) Hunt hand because-1.pl we dogs are b. *wa mia Hunt hand because we dogs are

In the final step, the new inflectional su‰x eventually appears on the verb in final position as well; cf. (20): (20) wama bmaid hama because-1.pl thought have-1.pl

(Kollmer 1987: 177)

This is the case in some variants of Central and Southern Bavarian (Kollmer 1987: 358; Wiesinger 1989). However, even in these dialects the new inflectional su‰x is restricted to auxiliaries and some modal verbs, but never occurs on lexical verbs in final position.

Pro-drop in the history of German

35

To recapitulate so far: The new inflectional morpheme is confined to C  (or to any other head within an extended C-projection which hosts complementizers or finite verbs in complementary distribution) and it has developed out of subject clitics. The last step, which consists of giving up the restriction to C  , occurred only rarely. However, it did occur, for example, in all (or most) German dialects in the 2nd person singular, and in Bavarian in the 2nd person plural (cf. Weiß 1998, 2005). 3.3. pro-drop The second relevant issue in connection with complementiser agreement is pro-drop. Though the Continental West Germanic dialects are only partial pro-drop-languages in the sense defined in Weiß (2001), they exhibit undisputable instances of pro-drop both in verb-second as well as in verbfinal sentences introduced by inflected complementisers. In Bavarian, for instance, pro-drop is possible in the 2nd person singular15, but not in the 1st person singular; cf. (21)-a, -b vs. (21)-c, -d (cf. also Bayer 1984; Fuß 2005; Axel (Weiß 2010).

15. 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: (i)

was fragst? what ask-2.sg ‘‘why are you asking?’’ ‘why are you asking?’ (Hans Sachs, Fasnachtsspiele 4, 34, 30; from Held 1903: 134)

(ii)

wie bist erplichen? how are-2.sg 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. In contrast to Old High German, 3rd person subject pronouns are hardly omitted in Early Modern High German texts.

36

Katrin Axel and Helmut Weiß

(21) a. b.

(I glaub,) moang bisd wieda gsund (I think) tomorrow are-2.sg again healthy wennsd moang wieda gsund bisd if-2.sg tomorrow again healthy are-2.sg

c. (I glaub,) moang bin-*(e) wieda gsund (I think) tomorrow am-I again healthy d.

wenn-*(e) moang wieda gsund bin if-I tomorrow again healthy am

Since the crucial di¤erence between the 1st and 2nd person singular in Bavarian is that only the latter exhibits complementiser agreement, we can assume that the presence of overt Agr-in-C is necessary for pro-drop. Whether Agr-in-C is the result of verb raising to C or of inflected complementisers turns out to be irrelevant for pro-drop; cf. (21)-a and -b. However, Agr-in-C alone does not su‰ce to license pro-drop. Consider the following contrast, which, according to Weise (1900: 56), shows up in the Thuringian dialect spoken in Altenburg (cf. Axel & Weiß 2010). (22) a.

schreib mir einmal den Brief, kriegst write-imp-2.sg me once the letter, get-2.sg pro auch einen Groschen too a penny ‘If you write me the letter, you’ll get a penny’ (Altenburg (Thuringian))

b.

schreibt mir einmal den Brief, *kriegt pro auch write-imp-2.pl me once the letter, get-2.pl too einen Groschen a penny

Though Altenburgian Thuringian exhibits complementiser agreement in the 2nd person singular and plural (and in the 3rd person plural, cf. Weise 1900: 57), pro-drop is only possible with the 2nd person singular. It is not possible with the 2nd person plural, as the contrast between (22)-a and -b reveals. The di¤erence between the inflectional marker -st in the 2nd person singular and the -t in the 2nd person plural is that only the former contains a pronominal element, namely /t/ (see above). As shown in Weiß (2005) in detail, it is precisely pronominal Agr-in-C that licenses pro. As Weise (1907: 204) already notes, pronominal subjects

Pro-drop in the history of German

37

can be absent in the 2nd person singular and plural in those Bavarian and East Franconian dialects where complementiser agreement is pronominal, cf. (23), in contrast to the 1st and 3rd person plural in East Franconian, which exhibit non-pronominal complementiser agreeement; cf. (24) (cf. Axel and Weiß 2010). (23) a. b.

wennsd pro mogsd if-2.sg like-2.sg wennds pro megds if-2.pl like-2.pl (Central Bavarian, likewise North Bavarian and East Franconian)

(24) a.

waaln because-1.pl b. waaln because-3.pl

*(mer)

graad besamn senn we at-the-moment together are-1.pl *(se) graad besamn senn they at-the-moment together are-3.pl (East Franconian, Weise 1907: 204)

On the basis of the data discussed here (and comparable ones, cf. Weiß 1998, 2005) we can postulate the following licensing principle for prodrop in modern German dialects: (25) pro must be c-commanded by pronominal agreement This principle explains why pro is also possible in Frisian and in West Flemish; cf. (26a, b):16 (26) a. b.

datst (do) juˆn komst that-2.sg you tonight come-2.sg da-n-*(ze) goan werk een that-3.pl-they go-3.pl work have

(Frisian) (West Flemish)

4. From Old High German to the modern dialects As we have seen, there are three major di¤erences between Old High German and the modern dialects: (i) in Old High German, pro-drop is 16. This does not exclude the existence of additional language or dialect specific conditions (as in Frisian, cf. Zwart 1993) and/or problematic cases (cf. Weiß 2005).

38

Katrin Axel and Helmut Weiß

confined to root sentences, whereas in the modern dialects, it also occurs in embedded sentences introduced by inflected complementisers; (ii) in the modern dialects pro-drop depends on pronominal agreement, whereas in Old High German it did not; (iii) the modern dialects exhibit complementiser agreement and Old High German did not. What happened in the development from Proto Germanic with unrestricted pro-drop to the modern German dialects with partial pro-drop was that the licensing conditions became more specific (cf. Axel & Weiß 2010). As we have seen in section 2, at first pro-drop in Old High German involved V-to-C raising (hence the restriction to root sentences), which made it possible for Agr-in-C to c-command and thus license pro. In modern dialects, this c-command requirement still holds, but it is more restricted because only pronominal Agr-in-C is able to license pro-drop. However, there was an additional change – the development of complementiser inflection – which cancelled the restriction to root sentences so that pro-drop can now occur in embedded sentences as well (in contrast to Old High German). The restriction to pronominal Agr and the emergence of complementiser inflection are thus the two developments in need of an explanation. Both developments were triggered by the same phenomenon – (the emergence of ) double agreement – as we will argue in the next sections. If this explanation is right, it would have been indeed a small change in the underlying grammar which yielded greater di¤erences on the surface. 4.1. The emergence of double agreement As we have seen in section 2, in Old High German there was already double agreement in the 1st person plural: an older long ending (-me#s) and a newer short one (-m). Regarding the syntactic distribution, both forms occurred in verb-second/first as well as in verb-final sentences. Whatever the origin of -me#s might have been, it is noteworthy that it correlates with pro-drop in verb-second/first contexts where the subject pronoun would occur post-finitely if it was overtly realised (i.e. in ‘inversion’ contexts). Though the distribution of the long and short ending of the 1st person plural (i.e. -me#s vs. -m) was initially not correlated with the position of the finite verb, this was the case with the successor forms. The short form -m, which was originally restricted to indicative present but occurred later elsewhere, was substituted by the short form -m in all contexts. The course

Pro-drop in the history of German

39

of the 9th century (in its reduced form -n).17 Soon after the replacement of -me#s, a new type of double agreement for the 1st person plural began to emerge, which was then characteristic for the Middle High German period (cf. Axel & Weiß 2010). When the pronominal subject immediately followed the verb in second/first position, the -n was dropped (cf. Paul 2007: § E 21, number 2).18 According to Braune (2004: 263), the first occurrences of n-drop, cf. (27) for an example from Tatian, can be traced back to the 9th century, that is, it seemed to occur as soon as -n had started to replace -me#s. (27) thesan niuuizuuuir uuanan her ist this-acc neg-know-we whence he is ‘but as to this man we do not know where he is from’ hunc autem nescimus unde sit.

(T 223, 10)

This new double agreement in the 1st person plural was then definitely related to verb position (and subject-verb-inversion). The most prominent case of double agreement that was related to both verb placement and subject clitics is the 2nd person singular, whose original ending consisted only of an -s to which then a -t (, the author C the Mary every book C he writes sofort chau¤t immediately buys ‘the author such that Mary immediately buys every book he writes’ (CNPC) d.

de Sa¨nger, won i mi fro¨i, the singer C I me be.happy

when one him on TV brings ‘the singer such that I am happy when they show him on TV’ (adjunct island)

Importantly, such structures are not perceived as repairs and should therefore not be equated with intrusive pronouns in English, cf. Chao & Sells (1983). Corresponding wh-extractions are strongly ungrammatical (and would not improve with resumptives), cf. Salzmann (2006b: 331): (3) a. *[Vo wem]1 bisch i d Schuel? of who.dat are with a sister in the school lit.: ‘Who did you go with a sister of to school?’ (PP-island) ¨ ber wele Sportler]1 ha¨t vil Erfolg ghaa? b. *[U about which athlete has the biography much success had lit.: ‘Which athlete did the biography about have a lot of success?’ (subject island)

146

Martin Salzmann

c. *[Wele Autor]1 chau¤t d Marie < jedes Buech, wo __1 schriibt>? Which author buys the Mary every book C writes lit.: ‘Which author does Mary buy every book that writes?’ (CNPC) d. *[Wele Sa¨nger]1 fro¨isch di, ? which singer be.happy.2s you when one on TV brings lit.: ‘Which singer are you happy when they show on TV?’ (adjunct island) 1.1.2. Other types of A’-movement Comparative deletion shows the same distribution of resumptive pronouns as relativization, cf. Salzmann (2006b). Here is a pair contrasting direct and indirect object: (4) a.

b.

Es sind mee Patiente choo als de Tokter there are more patients come than the doctor (*sie) ha¨t cho¨ne behandle. (them) has could treat ‘There came more patients than the doctor could treat.’

ACC

Es sind mee Lu¨u¨t choo als *(ine) there are more people come than they.dat de Tokter ha¨t cho¨ne Medikama¨nt ge¨e¨. the doctor has could medicine give ‘There came more people than the doctor could give medicine to.’

DAT

Wh-movement and topicalization, however, pattern di¤erently. Resumption is categorically ruled out in local dependencies for all grammatical relations. The following pair illustrates extraction of a dative object (Salzmann 2006b: 376f.):6

6. In long-distance movement, PPs and indirect objects leave gaps while with subjects and direct objects there is some variation, cf. Weber (1965: 304). The latter may, however, be an instance of the so-called A’-splits discussed in Salzmann (2006b: 376: fn. 297) and also in 4.2 below.

Silent resumptives in Zurich German possessor relativization

(5) a.

b.

147

Welem Maa ha¨sch __ /*em es Buech gge¨e¨? which.dat man have.2s he.dat a book given? ‘To which man did you give a book?’ Dem Bueb han i __/*em es Buech gge¨e¨. that.dat boy have I he.dat a book given ‘To that boy, I gave a book.’

1.2. Resumption as a last resort In the analysis of resumption one can, in principle, distinguish two perspectives: According to the first, resumption is simply taken to be a possibility in a given language (e.g. McCloskey 1990, Sun˜er 1998, Willis 2000, Bianchi 2004); in languages of this type, resumptive and gap relatives exist side by side, at least in certain environments. None of the strategies is thus privileged. Alternatively, resumption is analyzed as a last resort that only comes into play when other syntactic mechanisms (leading to gap relatives) fail (e.g. Shlonsky 1992, Pesetsky 1998, Toman 1998, Boeckx 2003, Alexopoulou 2006, Guilliot 2006, Rouveret 2008). I believe that both perspectives are necessary to account for cross-linguistic variation. In languages where gaps and resumptives have the same distribution, the first position seems more adequate (but see Shlonsky 1992) while in languages where gaps and resumptives are in complementary distribution the second view seems preferable (cf. also Salzmann 2009a). In ZG the situation is quite clear. Resumptives and gaps are in complementary distribution (but recall fn. 4), which is an argument for a last resort approach. Next to the asymmetry subject/direct object vs. the other positions there is a similar contrast when non-individual-denoting types are relativized. In the following pair a predicate is relativized on; in one case originating in a non-oblique position, in the second case within a PP (i.e. within an island). While resumption is impossible in the first case, it is obligatory in the second (cf. also Salzmann 2006b):7

7. In the b-example the resumptive is an R-pronoun, the pronominal part of a pronominal adverb. Pronominal adverbs occur if prepositions take an inanimate pronominal complement (cf. Salzmann 2006b for a more careful statement). Consequently, de-appears instead of das.

148

Martin Salzmann

(6) a.

b.

Er isch de gliich Idiot, wo scho sin Vatter (*das) gsii isch. he is the same idiot C already his father that been is ‘He is the same idiot his father already was.’ Isch de Hans wu¨rkli de Trottel, won en all *(de) fu¨u¨r haltet? is the John really the idiot C him all there.for hold ‘Is John really the idiot everyone regards him as?’

If resumptives occur as a last resort, we need to identify the factor that triggers their appearance. For Zurich German there are two di¤erent factors that will be discussed in turn. 1.2.1. Resumptives amnesty locality violations Of the contexts surveyed so far, resumptives inside PPs and other islands can be captured by the following generalization: They all occur in positions from where movement is (normally) impossible. Thus, they somehow amnesty locality violations (cf. the data in (2) vs. (3) and fn. 5). 1.2.2. Dative resumptives make oblique case visible The solution of the previous subsection cannot be extended to indirect objects because they do not represent an non-transparent domain, as shown by (5)a. Instead, the occurrence of dative resumptives can be traced back to a language-internal constraint that requires the overt realization of oblique case: The case system of ZG is reduced compared to the Standard German one. The genitive has been lost and nominative and accusative have fallen together (being distinct only in the personal pronoun paradigm). As a consequence, we obtain a system that distinguishes direct and oblique, the dative being the only morphologically oblique case. Bayer at al. (2001) have shown that the direct-oblique distinction plays an important role in the grammar of German and its dialects: Like genitives, datives are subject to special morphological licensing conditions. Bayer et al. (2001) discuss a number of contexts two of which we will repeat here.8 First, complement clauses in German cannot directly fill the slot of a dative argument:

8. Matching e¤ects in ZG dative relatives provide additional evidence, cf. Salzmann (2006a/b, 2009b, section 5.4).

Silent resumptives in Zurich German possessor relativization

(7) a.

b.

149

Wir bestritten (die Behauptung), we denied the.acc claim [dass wir verreisen wollten]. that we travel.away wanted ‘We denied (the claim) that we wanted to go away.’ Wir widersprachen *(der Behauptung), we objected the.dat claim [dass wir verreisen wollten]. that we travel.away wanted ‘We rejected the allegation that we wanted to go away.’ (Bayer et al. 2001: 471)

Since CPs cannot realize morphological case in German, a DP has to be inserted to rescue (7)b. The non-oblique cases nominative and accusative do not require this extra licensing, inserting a DP is optional (7)a. Second, Topic Drop is only possible with nominatives and accusatives, but not with datives, cf. Bayer et al. (2001: 489): (8) a.

b.

[acc ] Hab’ ich schon gesehen. have I already seen ‘I have already seen (it).’ *[dat] Wu¨rde ich nicht vertrauen. would I not trust ‘I wouldn’t trust (him).’

All these facts hold for Zurich German as well and other German and Swiss German dialects in general (see Salzmann 2006b: 374¤. for more empirical evidence). Consequently, the fact that the dative also stands out in ZG relativization does not come as a surprise, dative resumptives are simply another reflex of the constraint that requires oblique morphological case to be visible.9 The fact that there are no resumptives for subjects and

9. There are alternative possibilities to motivate dative resumptives. Some explanations (Boeckx 2003, Bianchi 2004) have linked their occurrence to inherent case. Van Riemsdijk (1989) has argued that datives are in fact PPs, so that dropping the resumptive would violate recoverability. See Salzmann (2006b, 2009b: 139–142) and Salzmann & Seiler (2010) for clear evidence that an explanation in terms of the morphological notion ‘‘oblique case’’ is superior.

150

Martin Salzmann

direct objects follows automatically: Since nominative and accusative are not oblique cases, no special morphological licensing is necessary.10 1.3. Movement or base-generation? Gap relatives can be straightforwardly analyzed as involving movement since they show the usual movement e¤ects (but cf. also fn. 61). The analysis of resumptive relatives, however, is less straightforward. Until the 1990ies it was more or less taken for granted that resumption involves basegeneration. In recent years, however, there have been a number of approaches that have claimed that resumption sometimes involves movement, e.g.: Demirdache (1991), Pesetsky (1998), Aoun et al. (2001), Boeckx (2003), Bianchi (2004). As a matter of fact, the empirical facts in ZG are equivocal; some favor movement while others point towards base-generation. 1.3.1. Locality As shown in (1)d–e and (2), resumptive relatives in ZG are completely insensitive to locality. This is a frequent though not universal property of resumptive constructions.11 It argues against movement analyses. 1.3.2. Reconstruction Reconstruction e¤ects are a classical diagnostic for movement. The following pairs show reconstruction for variable binding and Principle A in both gap and dative resumptive relatives (the external head is enclosed in brackets, the reconstruction site is indicated by means of underline; see Salzmann 2006b for more details):12 10. The Zurich German pattern is by no means exotic. The same has been observed for a number of languages in Keenan & Comrie (1977: 93), also for colloquial registers of Slavic languages, e.g. Czech (Toman 1998: 305); restrictive relatives in Greek also show the same pattern, cf. Alexopoulou (2006: 63). 11. See Boeckx (2003: 108¤.) on Vata and Swedish where resumptives are sensitive to locality. The same is observed by Goodluck & Stojanovic (1996) for Serbo-Croatian, by Georgopoulos (1991) for Palauan, by Rouveret (2008: 179) for Welsh, by Adger & Ramchand (2005) for Scottish Gaelic, and by Alexopoulou (2006: 85) for restrictive relatives in Greek. 12. As Josef Bayer has correctly pointed out to me, in (10) the resumptive is fronted to the Wackernagel position so that its surface position does not coincide with the reconstruction site. This implies that approaches based on NP-ellipsis, as discussed below, must assume that pronoun fronting is postsyntactic and thus irrelevant for reconstruction. This problem does not arise for semantically-based approaches to reconstruction. Cf. also 1.3.4.2 and 3.2.2.2 for the relevance of pronoun fronting.

Silent resumptives in Zurich German possessor relativization

151

(9) a.

Ich wett s [Fotti vo sinen i Eltere] gsee, I want the picture of his parents see wo jede Schu¨eleri __ am beschte findt. C every pupil the best likes ‘I would like to see the picture of hisi parents that every pupili likes best.’ DO

b.

s [Gru¨cht u¨ber siich i ], wo de Peter i __ no¨d chan ignoriere the rumor about self C the Peter not can ignore ‘the rumor about himselfi that Peteri cannot ignore’ DO

(10) a.

b.

de [Pricht u¨ber sini i Frau], won em ka¨n Politiker i the story about his wife C he.dat no politician __ wu¨rd Glaube scha¨nke would belief give ‘the story about hisi wife that no politiciani would believe’ IO de Peter i s [Buech u¨ber siich i ], won em the book about self C he.dat the Peter __ jede Wert abgsproche ha¨t every value denied has ‘the book about himselfi that Peteri denied any value’

IO

Since much of the literature on resumption took base-generation for granted, the issue of reconstruction under resumption has only recently become a major point of discussion. By now it seems widely accepted that reconstruction e¤ects can occur under resumption: Aoun et al. (2001) on Lebanese Arabic, Bianchi (2004/2008) on Romance, Belletti (2006) on Italian, Guilliot (2006) on Breton, Guilliot & Malkawi (2006) on Jordanian Arabic, Guilliot (2007) on French, Boeckx & Hornstein (2008) on Lebanese Arabic, and Rouveret (2008) on Welsh have all documented that resumption allows at least some reconstruction e¤ects, especially reconstruction for anaphor binding and variable binding. Aoun et al. (2001) and Boeckx & Hornstein (2008) have argued that reconstruction e¤ects with resumption require more care. They show that in Lebanese Arabic, the possibility of reconstruction in resumptive constructions correlates with locality: If the resumptive is in a position from where extraction is in principle possible – e.g. indirect objects as in (10) (and compare wh-extraction in (5)a. – reconstruction is possible, while in

152

Martin Salzmann

cases where the resumptive is inside an island there is no reconstruction. Resumption is called ‘apparent’ in the first case and ‘true’ in the latter.13 In ZG, however, there is no evidence for such an asymmetry. Reconstruction e¤ects are also found when the resumptive is located inside plain PPs, inside PPs which are embedded within another PP and even within classical strong islands such as a noun complement clause:14 (11) a.

b.

D [Ziit vo sim i La¨be], wo niemert i gern dru¨ber the time of his life C nobody likes.to there.about redt, isch d Puberta¨t. talks is the puberty ‘The time of his i life that nobodyi likes to talk about is puberty.’ s [Fotti vo sinere i the picture of his Gschwa¨tz dru¨ber> gossip there.about

Frau], wo ka¨n Politiker i dative > accusative > zero

37. For a di¤erent structural analysis of the possessor doubling construction based on an NP-over-DP approach, cf. Georgi & Salzmann (2010). 38. Weiss (2008: 383) gives an example with genitive from Northern Bavarian (Egerla¨nder dialect). Apparently, the same was possible in older stages of Swiss dialects. Brandstetter (1904: 75) gives the following example from Lucerne German (genitive was already strongly restricted back then): (i) s Wa¨rnis si Frau the.gen W.gen his wife ‘Werner’s wife’ Examples of this type can also be found in Stalder (1819: 81/82). Unfortunately, he does not indicate whether this type of construction was possible in all Swiss German varieties. More examples from other Swiss varieties can be found in the Idiotikon (XII: 1015), examples from all varieties of German can be found in Schirmunski (1962: 434f.) and historical data in Demske (2001: 263) and Grimm (10/1: 361f.). Apparently, genitive is also possible nowadays for speakers of (sub-)standard German; a recent example was uttered by detective Schmu¨gge in the sequel Rot ist eine scho¨ne Farbe (1998) of the Polizeiruf series: (ii) Wer wu¨rde anderer Leute ihr Bu¨ro streichen? who would other.gen people their o‰ce paint ‘Who would paint other people’s o‰ce?’ 39. The fact that some varieties use genitive argues against de Vries’ (2006: 24) analysis where the possessive pronoun results from incorporating an empty counterpart of von ‘of ’, which introduces the postnominal possessor, into D. The PP including the possessor then moves to Spec, DP. By assumption, the possessor is case-licensed by the abstract preposition: (i) D [ np N [ pp P DP]] ! [ dp1 [ pp __1 DP]2 P1 þ D [ np N __2]]

Silent resumptives in Zurich German possessor relativization

179

Consequently, the possessor can also be accusative in some languages (e.g. Low Saxon, cf. Strunk 2004) or unmarked (colloquial Dutch, West Flemish, Norwegian etc.).40 An important question in the present context is the nature of the dative case. Morphologically, it is certainly an oblique case. But when looking at abstract case, is it a structural or an inherent/lexical case? I believe there are good reasons to assume that it is a structural case, despite its morphological obliqueness: The term possessor doubling or external possessor construction is actually kind of a misnomer in that the pre-nominal dative can bear just about any relation to the possessum (cf. e.g. em Hans sini Angst lit. the John his fear ¼ ‘John’s fear’, where the relation is experiencer-like). Particularly relevant are arguments of derived nouns bearing the agent or theme role (the situation is thus basically the same as with pre-nominal genitives in Standard German, cf. Lindauer 1995/1998).41,42,43 (63) a. em Edison sini Erfindig vo de Glu¨elampe the.dat Edison his discovery of the light.bulb ‘Edison’s discovery of the light bulb’ agent b. Em Hans sini Abwaal ha¨t ali u¨berrascht. the.dat John his recall has everyone surprised ‘John’s recall surprised everyone.’ patient While objective/accusative or dative case can plausibly be linked to von since these are the cases it normally assigns, this is not the case with genitive. Obviously, the case of the possessor must have a di¤erent source. Accounts which analyze possessive pronouns as adjectives because adjectives normally assign dative/accusative/objective case (e.g. Lindauer 1995) are confronted with the same problem. 40. Pronominal possessors appear in the objective form in Dutch, cf. de Vries (2006: 22, fn. 24): (i) hem/ *hij zijn broer he.obj he.nom his brother ‘his brother’ 41. As pointed out to me by Josef Bayer, the pre-nominal ‘‘possessor’’ cannot correspond to oblique verbal arguments. The following example only allows the agent interpretation: (i) dem Hans seine Hilfe the.dat John his help ‘John’s help’ The possessor, just like pre-nominal genitives in the standard language, cannot correspond to the internal dative object argument, i.e. the one who benefits from the help.

180

Martin Salzmann

These roles can also be realized post-nominally by means of a PP: (64) a.

d Erfindig vom Edison the discovery of.the Edison ‘the discovery of Edison’

agent

b.

d Abwaal vom Hans the recall of.the John ‘the recall of John’ patient These facts clearly show that the type of case assigned in the Swiss German DP depends on the structural position, not on the semantic role. This is, of course, characteristic of structural case (cf. also Sternefeld 2006: 222). Typologically more distant languages with possessor doubling also allow various kinds of theta-roles on the ‘‘possessor’’, cf. the following example from Turkish: (65) mahkeme-nin bu karar-ı court-gen this decision-3s.pos ‘this decision of the court’ (Lewis 1967) If several arguments of N are projected, they are ordered according to the following hierarchy: Possessor > Agent > Theme (cf. Alexiadou et al. 2007: 585f.): (66) a. em Presida¨nt sini Begru¨essig vo de Gescht the.dat president his welcoming of the guests ‘the president’s welcoming of the guests’ Agent > Theme; *Theme > Agent b.

#de Gescht iri Begru¨essig vom Presida¨nt the.dat guests their welcoming of.the president lit.: ‘the guests’ welcoming of the present.’ *Theme > Agent; Agent > Theme

42. The same holds in West Flemish, cf. Alexiadou et al. (2007: 593, fn. 44). Strunk (2004: 195¤.) has no occurrences of non-agent arguments as pre-nominal possessors in his corpus (apart from genuine possessors, of course). It is certainly correct that there is a tendency for patients to occur post-nominally, but I do not take this to be a grammatical e¤ect. 43. Inanimate pre-nominal possessors are strongly dispreferred, cf. Weber (1964: 214); cf. also Weiss (1998: 77, fn. 59) on Bavarian and Strunk (2004/2005) on Low Saxon.

Silent resumptives in Zurich German possessor relativization

181

c. em Hans sis Bild vom Picasso the.dat John his painting of.the Picasso ‘John’s painting of Picasso’ Possessor > Agent/Theme; Agent > Theme *Theme > Agent/Possessor *Agent > Possessor This does not argue against the structural case approach, but rather simply follows from locality: The agent/possessor is base-generated above the theme so that it is closer to Spec, DP (and FP, cf. below) and will therefore be attracted. We basically get the same ordering as with subjects and objects, cf. Lindauer (1998). A pre-nominal patient with a postnominal agent is marginally possible if the agent is introduced by the preposition dur ‘through’. Since it is generated as an adjunct, it does not block movement of the theme: (67) ?em Hans sini Verhaftig dur d Polizei the.dat John his arrestation through the police ‘John’s arrestation by the police’ A further argument for treating the pre-nominal dative as structural is the fact that it is assigned by a functional element, namely the possessive pronoun, which we treat as an agreement marker. Evidence for this comes from the fact that dative is not licensed by normal determiners:44 44. Interestingly, Fischer (1960: 323, fn.) claims for Lucerne German that next to the regular construction with possessive pronoun one does find examples with definite determiners, especially if the DP additionally contains an adjective: (i) em Peter de (lieb) Vatter the.dat Peter the nice father I cannot assess how widespread such examples are/were. Brandstetter (1904: 75), who also describes Lucerne German, gives an example, and Stalder (1819: 82) gives one even without definite article. In similar vein, according to Stalder (1819: 81), genitive possessors as in fn. 38 can occur without the possessive pronoun. I have never come across such constructions in ZG, but since there were (residual) genitive possessors in earlier stages of the dialect (Weber 1964), it cannot be ruled out that such patterns were more widespread in Zurich and also Swiss German in general. Should such constructions turn out to be more frequent they would have important implications for the analysis of case assignment: For all these cases without possessive pronoun one either has to assume (with Weiss 2008) that D assigns dative/genitive case or one might want to adopt a silent F head to capture the parallelism with the possessor doubling construction (see the next subsection). I will leave this for further research.

182

Martin Salzmann

(68) *em Peter s/es Huus the.dat Peter the/a house Additionally, the dative case-feature on the possessive pronoun has to be checked. This accounts for the following asymmetry concerning PPpossessors:45 (69) a.

*vom Peter sis Huus of.the Peter his house ‘Peter’s house’

b.

*sis Huus vom Peter his house of.the Peter ‘lit.: his house of Peter’

c. vom Peter s Huus of.the Peter the house ‘Peter’s house’ A PP-possessor cannot co-occur with a possessive pronoun because this would leave the case-feature on the possessive unchecked: Even though the possessive pronoun has an EPP-feature that could potentially attract the PP-possessor, its case feature cannot be discharged/assigned in such a configuration. Co-occurrence of a definite article with a PP-possessor, however, is possible because the article does not have any case-features that have to be discharged/assigned. Fronting of a PP-possessor as in (69)c is possible under certain information structural conditions (and is arguably triggered by an edge-feature on D). We can therefore conclude that the possessive pronoun assigns structural case to the possessor and 45. (69)c argues against the proposal in Weiss (2008: 390) that von is the spell-out of an Agr-head. On his analysis of possessive constructions, possessor and possessum start out inside a small clause. In the analysis of postnominal possessors, the possessum moves to Spec, AgrP while the possessor remains inside the small clause (governed by von). To derive (69)b, one would have to move a non-maximal projection (Agr’) to Spec, DP, which I take to be undesirable: (i) D [ AgrP Possessum1 Agr(von) [ sc Possessor __1]] ¼ ‘The house of John’ As Peter Gallmann has pointed out to me, this construction is subject to curious restrictions. It seems possible only when the entire DP occupies the subject or direct object position or occurs as a predicate. As a complement of a preposition or as an oblique argument (dative/genitive), however, it is ruled out. This may suggest that its syntax is actually more complex.

Silent resumptives in Zurich German possessor relativization

183

attracts it to its specifier. This abstract case is sometimes referred to as POSS (e.g. Weiss 2008). A cross-linguistically consistent realizational rule within Germanic then specifies that POSS is expressed by the most oblique morphological case of the language.46,47 2.5. The structure of the possessor doubling construction The previous subsections have prepared the ground for the derivation of the possessor doubling construction in Zurich German. I will start with the NP-layer: 1. The possessor (both proper possessors and those realizing arguments of a relational noun) originates within the projection of N. This holds for overt dative possessors, those realized by means of pro, as well as oblique possessors realized by means of PPs. I will further assume that 46. Whether the post-nominal ‘of ’-‘‘possessor’’ is also assigned structural case is a question I will leave open here, as much depends on how grammatical prepositions are handled in Minimalist Syntax, a question beyond the scope of this paper. If it is taken to be structural, an additional functional category above N (but below the landing site of the possessum) may be necessary, cf. Lindauer (1998) and Weiss (2008). 47. Possessor datives are therefore di¤erent from at least some verbal datives/ genitives where the morphological case is to a large extent tied to a particular theta-role. There has been a long discussion on whether verbal datives should be treated as structural/inherent, cf. e.g. Gallmann (1992). It is clear that all dative objects are di¤erent from structural accusative objects in some respects. At the same time, there is evidence that not all datives are equally ‘‘inherent’’. Whether this should lead to a classification into inherent and structural datives is a question I will not pursue here. Given the requirement on overt realization of oblique morphological case (cf. 1.2.2) it is surprising that the dative case can remain unrealized when the possessor is a pro. This does not seem to be a quirk of ZG, though, but rather a general fact about pro: Even if the governor assigns oblique case, the argument can remain zero as long as its content can be recovered through agreement. And this is certainly the case in the possessor doubling construction. The constraint on the realization of oblique case can thus bet met in di¤erent ways: through overt realization by means of morphological case, via matching in relative clauses (cf. Salzmann 2006a/b) and via agreement. It is therefore best thought of as a recoverability requirement. Quirky-PRO in Icelandic is another example of a silent pronominal element with oblique case (cf. Bianchi 2004: 110, fn. 66). The licensing conditions will be di¤erent, though. Additionally, as far as I know, the requirement to realize oblique case cannot be directly transferred to Icelandic.

184

Martin Salzmann

possessors and external arguments of the noun are base-generated in Spec, NP (cf. Lindauer 1998: 119, Sternefeld 2006: 588) and internal arguments of N in its complement. In the case of dative possessors, the binding facts from 2.3.1 have already shown that dative possessors asymmetrically c-command the other NP-internal constituents. For oblique possessors, the same is shown by the following variable binding facts: (70) d Angscht vo jedem i Star voor sim i Abstieg the fear of every star before his decline ‘Every star’s i fear of his i decline’ Here, jedem Star ‘every star’ binds sim Abstieg ‘his decline’ and consequently must c-command it. The surface order follows if N moves across the possessor to some higher functional head, e.g. little n:48,49 (71) [ dp d [ nP [Angscht]1 þ n [ np vo jedem Star [ n’ __1 voor sim Abstieg]]]] 48. For plain possessors, base-generation in an N-complement position (Delsing 1998, de Vries 2006) or a small-clause analysis (den Dikken 1998, Weiss 2006/2008; cf. Alexiadou et al. 2007: 563f. for further references) would also work. However, once several arguments of N or an argument of N plus a possessor are involved, this is no longer su‰cient to derive the ordering restrictions described in 2.4. The same holds for the proposal in de Vries (2006: 21, fn. 20), who explicitly argues against base-generating anything in spec, NP, but suggests instead a double-object construction within NP for cases like (70) and more generally when 2 DP-internal constituents are involved. If the objects were projected on a right branch of NP: (i) [ NP [ N’ N Ag] Th] then movement of the agent to some functional specifier would be necessary so that it can c-command the theme. This might work for dative possessors as far as the linear order is concerned, but the ordering restrictions noted in 2.4 would remain mysterious, one would rather expect the opposite ordering since the theme asymmetrically c-commands the agent in the base. For PPagents as in (70), movement of N next to movement of the agent would be necessary to derive the correct order. Such a derivation would not only be less economical than the one proposed in the text, it would also incorrectly predict wrong ordering restrictions, since post-verbal arguments of N are also subject to the ordering restrictions described in 2.4. If, finally, the DP-internal constituents would both be generated on the left branch of N: (ii) [ NP Ag [ N’ Th N]]

Silent resumptives in Zurich German possessor relativization

185

2. The dative possessor moves out of NP and lands in the specifier of the possessive pronoun. It is clear that the possessive pronoun occupies D at the end of the derivation (this is necessary to account for the complementary distribution with determiners, cf. 2.2.3.3). What is less clear is whether the pronoun starts out in D (e.g. de Vries 2006, Weiss 2008) or in some intermediate functional head like e.g. Pos (e.g. Delsing 1998: 92), I (Haegeman 2004) or Agr (Alexiadou et al. 2007: 611) between D and little n (in our view, IP/AgrP are preferable labels because we assume that the possessor role is assigned inside NP so that the function of the possessive pronoun is just to check agreement with the possessor). In the first case, the pronoun agrees with the possessor, assigns case to it and attracts it to its specifier (Spec, DP). In the second case, the pronoun agrees with the possessor, assigns case to it and attracts it to its specifier, viz. Spec, PosP/IP. Subsequently, the pronoun undergoes head movement to D to check its definiteness feature. The possessor finally moves to Spec, DP for reasons of EPP (cf. Delsing 1998: 95). The two possible derivations then look as follows (the intermediate functional layer will henceforth be labeled FP; head-movement of the possessum as in (70) is not necessary with dative possessors): (72) a.

[ dp possessor1 [ d’ pos.pron2 [ FP __1 [ F’ __2 [ np __1 [ n’ possessum ]]]]]]

b. [ dp possessor1 [ d’ pos.pron [ np __1 [ n’ possessum ]]]] the correct ordering would follow, but movement of N would be necessary in any case (for pre- as well as postnominal agents). On the present account, on the other hand, movement of N to n is not necessary to derive the examples with dative possessors as in (63) and (66). Additional movement of N is also needed in accounts that base-generate the possessor/agent in Spec, nP (thereby treating possessors/agents like external arguments of verbs that are introduced by little (v): To derive cases like (70), N would not only have to move to n, but even higher so that it ends up preceding the PP agent. Such solutions are thus more complex than the one in the text and are therefore to be dispreferred unless one assumes that a designated functional projection is necessary to assign the possessor and agent interpretation. Cf. Alexiadou et al. (2007: 552¤.) for detailed discussion of the various options concerning possessors and Alexiadou et al. (2007: 477–541) for an overview over the projection of arguments in derived nominals. 49. Alternatively, N undergoes reprojective movement as in Georgi & Mu¨ller (2010).

186

Martin Salzmann

It is di‰cult to come up with solid Zurich German-internal evidence in favor of an intermediate functional projection because of the complementary distribution between determiners and possessive elements. Other languages allow the co-occurrence of a determiner with a possessive element; a well-known case is Italian where determiners (definite and indefinite) co-occur with possessive adjectives. Definiteness and possession are thus distributed over two elements.50 Similarly, Hungarian allows the cooccurrence of a determiner and a pre-nominal possessor in Spec, FP. The agreement marker on the noun indicates N-to-F movement (Delsing 1998: 104): (73) a.

b.

il mio libro the my book ‘my book’ az en vendeg-em the I.nom book-pos.1s ‘my book’

(Italian)

(Hungarian)

Within varieties of German and Germanic more generally, there is some evidence for an intermediate position. Weiss (2008: 386) presents an example from Bavarian with a dative possessor co-occurring both with an indefinite determiner and a possessive adjective (unfortunately, it is not clear in what sense or if at all such DPs are indefinite): (74) am Schlossbauern a seinige Tochta the.dat S.dat a his daughter The co-occurrence of determiner and possessive element is also found in the following construction with a possessive adjective and an elided possessum that is attested in Bavarian (Weiss 1998: 82f.), West Flemish (Haegeman 2004: 214), varieties of Dutch (Corver & van Koppen 2010), and Colloquial German (according to Weiss 2008: 391, fn. 17). Unfortunately, the construction is impossible in Zurich German as it does not have possessive adjectives (but for other dialects see Idiotikon VII: 1017): (75) a.

am Hans des sei the.dat Hans the his ‘John’s’

(Bav.)

50. Cf. Lyons (1999) for the distinction between determiner-genitive and adjectivalgenitive languages.

Silent resumptives in Zurich German possessor relativization

b.

Da zyn Vale`re de zyne. these are V. the his ‘These are Vale`re’s.’

c. dem Hans das seine the.dat John the his ‘John’s’

187

(WF)

(Coll. G)

Both constructions can be accommodated by assuming that the possessive adjective occupies F as in Italian. The adjective then assigns case to the possessor and attracts it to its specifier, viz. Spec, FP. Subsequently, the possessor moves on to Spec, DP (for reasons of EPP).51 De Vries’ (2006: 24) claim that such an intermediate position is unnecessary for Germanic may thus be too strong. For ZG though, such a layer does not seem to be necessary. To some extent the issue also depends on the solution to the double agreement issue: 3. As shown in 2.2.2, the possessive pronoun agrees both with the possessum and the possessor. Importantly, one has to assume two sets of features as the possessor and the possessum do not have to agree in features. In the following example (repeated from 2.2.2 above), the possessor is masculine singular while the possessum is feminine plural: (76) em Maa sin-i To¨chter the.msc.sg man msc.sg.P’or-fem.pl.P’sum daughter.fem.pl This was considered a problem for treating the pronoun as the head of the DP, cf. Olsen (1989). On a simplistic view of features, this indeed seems to be a problem. A lexical item can only be specified for one value of a given feature. Conflicting values are ruled out. Lo¨bel (1996) and Sternefeld (2006) argue that we are dealing with two fundamentally di¤erent types of features: On the one hand, the pronoun has syntactic features that express agreement with the possessum. On the other hand, it has inherent/ anaphoric/semantic features (that refer to a possessor). Only the syntactic features percolate to the maximal projection. This distinction is motivated 51. Weiss (2008: 386f.) argues instead that examples like (74) provide evidence that the case is not linked to the possessive pronoun, but rather is assigned in Spec, DP. But data like (68), which I take to be ungrammatical in Bavarian, show that the presence of a possessive element is crucial. Whether this is a possessive pronoun or a possessive adjective is apparently not relevant.

188

Martin Salzmann

on the basis of the Standard German construction where there is no overt possessor. But once the doubling construction is taken into account, it is no longer clear whether this distinction is warranted. We will instead assume that there really is double agreement: there are two sets of agreement features, one for the possessor and one for the possessum, the first optionally licensing a small pro. What has to be explained, though, is the fact that only the features that express agreement with the possessum project, but not those of the possessor:52 (77) [de Ma¨itli ir-es Sa¨il ] isch/*sind rot the.dat girl.pl fem.pl.P’or.-fem.sg.P’sum rope is/are red ‘the girls’ rope is red’ There are at least two possible solutions conceivable among which I will not choose: On the one hand, – similar to work in non-derivational frameworks such as HPSG (Sag & Wasow 1999, see also Demske 2001: 180¤., Strunk 2004: 74¤.) – one could posit a distinction between headfeatures that project to the maximal projection and selectional features that check agreement with complements and specifiers and are only projected until they are satisfied. In the case at hand, the complement features of D will be identical to the head-features while the specifier features will depend on those of the possessor. In more conventional Minimalist terms, one possibility consists in adopting two sets of agreement features on the possessive pronoun. It will undergo agreement with both the possessum and the possessor. This is possible without a PosP: (78) [ dp possessor1 [ d’ pos.pron [ np __1 [ n’ possessum]]]] Agree D – NP Agree D – DPPoss’or Move DPPoss’or The derivation is not without problems. Agreement between D and NP is unproblematic. Agreement between D and the possessor, however, presupposes that NP does not intervene. Since NPs case feature will not be valued until an external case-assigning probe is merged (v/T), NP may still count as active and therefore block Agree. I will provisionally assume that Agree between D and the possessor is possible because it is sensitive to the 52. It was for this reason that Olsen (1989) base-generates the stem of the possessive pronoun in Spec, DP and its inflectional ending in D.

Silent resumptives in Zurich German possessor relativization

189

category (D instead of N) of the goal so that there is no intervention. To make sure that dative case is assigned to the possessor and not to the possessum, the case feature has to be linked to the D-feature. But why do we get the agreement pattern in (77)? Adapting a suggestion from Erik Schoorlemmer (p.c.): The crucial di¤erence between the two sets of agreement features on the possessive pronoun is that those agreeing with N have an unvalued case-feature which is not valued until agreement is established with a DP-external probe (v/T). The features agreeing with the possessor, however, will all be valued. This is why only the phi-features agreeing with N will remain active and visible for an external probe so that we get the pattern in (77).53 The derivation for the examples in (74)–(75) is very similar: The possessive adjective has two sets of agreement features; it is merged in F, agrees with N and attracts the possessor. Thereafter, D is merged, attracts F and 53. An alternative way of handling the double agreement problem is to take the morphological complexity of the possessive pronoun seriously and basegenerate its parts in di¤erent head positions: the stem in F and the ending in D. The two agreement processes will thus take place independently, and this will guarantee that only the features agreeing with the possessum end up on DP (the stem finally incorporates into D). Otherwise, the derivation is almost the same. As in the previous derivations, non-trivial configurations occur with respect to intervention: F has to establish agreement with and attract the possessor across the possessum (i). This may be possible if Agree is sensitive to syntactic category (D instead of N) (cf. the discussion in the main text). The ending in D will have to establish agreement with the possessum across the possessor (ii), which is possible since the possessor does not have any unvalued features anymore. The resulting representation is shown in (iii): (i) [ dp [ d’ -Dending [ FP [ F’ Posstem

[ np possessor [ n’ possessum ]]]]]]

Agree

(ii) [ dp [ d’ -Dending [ FP possessor1 [ F’ Posstem [ np __1 [ n’ possessum ]]]]]] (iii) [ dp possessor2 [ d’ Posstem2þ -Dending [ FP __1 [ F’ __2 [ np __1 [ n’ possessum ]]]]]]

Agree

See Heck & Mu¨ller (2007) for a similar approach. Note that under an Agree approach movement of the possessor to an intermediate position like Spec, PosP as in this derivation or in (72)a is no longer necessary since c-command su‰ces for case assignment. Doreen Georgi (p.c.) has pointed out to me that in an NP-over-DP approach no di‰culties arise with respect to projecting the right features because N is the head of the noun phrase while the pronoun is just a specifier of N.

190

Martin Salzmann

then the possessor (for reasons of EPP), thus essentially as in (72)a. If one wants to adopt an F-layer for regular possessor doubling, the derivation is also as in (72)a.54 2.6. The syntax of possessor relativization The analysis of the possessor doubling construction proposed above has important consequences for the analysis of certain parts of the Swiss German resumption system. Since Spec, DP of the possessor construction can host either a full DP or a pro, there are in principle also two possible analyses of possessor relativization: Either there is extraction of a relative operator from Spec, DP or the operator is base-generated instead and 54. It is surprising that the double agreement issue is not addressed explicitly in many recent approaches as e.g. in Delsing (1998), Haegeman (2003/2004), de Vries (2006), and Weiss (2006/2008) even though the double agreement pattern also occurs in languages other than German (e.g. Norwegian, cf. Delsing 1998: 95, or West Flemish, cf. Haegeman 2004: 216). The agreement facts remain completely mysterious under de Vries’ (2006) approach (recall the derivation from fn. 39): (i) D [ np N [ pp P DP]] ! [ dp [ pp __1 DP]2 P1 þ D [ np N __ 2]] While agreement between D and the possessum (i.e. D’s complement) seems straightforward, I fail to see how the preposition could trigger agreement with the possessor: Prepositions do not agree with their complements or specifiers in German. Additionally, the agreement is dependent on incorporation of P into D (as no agreement occurs with post-nominal possessors), but nothing in de Vries’ account derives these facts. The same applies to the analysis in Weiss (2006/2008: 389¤.) where the possessive pronoun results from incorporation of an agr-head into D: (i) [ dp Possessor1 D þ Agr3 (possessive) [ AgrP possessum2 __3 [ sc __1 __ 2]]] Agreement with the possessum is plausible as it always moves to Spec, AgrP on Weiss’ analysis, cf. also fn. 45. However, it remains unclear to me how agreement with the possessor is established. On Weiss’ analysis, Spec, DP can be occupied by pre-nominal genitives as well as dative possessors, but there is agreement only in the second case. But in both cases, Agr incorporates into D so that the di¤erence is unexpected. Furthermore, I do not see how it can be ensured that the features expressing agreement with the possessum end up on the DP in such a system – if D is responsible for agreement with the possessor, one would expect these features to project, not those of the incorporated Agr. A very di¤erent analysis of the possessor doubling construction is proposed in Georgi & Salzmann (2010).

Silent resumptives in Zurich German possessor relativization

191

binds the pro in Spec, DP of the possessive phrase. I will therefore test whether a movement or a base-generation analysis is preferable. 2.6.1. Reconstruction and Strong Crossover e¤ects Like all other relatives, possessor relativization shows certain movement e¤ects such as reconstruction and SCO e¤ects, irrespective of whether the possessive phrase is located in a transparent domain or inside an island. The first pair illustrates reconstruction, the second one SCO e¤ects:55 (79) a.

b.

(80) a.

b.

d [Fru¨ndin vo sim i Soon], wo jede Vatter i __iri the girlfriend of his son C every father her Ellere wett ka¨ne leere acquaintance wants make ‘the girlfriend of his i son whose parents every father i wants to get to know’ d [Frau vo sine i Tra¨um], wo jede Maa i stolz the woman of his dreams C every man proudly za¨iget the album C her portrait in.it is shows lit.: ‘the woman of his i dreams that every man i proudly shows the album in which her portrait is’ *de [Bueb] i, won er i __ sini i Mueter gern ha¨t the boy C he his mother likes lit.: ‘the boy i whose i mother he i likes’ *de [Bueb] i, won er i de Maa, wo __ sini i Mueter ku¨sst ha¨t, the boy C he the man C his mother kissed has no¨d chan uusschtaa not can stand lit.: ‘the boy i whose i mother he i can’t stand the man that kissed’

As discussed in 1.3.4 above, these facts do not necessarily argue in favor of movement, but are in principle also compatible with base-generation.

55. It is very di‰cult to come up with naturally sounding examples. Most speakers will therefore find these examples rather awkward.

192

Martin Salzmann

2.6.2. Locality As in resumptive relatives, a movement analysis for ZG possessor relatives runs into di‰culties once locality is taken into account: Even though possessor extraction from a predicate is marginally possible (cf. 2.3.2), it is ungrammatical in all other contexts. Possessor relativization, however, is fully grammatical across the board. Consider the following contrast: (81) a.

Das deet isch de that there is the Vatter ka¨ne gleert father got.to.know

Schu¨eler, wo de Peter geschter sin student C the Peter yesterday his ha¨t. has

‘Over there is the student whose father Peter met yesterday.’ b.

??[Wem]1 ha¨t de Peter geschter [__1 sin Vatter] ka¨ne gleert? who.dat has the Peter yesterday his father got.to.know ‘Whose father did Peter meet yesterday?’

In relativization, the possessive DP can also be more deeply embedded, e.g. within another possessor, in a PP or even within a CNPC island: (82) a.

Das deet isch de Schu¨eler, won i geschter that there is the student C I yesterday bewundered han. his father his car admired have ‘Over there is the student whose father’s car I admired yesterday.’

b.

Das deet isch de Schu¨eler, won i geschter that there is the student C I yesterday gredt han. with his father talked have ‘Over there is the student with whose father I spoke yesterday.’

c. de Autor, wo d Ursi < jedes Buech, wo sini the author C the Ursi every book C his Kritiker verriissed >, chau¤t critics tear.apart buys lit.: ‘the author that Ursi buys every book that his critics tear apart’

Silent resumptives in Zurich German possessor relativization

193

2.6.3. Resumption involves base-generation In the discussion of the derivation of resumptive relatives in 1.3 above I dismissed most movement approaches because of the locality facts. The only possibility I did not fully reject was the PF-theory of locality as in Pesetsky (1998): Movement out of islands is possible because the tail of the chain is overt thanks to the resumptive. The overtness of resumptives would then be the crucial factor that makes movement out of islands possible. The facts from possessor relativization strongly argue against this since there is no overt resumptive even though the putative extraction site is within an opaque domain.56 In fact, adding an overt possessor resumptive leads to degradation: (83) Das deet isch de Schu¨eler, wo de Peter geschter (?im) that there is the student C the Peter yesterday he.dat sin Vatter ka¨ne gleert ha¨t. his father got.to.know has ‘Over there is the student whose father Peter met yesterday.’ Overtness of the tail of a movement chain thus cannot explain why possessor relativization is insensitive to locality. The only explanation that remains is that possessor relativization involves base-generation: An operator in Spec, CP binds the pro in Spec, DP of the possessive phrase.57 We are thus dealing with an empty resumptive pronoun as in languages with rich(er) agreement like e.g. Irish (Mc Closkey 1990), Welsh (Willis 2000) or Palauan (Georgopoulos 1985: 69): 56. It is incoherent to claim that the possessive pronoun is the resumptive given that it has been analyzed as a pure agreement marker in the possessor doubling construction. Cf. also Haegeman (2003: 253, fn. 15). 57. One cannot attribute the lack of an overt resumptive to the structural case of the possessor. As shown in (2) above, overt resumptives are necessary for structural cases as well once they occur inside islands. The same goes for accounts that relate the impossibility of an overt resumptive to the fact that it would be a strong pronoun in possessor relativization (like in (26)/(83)) rather than a weak/clitic one as in other areas of relativization in ZG. There is reason to believe that there is no general ban against strong resumptives in Zurich German as the following example shows where an object of comparison is relativized: (i) de a¨inzig Bueb i de Klass, won i gro¨o¨sser bin als e¨e¨r/*er the only boy in the class C I taller am than he/he ‘the only boy in my class that I am taller than’

194

Martin Salzmann

(84) a buik el k- illebed-ii [a  obok-ul pro] a se el-ik boy C ir.1s-pf.hit-3s older.brother-3s friend-1s ‘The boy whose brother I hit is my friend.’ To complete the argument, let me present independent evidence that silent resumptives are a possibility in ZG and can void island constraints.58 When the subject is 2nd person singular it can be zero (perhaps because 2nd person singular is expressed very clearly by verb morphology). In V-final sentences (including relative clauses) the complementizer additionally takes an agreement marker, arguably because 2nd person singular must be realized on C. I will assume that a pro is licensed whenever the verb is 2nd person singular. An overt subject pronoun can be optionally used as well (e.g. for emphasis):59 (85) . . . wo-t em pro/du das gsa¨it ha¨sch when-agr he.dat pro/you that said have.2s ‘when you said this to him’ Relatives are thus parallel to regular comparative expressions involving als ‘than’: (ii) Ich bi gro¨sser als e¨e¨r/*er. I am taller than him ‘I am taller than him.’ Even if such explanations for the silence of the resumptive could be upheld, they would still be in conflict with the PF-theory of locality: The insensitivity to islands would remain mysterious. 58. I am grateful to Werner Frey for discussion of this point. 59. As pointed out to me by Peter Gallmann, there is some indication that we are perhaps not dealing with complementizer agreement but rather with clitic doubling: The marker -t not only occurs after elements in C but also after elements in Spec, CP such as wh-adverbials: (i) Ich verschtaa no¨d, warum¼t em (du) no¨d ghul¤e ha¨sch. I understand not why¼2s he.dat you not helped have.2s ‘I don’t understand why you didn’t help him.’ For such cases one would have to assume that an empty C is inflected, which is rather unexpected from a morphological point of view. A clitic doubling analysis must, however, explain why no clitic appears under V2: (ii) Warum ha¨sch-(*t) em du ghul¤e? why have-2s he.dat you helped ‘Why did you help him?’

Silent resumptives in Zurich German possessor relativization

195

Crucially, when a second person singular is relativized, there is a strong preference for (resumptive) pro, even if the variable is inside an island; relativizing pronouns is awkward):60 (86) du, won ich glaube, dass es you would.like.2s help ‘you, such that I believe that there is no one who you would like to help’ We can thus conclude that resumption can involve silent elements in ZG. Since silent resumptives can occur inside islands, it is no longer possible to claim as in Pesetsky (1998) that overtness of the tail of an A’-dependency makes movement out of islands possible. Instead, the island-voiding nature of resumption can only be attributed to base-generation. Since base-generation is necessary to account for possessor relativization, it is most economical to assume the same for the other instances of resumption in ZG. While a base-generation analysis as such is straightforward, it raises a number of technical issues regarding economy that I will address in the following section.61

As pointed out to me by Peter Gallmann, there is good reason to believe that the -t falls victim to a general phonological rule of the language that deletes -t after -sch. The rule is stated explicitly in Weber (1964: 174). It applies to verb forms (there used to be a -t in the second person singular, cf. the standard language), but also more generally to monosyllables like fascht > fasch ‘almost’ and also to the verb-clitic sequence in (ii). 60. The fact that a silent resumptive is preferred over an overt one in (83) and (86) can be subsumed under the Avoid Pronoun Principle (Chomsky 1982), which prefers silent over overt pronouns (unless overt pronouns are independently required), cf. Salzmann (2009a: 46¤.) for detailed discussion. 61. The analysis is thus eventually similar in spirit to van Riemsdijk (1989), but crucially di¤ers from it in assuming that SU/DO-relatives involve movement rather than base-generation plus deletion of the resumptive. See Salzmann (2009a/2009b: 143¤./2009c) for arguments that SU/DO-relatives must involve movement. For general discussion of van Riemsdijk’s approach, cf. Salzmann (2006b: 368¤.). This di¤erence will become rather crucial below where the definition of the reference set is discussed.

196

Martin Salzmann

3. Analysis 3.1. The syntax of base-generation My implementation of base-generation is very simple: As in traditional analyses, an operator is base-generated, i.e. directly merged, in an operator position. The resumptive, i.e. a regular pronoun, is merged in an argument position. Finally, the operator binds the pronoun, creating an operator variable dependency. As a consequence, the operator must be unmarked for case. If it had a case-feature, there would be no way for it to be checked/valued in this configuration. The operator thus only has an iOp-feature that values the corresponding uOp feature on C62 while the resumptive is involved in Agree with v/T. In addition to the operator unmarked for case, there also is a (silent) case-marked operator. If that operator is chosen, a movement derivation occurs as it can value both uPhi of v/T and uOp of C. The two derivations schematically look as follows: (87) a.

[ cp Op i C

[ vP [ vp pron i V] v ]]

iOp[x] uOp[x] EPP

b.

[ cp Op iOp[x]

uCase[acc]] iPhi[z]

C uOp[x]

uCase[acc] EPP

uPhi[z]

base-generation

[ vP [ vp Op V] v ]] iOp[x] uCase[acc] iPhi[z]

uPhi[z]

movement

3.2. The distribution of movement and base-generation 3.2.1. Resumptive derivations succeed when movement derivations fail I have argued above that resumption is a last resort in ZG, occurring only when a gap-derivation fails. This happens when the extraction site is an indirect object or is located inside an island. In both cases, only the basegeneration derivation with resumptives will converge, albeit for di¤erent 62. This assumption may be somewhat non-standard. It is often assumed that uninterpretable/unvalued features of a probe can only be checked via internal merge, cf. McCloskey (2002: 204), Alexopoulou (2006: 80, 88). If one wants to uphold this restriction, one can assume that C does not have an Op-feature, but only an EPP-feature, or even no feature at all. In the last case, Merge of the operator would be purely semantically-driven.

Silent resumptives in Zurich German possessor relativization

197

reasons: In the case of the indirect object, a gap derivation will lead to a representation where the oblique case remains unexpressed in violation of the constraint discussed in 1.2.2. above. If the extraction site is within an island, the gap derivation crashes since movement out of islands violates locality. In these configurations there will thus be no competition between movement and base-generation. 3.2.2. Gap derivations are more economical Now that we have concluded that base-generation is a possibility in ZG, the following question arises: What blocks resumption when subjects or direct objects (or predicates in non-oblique positions, cf. (6)) are relativized? Given the reasoning above, resumption is blocked because it is not necessary. In other words, gap derivations are taken to be more economical. While easily stated in prose, it is quite di‰cult to implement this technically, at least in a Minimalist setting where Economy computations require that the derivations that are compared belong to the same candidate/ reference set. There are several options to determine the reference set (see Sternefeld 1997 for discussion). The original and possibly still most widely held assumption is that the reference set is based on the lexical material used in a derivation: Derivations belong to the same reference set if they are built from the same numeration, i.e. if they are made up of the same set of lexical (and possibly functional) elements. 3.2.2.1. Should the numeration determine the reference set? Aoun et al. (2001: 393¤.) adopt this definition of the reference set for their analysis of resumption. If gap/movement and resumptive/base-generation derivations compete, they have to be built from the same numeration, which implies that the resumptive cannot be part of the numeration; rather it must be generated during the derivation. The derivation proceeds as follows: First an operator is merged in the thematic position. When it is attracted by a C it will attempt to move to the specifier of that head. If there is no island, the operator will do so, the result being a gap derivation. If, however, the operator is inside an island, it cannot move out. Then, Aoun et al. (2001) argue, an operation termed Bind applies: The operator is demerged from the phrase marker and remerged in the operator position, and a resumptive pronoun is substituted for the demerged expression in the thematic position. Importantly, both derivations can in principle apply when the operator is in a transparent domain. However,

198

Martin Salzmann

the base-generation derivation is less economical because it involves more operations than the movement derivation (additionally, the pronominalization operation is interpreted as a violation of representational economy, cf. Aoun et al. 2001: 398): (88) a.

movement: Copy þ Merge

b. base-generation: Demerge (Copy þ Delete) þ Merge þ Pronominalize We are thus dealing with a trans-derivational economy constraint that blocks base-generation if there is a converging movement derivation. In the case of datives and non-transparent domains, the movement derivation fails so that the converging derivation involving Bind – even though it is less economical – results as the only grammatical possibility. There are a number of problems with such an approach if applied to ZG that make it worthwhile to think about alternative ways of determining the reference set. First, a constraint in terms of trans-derivational economy (perhaps subsumable under Fewest Steps) goes against the trend of the last decade to do away with such constraints, cf. Mu¨ller & Sternefeld (2001). Second, the resumptive always appears more or less (i.e. modulo pronoun fronting) in the thematic position. But in case the operator first undergoes A’-movement before it encounters an island, one would expect the resumptive in higher positions, contrary to fact: (89) [ cp Cþwh [ island [ cp Op i [ cp Op i . . . Op i]]]]. + res Here the operator has first moved to Spec, CP of the lowest clause, then to an intermediate Spec, CP. Then, when it attempts to move to the matrix Spec, CP, it encounters an island. As a consequence, this is where the operator would have to be replaced by a resumptive. But resumptives normally do not occur in intermediate positions. They certainly do not in ZG, but this fact holds more generally, perhaps with the exception of Hebrew (cf. e.g. Demirdache 1991).63 63. Perhaps the remerger theory of movement would help, cf. the discussion in Aoun et al. (2001: 399, fn. 31). Alternatively, one could argue that a derivation as in (89) is possible but generally blocked by a base-generation derivation that substitutes the pronoun at the very beginning. The latter derivation

Silent resumptives in Zurich German possessor relativization

199

Third, it is rather unclear to me how such an approach could be implemented in the more recent versions of the Minimalist Program of stronglyderivational character, e.g. Chomsky (2001): In such approaches, the ultimate C-probe often will not have been merged when the operator, i.e. the goal, would have to be de-merged. Demerging would have to apply pre-emptively (a form of look-ahead), and it is unclear what one would do with the copy resulting from it. It would have to be kept in storage somewhere until the relevant probe is merged, but how this should be done is by no means obvious (cf. Heck & Mu¨ller 2000: 35 for similar discussion). Fourth, the shape of the pronoun that is substituted for the demerged operator phrase depends on the syntactic context: With possessors it is a null-resumptive, in most other cases it is a weak overt pronoun and after the comparative particle it is a strong pronoun, cf. fn. 57. It is questionable whether this could be taken care of by the pronominalization rule involved in Bind. It is not per se impossible, but such an operation would be undesirably powerful.64

would arguably be more economical as it would involve fewer movement steps. This means that for the trans-derivational constraint to work, we would need another one. This is certainly not a very satisfactory solution. And since successive-cyclic wh-movement is often analyzed as counting just as one operation (one instance of Form Chain, cf. Mu¨ller & Sternefeld 2001: 12), not even this may work. 64. To be fair, Aoun et al. (2001: 396) actually implement the pronominalization operation in a somewhat di¤erent and rather perplexing way: The pronoun that is substituted is not the resumptive that can be seen on the surface but rather an additional empty resumptive pronoun. This is arguably necessary within their approach because Lebanese Arabic also features strong pronouns and epithets as resumptives, which cannot so easily be analyzed as the pronominalization of a DP. Furthermore, due to their distinction between true and apparent resumption (cf. 1.3.2), an overt resumptive is never indicative of base-generation so that an additional empty one has to be assumed for base-generation. I find this rather unattractive and certainly unnecessary for the ZG facts. Alternatively, one could argue that what is substituted are just pronominal features. The exact shape of the pronoun would then be determined postsyntactically on the basis of the syntactic context – basically as in spell-out approaches to movement. Since pronoun fronting depends on the weakness of the pronoun, it will have to take place after vocabulary insertion, certainly not an innocuous requirement.

200

Martin Salzmann

Fifth, all overt resumptives that are not governed by prepositions undergo weak pronoun fronting in ZG. This fact is even more fatal for Aoun et al’s (2001) base-generation approach than for movement approaches involving spell-out (cf. 1.3.4.2 above): It is unclear how such a movement should be possible given Aoun et al.’s assumptions: At the point where the pronoun is substituted, the derivation has already proceeded too far for pronoun fronting to apply, in fact the operator has been remerged in the final landing site so that the derivation is basically finished. Subsequent pronoun fronting, targeting a position inside the lowest CP, would then be ruled out by Cyclicity. Furthermore, if pronoun fronting is triggered by a syntactic feature, it is unclear how the pronoun that is substituted during the derivation could have such a feature.65 It is save to conclude, then, that an approach that adheres to numerations as the basis of the reference set is very undesirable when movement and base-generation compete. Quite apart from these technical problems, I follow Heck et al. (2002) in assuming that it is generally undesirable to base the candidate/reference set on the input: The more complex a sentence becomes, the more structured the numeration will have to be, often reaching a complexity that is almost identical to that of the sentence to be derived. As a consequence, an input-free definition of the reference set is called for. 3.2.2.2. The LF determines the reference set If the numeration cannot be the decisive factor that determines the reference set, we have to resort to other options. In the present context, there remain three options: Either, the derivations have to have the same semantic interpretation, the same S-Structure or the same LF. Basing the definition of the reference set on the semantic interpretation has been shown to be undesirable (cf. Sternefeld 1997: 89¤.) because this would rule out all movement operations that do not lead to a truthfunctional di¤erence with respect to the base structure: Topicalization and scrambling should always be blocked by a non-movement derivation. Furthermore, paraphrases of all kinds should no longer be possible. Using the S-Structure (the structure at Spell-out) as a criterion also does not work in the case at hand because there are substantial di¤erences between a movement and a base-generation derivation: 65. The only way out, it seems, would be to assume that pronoun fronting is phonological, cf. Salzmann (2006b: 304) for discussion.

Silent resumptives in Zurich German possessor relativization

(90) a.

[ cp Op . . .

b. [ cp Op . . .

__]

201

movement

pron]

base-generation

Clearly, the two derivations di¤er too much on the surface to belong to the same reference set. Identical LFs as a criterion, however, fares better, given certain assumptions: Intermediate traces will have been deleted, and the bottom copy of the movement derivation is converted into a variable. Furthermore, through binding by the operator the resumptive pronoun will also function as a variable: (91) a. b.

! lx . . . x

movement

[ cp Op i . . . pron i ] ! lx . . . x

base-generation

[ cp Op i . . . x i ]

I will consequently take the LFs of movement and base-generation derivations to be su‰ciently similar for both to be part of the same reference set.66,67 Suppose then that both a gap and a movement derivation converge in the relativization of a subject or a direct object. What blocks the resumptive derivation? The established (trans-derivational) Economy constraints Shortest Path or Fewest Steps (e.g. Chomsky 1995) are not easily applicable because the two derivations are syntactically quite di¤erent. Furthermore, Sternefeld (1997) showed that they can be replaced by derivational or representational constraints thereby strongly reducing the number of trans-derivational competitions. In the case at hand, however, there must

66. More evidence suggesting that the definition of the reference set should be based on identical LFs can be found in Sternefeld (1997), Broekhuis & Dekkers (2000) and Broekhuis & Klooster (2007). 67. It is a well-known fact that resumptive pronouns impose semantic restrictions on the external head. More precisely, they require it to be specific and block scope reconstruction, cf. Doron (1982), Sun˜er (1998), Sharvit (1999), Boeckx (2003), Bianchi (2004/2008), Guilliot (2007). In that case, there will be competition with movement derivations whose lowest copy is interpreted as a definite description. Importantly, however, the copy of movement can also be interpreted as an indefinite description and then allows for scope reconstruction (cf. Guilliot 2007, Bianchi 2008). In that case, there will be no competition between movement and base-generation. The impact of scope for the definition of the reference set is discussed in detail in Salzmann (2009a/2009c).

202

Martin Salzmann

be competition, since no principle of grammar rules out base-generation once we take it to be in principle available in a language.68 But why exactly is movement more economical in this case? The issue is far from trivial. For my present concerns it is su‰cient to adopt a representational constraint that simply penalizes resumptives (cf. also Salzmann 2008, section 5.1):69 (92) *Resumptive Resumptive pronouns are not allowed. This translocal constraint will block base-generation in favor of movement if both derivations converge. It would, of course, be desirable to derive the preference for movement over base-generation from more primitive notions. Since a detailed discussion of this issue is essentially orthogonal to my main interests, I will leave it at this here.70,71,72 68. Given what was said in 1.2 above this may sound like a contradiction. It was claimed that, in principle, there seem to be two types of languages with resumption. One where resumption is freely available and coexists next to movement and one where it is just a last resort. Saying that base-generation is freely available may give the impression that resumption in Zurich German is in fact not a last resort but rather similar to Irish or Hebrew. But this is not correct. It depends on the level where this free availability holds. In Irish or Hebrew, the optionality holds at the surface, i.e. both gap and resumptive derivations not only converge, they are also both grammatical. In ZG, however, the optionality only holds as far as convergence goes. Since movement is taken to be more economical than base-generation, only one of them will be grammatical, the movement derivation. See also fn. 72. 69. Given the possibility of silent resumptives, constraints referring to overtness like SilentTrace in Pesetsky (1998) or the Avoid Pronoun Principle in Chomsky (1982: 63f.), van Riemsdijk (1989), Heck & Mu¨ller (2000: 44), and probably the interpretation in Mu¨ller & Sternefeld (2001: 60) are therefore inadequate. Willis (2000) analyses resumption in Welsh, which has the same distribution of resumptives in main clauses. He does not refer to economy, but rather makes a proposal in terms of A’-disjointness. See Salzmann (2009a, 2009c, to appear) for detailed discussion of this issue. 70. In Salzmann (2009a/2009c) I relate the ban against resumption/base-generation to a general ban against External Merge. Cf. also Salzmann (2008: section 5.1/to appear) for discussion. 71. This type of constraint is reminiscent of Optimality-theoretic syntax. See Salzmann (2008/2009a/2009b/to appear) and Salzmann & Seiler (2010) for arguments that some aspects of OT and MP should be combined to attain descriptive adequacy. Given my assumption that the reference set should be based on identical LFs rather than identical numerations, an approach that penalizes resumption by means of faithfulness constraints like Fill by Le´gendre et al. (1998) is automatically ruled out.

Silent resumptives in Zurich German possessor relativization

203

4. The wh/relativization asymmetry The attentive reader will have noticed that the ungrammaticality of (81)b is actually quite puzzling: Since possessor relativization involves a small pro, one expects the same possibility for wh-movement, thereby making it possible to relate the wh-pronoun to a position inside an island. Interestingly, however, this is not possible. The restriction holds more generally, wh-movement (and also topicalization) is incompatible with resumption throughout Zurich German, cf. (5). I will briefly discuss two possibilities to account for this wh-movement/relativization asymmetry. 4.1. Semantic factors One rather obvious explanation would relate the asymmetry to semantic di¤erences between the two constructions. It seems to be the case that cross-linguistically resumption is more widespread with relativization than with wh-movement (cf. e.g. Boeckx 2003, Bianchi 2004). Whether this is really due to semantic reasons, is, however, unclear. For some languages, this seems to be correct. In Hebrew (cf. e.g. Sharvit 1999: 591) for instance, resumption is possible with D-linked wh-operators but not with non-D-linked ones, suggesting that resumption requires a – in a sense yet to be made precise – specific operator:73 (93) a.

b.

*Mi nifgasˇta ito? who met.you with.him ‘Who did you meet with?’ Eyze student nifgasˇta ito? which student met.you with.him ‘Which student did you meet with?’

72. Once it is assumed that movement and base-generation compete, the question arises why some languages, e.g. Hebrew and Irish, allow both gap and resumptive structures in certain grammatical relations such as in the relativization of matrix direct object. Given what has been said so far, one would expect both derivations to compete, and only the gap-derivation should emerge as grammatical. In Salzmann (2009a/2009c) I argue that economy constraints such as *Resumptive do not hold universally, but rather interact with other constraints, so that di¤erent patterns of interaction between movement and base-generation become possible, just like the optionality in Irish/ Hebrew. 73. See Hoekstra (1991: 71f.), who observes that resumption in Frisian is restricted to definite wh-phrases.

204

Martin Salzmann

This explanation does not work for Zurich German, though, as there is no such e¤ect: Resumption with wh-possessors is equally unacceptable with simplex and D-linked wh-operators: (94) ??[Wem] i / [Welem Schu¨eler] i ha¨sch geschter who.dat which.dat student have.2s yesterday [pro i sin Vatter] ka¨ne gleert? his father got.to.know ‘Whose/Which student’s father did you meet yesterday?’ 4.2. The case of the operator Consider again the asymmetry in (81), repeated here in slightly adapted form: (95) a.

de Schu¨eler, Op i wo de Peter geschter the student C the Peter yesterday [pro i sin Vatter] ka¨ne gleert ha¨t his father got.to.know has ‘the student whose father Peter met yesterday’

b.

??[Wem] i ha¨t de Peter geschter who.dat has the Peter yesterday [pro i sin Vatter] ka¨ne gleert? his father got.to.know ‘Whose father did Peter meet yesterday?’

There is one obvious di¤erence between the two examples: In the whexample, the operator is overt and case-marked while in relativization it is zero and, as proposed in 3.1, unmarked for case. Base-generation is only possible with operators unmarked for case because otherwise their case feature cannot be checked/valued. In (95)b, the operator is casemarked and thus cannot be directly inserted into the operator position. While this may seem like a stipulation at this point, I believe that there is good reason to believe that a constraint along these lines is operative in many languages: Merchant (2004: ex. 24) makes the same claim for Greek. Resumption is possible with that-relatives but not with phrasal

Silent resumptives in Zurich German possessor relativization

205

wh-relatives and generally impossible with wh-movement, which also involves a phrasal antecedent:74 (96) a.

b.

o andras tou opiou (*tou) edhosa ta klidhia mou the man the which.gen him.gen gave.1sg the keys.acc mine ‘the man to whom I gave my keys’ o andras pou (tou) edhosa ta klidhia mou the man that him.gen gave.1sg the keys.acc mine ‘the man that I gave my keys to’

Here, the asymmetry is not between di¤erent types of A’-movement but within a given type, clearly suggesting that semantic factors cannot be decisive (both relatives are restrictive). Additionally, there is independent evidence internal to Zurich German: First, the only other A’-movement construction that allows resumptives in local movement is comparative deletion, as shown in (4). Crucially, the operator is also zero, so that it is possible that it is unmarked for case. Secondly, next to regular wh-movement that leaves gaps, Zurich German also has a construction which I have termed A’-splits in Salzmann (2006a). In that construction, the wh-phrase (and, for some speakers, a topic) occurs unmarked for case in Spec, CP while the putative extraction site is occupied by a resumptive pronoun (and, if PPs are involved, the preposition; interestingly, the dative resumptive is also obligatory for speakers who can drop them in local relativization; A’-splits thus pattern with long-distance relativization, recall fn. 4): (97) a.

Wele Maa i ha¨sch behauptet, dass t *(em i) which man have.2s claimed that you he.dat es Buech gge¨e¨ ha¨sch? a book given have.2s ‘To which man did you claim that you had given a book?’

74. According to Alexopoulou (2006: 69¤.) things are more complex in Greek. While this asymmetry is correct for restrictive relatives, it is not observed in non-restrictives where resumptives are claimed to be compatible with both that- and wh-relatives. Interestingly though, she mentions (2006: 86) that in non-restrictives the relative pronoun may appear in the nominative irrespective of the position relativized on. This may, of course, be an indication of base-generation, as in the A’-splits to be discussed below.

206

Martin Salzmann

b.

Weli Frau i ha¨sch behauptet, dass t i d which woman have.2s claimed that you in the Schuel bisch mit *(ere i)? school are with her ‘With which woman did you claim that you went to school?’

The wh-pronouns formally bear the direct case – recall that ZG only distinguishes direct case and one oblique case, dative –, but since the direct case is formally unmarked in Zurich German, it seems plausible to analyze these operators as appearing in a default form not carrying a case feature. Therefore, these examples are amenable to a base-generation analysis as well. Importantly, such A’-splits are also possible with possessors (thus showing insensitivity to locality): (98) Wele Maa i ha¨sch gsa¨it, dass geschter which man have.2s said that yesterday [pro i siis Huus] abbra¨nnt isch? his house burned.down is ‘Which man’s house did you say burned down yesterday?’ Obviously, the base-generated operator is unmarked for case, therefore making binding of the case-marked pro possible, in contrast with (95)b. Given the possibility of A’-splits, one expects the A’-split version of (95)b to be grammatical as there is no case problem anymore. However, this prediction is not borne out (wh-pronouns for nominative/accusative are identical): (99) *[We¨e¨r] i ha¨t de Peter geschter [pro i sin Vatter] ka¨ne gleert? who has the Peter yesterday his father got.to.know ‘Whose father did Peter meet yesterday?’ The restriction is more general: A’-splits are limited to cross-clausal dependencies, i.e. the resumptive may not occur within the same CP as its binder: (100) *Wele Maa i ha¨sch em i es Buech gge¨e¨? which man have.2s he.dat a book given ‘To which man did you give a book?’ Thus while relativization and wh-movement do not di¤er with respect to the availability of resumption as such, they do di¤er with respect to local resumption.

Silent resumptives in Zurich German possessor relativization

207

Intriguingly, West Flemish, which has also a very similar construction at least with possessors, is subject to exactly the same restriction (Haegeman 2003: 223; wh-pronouns are never case-marked overtly in West Flemish, possessor extraction is impossible even out of predicates): (101) a.

Wekken verpleegster zei-je gie dan-ze gisteren which nurse said-you you that-they yesterday eur us verkocht een? her house sold have ‘Which nurse’s house did you say that they sold yesterday?’

b. ??Wien was da doa gisteren zenen oto? who was that there yesterday his car ‘Whose car was that yesterday?’ Given the logic so far, a simple solution suggests itself: Base-generation is blocked because the more economical movement derivation is available. Since both derivations converge and since both have the same LF, they compete. For cases like (100), this is straightforward, the A’-split version will be blocked by the movement derivation in (5)a due to the economy constraint in (92). The possessor cases are more interesting: I repeat the relevant examples for convenience: (102) a.

*[Wem]1 ha¨t de Peter geschter [__1 sin Vatter] ka¨ne gleert? who.dat has the Peter yesterday his father got.to.know ‘Whose brother did Peter get to know yesterday?’

b. *[We¨e¨r] i ha¨t de Peter geschter [pro i sin Vatter] ka¨ne gleert? who has the Peter yesterday his father got.to.know ‘Whose father did Peter meet yesterday?’ c. [Wem sin Vatter]1 ha¨t de Peter geschter __1 ka¨ne gleert? who.dat his father has the Peter yesterday got.to.know ‘Whose father did Peter get to know yesterday?’ (102)a is impossible under a movement analysis because there is no possessor extraction in Zurich German. It is also ruled out under basegeneration because the wh-operator would end up with an unvalued casefeature. Consequently, there simply is no converging derivation for (102)a. (102)b, however, would then be expected to be possible under basegeneration: Since the operator is unmarked for case, it would be licensed in Spec, CP, like the operators in (97)–(98). Obviously, base-generation is blocked again by the more economical movement derivation in (102)c.

208

Martin Salzmann

Even though the two sentences seem to be very di¤erent on the surface, their corresponding LFs will be identical: Pied-piped material in (102)c will be reconstructed into the base-position with only the operator remaining in Spec, CP. (92) then singles out (102)c as grammatical. To conclude this section, let me point out that while in wh-movement movement derivations always block resumptive derivations in local dependencies, this is not the case in relativization. There is a simple explanation for this asymmetry: There is no overt relative pronoun that would make movement derivations involving oblique relations (datives and PPs) possible while in wh-movement (and topicalization) there are such pronouns.75,76,77 75. Interesting questions remain concerning possessor extraction from predicates (cf. (50)) where both possessor extraction and pied-piping are possible. Since both have the same LF, they must compete. But since both are grammatical, this also implies that they are equally economical. 76. There is one aspect of this proposal that remains problematic: Next to the A’-splits in (97), it is of course possible to have regular long-distance A’movement (generally without resumption, even though there is some speaker variation): (i) [Welem Maa]1 ha¨sch behauptet, dass t __1/ which.dat man have.2s claimed that you ??em es Buech gge¨e¨ ha¨sch? he.dat a book given has.2s ‘To which man did you claim that you had given a book?’ Long-distance movement and the A’-splits in (97) arguably have the same LFs. As a consequence, one would expect the base-generation derivations to be blocked by the economy constraint in (92), contrary to fact: Both movement and base-generation are possible. I will have to leave this for further research. No such optionality if found in relativization where resumptives appear across the board in long-distance relativization. In Bayer & Salzmann (in prep.) we relate this to Improper Movement resulting from movement of specific relative operators to the spec of the complementizer dass, which is only compatible with contrastive elements such as wh-operators, but crucially not relative operators. 77. Much of this analysis can probably be extended to West Flemish. The major di¤erence between West Flemish and Zurich German seems to be that while there is a fully fledged resumptive system in ZG, resumption in West Flemish is limited to possessors. Resumptives for subjects and objects in transparent positions do not exist; overt resumptives are only found inside islands. I do not know whether these are regular resumptives or rather intrusive pronouns as in English. The discussion in Haegeman (2003: 237, 239) seems to suggest the latter.

Silent resumptives in Zurich German possessor relativization

209

5. The derivation of relative clauses To complete the description of ZG relativization, let me briefly elucidate my assumptions about how the relative clause is related to the head noun. Since reconstruction e¤ects were shown to occur in both movement and base-generation derivations, they will either result from interpreting the lower copy of a movement chain or from semantic reconstruction/NP ellipsis. There is one aspect of reconstruction, however, that requires special attention. As discussed in detail in Salzmann (2006b), while reconstruction e¤ects are generally very robust in ZG relativization, there is no reconstruction for Principle C, neither in gap nor in resumptive relatives: (103) a.

d [Naaforschige u¨ber de Peter i ], won er i the investigations about the Peter C he mer __ lieber verschwige het me.dat prefer conceal had.subj ‘the investigations about Peter i that he i would have preferred to concealed from me’

b. s [Buech u¨ber de Peter i ], won er em the book about the Peter C he he.dat jede Wert abgsproche ha¨t every value denied has ‘the book about Peter i that he i denied any value’ I have interpreted these facts as arguing in favor of a Matching Analysis of relative clauses (cf. e.g. Citko 2001, Sauerland 2003) where the relative operator (or rather: its NP-complement) is deleted under identity with the external head (outline indicates PF-deletion):78 (104) s [Buech] j [cp [Op [Buech] j ]1 wo de Peter __1 gern ha¨t] the book book C the Peter likes ‘the book that Peter likes’

78. For sake of concreteness, I adopt the traditional NP-adjunction approach, but the data are fully compatible with approaches based on complementation as e.g. Sternefeld (2006), where the external determiner takes the head noun as its sister and the relative clause as its right-hand specifier.

210

Martin Salzmann

Since an ellipsis operation is involved, we also expect special properties of ellipsis to emerge. Vehicle Change, the possibility to turn an R-expression into a pronoun, is such an e¤ect: (105) a. b.

*John likes Mary i, and she i does (like her i), too. John likes Mary i, and she i knows that I do (like her i), too.

In the first example, Vehicle Change does not lead to an improvement as there is still a Principle B violation. In the second one, however, with an additional level of embedding, the sentence is well-formed. Since relative clauses also involve an ellipsis operation under the matching analysis, Vehicle Change also takes place and turns an R-expression into a pronoun, thereby voiding the Condition C violation. A sentence like (103)a will therefore receive the following LF (strikethrough indicates LFdeletion, the operator is converted into a variable in the lower copy): (106) a.

d [Naaforschige u¨ber de Peter i ] j, the investigations about the Peter [ cp [Op [Naaforschige u¨ber in i ] j ]1 won investigations about him C

b. er i mer [x [Naaforschige u¨ber in i ] j ]1 he me.dat investigations about him lieber verschwige het prefer conceal had.subj The relative clause is therefore just as grammatical as the following simple clause: (107) Er i het mer die Naaforschige u¨ber in i lieber verschwige. he has me.dat these investigations about him prefer concealed ‘He would have preferred to conceal these investigations about him from me.’ Importantly, the matching analysis (including possible Vehicle Change) also applies under base-generation so that the semantic content of the (complement of the) operator and not that of the external head is copied into the reconstruction site. This means that the relative operator is just a D-element taking a case-unmarked NP-complement. A sentence like

Silent resumptives in Zurich German possessor relativization

211

(103)b will thus have the following representation, which explains the absence of a Condition C e¤ect:79 (108) s [Buech u¨ber de Peter i ] j, [ cp [Op [Buech u¨ber in i] j ] won the book about the Peter book about him C er i [dp em [Buech u¨ber in i ] j ] jede Wert abgsproche ha¨t he the.dat book about him every value denied has The meaning is thus equivalent to the following simple clause: Buech u¨ber in i jede Wert abgsproche. (109) Er i ha¨t em he has the.dat book about him every value denied ‘He denied the book about him any value.’

6. Conclusion Resumptive relatives in Zurich German relatives show mixed results with respect to movement properties. On the one hand, they are insensitive to locality, on the other, they show movement e¤ects like reconstruction and SCO e¤ects. While the latter e¤ects are directly compatible with movement, the locality facts argue in favor of base-generation. Recent approaches have shown that there are ways of handling reconstruction under base-generation, but it is still very unclear whether movement approaches can explain why resumption voids locality violations. After dismissing most movement approaches, we finally entertained the possibility that it is simply the overtness of the tail of the movement chain that makes movement out of islands possible, e.g. as in Pesetsky (1998). While resumptives are mostly overt in ZG and are thus basically compatible with this approach, possessive relatives provide crucial evidence against this position: Since the underlying possessive construction can involve a small pro in the position of the possessor and since possessive relatives are insensi-

79. Principle C e¤ects do not re-emerge if reconstruction is independently required for positive binding conditions such as for variable binding, as has been observed e.g. by Citko (2001), Sauerland (2002), Guilliot & Malkawi (2006), Guilliot (2007), and Rouveret (2008). This is a general property of relatives in (Swiss) German, cf. Salzmann (2006b). It shows that the absence of Condition C e¤ects cannot be explained by assuming that reconstruction only takes place if necessary. Rather, absence of Condition C e¤ects in Zurich German is the result of systematic Vehicle Change.

212

Martin Salzmann

tive to locality constraints without there being an overt resumptive for the possessor, there must be an empty resumptive pronoun. As a consequence, only base-generation is a possibility for possessive relatives. It is then most economical to adopt a base-generation analysis for all the other resumptive relatives. Resumptives are analyzed as a last resort when gap derivations fail: They either occur to make oblique case visible (dative resumptives) or to amnesty locality violations (resumptives inside islands). On the other hand, resumption must be blocked for subjects and direct objects. For this to be possible, gap and resumptive relatives must belong to the same reference set. We have argued that competition in ZG relatives provides evidence that the reference set should be determined on the basis of identical LFs, contrary to what is often assumed in the literature. Resumptives are penalized by a general representational economy constraint *Resumptive so that movement derivations emerge as grammatical. Finally, there is an intriguing wh-movement/relativization asymmetry with respect to resumption in ZG. It was argued that the limited availability of resumption with wh-movement is due to case-marking: Basegenerated case-marked wh-operators are not licensed, therefore ruling out resumption in most cases. Only an alternative wh-construction (so-called A’-splits) involving operators unmarked for case is compatible with resumption.

References Adger, David and Ramchand, Gillian 2005 Merge and Move: Wh-Dependencies Revisited. Linguistic Inquiry 36, 161–193. Alexiadou, Artemis; Haegeman, Liliane and Melita Stavrou 2007 Noun Phrase in the Generative Perspective. Mouton de Gruyter: Berlin/New York. (¼Studies in Generative Grammar 71). Alexopoulou, Theodora 2006 Resumption in Relative Clauses. NLLT 24, 57–111. Aoun, Yoseph; Choueiri, Lina & Norbert Hornstein 2001 Resumption, Movement, and Derivational Economy. Linguistic Inquiry 32, 371–403. Aoun, Yoseph and Yafei Li 1993 Wh-Elements in Situ: Syntax or LF? Linguistic Inquiry 24, 199– 238.

Silent resumptives in Zurich German possessor relativization

213

Authier, J. Marc P. 1992 Is French a Null Subject Language in the DP? Probus 4, 1–16. Barbiers, Sjef; Koeneman, Olaf; Marika Lekakou and Margreet van der Ham 2008 Microvariation in Syntactic Doubling. Emerald: Bingley. Bayer, Josef; Markus Bader and Michael Meng 2001 Morphological Underspecification meets oblique Case: Syntactic and processing E¤ects in German. Lingua 111, 465–514. Bayer, Josef and Martin Salzmann in prep. That-trace e¤ects and resumption as improper movement. Talk presented at the DGfS 2009 in Osnabru¨ck, Ms. University of Konstanz. Belletti, Adriana 2006 Extending doubling to non-local domains: complete vs. partial copying þ deletion and related reconstruction issues. In Patrick Brandt and Eric Fuss (eds.), Form, Structure, and Grammar. Akademie-Verlag: Berlin, 129–136. Bianchi, Valentina 2004 Resumptive Relatives and LF Chains. In Luigi Rizzi (ed.), The Cartography of syntactic Structures. Volume 2. OUP: New York/Oxford, 76–114. Bianchi, Valentina 2011 Some notes on the ‘specificity e¤ects’ of optional resumptive pronouns. In Alain Rouveret (ed.), Resumptive Pronouns at the Interfaces. Amsterdam: John Benjamins, 319–342. Boeckx, Cedric 2003 Islands and Chains. Resumption as Stranding. John Benjamins: Amsterdam. Boeckx, Cedric and Norbert Hornstein 2008 Superiority, reconstruction and islands. In Robert Freidin, Carlos P. Otero and Maria Luisa Zubizarreta (eds.), Foundational Issues in Linguistic Theory. Essays in Honor of Jean-Roger Vergnaud. MIT Press: Cambridge, Massechusetts, 197–225. Brandstetter, Renward 1904 Der Genitiv der Luzerner Mundart in Gegenwart und Vergangenheit. Zu¨rcher und Furrer: Zu¨rich. Bresnan, Joan and Sam A. Mchombo 1987 Topic, Pronoun, and Agreement in Chichewa. Language 63, 741–782. Broekhuis, Hans 2008 Derivations and Evaluations: object shift in the Germanic languages. Mouton: Berlin etc. Broekhuis, Hans and Joost Dekkers 2000 The Minimalist Program and Optimality Theory: Derivations and Evaluations. In Joost Dekkers, Frank van der Leeuw and Jeroen van der Weijer (eds), Optimality Theory. Phonology, Syntax, and Acquisition. OUP: New York/Oxford, 386–422.

214

Martin Salzmann

Broekhuis, Hans and Wim Klooster 2007 Merge and Move as costly operations. GAGL 45, 17–37. Cecchetto, Carlo 2005 Reconstruction in relative clauses and the copy theory of traces. In Pierre Pica, Johan Rooryck and Jeroen van Craenenbroeckx (eds.), Linguistic Variation Yearbook 2005. John Benjamins: Amsterdam, 3–35. Chao, Wynn and Peter Sells 1983 On the Interpretation of Resumptive Pronouns. NELS 13, 47– 61. Chomsky, Noam 1982 Some Concepts and Consequences of the Theory of Government and Binding. MIT Press: Cambridge, Massechusetts. Chomsky, Noam 1995 The Minimalist Program. MIT Press: Cambridge, Massechusetts. Chomsky, Noam 2001 Derivation by Phase. In Michael Kenstowicz (ed.), Ken Hale: a Life in Language. MIT Press: Cambridge, Massechusetts, 1–52. Chung, Sandra 1982 On Extending the Null Subject Parameter to NPs. In D. Flickinger, M. Macken, and N. Wiergard (eds.), Prodeedings of the First West Coast Conference on Formal Linguistics, Stanford. Citko, Barbara 2001 Deletion under Identity in Relative Clauses. NELS 31, 131–145. Corver, Norbert 1990 The Syntax of Left Branch Extraction. Ph.D. dissertation: Tilburg. Corver, Norbert, and van Koppen, Marjo 2010 Ellipsis in Dutch possessive noun phrases: a micro-comparative approach. The Journal of Comparative Germanic Linguistics 13: 99–140. Delsing, Lars-Olof 1998 Possession in Germanic. In Artemis Alexiadou and Chris Wilder (eds.), Possessors, Predicates and Movement in the Determiner Phrase. John Benjamins: Amsterdam, 87–108. Demske, Ulrike 2001 Merkmale und Relationen. Diachrone Studien zur Nominalphrase im Deutschen. de Gruyter: Berlin/New York. (¼Studia Linguistica Germanica 56). Demirdache, Hamida 1991 Resumptive Chains in Restrictive Relatives, Appositives and Dislocation Structures. Ph.D. Dissertation: MIT. Dieth, Eugen 1938 Schwyzertu¨tschi Diala¨ktschrift. Leitfaden. Zu¨rich: Orel Fu¨ssli Verlag.

Silent resumptives in Zurich German possessor relativization

215

Dikken, Marcel den 1998 Predicate Inversion in DP. In Artemis Alexiadou and Chris Wilder (eds.), Possessors, Predicates and Movement in the Determiner Phrase. John Benjamins: Amsterdam, 177–214. Dikken, Marcel den 2006 Specificational Copular Sentences and Pseudoclefts. In Martin Everaert and Henk van Riemsdijk (eds.), The Blackwell Companion to Syntax. Blackwell: Oxford, 292–409. Dikken, Marcel den; Andre´ Meinunger and Chris Wilder 2000 Pseudoclefts and ellipsis. Studia Linguistica 54, 41–89. Doron, Edith 1982 On the syntax and semantics of resumptive pronouns. Texas Linguistic Forum 19, 1–48. Elbourne, Paul 2001 E-type anaphora as NP deletion. Natural Language Semantics 9, 241–288. Fischer, Ludwig 1960 Luzerndeutsche Grammatik. Comenius: Hitzkirch. Fiva, Toril 1987 Possessor chains in Norwegian. Novus: Oslo. Fleischer, Ju¨rg 2002 Die Syntax von Pronominaladverbien in den Dialekten des Deutschen: Eine Untersuchung zu Preposition Stranding und verwandten Pha¨nomenen. Steiner: Wiesbaden/Stuttgart. Fox, Danny 2002 Antecedent-Contained Deletion and the Copy Theory of Movement. Linguistic Inquiry 33, 63–96. Gallmann, Peter 1992 Dativanhebung? GAGL 35, 92–122. Gallmann, Peter 1996 Die Steuerung der Flexion in der DP. Linguistische Berichte 164/ 199, 283–314. Georgi, Doreen and Gereon Mu¨ller 2010 Noun phrase structure by reprojection. Syntax 13, 1–36. Georgi, Doreen and Martin Salzmann 2010 DP-internal double agreement is not double Agree: a unification with possessor agreement GLOW-in-Asia VIII, Beijing. Georgopoulos, Carol 1985 Variables in Palauan Syntax. Natural Language and Linguistic Theory 3, 59–94. Georgopoulos, Carol 1991 Syntactic Variables: Resumptive Pronouns and A’ Binding in Palauan. Kluwer: Dordrecht/Boston/London. Grimm, Jacob & Wilhelm Grimm 1854–1960 Deutsches Wo¨rterbuch. S. Hirzel: Leipzig.

216

Martin Salzmann

Guilliot, Nicolas 2006 A top-down analysis for reconstruction. Lingua 116, 1888–1914. Guilliot, Nicolas 2007 Reconstructing resumption. Handout from the workshop on ‘‘Resumption at the interfaces’’, June 22, 2007. Guilliot, Nicolas and Nouman Malkawi 2006 When Resumption determines Reconstruction. Proceedings of WCCFL 25, 168–176. Haegeman, Liliane 2003 The external Possessor Construction in West Flemish. In Martine Coene and Ives d’Hulst (eds.), From NP to DP. Volume 2. John Benjamins: Amsterdam. (¼Linguistik Aktuell / Linguistics Today 56), 221–256. Haegeman, Liliane 2004 DP-Periphery and Clausal Periphery: Possessor Doubling in West Flemish. In David Adger, Ce´cidle de Cat and George Tsoulas (eds.), Peripheries. Syntactic Edges and their E¤ects. Kluwer: Dordrecht/Boston/London, 211–240. Haegeman, Liliane 2009 Three positions for prenominal possessors in West Flemish. Paper presented at the University of Utrecht, OTS. Atoms and Lasws of the NP. 2–3 July 2009. Haider, Hubert 1988 Die Struktur der deutschen Nominalphrase. Zeitschrift fu¨r Sprachwissenschaft 7/1, 32–59. Haider, Hubert 1992 Die Struktur der Nominalphrase – Lexikalische und funktionale Strukturen. In Ludger Ho¤mann (ed.), Deutsche Syntax: Ansichten und Aussichten. Jahrbuch 1991 des Instituts fu¨r deutsche Sprache. de Gruyter: Berlin/New York, 304–333. Heck, Fabian and Gereon Mu¨ller 2000 Repair-Driven Movement and the Local Optimization of Derivations. Ms. University of Stuttgart/Mannheim. Heck, Fabian, and Gereon Mu¨ller 2007 Extremely Local Optimization. Proceedings of WECOL 17: 170–182. Heck, Fabian; Mu¨ller, Gereon; Fischer, Silke; Vikner, Sten and Tanja Schmid 2002 On the nature of the input in optimality theory. Linguistic Review 19, 345–376. Hoekstra, Jarich 1991 Expletive der and resumptive pro in Frisian. Leuvense Bijdragen 80, 61–80. Idiotikon 1881¤. Schweizerisches Idiotikon. Wo¨rterbuch der schweizerdeutschen Sprache, gesammelt auf Veranstaltung der Antiquarischen Gesellschaft in Zu¨rich. Huber, Frauenfeld, 1881¤.

Silent resumptives in Zurich German possessor relativization

217

Kappus, Martin 1999 Possessive Structures and the DP in Swabian. In Hans Otto Spillmann and Ingo Warnke (eds.), Internationale Tendenzen der Syntaktik, Semantik und Pragmatik: Akten des 32. Linguistischen Kolloquiums in Kassel 1997. Peter Lang: Frankfurt, 203–210. Keenan, Edward and Bernard Comrie 1977 Noun phrase accessibility and Universal Grammar. Linguistic Inquiry 8, 63–99. Koptjevskaja-Tamm, Maria 2002 Adnominal possession in the European languages: form and function. Sprachtypologie und Universalienforschung (STUF) 55, 141–172. Koptjevskaja-Tamm, Maria 2003 Possessive noun phrases in the languages of Europe. In Frans Plank (eds.), Noun Phrase Structure in the Languages of Europe. Mouton de Gruyter: Berlin/New York. (¼Empirical Approaches to Language Typology, Eurotyp 20-7), 621–722. Legendre, Ge´raldine; Smolensky, Paul and Colin Wilson 1998 When is less more? Faithfulness and Minimal Links in whchains. In Pilar Barbosa, Danny Fox, Martha McGinnis and David Pesetsky (eds.), Is the best good enough? Optimality and Competition in Syntax. MIT Press: Cambridge, Massechusetts, 249–289. Leu, Thomas 2008 The internal syntax of determiners. PhD Dissertation: New York University. Lindauer, Thomas 1995 Genitivattribute. Eine morphosyntaktische Untersuchung des deutschen DP/NP-Systems. Niemeyer: Tu¨bingen. (¼Reihe Germanistische Linguistik 155). Lindauer, Thomas 1998 Attributive Genitive Constructions. In Artemis Alexiadou and Chris Wilder (eds.), Possessors, Predicates and Movement in the Determiner Phrase. John Benjamins: Amsterdam. (¼Linguistik Aktuell / Linguistics Today 22), 109–140. Lo¨bel, Elisabeth 1996 Kategorisierung der Possessiva als Adjektive in der NP/DP. In Thilo Tappe and Elisabeth Lo¨bel (eds.), Die Struktur der Nominalphrase. Universita¨t Wuppertal: Wuppertal. (¼Wuppertaler Arbeitspapiere zur Sprachwissenschaft 11), 48–94. Lyons, Christopher 1999 Definiteness. CUP: Cambridge. McCloskey 1990 Resumptive Pronouns, A’-Binding and Levels of Representation in Irish. In Randall Hendrick (ed.), The Syntax of the Modern

218

Martin Salzmann

Celtic Languages. Volume 23 of Syntax and Semantics, Academic Press: 199–248. McCloskey, James 2002 Resumption, Successive Cyclicity, and the Locality of Operations. In Samuel David Epstein and T. Daniel Seely (eds.), Derivation and Explanation in the Minimalist Program. Blackwell: Oxford, 184–226. Merchant, Jason 2004 Resumptivity and Non-Movement. Studies in Greek Linguistics 24, 471–481. Mu¨ller, Gereon and Wolfgang Sternefeld 2001 The Rise of Competition in Syntax: A Synopsis. In Gereon Mu¨ller and Wolfgang Sternefeld (eds.), Competition in Syntax. De Gruyter: Berlin/New York, 1–68. Olsen, Susan 1989 Das Possessivum: Pronomen, Determinans oder Adjektiv? Linguistische Berichte 120, 133–153. Olsen, Susan 1996 Dem Possessivum seine Eigentu¨mlichkeit. In Thilo Tappe and Elisabeth Lo¨bel (eds.), Die Struktur der Nominalphrase. Universita¨t Wuppertal: Wuppertal. (¼Wuppertaler Arbeitspapiere zur Sprachwissenschaft 11), 112–143. Pesetsky, David 1998 Some Optimality Principles of Sentence Pronunciation. In Pilar Barbosa, Danny Fox, Martha McGinnis & David Pesetsky (eds.), Is the best good enough? Optimality and Competition in Syntax. MIT Press: Cambridge, Massechusetts, 337–383. Perlmutter, David 1972 Evidence for shadow pronouns in French relativization. In Paul Peranteau, Judith Levi, & Gloria Phares (eds.), The Chicago which hunt: Papers from the relative clause festival. Chicago: CLS. Riemsdijk, Henk van 1989 Swiss Relatives. In Dany Jaspers, Wim Klooster, Yvan Putseys & Pieter Seuren (eds.), Sentential Complementation and the Lexicon. Studies in Honour of Wim de Geest. Foris: Berlin. (¼Linguistic Models 13), 343–354. Riemsdijk, Henk van 2008 Identity Avoidance: OCP e¤ects in Swiss relatives. In Robert Freidin, Carlos P. Otero and Maria Luisa Zubizarreta (eds.), Foundational Issues in Linguistic Theory. Essays in Honor of Jean-Roger Vergnaud. MIT Press: Cambridge, Massechusetts, 227–250. Rouveret, Alain 2008 Phasal agreement and reconstruction. In Robert Freidin, Carlos P. Otero und Maria Luisa Zubizarreta (eds.), Foundational Issues

Silent resumptives in Zurich German possessor relativization

219

in Linguistic Theory. Essays in Honor of Jean-Roger Vergnaud. MIT Press: Cambridge, Massechusetts, 167–195. Sag, Ivan A. und Wasow, Thomas 1999 Syntactic Theory. A formal Introduction. CSLI Publications: Stanford. Salzmann, Martin 2006a Resumptive Pronouns and Matching E¤ects in Zurich German Relative Clauses as Distributed Deletion. In Noureddine Elouazizi, Frank Landsbergen, Maika Poss & Martin Salzmann (eds.), Leiden Papers in Linguistics 3.1, 17–50. Salzmann, Martin 2006b Resumptive Prolepsis. A Study in indirect A’-dependencies. LOT: Utrecht. (¼LOT Dissertation Series 136). online: www.lotpublications.nl/publish/articles/001910/bookpart.pdf Salzmann, Martin 2008 Variation in resumption requires violable constraints – A case study in Alemannic relativization. In Optimality Theory and Minimalism: Interface Theories (Linguistics in Potsdam 28), ed. by Hans Broekhuis and Ralf Vogel, 99–132. Salzmann, Martin 2009a When movement and base-generation compete – on the definition of the reference set, the typology of resumption, and ranked economy constraints. Groninger Arbeiten zur Germanistischen Linguistik 48, 27–63. Center for Language and Cognition Groningen. http://gagl.eldoc.ub.rug.nl Salzmann, Martin 2009b Di¤erent notions of variation and their reflexes in Swiss German relativization. In Andreas Dufter, Ju¨rg Fleischer & Guido Seiler (eds): Describing and Modeling Variation in Grammar. Berlin: Mouton de Gruyter, 135–161. Salzmann, Martin 2009c When movement and base-generation compete. The definition of the reference set and parameterized preferences for elementary operations. Linguistics in the Netherlands 2009, 64–77. Salzmann, Martin & Guido Seiler 2010 Variation as the exception or the rule? Swiss relatives, revisited. Sprachwissenschaft 35, 79–117. Salzmann, Martin to appear On three types of variation in resumption – evidence in favor of violable and ranked constraints. To appear in Hans Broekhuis & Ralf Vogel (eds.), Derivation and filtering. Equinox. Sauerland, Uli 2003 Unpronounced Heads in relative Clauses. In Kerstin Schwabe & Susanne Winkler (eds.), The Interfaces: deriving and interpreting omitted structures. John Benjamins: Amsterdam, 205–226.

220

Martin Salzmann

Schirmunski, Viktor 1962 Deutsche Mundartkunde. Berlin: Akademie Verlag. Sharvit, Yael 1999 Resumptive Pronouns in Relative Clauses. Natural Language and Linguistic Theory 17, 587–612. Simon, Horst 2008 Methodische Grundfragen zu einer Vergleichenden Syntax deutscher Dialekte. In Karin Donhauser, Elvira Glaser & Marcel Vuillaume (eds.), Empirische Grundlagen moderner Grammatikforschung. [part of: Akten des XI. Internationalen Germanistenkongresses Pairs 2005 ‘‘Germanistik im Konflikt der Kulturen’’, ed. by Jean-Marie Valentin and He´le`ne Vinckel.] Bern: Lang, 59–70. Shlonsky, Ur 1992 Resumptive Pronouns as a Last Resort. Linguistic Inquiry 23, 443–468. Stalder, Franz Joseph 1819 Die Landessprachen der Schweiz oder schweizerische Dialektologie, mit kritischen Sprachbemerkungen beleuchtet. Sauerla¨nder: Aarau. Sternefeld, Wolfgang 1997 Comparing reference sets. In Chris Wilder, Hans-Martin Ga¨rtner and Manfred Bierwisch (eds.), The Role of Economy Principles in Linguistic Theory. AkademieVerlag, Berlin, 81–114. Sternefeld, Wolfgang 2000 Semantic vs. Syntactic Reconstruction. SfS-Report-02-00 Sternefeld, Wolfgang 2006 Syntax. Eine morphologisch motivierte Generative Beschreibung des Deutschen. Stau¤enburg: Tu¨bingen. Strunk, Jan 2004 Possessive Constructions in Modern Saxon. MA thesis: Stanford. Strunk, Jan 2005 Pro-drop in nominal possessive Constructions. In Miriam Butt and Tracy Holloway King (eds.), Proceedings of the LFG05 Conference. CSLI Publications: Stanford. 447–467. Sun˜er, Margerita 1998 Resumptive Restrictive Relatives: A crosslinguistic Perspective. Language 74, 335–364. Toman, Jindrich 1998 A discussion of resumptives in Colloquial Czech. In Zeljko Boskovic, Steven Franks & William Snyder (eds.), Formal Approaches to Slavic Linguistics. The Connecticut Meeting 1997. Michigan Slavic Publications: Ann Arbor, 303–318. Vries, Mark de 2006 Possessive Relatives and (Heavy) Pied-Piping. Journal of Comparative Germanic Linguistics 9:1, 1–52.

Silent resumptives in Zurich German possessor relativization Weber, Albert 1964

221

Zu¨richdeutsche Grammatik: Ein Wegweiser zur guten Mundart. Schweizer Spiegel Verlag: Zu¨rich. (¼Grammatiken und Wo¨rterbu¨cher des Schweizerdeutschen 1). Weerman, Fred and Petra de Wit 1999 The decline of the genitive in Dutch. Linguistics 37, 1155–1192. Weiss, Helmut 1998 Syntax des Bairischen. Studien zur Grammatik einer natu¨rlichen Sprache. Niemeyer: Tu¨bingen. (¼Linguistische Arbeiten 391). Weiss, Helmut 2006 The Possessor that appears twice. Variation, Structure and Function of possessive Doubling in German. In Sjef Barbiers, Margreet van der Ham, Olaf Koeneman and Marika Lekakou (eds.), Papers presented at Workshop on Syntactic Doubling, Amsterdam, March 2006. Meertens Institut: Amsterdam. Weiss, Helmut 2008 The possessors that appears twice, variation, structure and function of possessive doubling in German. In Sjef Barbiers, Olaf Koeneman, Marika Lekakou, and Margreet van der Ham (eds.), Microvariation in Syntactic Doubling. Syntax and Semantics Volume 36. Bingley: Emerald, 381–401. Willis, David 2000 On the distribution of resumptive pronouns and wh-trace in Welsh. Journal of Linguistics 36, 531–573. Zifonun, Gisela 2003 Dem Vater sein Hut. Der Charme des Substandards und wir wir ihm gerecht werden. Deutsche Sprache 2/03, 97–126.

Anti-agreement with subjects and possessors from a typological perspective: A case for null pronouns or for economy? Albert Ortmann 1. Introduction1 The phenomenon of so-called ‘anti-agreement’ in Hungarian adnominal possession has received some attention in the literature, usually from a syntactic perspective; see Szabolcsi (1994) and E´. Kiss (2002: 170f ) as well as the references in these works. Most notably, den Dikken (1999) considers the facts in their entire complexity, and, using a more recent version of Principles and Parameters Theory, points out the parallels to anti-agreement at the clause level in Welsh. The approach that he puts forward, however, faces a number of non-trivial intricacies. Regarding non-realizations of a possessor phrase as null pronouns, and accordingly treating them on a par with overt pronouns, furthermore, none of the authors draws the connection to (i) the cross-linguistic role of salience and economy in agreement splits, or to (ii) the role of economy in Hungarian noun plurals in contexts other than possessor agreement. I argue that an economy-based constraint analysis has at least the same descriptive adequacy as the assumption of null pronouns; it can, moreover, claim a higher degree of explanative adequacy – provided that it takes into account the typological background of related agreement asymmetries that are ruled by salience, comprising, in addition to the distinction pronominal vs. lexical NP, such dimensions as animacy and definiteness. As 1. The work reported here was carried out in the research Unit FOR 600 ‘‘Functional concepts and frames’’, sponsored by the German Research Foundation (DFG). I would like to thank Doris Gerland for her support with the Hungarian data, an anonymous reviewer for valuable criticism, and the editors of the present volume, Peter Gallmann and Melani Wratil, as well as the participants of the workshop ‘Nullpronomen’, held at the Institut fu¨r Germanistische Sprachwissenschaft, Friedrich-Schiller-Universita¨t Jena in September 2006, for helpful feedback. Special thanks go to Marcel den Dikken for showing me his readiness to regard my criticism of his work as constructive criticism.

224

Albert Ortmann

I will show, a natural solution of Hungarian anti-agreement lends itself once the interplay of two di¤erent economy splits is acknowledged, namely a salience-driven plural agreement split and a DP-internal plurality split, both of which ‘conspire’ to yield the complex pattern of Hungarian. A comparison with anti-agreement in Turkish, which is accounted for by slightly re-ranking the constraints in the sense of factorial typology, underpins the analysis. The paper is organised as follows: In section 2, the empirical generalizations of Hungarian possessor (anti-)agreement are stated. In section 3, the account by den Dikken (1999) is appraised, both from a theoretical and from a typological perspective. Section 4 contrasts the economic strategy of Hungarian number marking in the noun phrase with that of other languages in terms of an Optimality-Theoretical analysis. Section 5 assesses salience-governed plurality asymmetries in verbal agreement. Section 6 provides OT analyses of these asymmetries. Against this background, constraint analyses of Hungarian and Turkish possessor antiagreement are presented in section 7. Section 8 concludes with the major results.

2. Hungarian possessor agreement In Hungarian, possessed nouns display morphological agreement with respect to the category of person.2 (1) Hungarian a. az e´n ha´z-am df pron1sg house-por1sg ‘my house’

b. a mi ha´z-unk df pron1pl house-por1pl ‘our house’

c. a te ha´z-ad df pron2sg house-por2sg ‘your(sg) house’

d. a ti ha´z-atok df pron2pl house-por2pl ‘your(pl) house’

2. I will use the following abbreviations for grammatical categories: anim ‘animate’, cl ‘noun class’, cop ‘copula’, det ‘determiner’, df ‘definite article’, f ‘feminine’, indef ‘indefinite article’ irr ‘irrealis’, m ‘masculine’, nom ‘nominative’, obj ‘object’, past ‘past tense’, perf ‘perfective’, pl ‘plural’, por ‘possessor’, poss_pron ‘possessive pronoun’, prog ‘progressive, pron ‘personal pronoun’, sg ‘singular’, subj ‘subject’; 1, 2 and 3 represent first, second and third person, respectively.

Anti-agreement with subjects and possessors

225

e. az o kalap-ja df pron3sg hat-por3sg ‘his/her hat’ Note that with non-lexical possessors the additional realization of a personal pronoun preceding the possessed noun is optional;3 thus a ha´z-am, a ha´z-ad, and a kalap-ja, respectively, would be variants with non-focus possessors.4

3. As far as colloquial Hungarian is concerned, by contrast, Doris Gerland (p.c.) observes that the presence vs. absence of pronominal possessors tends to be governed by the dimension of alienability; that is, with alienable possession a pronoun is allowed even for non-focus possessors: (i) inalienable: a. az ap-a´d df father-2sg ‘your father’

b. a fe´rfj-em df husband-1sg ‘my husband’

alienable: c. az e´n ha´z-am df pron1sg house-1sg ‘my house’

d. a te kalap-od df pron2sg hat-2sg ‘my hat’

Gerland & Ortmann (2009) propose an analysis that is coached in the typology of adnominal possession. 4. Besides the nominative (or, according to E´. Kiss 2002: 157, the ‘‘caseless’’) possessor construction illustrated in (2) and (ia), there is also the possibility of realising a possessor phrase in the dative (ib). (i) a. az egyetemista´-k proble´ma´-i df student-pl problem-pl.por3 ‘the problems of the students’ b. az egyetemista´-k-nak a proble´ma´-i df student-pl-dat df problem-pl.por3 ‘the problems of the students’ c. az egyetemista´-k-nak ezek a proble´ma´-i df student-pl-dat dem df problem-pl.por3 ‘these problems of the students’ Note that the dative possessor structurally di¤ers from the nominative possessor. This is obvious from the fact that the former precedes both the article and the demonstrative pronoun, as (ic) shows, and furthermore from the possibility of extraction; see Szabolcsi (1994: 198¤ ) and E´. Kiss (2002: 157f ) for empirical and analytical details.

226

Albert Ortmann

In addition to the specification of person, the possessor agreement morphology also includes a specification of number. Crucially, however, plural agreement only occurs with pronominal possessors, not with lexical possessors. Compare (2a, b) and (2c, d): in (2b) the possessor agreement displays the plural of the possessor, whereas in (2d) the unmarked ‘singular’ is used as the default form. (2) a. az o ha´z-a b. az o ha´z-uk df pron3sg house-por3sg df pron3sg house-por3pl ‘his/her house’ ‘their house’ c. a nagyne´ni ha´z-a df aunt house-por3sg ‘the aunt’s house’

d. a nagyne´ni-k ha´z-a df aunt-pl house-por3sg ‘the aunts’ house’

We are thus dealing with an agreement split of the sort ‘‘no number agreement with lexical arguments’’. Typologically, such splits are quite common and are often referred to as ‘anti-agreement’. In Hungarian the domain of this split is adnominal possession.5 This is obvious because of the fact that we always encounter ‘full’ agreement at the clause level, that is, the verb obligatorily exhibits number agreement, regardless of whether the subject is pronominal or not: (3) a. A nagyne´ni so¨rt isz-ik. b. A nagyne´ni-k so¨rt isz-nak df aunt beer drink-3sg df aunt-pl beer drink-3pl ‘The aunt is drinking beer.’ ‘The aunts are drinking beer.’ One other crucial fact of Hungarian possessor agreement to be noted concerns the choice of the third person pronoun. (2a, b) illustrate that one and the same form, o, is invariably used for both singular and plural reference (cf. also E´. Kiss 2002: 170f ). In contexts other than that of possession, by contrast, the form ok is obligatory in case of plural reference. (4) a. O isz-ik. b. O k isz-nak. pron3sg drink-3sg pron3pl drink-3pl ‘S/he drinks.’ ‘They drink.’ 5. For Hungarian, the domain of adnominal possession must be assumed to also comprise agreement of postpositions as well as – apparently as an oddity of the language – of case markers with their noun phrase complements. This assumption will account for the fact that the agreement that is otherwise displayed by the two categories is precluded with lexical noun phrases – an obvious parallel to possessed nouns (except that the latter still show person agreement). Furthermore, it makes allowance for the nominal source of both categories; see Szabolcsi (1994: 207f ) and references there.

Anti-agreement with subjects and possessors

227

Yet another fact to be observed is that person and number agreement on the possessed noun are, like in most other head-marking languages, found irrespective of the presence or absence of any syntactic realization of a non-lexical possessor. In other words, the optionality of the personal pronoun does not a¤ect the articulation of number agreement; compare (5a, b) with (2a) and (5c, d) with (2b), respectively. a ha´z-a df house-por3sg ‘his/her house’

b.

a sze´k-e df chair-por3sg ‘his/her chair’

c. a ha´z-uk df house-por3pl ‘their house’

d.

a sze´k-u¨k df chair-por3pl ‘their chair’

(5) a.

The generalization that most adherents of Chomsky’s Minimalist Program or its predecessor versions would immediately suggest can rather safely be assumed to be approximately of the following kind: Possessed nouns without an overt personal pronoun involve a null pronoun in the same position. Since null pronouns share the essential semantic and syntactic properties of overt pronouns, their parallel behaviour with respect to number agreement is expected. In the following section, I will discuss the most profound account of Hungarian possessor anti-agreement along these lines that has been put forward.

3. Syntactic accounts of anti-agreement: Rouveret on Welsh and den Dikken on Hungarian In den Dikken’s (1999) account of Hungarian, possessed nouns that are not preceded by an overt personal pronoun, as it is the case with the examples in (5), are conceived of as involving a ‘‘null pronominal possessor’’ (l.c.: 138) in the prenominal position. At first sight, the assumption of a null pronoun appears to be well-motivated since it allows for a simple generalization, according to which plural agreement is found if and only if the possessor is an – overt or non-overt – pronoun. According to this view no overt pronoun is, so to speak, still a pronoun. Den Dikken relates his analysis of Hungarian possessor anti-agreement to a parallel split in the agreement of verbs with postnominal subjects in Welsh. In this VSO language, finite verbs are specified for the person of the subject. As far as number agreement is concerned, a plural specifica-

228

Albert Ortmann

tion is found either with pronominal subjects or in the absence of an overt subject, whereas such a specification fails to occur with lexical subjects (Sadler 1988: 50f; Borsley, Tallerman & Willis 2007: 17f ).6 The following contrastive pairs show that clauses with lexical subjects allow only for default agreement, that is, a third person singular specification of the verb:7 (6) Welsh (Sadler 1988: 50f ) a. Mae y plant yn canu. cop.3sgl df children prog sing ‘The children are singing.’ b.

Maent hwy yn canu. cop.3pl pron3pl prog sing ‘They are singing.’

c. Maent yn canu. cop.3pl prog sing ‘They are singing.’ (7) Welsh (Rouveret 1991, 1994, quoted after den Dikken 1999: 155) a. Darllen-asant hwy y llyf read-past.3pl pron3pl df book ‘They read the book.’ b.

Darllen-asant pro y llyfr read-past.3pl pro df book ‘They read the book.’

c. *Darllen-asant y plant y llyfr read-past.3pl df children df book 6. In case of (marked, cleft-like) SVO word order, there is no agreement at all in Welsh, not even with pronominal subjects. 7. The latter is therefore traditionally referred to as ‘analytic verb form’, while fully specifed forms are known as ‘synthetic’; see Wratil (this volume), who also points out a crucial empirical di¤erence between Welsh and its Celtic relative Irish: in the latter language, synthetic forms do not even combine with overt pronouns. This implies that the supposed generalization in terms of a null pronoun to be discussed here cannot be carried over to Irish; consequently, Wratil assumes the Irish ‘synthetic’ forms to come about by incorporation of pronominal subjects into the main verb without being further grammaticalised to agreement su‰xes.

Anti-agreement with subjects and possessors

d.

229

Darllen-odd y plant y llyfr read-past.3sgl df children df book ‘The children read the book.’

It will be noticed that ‘anti-agreeing’ (6a) is analogous to (2d), while (6b), in which the copula shows plural agreement, is analogous to (2b), and subjectless (6c) is analogous to (5c, d). To account for this pattern, Rouveret (1991, 1994) assumes a null pronoun ‘pro’ for clauses without an overt subject, in analogy to overt pronouns, as indicated in example (7b) as well as in the structure in (8b) (cf. also den Dikken 1999: 156f ): (8) a. [ AgrSP AgrS [TP DP [T’ T . . .]]] ! anti-agreement b. [ AgrSP AgrS þ Num i [TP [ NumP ti [ NP pro/pronoun]] [T’ T . . .]]] ! agreement More precisely, Rouveret’s (1991, 1994) approach to the asymmetry in plural agreement is based on the following assumptions: (i)

The VSO word order emerges because the subject does not move to SpecAgrP but only to SpecT. Since SpecAgrP is not targeted by movement, there is no Spec-Head agreement with lexical postverbal subjects. (ii) In contrast to lexical subjects, however, for both null and overt pronominal subjects the conditions for agreement are supposed to be met even though the pronoun does not end up in the checking domain SpecAgrS. Instead, the agreement relationship emerges by head-movement, namely of Num in SpecT, to AgrS in order for Num to be licensed. (iii) Lexcial noun phrases are ‘full’, in the sense that they project DPs, whereas pronouns are only NumPs. The Num head of a NumP with an overt or zero pronoun does not find a D on top of it, hence it moves to AgrS. This will result in full agreement as indicated in (8b). With lexical subjects, by contrast, AgrSP is not targeted by movement: Since Num only moves to D and since subjects do not raise as high as to SpecAgrS, there is no Spec-Head agreement. As a consequence, the unmarked singular form is used as the default form; cf. (8a). Den Dikken carries this account over to possessor anti-agreement in Hungarian. The following parallels between noun phrases and clauses are assumed to hold:

230

Albert Ortmann

(i) (ii)

Like clauses, noun phrases have an Agr projection, too. The possessor occupies the same hierarchical position as postverbal subjects in Welsh: SpecF and SpecT, respectively; that is, the one immediately below SpecAgr. (iii) Plural agreement is captured by movement from Num in SpecF into Agr. Accordingly, for possessed noun phrases without an overt syntactic possessor den Dikken assumes a null pronoun. The contrast between a lexical possessor as in a no kalapja ‘the woman’s hat’, a no(k) kalapja ‘the women’s hat’, both without plural agreement (‘anti-agreement’), and a pronominal possessor (either overt az o kalapjuk ‘their hat’ or, as it is supposed, covert as in a pro kalapjuk (id.)) with plural agreement is represented by the derivations in (9a) and (9b), respectively (den Dikken 1999: 157f ). (9) a.

[ DP a [ AgrP Agr [FP [ PP PØ [ DP no(k)]] [F’ F . . . kalap . . .]]]] !

(9) a.

anti-agreement

b. [ DP a(z) [ AgrP Agr þ Num n [ FP [ PP PØ [ NumP tn [ NP pro /o ]]] [ F’ F . . . kalap . . .]]]] ! agreement As for the additional riddle of Hungarian – the preference of the default ‘singular’ form o over the plural form ok in possessor contexts – the explanation is based on the following assumption: ‘‘In recognition of the concatenative nature of Hungarian plural marking, I analyze the plural morpheme -k as an independent syntactic constituent, harbored by the Num-head of the noun’s extended projection’’ (den Dikken 1999: 159). Noun plural (e.g., ha´z-ak house-pl) is in general analysed as coming about by a‰x hopping, under the assumption that there is no overt N-to-Agr because of prenominal adjectives and possessor. This is assumed to hold for pronouns, such as o ‘s/he’ and ok ‘they’, as well – except where the pronoun functions as a nominative possessor of a head-noun. In this case, ‘‘Num-to-Agr movement severs the pronoun from the number morpheme -k, which a‰xes to the Agr-head which is itself unmarked for number, due to the absence of a Spec-Head relationship between it and some number-marked nominal constituent’’ (den Dikken 1999: 159). In other words, in order for Num to be licensed, it moves to Agr, the e¤ect being ‘‘physical displacement of the number marker -k . . . [T]he lexical pronoun is severed from the associated Num-head’’ (den Dikken 1999: 159): (10) [ DP D [ AgrP -ju þ [Num-k ]n [ FP [ PP PØ [ NumP tn [ NP o ]]] [ F’ F . . . kalap . . .]]]]

Anti-agreement with subjects and possessors

231

As a consequence, the stem of the pronoun surfaces by itself, and the agreement su‰x -( j )uk emerges, ending up via a‰x hopping on the head noun.8 An analysis along these lines, however, appears to bring about a couple of conceptual and empirical intricacies, some of which I will formulate directly, while others will be formulated as open questions: 1. Leaving aside the issue of whether movement out of the specifier position of F as in (10) is theory-internally legitimate with regard to locality, the more severe question that remains is what this movement is triggered by. The reason provided by den Dikken (1999: 156), following Rouveret, is that Num must be licensed – but then this would, by parity of reasoning, also have to be a necessary derivation for pronouns in non-possessed contexts, since it is taken for granted that pronouns always project only Num rather than D. However, in non-possessive contexts -k is never separated from the stem o: neither when the pronoun features as the subject of a clause (cf. (4b)), nor as some other verbal argument, nor when the pronoun occurs in isolation. Instead, we encounter the form ok throughout in those contexts. This would mean that Num is licensed di¤erently in those cases, so that a‰x hopping of the Num-head to the pronoun’s stem in N is possible here. How could this be the case here but not in possessive contexts? Does Num generally want to raise to Agr in the absence of D, or only if Agr is unmarked for number as a consequence of SpecAgrP not being targeted by movement, or does Num generally want to ‘hop’ to the stem of the pronoun in N? 2. If – in accordance with its agglutinative status – the plural su‰x -k is analysed as an independent syntactic constituent, the question is what this means exactly for the supposed null pronouns featuring as subjects. The representation provided by den Dikken is only concerned with the overt pronoun o, although the schema is explicitly intended to account for ‘pro’ as well. One would therefore have to assume that in this case -k, or its equivalent in Turkish, is also present in Num. 8. As far as the exact phonological shape of the third person possessor su‰xes -( j)a and -( j)uk is concerned, note that the palatal glide occurs (i) as long as the noun is alienably possessed, and (ii) only after stems that do not end in any of the strident or palatal consonants [s, z, §, j, , ] (Olsson 1992); thus kalap-ja ‘his/her hat’, kalap-juk ‘their hat’, but ha´z-uk ‘their house’, has-a ‘his/her/its belly’, toja´s-a ‘its egg’.

232

Albert Ortmann

Now given that in contexts other than possession -k is not ‘‘displaced’’ from NP by movement to Agr, where would the su‰x end up in order to be realised? 3. As for the predicative power of the analysis, one would expect the same ‘‘physical displacement of the number marker’’ for other languages whose number morphology is as concatenative and transparent as, or even more transparent than that of Hungarian. Turkish is one of these cases, with the third person pronoun forms o(n) pron3sg ‘s/he’ and on-lar pron-3pl ‘they’, where -lar/-ler is the regular and sole plural su‰x for nouns as well. However, ‘‘displacement’’ e¤ects of the pronoun’s plural su‰x, comparable to those of Hungarian, are absent from Turkish: the singular form o(n) never appears in contexts of plural, be they possessive or clausal.9 4. Another point of discussion concerns the treatment of pronouns as being ‘defective’, hence structurally less articulated in comparison to lexical noun phrases, in that they project only NumP rather than DP. This appears to be in contradiction to the referential properties of pronouns, as well as to their position on the Definiteness Scale, which will play a major role in the analyses I develop in sections 6 and 7.

9. Interestingly, what does exist in Turkish is an anti-agreement e¤ect which looks like the exact converse of that of Hungarian. Kornfilt (1997: 540) reports that ‘‘[w]here a third person plural possessor or subject is very close to the agreement, the third person singular form is preferred over the third person plural form.’’ Crucially, what is meant here is singular on the agreeing verb or possessed noun, rather than on the pronoun, as it is the case with Hungarian. Note first that in accordance with my review of den Dikken, we are not dealing with ‘‘displacement’’ of the plural su‰x from the stem in Turkish, the form of the third pronoun being onlar rather than o(n). Second, that in Hungarian it is the plural of the pronoun that is not realised, while in Turkish it is that of the agreeing possessed noun, would require additional assumptions: Why does the derivation in (9b) only converge for Hungarian? What seems to be even harder for a null pronoun account is the fact that the putative third person plural null pronun of Turkish behaves di¤erently rather than analogously to its overt counterpart, in that only the former and not the latter combines with full agreement on the possessed noun. Under the approach taken in this paper, the di¤erent anti-agreement versions of Turkish and Hungarian reduce to a slight di¤erence in the constraint ranking, which will be the object of section 7.2.

Anti-agreement with subjects and possessors

233

5. With regard to the typological setting of the phenomena under discussion, den Dikken’s analysis does not make use of the fact that agreement asymmetries between lexical and pronominal arguments are only one instance of a typologically wide-spread type of economy splits. No appeal is made to salience splits along various ‘animacy’ hierarchies as they are well-known from the typological literature, which in fact are to be seen as the functional foundation of the phenomenon (see section 5). 6. Finally, with respect to Hungarian itself, den Dikken’s analysis does not refer to the fact that the motivation of avoiding the plural form ok is only one instance of economy in the realization of number morphology, and that economy generally determines the distribution of plural marking in the Hungarian noun phrase. This aspect is especially crucial for the economy account I eventually propose, and will be dealt with in the next section.

4. Economy-based plural splits in the DP 4.1. Economy vs. redundancy in plural marking in the noun phrase Languages di¤er according to whether number is established as an inflectional category at all. For those languages where it is present, there is variation as to the economy or redundancy of its realization. The major types are amply illustrated in Ortmann (2000, 2004): On the one hand, there are inflectional languages such as German or Latin that allow for more than one overt explicit articulation of plural within the noun phrase. The Bantu languages with their concord prefixes also fall into this group, since every modifier displays, among other things, number information. Below is an example from the Papuan language Yimas, with the head noun being marked for noun class as well as for dual and plural, respectively. Both modifying adjectival verbs and possessives display full number agreement: (11) Yimas (Ramu-Lower Sepik; Foley 1991: 140)10 a.

tnknt-k-ntrm tkt-ntrm ama-na-ntrm heavy-irr-cl5.dl chair-cl5.dl 1sg-poss-cl5.dl ‘my two heavy chairs’

10. The orthography that Foley uses renders underlying rather than surface forms, hence it does not reflect the rather complex processes of vowel insertion.

234

Albert Ortmann

b.

nawray ama-na-ra urkpwica-k-ra armband.cl5.pl 1sg-poss-cl5.pl black-irr-cl5.pl ‘my black armbands’

The opposite is found in those numerous languages in which there is no plural agreement within the noun phrase; this is also the case in the languages under consideration, Hungarian and Turkish. As a matter of fact, if the concept of ‘plurality’ is signalled by a di¤erent category such as numerals or quantifiers there is no plural marker at all, that is, not even on the noun itself. The combination of a numeral with a noun yields ke´t ha´z as in (12c), literally ‘two house’. (12) Hungarian a. (egy) ha´z b. ha´z-ak c. ke´t ha´z d. sok ha´z (one) house house-pl two house many house ‘a/one house’ ‘houses’ ‘two houses’ ‘many houses’ e. (egy) zo¨ld ha´z f. zo¨ld ha´z-ak g. A ha´z-ok zo¨ld-ek. indef green house green house-pl df house-pl green-pl ‘a green house’ ‘green houses’ ‘The houses are green.’ Languages such as Hungarian (as well as, to mention but a few, Turkish, Georgian, and the northeast Caucasian languages) can thus be said to display an economy split since plural is marked on the noun depending on whether or not the concept of plurality is expressed otherwise. Moreover, in Hungarian an attributive adjective as in (12f ) does not show plural agreement even though the adjective is in principle capable of bearing a plural specification, namely in predicative use, as can be seen from (12g). This split, too, is motivated by the fact that the attributive adjective is combined with a noun that already bears a plural marker. 4.2. The economy account of number inflection The analysis developed in Ortmann (2000, 2004) is couched in the framework of Optimality Theory (henceforth OT) as developed by Prince & Smolensky (1993; eventually published as Prince & Smolensky 2004). The tension of morphological economy and explicitness is implemented in terms of the interaction of a markedness constraint and the relevant faithfulness constraint in the sense of Correspondence Theory, put forward by

Anti-agreement with subjects and possessors

235

McCarthy & Prince (1995). In (13), the markedness (or ‘avoid’) constraint *pl(dp) is introduced: (13) *pl(dp): Avoid the realization of the specification [þpl] within the DP. This multiply violable constraint stars any plural marking within a syntactic domain, the DP. *pl(dp) is in conflict with the requirement of explicit plural marking, which is conceived of as a member of the max subfamily of faithfulness constraints: (14) Max(pl): An underlying specification [þpl] has a correspondent in the output. The general constraint Max(pl) stars any ‘‘missing’’ exponent of plural, including non-marked modifiers, in contrast to Max-head(pl), a constraint with a context specification, which will also be needed. (15) Max-head(pl): A specification [þpl] on a syntactic head in the input has a correspondent in the output. max-head(pl) implements the typological observation that inflectional categories are usually expressed either on the lexical head (noun or verb) or on a functional head (article or auxiliary), whereas any realization on the first/last/second element of a phrase, that is, as phrasal clitics such as the English possessor ’s, is more marked and requires specific high-ranked alignment constraints. Due to its more specific context, max-head(pl) is ranked higher than the general version max(pl), thus max-head(pl) X max(pl). Furthermore, I assume a constraint that is grounded in the semantics and requires plurality to be ‘detectable’ on the surface: (16) ExpressPlurality: The semantic concept of plurality is discernible in the output. The facts of Hungarian as exemplified in (12) are correctly predicted by the ranking in (17). What is crucial is that the Max constraints are dominated by *pl(dp) and that ExpressPlurality is undominated: (17) Ranking for Hungarian: ExpressPlurality X *pl(dp) X Max-head(pl) X Max(pl)

236

Albert Ortmann

The e¤ect is that number is not articulated redundantly within the DP,11 which captures both of the plurality splits exemplified in (12), namely the ones involving numerals and adjectives, as the following evaluations will demonstrate. Here and in general, I conceive the input as a semantic form associated with the morphological specification that is derived from it. Thus, the information that something is individuated as an aggregate of objects gives rise to the input specification [þpl] per default, much in the same way that the information ‘female’ gives rise to the gender feature specification [þfeminine] in the input, provided there is no idiosyncratic specification that overrides the default, as is the case with singularia tanta and grammatical gender, respectively. I furthermore assume that numerals higher than ‘one’, by virtue of their cardinality value, express the semantics of ‘plurality’, which I represent by the semantic predicate ‘AGGR(egate)’ in the input of (18). As a result, even in combination with singular nouns the undominated constraint ExpressPlurality is satisfied. (18) shows the evaluation of candidates for the combination of noun and numeral: (18) Input: lx [house(x) & aggr(x) & card(x) ¼ 2]; noun [þpl] Express Plurality a.

ke´t ha´z-ak two house-pl

b. F ke´t ha´z two house

*pl(dp)

Maxhead(pl)

Max(pl)

*

*

*!

Candidate (18a), being marked for plural, brings about a violation of *pl(dp) a violation, which is fatal given the ranking in (17). Candidate (18b) is therefore optimal, at the cost of violating the Max constraints since it does not realise the underlying specification [þpl]. Numerals of languages that adhere to the ‘Type Hungarian’ ranking are thus correctly predicted to combine with singular nouns. At the same time, the analysis also naturally accounts for the fact that plural is marked 11. The only case of redundant plural (as well as case) marking in Hungarian is constituted by the demonstrative pronoun. There is, however, good reason to assume that demonstratives feature outside the DP and project a higher functional category Dem, which means that they cannot violate *pl(dp); see Ortmann (2002).

237

Anti-agreement with subjects and possessors

rather than ‘saved’ with referential bare nouns, since in the latter case we would be dealing with a violation of undominated ExpressPlurality. (19) Input: lx [house(x) & aggr(x)]; noun [þpl] Express Plurality

*pl(dp)

a. F ha´z-ak house-pl b.

ha´z house

Maxhead(pl)

Max(pl)

*

*

* *!

As regards the combination of plural nouns and adjectives, illustrated above in (12f ), I consider the input to be the semantic representation of the conjunction of the predicates ‘ship’, ‘fast’, and ‘aggr’ associated with the morphological specification that is derived from this semantics. I assume that the semantic predicate aggr brings about the input specification [þpl] on each part-of-speech that involves a paradigm with the respective feature contrast. (20) Input: lx [house(x) & green(x) & aggr(x)]; noun [þpl], adjective [þpl] Express Plurality

*pl(dp)

a.

zo¨ld-ek ha´z-ak green-pl house-pl

b.

zo¨ld-ek ha´z green-pl house

*

c. F zo¨ld ha´z-ak green house-pl

*

d.

zo¨ld ha´z green house

maxhead(pl)

max(pl)

!*

*

**!

*!

* *

**

Plural on the adjective as well as on the noun in (20a) is excluded as an additional and therefore fatal violation of high-ranked *pl(dp). By con-

238

Albert Ortmann

trast, candidates (20b) and (20c) with unmarked adjective or noun, respectively, incur violations of the low-ranked Max constraint. The decision between the two candidates is made by the constraint Max-head. The undominated ExpressPlurality makes sure that there is a plural marker; otherwise the totally unmarked candidate (20d) would be optimal. The account furthermore correctly predicts that the adjective bears a plural inflection in its predicative use, which was shown in (12g). Since *pl(dp) is vacuously satisfied in DP-external environments such as predicative adjective phrases, any candidate without a plural su‰x would fatally violate max(pl) and the candidate with plural on the adjective will be preferred. The role of economy, then, is implemented in terms of the ranking *pl(dp) X Max-head(pl), Max(pl). Those languages that in contrast to Hungarian do allow redundant plural marking in the DP have the reverse ranking (see Ortmann 2004 for the details of a factorial typology). It will be shown in section 7 how the account correctly predicts the choice of the form of the pronoun in Hungarian possessive constructions. In order to give a comprehensive account of Hungarian and Turkish possessor anti-agreement, I will now turn to plurality asymmetries that are governed by salience.

5. Salience-driven economy splits If one looks at argument linking – that is, the projection of semantic arguments into syntax – from a cross-linguistic perspective, one finds that the realization of case and agreement morphology is often subject to specific conditions. It is particularly typical of object case and agreement that they are restricted to objects with human or animate referents, or to those with definite or specific reference. 5.1. Di¤erential object marking From a cross-linguistic point of view it can be said to be the rule rather than the exception for a language to use morphological markers of agreement and case only when the object has a certain degree of ‘salience’. In Swahili, for example, object agreement prefixes, which immediately precede the verb stem, are only found if the object is either definite or human, as in (21b–d):

Anti-agreement with subjects and possessors

239

(21) Swahili (Bantu < Niger-Congo; Givo´n 1976: 159) a. Ni-li-soma ki-tabu. b. Ni-li-ki-soma ki-tabu. 1sg-past-read 7-book 1sg-past-cl7-read 7-book ‘I read a book.’ ‘I read the book.’ c. Ni-li-mw-ona m-tu m-moja. 1sg-past-cl1-see 1-person 1-one ‘I saw one person.’ d.

Ni-li-mw-ona yula m-tu. 1sg-past-cl1-see def 1-person ‘I saw the person.’

Similarly, the distribution of case markers for direct objects tends to be restricted in terms of definiteness and animacy. An example is Maltese, whose object case marker lil fails to occur with non-human direct objects: (22) Maltese (Semitic < Afro-Asiatic; Fabri 1993: 117) a.

Raj-t lil Pawlu. see-1sg case Paul ‘I saw Paul.’

b.

Xtraj-t il-ktieb. buy-1sg df-book ‘I bought the book.’

The functional explanation of di¤erential object marking lies in the tendency towards language economy. Object markers tend to be ‘saved’ when the object lacks such properties as animacy or definiteness that are typical of subjects, that is, they can be left out in those contexts where ambiguity is unlikely (Plank 1980; also cf. Aissen 2003). To account for this state of a¤airs, reference is made to salience hierarchies such as the ones in (23). The minor di¤erences between the various versions need not concern us here. (23) a.

‘semantic hierarchy’ (Silverstein 1976): 1st > 2nd > 3rd human > 3rd animal > 3rd inanimate

b. ‘personal hierarchy’ (Siewierska 1988): 1st > 2nd > 3rd human > higher animals > other organisms > inorganic matters > abstracts In sum, the more specifically individuated, or the more ‘animate’, or in other words the more salient, the referent of the direct object is, the more likely is a morphological case or agreement specification of its grammatical role.

240

Albert Ortmann

5.2. Plurality splits in subject agreement This additional type of salience-driven economy split in number morphology will turn out to be most relevant for the actual issue of the paper introduced in section 2. I claim that Hungarian anti-agreement can be explained as a combination of two di¤erent common types of splits, one of which is a plurality split in subject agreement. Crucially, this split cannot be functionally motivated by the (non-)likelihood of confusing two individuals, for the simple reason that it also occurs with intransitive verbs, where only one argument is involved. In many languages full subject-verb-agreement specifications with respect to number are only found with ‘salient’ subjects. In Persian (as well as, for example, in Ossetic and in Turkish) number agreement is only obligatory with animate subjects, while it is optional with inanimate subjects; compare (24a) and (24b): (24) Persian (Iranian < Indo-European; Amin-Madani & Lutz 1972: 329) a. Baradar-an-am nazde-u bud-and. brother-pl-por1sg at-3sganimcop.past-3pl ‘My brothers were with him.’ b. Sˇaxe-ha sˇekast-and / sˇekast branch-pl crack.past-3pl / 3sg ‘The branches broke.’ Likewise, in Georgian number agreement on the verb is restricted to animate subjects for many speakers, while inanimate plural subjects combine with the ‘singular’ form of the verb. Consider (25a) vs. (25b): (25) Georgian (South Caucasian; Harris 1981: 21) a. knut-eb-i gorav-en. b. burt-eb-i gorav-s. kitten-pl-nom roll-3pl ball-pl-nom roll-3sg ‘The kittens are rolling.’ ‘The balls are rolling.’ Plurality splits in verbal agreement are also instances of economy in the sense that morphological material is saved in cases of little salience. Since in contrast to object linking splits this type of split cannot be traced back to the similarities or di¤erences of prototypical objects and subjects, it must be concluded that salient arguments are in general more likely to trigger fully articulated verbal agreement than less salient arguments,

Anti-agreement with subjects and possessors

241

irrespective of their grammatical role. I will now turn to a constraint analysis of agreement splits that implements the notion of economy. 6. Constraint-based analyses of economy splits in argument linking 6.1. Di¤erential Object Marking as economy vs. transparency: Aissen (2003) In OT, the predictions expressed by typological hierarchies such as the ones discussed in the preceding chapter are directly implemented into constraint hierarchies in terms of Harmonic Alignment. Harmonic Alignment is a formal device that takes two scales and aligns each element of one scale with each element of the other. This yields a set of markedness constraints in an order which is universally fixed, rather than being subject to language-specific ranking. (26) Harmonic Alignment (Prince & Smolensky 1993: 136): Given a binary dimension D1 with a scale X > Y and another dimension D2 with a scale a > b > . . . > z, the harmonic alignment of D1 and D2 is the following pair of harmony scales: Hx: X/a > X/b > . . . > X/z

Hy: Y/z > . . . > Y/b > Y/a

The constraint alignment is the following pair of constraint hierarchies: Cx: *X/z X . . . X *X/b X *X/a

Cy: *Y/a X *Y/b . . . X *Y/z

For the case of di¤erential object marking, Aissen (2003) proposes that the two scales that are to be aligned are instantiated by the relational scale ‘subject higher than object’ on the one hand, and by the Animacy Scale, or Definiteness Scale, respectively, on the other hand.12 (27) Relational Scale: Su(bject) > Ob( ject) Animacy Scale: Hum(an) > Anim(ate) > Inan(imate) Definiteness Scale: Pronoun > Name > Definite > Indefinite Specific > NonSpecific 12. In assuming a scale that refers to grammatical relations, Aissen follows the tradition of the framework of Lexical Functional Grammar (LFG), where these notions are assigned a primary status rather than being derived notions. Alternatively, one may follow the approach by Stiebels (2002) (to be mentioned below), which instead refers to argument positions that are derived from the Semantic Form of the verb’s (or noun’s) lexical representation.

242

Albert Ortmann

Harmonic alignment with the latter scale results in the following two markedness hierarchies. (28a) states that pronominal subjects are more harmonic than proper name subjects, these in turn are more harmonic than definite subjects, and so forth. Likewise, (28b) states non-specific objects are more harmonic than all other objects: (28) Harmonic alignment of Relational and Definiteness Scale: a. Su/Pro L Su/PN L Su/Def L Su/Spec L Su/Nspec b. ObNSpec L Ob/Spec L Ob/Def L Ob/PN L Ob/Pro If the alignment of the scales is inverted according to the definition in (26), we arrive at the following ranking hierarchy of ‘avoid’ constraints: (29) Hierarchy of avoid (or ‘markedness’) constraints: a. *Su/NSpec X *Su/Spec X *Su/Def X *Su/PN X *Su/Pro b. *Ob/Pro X *Ob/PN X *Ob/Def X *Ob/Spec X *Ob/Nspec Unlike constraints that require a language-specific ranking, constraint hierarchies, being derived from scales that are presumably grounded in cognition, hold universally. (29) says that a non-specific subject invokes a higher violation than a specific subject, and so on; likewise, a pronominal object is worse than a proper name object, and so on. In order to implement the balance of morphological explicitness and economy, Aissen (2003: 447¤) defines local conjunctions with two kinds of constraints of a very general nature: *Ø and *struc, respectively, both of which come along in their instantiations for case, noted by subindeces; thus *ØC (‘‘Penalizes the absence of a value for the feature CASE’’) and *strucC (‘‘Penalizes a value for the morphological category CASE’’). Consequently, the constraint hierarchies that account for di¤erential object marking are as in (30): (30) a. b.

*Ob/Hum & *ØC X *Ob/Anim & *ØC X *Ob/Inan & *ØC *Ob/Pro & *ØC X *Ob/Name & *ØC X *Ob/Def & *ØC X *Ob/Spec & *ØC X *Ob/NSpec & ØC

The interpretation of (30) is that realising a human object without case amounts to a higher violation than leaving a non-human object caseless. This holds universally, and it is in this sense that we can say that hierarchies like the ones in (30) account for the observed cross-linguistic markedness findings: If a language leaves animate objects unmarked, it will always

Anti-agreement with subjects and possessors

243

leave inanimate objects, and everything ‘further below’, unmarked as well. The same relation holds true for pronominal objects and indefinite objects. An individual language’s response to the tension of markedness and economy is expressed by the interpolation of *strucC, into the constraint hierarchy. This economy constraint says that overt structure – the articulation of morphological case – is dispreferred. The interpretation of this interpolation is that a given language saves agreement morphology below a certain cut-o¤ point of the hierarchy, and uses agreement morphology above this threshold. A related account, coached in the framework of Lexical Decomposition Grammar as outlined in Wunderlich (1997), is provided by Stiebels (2002). Like Aissen, Stiebels implements the role of economy by harmonic alignment. However, Aissen’s somewhat informal constraint *ØC is replaced by a Max constraint. This is more in line with the practice of Correspondence Theory (McCarthy & Prince 1995), according to which the subfamily of Max constraints requires underlying structure to be visible in the output. Thus, under Stiebels’s markedness approach the input encompasses the information which case and agreement features should be morphologically specified. This modification allows for the familiar way of computing output forms in terms of Correspondence Theory and I will therefore adopt it for my analysis of linking splits. 6.2. Constraint analysis of subject anti-agreement The theoretical background given so far enables a natural account not only of those salience-driven splits of the di¤erential object marking type in 5.1, but also those of the subject agreement type introduced in 5.2. In the following, I summarise the approach to salience-driven plural subject agreement splits as developed in Ortmann (2002) and then show how it can be extended so as to capture the pronominal/lexical agreement split of Welsh introduced in section 3. The characteristic of the plural subject agreement splits of Persian and Georgian illustrated above in (24b) and (25b) is that the more articulated agreement specification is reserved for more salient subjects. This implies that this type of split is not among the ones that are accounted for by Aissen’s (1999, 2003) approach, under which di¤erential subject marking is predicted to be the mirror image of di¤erential object marking: More typical (‘harmonic’) subjects are expected to involve less explicit morpho-

244

Albert Ortmann

logical specification than less typical ones. This prediction is indeed borne out for linking splits,13 not, however, with plurality agreement splits.14 What I suggest, therefore, is to build in another scale, conceived of in terms of informativity, that accounts for the markedness findings with respect to plural splits in subject-verb-agreement: (31) Plurality Scale: AGR[þpl] > AGR[ ] ‘‘Morphologically specified plural agreement is more expressive than an unspecified form.’’ This scale can be seen as instantiating the Elsewhere Principle, which prefers the most specific form that fits a given context.15 The Plurality Scale will now be harmonically aligned with the Animacy Scale ‘‘Hum(an) > Anim(ate) > Inan(imate)’’, which yields the markedness scale in (32): (32) AGR[þpl]/Hum L AGR[þpl]/Anim L AGR[þpl]/Inan The scale states that full number agreement with human referents is more harmonic than with non-human animate referents, and the latter is in turn

13. For many languages the ergative case, or ergative agreement, respectively, is confined to less salient subjects. 14. In the light of the opposite direction of subject linking splits and subject plurality agreement splits, an anonymous reviewer points out, drawing on Zeevat & Ja¨ger (2002), that Harmonic Alignment approaches involve the risk of arbitrariness: The technique as such also allows for deriving the reverse of the actual distribution. Although Zeevat & Ja¨ger (2002: 5) characterise the problem of ad hoc combinations and the potential of describing mirror image languages as resulting from the concept of constraint conjunctions, of which I do not make any use, I concede to the reviewer that external functional justification for the respective alignment of scales is necessary. As for the present issue, the two types of splits indeed seem to be motivated di¤erently. Linking splits crucially involve person agreement, which is decisive for identifiying arguments and their grammatical role. By contrast, plurality agreement splits involve only number agreement, whereas person agreement is untouched. Therefore the alignment I am going to propose for this type models the intuition that more salient referents receive more articulated number specifications. 15. The morphologically unmarked ‘singular’ form of those categories that display agreement splits must be analysed as being unspecified ([ ]), rather than negatively specified ([–pl]), for otherwise it would not be able to combine with a positively specified constituent ([þpl]); see Ortmann (2000) for details.

Anti-agreement with subjects and possessors

245

more harmonic than full number agreement with inanimate referents. Inversion of the alignment attains a hierarchy of markedness constraints: (33) *AGR[þpl]/Inan X *AGR[þpl]/Anim X MaxAgr(pl) X *AGR[þpl]/Hum This constraint hierarchy is to be combined with the relevant Max constraint, which I define as follows: (34) MaxAgr(pl): An agreement specification [þpl] in the input has a correspondent in the output. Interpolating this Max constraint, for example, above *AGR[þpl]/Hum, we arrive at the ranking in (35). (35) *AGR[þpl]/Inan X *AGR[þpl]/Anim X MaxAgr(pl) X *AGR[þpl]/Hum The outcome is that it is worse to realise plural agreement with inanimates than with animates, which is also worse than ‘unfaithfully’ not realising plural.16 Obviously, in accordance with the data in (24) and (25), for Persian and Georgian the language-specific cut-o¤ for the Max constraint that is to be inserted into the scale must be as in (36), for it is only animate subjects for which plural is obligatorily cross-referenced on the verb. (36) Ranking for Persian and Georgian: *AGR[þpl]/Inan X MaxAgr(pl) X *AGR[þpl]/Anim X *AGR[þpl]/Hum According to this ranking, verbs in combination with inanimate subjects will bear no plural agreement, since the candidate that proves to be optimal violates only the lower ranked Max, for the benefit of the higher ranking constraint *AGR[þpl]/Inan; see tableau (37ii).

16. Due to the complementary extension of [þpl] and [ ] (unspecified for plural), the ranking (35) is equally expressive as a ranking with a markedness constraint, rather than a Max constraint, interpolated in the reversed order of agreement constraint, thus *AGR[ ]/Hum X AvoidAgr(þpl) X *AGR[ ]/Anim X *AGR[ ]/Inan. My preference of (35) is only due to the higher degree of intuition.

246

Albert Ortmann

(37) Evaluation of Georgian subject agreement: (i) Input: roll(ix (kitten(x) & aggr(x))); subject[þpl], agreement [þpl] *AGR[þpl]/ Inan

MaxAgr (pl)

a. F knut-eb-i gorav-en. kittens roll-3pl b.

*AGR[þpl]/ Anim

*AGR[þpl]/ Hum

*

knut-eb-i gorav-s. kittens roll-3sg

*!

(ii) Input: roll(ix (ball(x) & aggr(x))); subject[þpl], agreement [þpl] a.

burt-eb-i gorav-en. balls roll-3pl

*!

b. F burt-eb-i gorav-s. balls roll-3sg

*

The general implication is that for a language to display a plurality split means that maxAgr(pl) is dominated by at least the highest markedness constraint in the hierarchy, that is, by *AGR[þpl]/Inan, and furthermore that maxAgr(pl) dominates at least the lowest markedness constraint, namely *AGR[þpl]/Hum. With the device developed so far we are able to return to the verbal agreement split of Welsh discussed in section 3. The crucial set of data of Welsh anti-agreement is repeated here from (6). (38) Welsh (Sadler 1988: 50f ) a.

Mae y plant yn canu. cop.3sgl df children prog sing ‘The children are singing.’

b.

Maent hwy yn canu. cop.3pl pron3pl prog sing ‘They are singing.’

c. Maent yn canu. cop.3pl prog sing ‘They are singing.’

Anti-agreement with subjects and possessors

247

Plural agreement on the verb, then, is found with either pronominal subjects or in the absence of an overt subject, not, however, with lexical subjects. Now in order to account for (6a) vs. (6b) (the ‘lexical vs. pronominal’ part of the split), what is needed is to adjust the strategy pursued for Georgian so as to employ a constraint hierarchy that derives from the Definiteness Scale, the relevant section being . . . *Agr[þpl]/Specific X *Agr[þpl]/Def X MaxAgr(pl) X *Agr[þpl]/Pron (more details of the derivation will follow immediately). But how can an approach like the present one which rejects null pronouns explain the analogous agreement specification in (6a) and (6c)? After all, no pronominal or other noun phrase is assumed for (6c), so how can any of the constraints that can be derived from the scales in (27) be successfully appealed to in order to punish the lack of plural on the verb? The solution I would like to suggest does not build on the a‰nity between pronominal and supposed null arguments, but rather on their neighbouring, but yet di¤erent positions on a scale. The lack of any overt syntactic specification is, under normal circumstances, restricted to those arguments that have the highest possible salience in terms of information structure. Concretely, anaphoricity or ‘pronominality’ by itself does not su‰ce for an argument to be dropped and realised only morphologically in terms of agreement. Rather, its discourse function must be that of a topic, and not that of a focus, for this latter function typically gives rise to the use of a stressed pronoun even in pro-drop languages. In other words, topicality amounts to even higher salience than pronominality as such: Of all individuals, the one that is ‘closest’ to the discourse participants is either expressed by a reduced pronoun, typically a pronominal clitic, or it does not have to be realised at all in the syntax.17 Therefore, I propose to revise the Definiteness Scale by supplementing it in terms of a top-most element ‘‘NonFocus’’, to be understood as an 17. This also means that the term agreement is not to be understood in its most literal sense of two syntactic constituents agreeing with each other, for there is nothing that the verb ‘agrees’ with (except under a null pronoun account, where an empty agreement controller would be assumed). Rather one should conceive of NP-external agreement in the sense of ‘cross-reference’, that is, a morphological specification of the person/number/gender features of an argument. This is in line with the typological notion of ‘head-marking’ as introduced by Nichols (1988), who states that ‘‘in languages with consistently head-marked clauses, the verb itself normaly constitutes a complete sentence; full NPs are included only for emphasis, focus, disambiguation, etc’’ (l.c.: 107).

248

Albert Ortmann

afore-mentioned individual that is topic rather than focus, hence typically not realised as a free personal pronoun in head-marking languages18: (39) Definiteness Scale (revised): NonFocus > Pronoun > Name > Definite > Indefinite Specific > NonSpecific Harmonic alignment of the Plurality Scale defined in (31) above (AGR[þpl] > AGR[ ]) and the revised Definiteness Scale yields the scale in (40). (40) AGR[þpl]/NonFocus L AGR[þpl]/Pron L AGR[þpl]/Def L AGR[þpl]/Specific L AGR[þpl]/NonSpecific Inversion of this scale results in the following universal hierarchy of markedness constraints: (41) *Agr[þpl]/NonSpecific X *Agr[þpl]/Specific X *Agr[þpl]/Def X *Agr[þpl]/Pron X *agr[þpl]/NonFocus Thus, agreement with non-specific subjects is worse than with specific indefinites, and so on. The conflicting Max constraint must be interpolated above *Agr[þpl]/Pron, which we know from the above-illustrated fact that the Welsh agreement split is governed by the opposition pronominal vs. lexical as opposed to the animacy opposition found with Persian and Georgian. As far as my notation is concerned, for convenience the constraints higher on the hierarchy in (41) than *Agr[þpl]/Pron are simply abbreviated as *AGR[þpl]/Lexical in the ranking and in the tableau, rather than being listed individually as *Agr[þpl]/Def, *Agr[þpl]/Specific, and *Agr[þpl]/NonSpecific.19 (42) Ranking for Welsh: *Agr[þpl]/Lexical X MaxAgr(pl) X *Agr[þpl]/Pron X *AGR[þpl]/NonFocus

The notion cross-referencing also includes those systems where agreement is in complementary distribution with (doubled) clitics, as analysed in Woolford (2003). Typically, the former comes about by reanalysis of the latter; see Fuß (this volume). 18. In what follows, I ignore proper names, since plural is of little relevance for them. 19. Note that the latter two constraints are vacuously satisfied in the evaluations in (43), since it is only dealt with definite subjects.

Anti-agreement with subjects and possessors

249

The following tableau shows that the analysis makes the right predictions for each of the three di¤erent scenarios of Welsh subject-verb-agreement: (43) Evaluation of Welsh subject agreement: (i) Input: sing(ix (child(x) & aggr(x))); subject[þpl], agreement [þpl] *AGR[þpl]/ Lexical a.

Maent y plant yn canu cop.3pl the children . . .

MaxAgr (pl)

*AGR[þpl]/ Pron

*AGR[þpl]/ NonFocus

*!

b. F Mae y plant yn canu cop.3sg the children . . .

*

(ii) Input: sing(x); subject [þpl], agreement [þpl] a. F Maent hwy yn canu cop.3pl pron3pl . . . b.

Mae y plant yn canu cop.3sg pron3pl . . .

* *!

(iii) Input: sing(x); agreement [þpl] a. F Maent yn cop.3pl . . .

canu

b.

canu

Mae yn cop.3sg . . .

* *!

In both (43ii) and (43iii) the candidate with default ‘singular’ form of the copula fails because of the faithfulness constraint MaxAgr(pl), which can be satisfied at the cost of violating one of the low-ranking economy constraints. By contrast, in case of a lexical subject (43i) the candidate with a plural agreement specification faces a fatal violation of the dominating economy constraint.20 20. The proposed addition of a topmost element NonFocus to the Definiteness Scale entails the prediction that it is possible that there is a grammar in which the faithfulness constraint is interpolated precisely above *AGR[þpl]/NonFocus, thus . . . X *Agr[þpl]/Pron X MaxAgr(pl) X *AGR[þpl]/NonFocus. This ranking characterises a language where head-marking is not even found with pronominal subjects, but rather is in complementary distribution to a syntactic argument realization. Subject markers that behave like this are known as anaphoric clitics. Irish with either its synthetic forms or with free pronouns

250

Albert Ortmann

Note in passing that the situation is at first sight similar but in actual fact more complex in Classical and Modern Standard Arabic. This language also displays the pronominal-vs.-lexical split at issue in its verbal agreement: for lexical subjects the verb only specifies gender, for pronominal subject it specifies number as well. Contrary to Welsh, however, the same split is also manifest in word order: full agreement, that is, number and gender agreement with human plural subjects, is found with all subjects, including lexical ones, if the word order is SVO.21 In this respect Arabic exhibits almost the opposite situation of Welsh, where there is no agreement at all with prenominal subjects. This di¤erence between the two languages, however, may be owing to the fact that in Welsh preverbal subjects are in fact located in the superordinate clause of a cleft-construction (cf. Sadler 1988, Wratil this volume), whereas in Arabic they apparently result from movement of the subject to the position of the topic at the left periphery of the clause. An OT account of the two-fold split of Classical Arabic is provided in Ortmann (2002), which implements the requirement that structural arguments are overtly realised in the syntactic domain of the finite verb by assuming that only full (as opposed to default) agreement is capable of saturating the subject y-role. To summarise the results of this section, the plurality split in the subjectverb-agreement of Welsh involves the opposition between pronominal and lexical arguments, and therefore resembles Hungarian possessor antiagreement. Making use of the machinery developed by Aissen (2003) and Stiebels (2002), I have proposed a Plurality Scale, harmonically aligned with the Definiteness Scale, thus implementing the role of economy. The Definiteness Scale was extended so as to comprise a top-most segment, ‘‘NonFocus’’, in order to enable the evaluation of arguments that are only morphologically specified, rather than being realised as a syntactic phrase.

but not both (cf. note 7) comes very close to the ranking in question, except that the subject markers of synthetic forms are capable of bearing focus, and that their availablity is not ruled by referentiality or information structure but rather by the features sentence mood, person, and number (Wratil this volume). 21. See Benmamoun (2003) for an interesting account of that split which is based on the a‰nity of VSO clauses to the possessive construction known as construct state.

Anti-agreement with subjects and possessors

251

7. Economy and faithfulness in possessor anti-agreement In this section I return to the two-fold asymmetry of Hungarian possessor agreement and, building upon Ortmann (2002), I show that it receives a natural explanation in terms of the already established analyses of salience-based splits on the one hand, and of plural economy in the noun phrase on the other hand. I then turn to the anti-agreement e¤ect of Turkish, which looks like the converse of Hungarian, and argue that it follows equally naturally from a slight di¤erence in the constraint ranking. 7.1. Hungarian anti-agreement as the interplay of two di¤erent economy splits The anti-agreement e¤ect that was introduced in section 2 (cf. the data under (2) and (5); the examples in (44) exemplify the same contrasts) can be recapitulated as follows: First, there is no plural agreement on the head-noun if the possessor is lexical; see (44b).22 Second, the third person plural pronoun ok fails to occur as a possessor; instead, the singular form is used, as in (44d). (44) Hungarian a. a professzor dia´k-ja-i df professor student-poss(3sg)-pl ‘the professor’s students’ b. a professzor-ok dia´k-ja-i df professor-pl student-poss(3sg)-pl ‘the professors’ students’ c. az o dia´k-ja-i df pron3sg student-poss(3sg)-pl ‘his/her students’ d. az o dia´k-ja-i-k df pron3sg student-poss-pl-p’or3pl ‘their students’ 22. Note that the su‰x -i in dia´k-ja-i is the plural allomorph used for possessed nouns. Crucially, it denotes the plural of the possessed, rather than that of the possessor, hence the statement that the form displays ‘singular’ agreement, in contrast to dia´k-ja-i-k in (44d, f ) with the final su‰x marking plural of the possessor in addition to that of the possessed.

252

Albert Ortmann

e. a dia´k-ja-i df student-poss(3sg)-pl ‘his/her students’ f. a dia´k-ja-i-k df student-poss-pl-p’or3pl ‘their students’ It was pointed out several times that the motivation for the anti-agreement e¤ect is to be seen in the generalization that pronominal possessors are higher on the Definiteness Scale, thus su‰ciently salient to be crossreferenced by number agreement, whereas lexical possessors are not, hence the form dia´kjai, rather than dia´kjaik, in (44d). As for the other puzzle that was addressed, namely the preference of the default ‘singular’ form o over the plural form ok, I propose to conceive of it as another instance of what was argued to be crucial for Hungarian noun phrases in section 4, and shown to follow from the constraint ranking: the avoidance of multiple realizations of plural for one and the same referent, the possessor, within the DP.23 These two motivations of the facts can be implemented straightforwardly by employing a constraint hierarchy of the plurality agreement split type as used in section 6, combined with the analytic tool of economybased plural splits in the DP, the ranking *pl(dp) X max-head(pl), max(pl) as developed in section 4. The universal hierarchy of markedness constraints resulting from harmonic alignment of the Plurality Scale and the revised Definiteness Scale as introduced in the preceding section is repeated here: (45) *Agr[þpl]/NonSpecific X *Agr[þpl]/Specific X *Agr[þpl]/Def X *Agr[þpl]/Pron X *AGR[þpl]/NonFocus Two comments on the notation are in order. First, it was already pointed out that there is no plurality split in Hungarian subject agreement. We are therefore dealing with a fact of possessor agreement. Therefore, strictly 23. In motivating the two-fold asymmetry of Hungarian possessor agreement by economy rather than by null pronouns my analysis is in harmony with Szabolcsi (1994), who remarks in a foornote (p. 271): ‘‘I assume that this agreement pattern can be accounted for with reference to some strategy like ‘Mark plurality only once within a constituent’ and need not mean [. . .] that whenever we see the form kalap-juk, we have a dropped pronoun.’’

Anti-agreement with subjects and possessors

253

speaking, the markedness constraints for plural agreement call for a context restriction to possessor agreement. Such a restriction would have to be noted as *agr-Por[þpl]/NonPron, *Agr-Por[þpl]/Pron, and so on.24 It is only for convenience that I refrain from this full accuracy. Second, as in the previous section the constraints higher on the hierarchy in (45) than *Agr[þpl]/Pron are abbreviated as *AGR[þpl]/Lexical in the tableau, rather than being listed individually as *Agr[þpl]/Def, *Agr[þpl]/Specific, and *Agr[þpl]/NonSpecific. The facts can now be captured by only those constraints that have been employed in the two preceding sections. The ranking in (46) can be motivated as follows: the faithfulness constraint MaxAgr(pl) is interpolated in the economy constraints hierarchy in (45) in accordance with the fact (44d), so that it dominates *AGR[þpl]/Pron. Furthermore, the ranking established in section 4 (stated in (17)) is integrated in accordance with the facts; in particular, both ExpressPlurality and *AGR[þpl]/Lexical are undominated since they are obeyed by any output. (46) Ranking for Hungarian: ExpressPlurality, *AGR[þpl]/Lexical X MaxAgr(pl), *pl(dp) X Max-head(pl) X Max(pl), *AGR[þpl]/Pron X *AGR[þpl]/NonFocus The relevant candidates, that is, the four logically possible combinations of plural and unmarked for pronominal and lexical possessors, as well as the two logical possibilities for non-focus possessors, those that are not realised as a syntactic phrase, are evaluated in the tableaus in (47i–iii).25 Recall that inputs are provided with full number specification in accordance with their aggregate semantics.

24. Likewise, *Agr-S[þpl]/. . . would be the more accurate notation for the constraints used in section 6. 25. The constraint Max-head(pl), which punishes the lack of inflectional morphology expressing plural on the head noun (rather than on its modifiers) is omitted from the tableau because it is always vacuously fulfilled as long as we are only concerned with the number value of the possessor rather than with that of the head noun, which is kept in the singular for the sake of simplicity.

254

Albert Ortmann

(47) Evaluation of Hungarian possessor agreement: (i) Input: ix [house(x) & poss(iy(student(y) & aggr(y)),x)]; possessor[þpl], agreement[þpl] Express Plurality

a.

a dia´k ha´z-a df student(sg) house-por3sg

*AGR [þpl]/ Lexical

*pl (dp)

*!

b. F a dia´k-ok ha´z-a df student-pl house-por3sg

*

c.

a dia´k ha´z-uk df student(sg) house-por3pl

*!

*

d.

a dia´k-ok ha´z-uk df student-pl house-por3pl

*!

**

Max Agr (pl)

Max (pl)

*

**

*

*

*AGR [þpl]/ Pron

*AGR [þpl]/ NonF.

*

(ii) Input: ix [house(x) & poss(y,x)]; possessor [þpl], agreement [þpl] a.

az o df pron3

ha´z-a house-por3sg

*!

b. F az ok df pron3pl

ha´z-a house-por3sg

*

c.F az o df pron3

ha´z-uk house-por3pl

*

ha´z-uk house-por3pl

**!

d.

az ok df pron3pl

*

**

*!

* *

* *

(iii) Input: ix [house(x) & poss(y,x)]; agreement [þpl] a df

ha´z-a house-por3sg

b. F a df

ha´z-uk house-por3pl

a.

*!

*

*

Max(pl) punishes all plural specifications of the input that are missing in the output, that is, those of the possessor itself as well as those of the agreement morphology. Any candidate with no plural specification at all fatally violates the undominated ExpressPlurality, as is the case with

*

Anti-agreement with subjects and possessors

255

the (a.) candidates of (47i–iii). For possessed nouns without a possessor phrase this already makes the decision as to the agreement specification. For agreement with lexical possessors, the evaluation in tableau (47i) shows that plural agreement on the noun as in (47i c, d) invokes a violation of *AGR[þpl]/Lexical. The ranking of this constraint on top of MaxAgr(pl) and Max(pl) correctly predicts that the noun exhibits no plural agreement when combined with a lexical possessor. For pronominal possessors as in (47ii), candidates with the singular form of the possessor su‰x on the noun, that is, those that fail to show plural agreement in the presence of a pronoun, are excluded as violations of MaxAgr(pl). Note that *pl(dp) can involve two violations, because the possessor phrase and its agreement specification on the head noun involve two plural specifications within one DP that refer to the same individual. It now becomes obvious why the unmarked ‘singular’ form of the pronoun, o, is preferred over plural ok: since in (47ii d) both the pronoun and the agreement su‰x on the noun specify plural, the constraint *pl(dp) is violated twice, rather than merely once, and therefore minimally, as is the case with (47ii c).26 The reverse pattern, represented by candidate (47ii b), with plural on the pronoun and the unmarked form on the noun, 26. Since first and second person require the plural form throughout in both clausal and possessor contexts (see exs. (1b, d) above), it must be ensured that *pl(dp) is vacuously satisfied with non-third person. This does, however, not immediately follow from the definition that was provided for this constraint. For the present purpose one may simply state that it only refers to third person, in accordance with the cross-linguistic tendency of avoiding redundant plural on nouns as analysed in section 4. More formally, though, I assume a proliferation of *pl(dp) along the definiteness scale. (In doing so, I build on a proposal I made in Ortmann (2004) in order to account for a split between specifcic and non-specific nouns: In some languages, including Hungarian and Turkish, the latter but not the former may occur without a plural marker in the absence of a numeral or quantifer.) This way we can distinguish *pl(dp)/3rd_Pron from *pl(dp)/Non3rd_Pron, which gives rise to the following hierarchy: *pl(dp)/NonSpec X *pl(dp)/Spec X *pl(dp)/Def X *pl(dp)/3rd_Pron X *pl(dp)/Non3rd_Pron. Now *pl(dp)/3rd_Pron should be taken to replace the general version of the constraint as it is used so far, hence it must dominate the Max constraints. By contrast, *pl(dp)/Non3rd_ Pron, which notes plural marking for first and second person pronouns and agreement markers, is violated for the benefit of Max(pl) in Hungarian and must therefore be ranked below the latter; thus in total: *pl(dp)/NonSpec X ExpressPlurality, *agr[þpl]/Lexical X MaxAgr(pl), *pl(dp)/Spec X *pl(dp)/Def X *pl(dp)/3rd_Pron X Max-head(pl) X Max(pl), *agr[þpl]/ Pron X *pl(dp)/Non3rd_Pron, *agr[þpl]/NonFocus.

256

Albert Ortmann

violates MaxAgr(pl) fatally. It is in this sense that the plurality agreement split and the economic plural split conspire in Hungarian possessor agreement. The fact that plural agreement is found both with pronominal possessors and with no possessor phrase need not be explained in terms of a generalization of the sort ‘‘no pronoun is still a pronoun’’, as most syntactic approaches would assume. Rather, the generalization is expressed in terms of the ranking of explicitness constraints above economy constraints: MaxAgr(pl) (in case of pronominal possessors) and ExpressPlurality (in case of no possessor phrase), respectively, will exclude any candidate without plural agreement; see (47ii, iii). Finally, what is also correctly predicted is the fact addressed in section 2 that in clausal contexts we encounter the pronoun’s plural form ok rather than ‘singular’ o: in domains other than the noun phrase, *pl(dp) is vacuously satisfied, so a violation of max(pl) induced by o will be fatal. Since in Hungarian verbs exhibit full plural agreement with lexical subjects as well, we must assume that MaxAgr(pl) ranks so high as to dominate even the highest economy constraint, thus MaxAgr(pl) X *Agr-S[þpl]/NonSpecific, which corresponds to the general ranking scenario under which no agreement split emerges. 7.2. Turkish anti-agreement I now turn to the anti-agreement e¤ect of Turkish, which looks like the exact converse of Hungarian: in contrast to the latter language, Turkish uses the ‘singular’ on the possessed noun rather than on the pronoun. This is illustrated by the example in (48) involving a wh-pronoun, as well as by the minimal pair with pronominal as opposed to ‘‘dropped’’ possessor in (49c, d). (48)

Turkish (Kornfilt 1997: 26) Ahmet [kim-ler-in kitab-ın-ı ] sat-tı? Ahmet who-pl-gen book-3sg-acc sell-past ‘Whose(pl.) book did Ahment sell?’

(49)

Turkish (George & Kornfilt 1981: 116) a. on-un viski-si b. viski-si pron3sg-gen whisky-por3sg whisky-por3sg ‘his/her whisky’ (id.) c. onlar-ın viski-si pron3pl-gen whisky-por3sg ‘their whisky’

d.

viski-leri whisky-por3pl (id.)

Anti-agreement with subjects and possessors

257

It will be noticed that the data defy any analysis that draws on the a‰nity between an overt and a supposed null pronoun of the third person plural, because only the latter and not the former triggers full number agreement on the possessed noun. Now the plural pronoun in combination with ‘singular’ agreement of the possessor kitab-ın-ı, viski-si rather than plural kitab-ların-ı, viski-leri appears at first sight not only to run counter to the predictions of the account by den Dikken but also to those of the present account as developed in the previous subsection: On the one hand, there is no plurality split in Turkish pronouns, that is, in plural contexts the pronoun will always be plural no matter whether it features as a subject or possessor as in (49a). On the other hand, moreover, the possessed noun does not display plural agreement with pronominal possessors. However, relying on the same constraints that were used thus far in fact predicts qua factorial typology the possibility of a grammar that licenses an anti-agreement e¤ect such as that of Turkish, and will explain it in terms of a di¤erent ranking. Two additional observations are crucial: (i) like in Hungarian, lexical plural possessors obligatorily combine with singular agreement: (50)

Turkish (George & Kornfilt 1981: 116) yazar-lar-ın viski-si author-pl-gen whisky-por3sg ‘the authors’ whisky’

(ii) Unlike Hungarian, ‘‘dropped’’ possessor arguments behave di¤erently from pronominal possessors, in that only the former trigger agreement. Taken together, plural of the possessor will be marked either on the possessor itself, and only there, as long as the possessor is syntactically realised; otherwise, it is solely marked in terms of agreement on the possessed noun.27 I conceive of this complementary distribution as an economy-split similar to the ones discussed so far, but with the faithfulness constraint MaxAgr(pl) being dominated by the entire hierarchy of *AGR[þpl].

27. This complementary distribution is equivalent to that of overt subjects and ‘synthetic’ verb forms in Irish as addressed in notes 7 and 20.

258

Albert Ortmann

In other words, no scalar context restriction *AGR[þpl]/Lexical X *AGR[þpl]/ > Pron for this markedness constraint is activated in this language (except for first and second person, which display full agreement; see below); thus: *AGR[þpl] X MaxAgr(pl). The reason that plural agreement nevertheless occurs in case of ‘‘dropped’’ possessors is that otherwise there would be no indication of plurality at all, which is captured by ExpressPlurality X *AGR[þpl]. That *pl(dp) must be dominated by ExpressPlurality and must in turn dominate the general Max constraint is clear from the fact that like in Hungarian, plural on the head noun is avoided where the concept is already expressed by a numeral or quantifier (see Ortmann 2004 for details). We thus arrive at the following ranking: (51) Ranking for Turkish: ExpressPlurality X *pl(dp), *AGR[þpl] X MaxAgr(pl) X Max(pl) The tableaus in (47i–iii), then, show the evaluations of all logically possible combinations of plural and unmarked for the possessor phrase and for the agreement, with lexical, pronominal, and purely morphologically realised possessor in turn: (52) Evaluation of Turkish possessor agreement: (i) Input: ix [whisky(x) & poss(iy(author(y) & aggr(y)),x)]; Por[þpl], agreement[þpl] Express Plurality a.

yazar-in visky-si author(sg)-gen whisky-por3

visky-si b. F yazar-lar-i n author-pl-gen whisky-por3

*pl (dp)

*AGR [þpl]/

*! *

c.

visky-leri yazar-i n author(sg)-gen whisky-por3pl

*

*!

d.

yazar-i n visky-leir author-pl-gen whisky-por3pl

**!

*

Max Agr (pl)

Max (pl)

*

**

*

* *

259

Anti-agreement with subjects and possessors

(ii) Input: ix [whisky(x) & poss(y,x)]; possessor [þpl], agreement [þpl] a.

on-un pron3sg-gen

visky-si whisky-por3

b. F onlar-i n pron3pl-gen

visky-si whisky-por3

*

c.

on-un pron3sg-gen

visky-leri whisky-por3pl

*

*!

d.

onlar-i n pron3pl-gen

visky-leri whisky-por3pl

**!

*

*!

*

**

*

* *

(iii) Input: ix [whisky(x) & poss(y,x)]; agreement [þpl] a.

visky-si whisky-por3

b. F visky-leri whisky-por3pl

*!

* *

*

*

It is obvious from the tableaus that where MaxAgr(pl) is violated, this is only for the benefit of the dominating markedness constraint *AGR[þpl]. This accounts for the ‘singular’ agreement with lexical and pronominal possessors, respectively. The respective candidate that does not plural-mark the possessor itself is excluded as a fatal violation of ExpressPlurality. For ‘‘dropped’’ topical possessors, the latter constraint is satisfied at the expense of *AGR[þpl].28 In sum, the solution that I propose for Hungarian and Turkish possessor anti-agreement employs only a few constraints, all of which are typologi-

28. Like in Hungarian (cf. note 26), first and second person pronouns obligatorily require plural to be expressed both on the pronoun and on the agreement, in both clausal and possessor contexts: biz-im viski-miz 1pl-gen whisky-1pl, siz-in viski-niz 2pl-gen whisky-2pl ‘our/your(pl) whisky’ (George & Kornfilt 1981: 116). Due to its dominance over MaxAgr(pl) one must assume that (unlike with Hungarian) *AGR[þpl] splits into a /3rd and /Non3rd contextual version. Since the latter is violated for the benefit of Max(pl), the complete ranking is in fact ExpressPlurality X *pl(dp)/3rd_Pron, *AGR[þpl]/ 3rd_Pron X MaxAgr(pl) X Max(pl) X *AGR[þpl]/Non3rd_Pron, *pl(dp)/ Non3rd_Pron.

260

Albert Ortmann

cally motivated, independently of the phenomenon under consideration. The morphological findings are explained functionally in terms of economy: Plural marking is non-redundant in Hungarian noun phrases, and therefore avoided on pronouns featuring as possessor and plural agreement is only found with ‘salient’ possessors. The similar, but converse anti-agreement e¤ect of Turkish follows from the ranking ExpressPlurality X *AGR[þpl] X MaxAgr(pl), which restricts plural agreement to only those cases where the possessor is dropped. 8. Results I have argued that an analysis of anti-agreement splits that generalises over real pronouns and ‘pro-drop’ (i.e., no syntactic realization of an argument) in terms of null pronouns will miss essential generalizations concerning (i) the cross-linguistic role of salience, and (ii) the distribution of plural morphology in Hungarian in contexts other than possessor agreement. Instead, an economy-based constraint analysis was proposed that is at least as explanatory as the assumption of null pronouns. According to this analysis, both Hungarian and Turkish possessor agreement is not only empirically adequately captured, but in addition functionally motivated by relying on a few assumptions that are independently justified. The essential characteristics are: – The account relies primarily on morphological typology and semantics. To the extent that the syntactic structure is crucial it is referred to in the constraints. – Crucial reference is made to economy-based phenomena that are relevant both cross-linguistically (di¤erential object marking and plural agreement splits) and for the Hungarian and Turkish DP (non-redundant plural marking). – The Hungarian data are best understood as a combination of a saliencebased split and a DP-internal plural agreement split, both in terms of constraint hierarchies derived from cognitively grounded scales. – The opposite anti-agreement e¤ect of Turkish clearly contradicts any explanation in terms of the parallel behaviour of third person null and overt pronoun, since the latter does not trigger plural agreement. According to the current account the facts are accounted for by reranking the constraints so as to implement the generalisation that third person plural agreement is generally avoided, except where it is enforced in the absence of any other plurality indicator.

Anti-agreement with subjects and possessors

261

By no means do the facts call for the assumption of zero/null/empty elements, even if one considers null pronouns in general to be part of the inventory of theoretical description: No empirical advantages could be detected that would be lost otherwise; neither is it clear how null elements could be independently motivated, other than by reasons internal to the chosen syntactic framework.

References Aissen, Judith 1999 Aissen, Judith 2003

Markedness and subject choice in Optimality Theory. Natural Language and Linguistic Theory 17, 673–711.

Di¤erential Object Marking: Iconicity vs. Economy. Natural Language and Linguistic Theory 21, 453–483. Amin-Madani, Sadegh & Dorothea Lutz 1972 Persische Grammatik. Heidelberg: Groos. Benmamoun, Elabbas 2003 Agreement parallelism between sentences and noun phrases: a historical sketch. Lingua 113, 747–764. Borsley, Robert D., Maggie Tallerman & Dave Willis 2007 The syntax of Welsh. Cambridge: Cambridge University Press. Dikken, Marcel den 1999 On the structural representation of possession and agreement. The case of (anti-)agreement in Hungarian possessed nominal phrases. In: Istva´n Kenesei (ed.), Crossing Boundaries. Advances in the theory of Central and Eastern European languages, 137– 178. Amsterdam/Philadelphia: John Benjamins. E´. Kiss Katalin 2002 The syntax of Hungarian. Cambridge: Cambridge University Press. Fabri, Ray 1993 Kongruenz und die Grammatik des Maltesischen. Tu¨bingen: Niemeyer. Foley, William A. 1991 The Yimas Language of New Guinea. Stanford: Stanford University Press. Fuß, Eric this volume Some remarks on the historical origin of pro-drop. George, Leland M. & Jaklin Kornfilt 1981 Finiteness and boundedness in Turkish. In: Frank Heny (ed.), Binding and filtering, 105–127. London: Croom Helm.

262

Albert Ortmann

Gerland, Doris & Albert Ortmann 2009 ‘Alienability splits in Hungarian’. Paper presented at the workshop ‘‘Nominal and verbal possession’’, University of Du¨sseldorf, 29.1.2009. Givo´n, Talmy 1976 Topic, Pronoun, and Grammatical Agreement. In: Cahrles N. Li (ed.), Subject and Topic. New York: Academis Press. Harris, Alice C. 1981 Georgian Syntax. A study in relational grammar. Cambridge: Cambridge University Press. Kornfilt, Jaklin 1997 Turkish. London: Routledge. McCarthy, John & Alan Prince 1995 Faithfulness and Reduplicative Identity. In: J. Beckmann, S. Urbanczyk & L. Walsh (eds.), Papers in Optimality Theory, 249–384. Amherst: GLSA (UMass Occasional Papers in Linguistics 18). Nichols, Johanna 1986 Head-marking and dependent-marking grammar. Language 62, 56–119. Olsson, Magnus 1992 Hungarian Phonology and Morphology. Lund: Lund University Press. Ortmann, Albert 2000 Where Plural refuses to agree: feature unification and morphological economy. Acta Linguistica Hungarica 47, 249–288. Ortmann, Albert 2002 Economy-based splits, constraints and lexical representations. In: Ingrid Kaufmann & Barbara Stiebels (eds.), More than words. A Festschrift for Dieter Wunderlich, 147–177. Berlin: Akademie-Verlag. Ortmann, Albert 2004 A factorial typology of number marking in noun phrases: the tension of economy and faithfulness. In: Gereon Mu¨ller, Lutz Gunkel & Gisela Zifonun (eds.), Explorations in Nominal Inflection, 229–267. Berlin: Mouton de Gruyter. Plank, Frans 1980 Encoding grammatical relations: acceptable and unacceptable non-distinctions. In: Fisiak, Jacek (ed.), Historical Morphology, 289–325. The Hague: Mouton. Prince, Alan & Paul Smolensky 1993 Optimality Theory: Constraint Interaction in Generative Grammar. Ms., Rutgers University, New Brunswick and University of Colorado, Boulder.

Anti-agreement with subjects and possessors

263

Prince, Alan & Paul Smolensky 2004 Optimality Theory. Constraint Interaction in Generative Grammar. Oxford: Blackwell. Rouveret, Alain 1991 Functional Categories and Agreement. The Linguistic Review 8, 353–387. Rouveret, Alain 1994 Syntaxe du gallois. Principes ge´ne´raux et typologie. Paris: CNRS Edotions. Sadler, Louisa 1988 Welsh syntax. A government-binding approach. London: Croom Helm. Siewierska, Anna 1988 Word Order Rules. London: Croom Helm. Silverstein, Michael 1976 Hierarchy of features and ergativity. In: R.M.W. Dixon (ed.), Grammatical Categories in Australian Languages, 112–171. New Jersey: Humanities Press. Stiebels, Barbara ¨ konomie und Expressivita¨t. 2002 Typologie des Argumentlinkings: O Berlin: Akademie Verlag. Szabolcsi, Anna 1994 The Noun Phrase. In: Ferenc Kiefer & Katalin E´. Kiss (eds.), The Syntactic Structure of Hungarian. Syntax and Semantics 27, 179–274. San Diego/New York: Academic Press. Woolford, Ellen 2003 Clitics and Agreement in Competition: Ergative Cross-Referencing Patterns. In: Carpenter, Angela, Andries Coetzee and Paul de Lacy (eds.), Papers in Optimality Theory II, 421–449. University of Massachusetts Occasional Papers (UMOP) 26. Amherst, MA: GLSA Publications. Wratil, Melani this volume Uncovered pro: On the Development and Identification of Null Subjects. Wunderlich, Dieter 1997 Cause and the Structure of Verbs. Linguistic Inquiry 28, 27–68. Zeevat, Henk & Gerhard Ja¨ger 2002 A reinterpretation of syntactic alignment. In: Dick de Jongh, Henk Zeevat & Maria Nilsenova´ (eds.), Proceedings of the 3rd and 4th Symposium on Logic, Language and Linguistics. Amsterdam/Tbilisi: ILLC Scientific Publications. (Downloadable at http://www.illc.uva.nl/Tbilisi/Borjomi2001/Proceedings/ borj-proc.pdf )

Language index Abkhaz 11 Alemannic 13, 31, 142 Bavarian 3, 9–10, 22–23, 28, 30, 32– 37, 40, 42–43, 54–57, 59–64, 66– 71, 74–75, 78, 80, 89, 165, 178, 180, 186–7 Bhojpuri 131 Brythonic languages 109, 111 Cape Verdean Creole 113 Cebuano 87–89 Celtic languages 9, 109, 114–117 Brythonic languages 109, 111 Goidelic languages 109 Irish 109–111, 117, 119, 193, 202– 203, 228, 249, 257 Welsh 11, 109, 111–114, 117, 150– 151, 155, 193, 202, 223, 227–228, 230, 243, 246, 248–250 Chabacano 55, 84–90 Chinese 5–7, 62–63, 125–126, 130 Dutch 5, 40, 166, 169, 172, 179, 186 East Netherlandic 32 West Flemish 37, 43, 59, 64, 161, 166, 169, 172, 177, 179–180, 186, 190, 207–208 East Franconian 22, 30–31, 37, 44– 45 East Netherlandic 32 English 1, 5–6, 22, 39–40, 42, 62, 67, 80, 82, 145, 155, 208, 235 Finnish 3, 56, 64–67, 115–119, 130, 132 French 3, 9–11, 25, 53, 55, 70–80, 82, 89, 103–108, 115, 117, 119, 122, 131–132, 151, 153, 163 Frisian 31, 37, 56, 203

German 4–10, 13–14, 21, 25, 30, 35, 42, 53, 118, 148–149, 159–169, 172–174, 176, 178, 186, 188, 190, 233 Alemannic 13, 31, 142 Bavarian 3, 9–10, 22–23, 28, 30, 32–37, 40, 42–43, 54–57, 59–64, 66–71, 74–75, 78, 80, 89, 165, 178, 180, 186–187 East Franconian 22, 30–31, 37, 44– 45 German dialects 9–10, 21, 30–31, 35, 37–38, 40, 42, 53, 143, 149 Middle High German 30, 39, 42 Modern High German 30, 35 Old High German 10, 21–30, 35, 37–46 Thuringian 36 Upper Saxon 31 West Flemish 37, 43, 59, 64, 161, 166, 169, 172, 177, 179–180, 186, 190, 207–208 Zurich German 13–14, 141–222 Germanic languages 7, 23–25, 27, 30, 35, 38, 43, 59, 99–100, 118, 125– 126, 129, 159, 172, 183, 186– 187 see also Dutch, English, Frisian, German, Icelandic Goidelic languages 109 Greek 130, 150, 204–205 Haitian Creole 78, 113 Hungarian 11–13, 186, 223–236, 238, 240, 250–260 Icelandic 7–8, 28, 126–130, 132, 183 Irish 109–111, 117, 119, 193, 202–203, 228, 249, 257 Latin 22, 24–26, 233

266

Language index

Malagasy 55, 82–84, 87–88, 90, 131 Maltese 11, 239 Mauritian Creole 11, 54–55, 79–81, 89, 122–126, 129–132 Mon-Khmer languages 125 Nez Perce 120–122, 125–126, 129, 132 Palauan 150, 193 Papuan languages 130, 233 Penutian Languages 120 Picard 73–74, 77–78 Pied-Noir 73, 77–78 Romance languages 2, 10, 102–103, 108 Cape Verdean Creole 113 French 3, 9–11, 25, 53, 55, 70–80, 82, 89, 103–108, 115, 117, 119, 122, 131–132, 151, 153, 163 Haitian Creole 78, 113 Mauritian Creole 11, 54–55, 79–81, 89, 122–126, 129–132 Spanish 2, 4–5, 55, 67, 79, 84–85, 89 Trentino 108 Veneto 108

Saramaccan 78, 113–115 Sino-Tibetan languages 125 Spanish 2, 4–5, 55, 67, 79, 84–85, 89 Tagalog 55, 83, 86–90 Thuringian 36 Trentino 108 Turkish 11–13, 163, 180, 224, 231– 232, 234, 238, 240, 251, 255–260 Upper Saxon 31 Veneto 108 Welsh 11, 109, 111–114, 117, 150– 151, 155, 193, 202, 223, 227–228, 230, 243, 246, 248–250 West Flemish 37, 43, 59, 64, 161, 166, 169, 172, 177, 179–180, 186, 190, 207–208 Zurich German 13–14, 141–222

Subject index A’-splits 146, 205–206, 208, 212 accessibility hierarchy 143 accusative 14, 148–150, 167, 178–179, 183, 206 see also direct object adjectival inflection 168 Agree 7, 156–157, 188–189, 196 agreement agreement in C 41, 59 anti-agreement 223, 227–232, 243, 251–261 complementizer agreement 30, 44, 56–57, 59, 64, 68, 194 double agreement 32–33, 38–45, 165, 187–190 possessor agreement 12, 15–16, 223, 226, 252, 260 pronominal agreement 28, 37–38, 41, 46, 59, 112, 163–164 verbal agreement 5–11, 32, 44, 53, 55, 59, 61, 64, 66, 68–72, 75–76, 78, 89, 100, 102–103, 108–109, 112–118, 125, 129–130, 224, 240, 246, 250 anti-agreement 223, 227–232, 243, 251–261 base-generation 141–142, 150–158, 184, 191, 193–212 Big-DP 157 big PRO 1, 125 bind 197–199 binding Binding Theory 1, 82, 100, 108, 112–113, 117, 132, 163 indirect binding 154 Principle A 1, 150 Principle B 1, 210 Principle C 82, 113, 163, 209, 211 variable binding 150–151, 184, 211 blocking 10, 54, 66–70, 75–78, 89

case oblique case 14, 148–150, 178–179, 183, 197, 206, 212 structural case 7, 179–183, 193 see also nominative, accusative, dative, genitive cleft 104, 113, 250 pseudoclefts 154 clitic 9–10, 27–28, 33–35, 39–41, 43– 44, 53–78, 81, 89, 103–104, 113, 121, 143, 162, 193–195, 235, 247– 250 see also pronoun comparative deletion 146, 205 constituency 169, 171, 173 coordination 173–176 Crossover E¤ects 152, 191 dative 88, 143, 146, 148–150, 159, 161, 164–166, 171, 176, 178–186, 189–190, 198, 205–206, 208, 212, 225 see also indirect object deblocking 10, 54, 60–61, 67, 70, 74, 89 deletion 155, 209 direct object 86, 88, 141–144, 146– 147, 150, 172, 182, 197, 201, 203, 212, 239 see also Accusative Distributed Morphology 8 Dmin/max 62–67 DP 157 economy 6, 11–15, 102, 195–212, 223, 233–261 trans-derivational economy 198, 201 ellipsis 150, 155–156 Elsewhere Condition 63 Elsewhere Principle 244

268

Subject index

empathy principle 119, 122 gap strategy 141 genitive 148, 159, 166–167, 174, 176, 178–179, 181–183, 186, 190 pre-nominal genitive 176, 179, 190 Government and Binding Theory 99, 108, 112 grammaticalization 55, 69, 73–77, 90, 132 incorporation 102, 109, 190, 228 indirect binding 154 indirect object 141, 143–144, 146, 148, 151, 196–197 see also dative inflection 4–9, 21, 32–36, 38, 40–45, 53–57, 60, 64, 67–70, 75, 80, 84– 85, 89, 101, 111, 115, 117, 168, 234, 238 adjective 168 rich inflection 26, 109 islands 141–142, 144, 148, 152, 155– 157, 193–195, 197, 208, 211–212 language acquisition 3, 15, 41, 67, 69, 81, 101, 118 last resort 14, 142, 147–148, 196, 202, 212 Late Insertion 61, 89 LF-movement 156 locality 14, 141, 148, 150–153, 156– 158, 181, 192–194, 197, 206, 211– 212, 231 PF-theory of locality 156–158, 193– 194 main/subordinate asymmetry 24 Matching Analysis 142, 209–210 -me#s vs. -m/-n in Old High German 26–28, 38–42 movement LF-movement 156 pronoun fronting 150, 157–158, 198–200

V-to-C-movement 23, 104 wh-movement 99, 146, 157, 199, 203, 205–206, 201, 212 NP-ellipsis 150, 155–156 nominative 62, 148–150, 165–167, 205–206, 224–225, 230 see also subject null object 79, 104, 120, 122, 125–126, 129 null possessor 11–12, 14–16 see also possessor null subject null subject cycle 102, 108, 114, 117, 132 null subject parameter 3, 99–100, 118 null topic 119, 122, 126–127 numeration 142, 197, 200, 202 object null object 79, 104, 120, 122, 125– 126, 129 see also direct object, indirect objekt oblique case 14, 148–150, 178–179, 183, 197, 206, 212 Optimality Theory 13, 202, 224, 234– 261 overtness 141, 193, 195, 202, 211 PF-deletion 155, 209 PF-theory of locality 156–158, 193– 194 possessive possessive adjective 165, 186–187, 189 possessive relativization 143 possessive construction 12–14, 161, 164, 182, 211, 238, 250 possessor null possessor 11–12, 14–16 possessor agreement 12, 15–16, 223, 226, 252, 260

Subject index possessor doubling 13, 141–142, 159–164, 169, 178–181, 183, 190, 193 possessor extraction 169, 171–173, 192, 207–208 PP-possessor 182 pre-nominal genitive 176, 179, 190 Principle A 1, 150 Principle B 1, 210 Principle C 82, 113, 163, 209, 211 Principles and Parameters 1–2, 99, 101, 125, 130, 132, 224 PRO PRO theorem 1 big PRO 1, 125 pro-drop full pro-drop 5, 25, 67 partial pro-drop 3, 9–10, 35, 38, 53, 55–56, 60–61, 64, 66, 70, 118 pro-drop parameter (see null subject parameter) pronominal agreement 28, 37–38, 41, 46, 59, 112, 163–164 pronoun clitic pronoun 9, 28, 43, 54–56, 60, 63, 66, 68, 70, 81, 89 pronoun fronting 150, 157–158, 198–200 weak pronoun 2, 9, 15, 54, 60–63, 67, 89, 102, 108, 115, 126, 132, 157, 200 see also clitic, null subject, PRO pseudoclefts 154 Quantifier Quantifier Floating 176 Quantifier Raising 154 R-expression 113, 163, 210 reanalysis 9, 33, 41, 53–55, 59–60, 64, 66–71, 74, 76, 78, 84, 89, 248 reconstruction reconstruction e¤ects 141, 150–152, 154–156, 209 scope reconstruction 154, 201

269

semantic reconstruction 154–155, 209 reference set 142, 195, 197–178, 200– 202, 212 relative clause 14, 141–142, 155 resumption resumptive pronoun 14, 141–143, 146, 156, 158–159, 193, 197, 199, 201–202, 205, 212 resumptive relatives 141–142, 150, 152–153, 156–158, 192–193, 209, 211–212 resumptive strategy 141 rich inflection 26, 109 scope reconstruction 154, 201 semantic reconstruction 154–155, 209 Strong Crossover E¤ects 152, 191 structural case 7, 179–183, 193 subject null subject cycle 102, 108, 114, 117, 132 null subject parameter 3, 99–100, 118 see also Nominative Subset Principle 63, 67 *[that-t]-e¤ect 3 *[that-t]-filter 99–100 TMA marker 80–81, 84–86, 88, 114, 123–124, 131 topic null topic 119, 122, 126–127 topic drop 118, 125–126, 149 topic-worthiness 13, 15, 119–120, 122, 126 trans-derivational economy 198, 201 translocal constraint 202 underspecification 67–69, 75–76, 116 V-to-C-movement 23, 104 V2 7–8, 100, 103–104, 106, 125–127, 129, 169, 194 variable binding 150–151, 184, 211

270

Subject index

Vehicle Change 210–211 verb verb movement 23, 25 verb placement 23, 39 verb second/first 38–40, 44 V-to-C-movement 23, 104 V2 7–8, 100, 103–104, 106, 125– 127, 129, 169, 194 VSO 85–86, 109, 227, 229, 250

weak pronoun 2, 9, 15, 54, 60–63, 67, 89, 102, 108, 115, 126, 132, 157, 200 wh-movement 99, 146, 157, 199, 203, 205–206, 201, 212 Zurich German relative clause 14, 141–142, 155