Describing and Modeling Variation in Grammar 9783110216097, 9783110205909

Citation preview

Describing and Modeling Variation in Grammar

≥

Trends in Linguistics Studies and Monographs 204

Editors

Walter Bisang Hans Henrich Hock Werner Winter

Mouton de Gruyter Berlin · New York

Describing and Modeling Variation in Grammar

edited by

Andreas Dufter Jürg Fleischer Guido Seiler

Mouton de Gruyter Berlin · New York

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter GmbH & Co. KG, Berlin.

앝 Printed on acid-free paper which falls within the guidelines 앪 of the ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication Data Describing and modeling variation in grammar / edited by Andreas Dufter, Jürg Fleischer, Guido Seiler. p. cm. ⫺ (Trends in linguistics. Studies and monographs ; 204) Includes bibliographical references and index. ISBN 978-3-11-020590-9 (hardcover : alk. paper) 1. Language and languages ⫺ Variation. 2. Grammar, Comparative and general. 3. Sociolinguistics. I. Dufter, Andreas. II. Fleischer, Jürg. III. Seiler, Guido. P120.V37D44 2009 4171.2⫺dc22 2009003855

ISBN 978-3-11-020590-9 ISSN 1861-4302 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. ” Copyright 2009 by Walter de Gruyter GmbH & Co. KG, D-10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording or any information storage and retrieval system, without permission in writing from the publisher. Cover design: Christopher Schneider, Laufen. Printed in Germany.

Editors’ Preface The present volume originates from a workshop on “Describing and Modeling Variation in Grammar” that we convened at the 28th Annual Meeting of the German Linguistics Society (DGfS) in Bielefeld, Germany, February 22–24, 2006. This workshop had a predecessor, entitled “Variation und ihre Erkl¨arung”, which was organized by J¨urg Fleischer and Guido Seiler at the 3rd Biannual Meeting of the Swiss Linguistic Society (SSG / SSL) in Berne, Switzerland, October 1, 2004. Both workshops stimulated lively discussions between linguists of different theoretical persuasions and areas of expertise. The present volume offers a selection of the papers presented at the Bielefeld workshop as well as a number of invited additional contributions. We would like to take this opportunity to acknowledge our debt of gratitude to a number of people: First a heartfelt thanks to all contributors for their commitment, dedication and unfailing patience. Special thanks are also due to our anonymous reviewers, who readily agreed to furnish their expert advice despite numerous other obligations. Our impression, which has been confirmed by explicit feedback from several of our contributors, is that their critical comments were of great help in improving and refining the argumentation put forth in the papers. Finally, we wish to thank the managing editors of the Trends in Linguistics. Studies and Monographs series for accepting the volume for publication as well as the editorial staff at Mouton de Gruyter, especially Birgit Sievert, for their cooperation and support. Munich, Marburg and Manchester September 2008

Andreas Dufter, J¨urg Fleischer, Guido Seiler

Table of contents Editors’ Preface

v

Introduction Andreas Dufter, Jürg Fleischer and Guido Seiler

1

Parameter-based and Minimalist approaches Methodological considerations on grammar variation. The right periphery as an OV/VO deciding parameter more so than the left periphery: Gradience in the verb cluster Werner Abraham

21

Variation as lexical choice: have, got and the expression of possession Faye Chalcraft

59

Variation in Icelandic morphosyntax Thórhallur Eythórsson and Jóhannes Gísli Jónsson

81

Constraint-based approaches The predicative as a source of grammatical variation Judith Berman

99

Morphological variation: A declarative approach Martin Neef

117

Different notions of variation and their reflexes in Swiss German relativization Martin Salzmann

135

From documentation to grammatical description: Prepositional phrases in Ruhrdeutsch René Schiering

163

viii

Table of contents

Usage-based, construction-based and statistical approaches Can we factor out free choice? Bert Cappelle

183

Empirical syntax: Idiolectal variability in two- and three-verb clusters in regional standard Dutch and Dutch dialects Leonie Cornips

203

Towards a multivariate model of grammar: The case of word order variation in Dutch clause final verb clusters Gert de Sutter

225

Transplanted dialects and language change: Question formation in Québec Martin Elsig and Shana Poplack

255

Agreement in English dialects Verena Haser and Bernd Kortmann

271

Semi-modal variation Katarina Klein

297

Variation in Komi object marking Gerson Klumpp

325

How lexicalization reflected in hyphenation affects variation and word-formation Britta Mondorf

361

Variation in German adjective inflection: A corpus study Said Sahel

389

Index

407

Introduction∗ Andreas Dufter, J¨urg Fleischer and Guido Seiler 1. The discovery of variation For those who are willing to think about grammar as an algorithmic device that relates mental representations to well-formed linguistic expressions, variation might appear as a rather unwelcome facet of parole, impeding the discovery of interesting generalizations about individual languages and, ultimately, about language as part of our biological and cognitive endowment. In fact, it has been stated that “the general program of all linguists begins with the search for invariance” (Labov 1975: 7). This research program seems to imply that variation, rather than its absence, is something to be expected in linguistic behavior. Conversely, it suggests that identifying invariant aspects behind surface variability of linguistic form will not always be a trivial task. Rough-and-ready definitions of variation as “different ways of saying the same thing” or as “differences that don’t make a difference” are grounded in denotational or truth-conditional equivalence relations and have been put to much use in structuralist theorizing about allo-relations in phonology and morphology and about the dichotomy between emic and etic levels of analysis. In earlier generative terms, this distinction has been recast in terms of input-output correspondences, with variation manifesting itself as different outputs generated from a single input. In syntax, not all linguists would agree that variation can insightfully be analyzed as arising from non-deterministic input-output mappings, all the more so since the exact nature of the syntactic input continues to be under heavy debate. Lexicalist and feature-driven approaches set out to restore determinism by enriching syntactic inputs with specifications related, among other things, to the structuring of information and discourse. At the same time, syntactic variation has been described in more surface-oriented ways, by appealing to underspecified constructions or construction hierarchies, or “re-arrangements” of ‘canonical’ constituent orders and argument structures. Whatever strategy one chooses to adopt, variation in syntax, but also in other components of grammar, might appear at first sight to constitute just another aspect of linguistic diversity. Indeed, in much recent theorizing, it has become customary to treat “macro-variation” – i.e., differences across languages – and “micro-variation” – i.e., linguistic variation observed within an individual language – on a par, subject to the same universal principles and,

2

Andreas Dufter, J¨urg Fleischer and Guido Seiler

possibly, typological tendencies. At the same time, linguists of different theoretical backgrounds have argued that macro-variation and micro-variation are fundamentally different aspects of linguistic diversity, since only in the former case variants necessarily co-exist within a speech community and even within the grammar of individual speakers. Whereas geographical, social and stylistic determinants of variation have figured prominently in variationist studies since their inception, the implications of variation for grammatical description and the consequences of variationist research for grammatical theory continue to be largely unexplored. It is to these issues that this book is devoted. The identification of grammatical variables and the description of their determinants will be the topic of Section 2. In particular, four descriptive key issues will be pointed out. Section 3 will contrast some characteristics of formal and variationist strands of linguistic inquiry (3.1) and offer a survey of the different ways in which grammatical theory has tried to come to grips with variation (3.2). Throughout both sections, we will refer to the contributions in this volume, and indicate points of convergence and divergence between them. Section 4 will conclude with a plea for bridging the gap between formal and variationist linguistics.

2. Describing variation in grammar Whereas variation in phonology and morphology typically involves paradigmatic choices from a limited set of categorially distinct elements, syntacticians often have found it surprisingly difficult to identify two or more syntactic structures as variants of each other, and selecting one of these elements as the mentally privileged, ‘basic’ underlying unit. Many linguists focus on derivational relationships between sentential variants. In doing so, they implicitly acknowledge the need for a constrained theory of grammatical variation, since in many instances, grammar offers a wealth of “different ways of saying the same thing”, i.e., equivalent codings of both propositional content and illocutionary force. However, not just every such pairing of equivalent sentences seems to be equally interesting as a candidate for syntactic variation (cf. Cheshire 2005: 85–86, and references cited there). Consider the sentences in (1) to (4): (1)

John entered the room. It was full of people.

(2)

John entered the room, which was full of people.

(3)

When John entered, the room was full of people.

(4)

The room which John entered was full of people.

Introduction

3

The analysis of such different ways of clause combing may be of great interest to textual analysis. However, few linguists would probably want to study them as grammatical variants obtained from a common syntactic input. Rather, it has been common practice to concentrate on cases of “intimate variation” (Weinreich, Labov and Herzog 1968: 167), i.e., syntactic variants which exhibit only minimal differences, such as the presence versus absence of certain functional elements, or alternative linear constituent orderings. This implies that variation is amenable to analyses in terms of parametric choices, or of optional applications of some elementary rule of deletion, insertion, or movement. The contributions in this volume reflect this preference for in-depth descriptions of such intimate variants. As it turns out, they also reflect vibrant research activities on grammatical variation in West Germanic languages (cf. also the predominance of West Germanic in Kortmann (ed.) 2004 and Delbecque, van der Auwera and Geeraerts (eds.) 2005, both from the same series as the present volume): Seven of the sixteen contributions in this volume are devoted, entirely or in part, to varieties of German (Abraham, Berman, Klein, Neef, Sahel, Salzmann, Schiering), another four to English (Cappelle, Chalcraft, Haser and Kortmann, Mondorf), and three to Dutch (Abraham, Cornips, De Sutter). In addition, the volume contains contributions on Icelandic (Eyth´orsson and J´onsson), French (Elsig and Poplack), and Komi, a Uralic language spoken mainly in the northeastern part of European Russia (Klumpp). Morphophonological and morphological variation is the subject of three contributions (Neef, Sahel, Schiering), while the bulk of the papers addresses morphosyntax and syntax. Within these fields, variation is studied in a wide range of topics, from agreement patterns (Eyth´orsson and J´onsson, Haser and Kortmann), object markings (Klumpp), subcategorization (Eyth´orsson and J´onsson, Klein), copula predicates (Berman), choices between different verb forms encoding possession (Chalcraft) or between synthetic and analytic comparative forms (Mondorf), alternations in particle placement (Cappelle), variable serializations within verb clusters (Abraham, Cornips, De Sutter), and different strategies in relativization (Salzmann) and in question formation (Elsig and Poplack). All of the grammatical variables examined are discrete, some of them even have only two possible variants. Nonetheless, the range of determinants of variation studied in the individual contributions to this volume differs considerably. On the one hand, linguists continue to study the extra-linguistic conditioning of grammatical variables. Differences between national varieties of English and French are investigated by Mondorf and by Elsig and Poplack, respectively. Grammatical variation between standard languages, dialects, and intermediate varieties on the dialect–standard continuum is discussed by Abraham, Cornips, Haser and Kortmann, Salzmann, and Schiering. In addition, Schiering takes into

4

Andreas Dufter, J¨urg Fleischer and Guido Seiler

account sociolinguistic parameters such as age, as do Eyth´orsson and J´onsson. Recent years have seen a renewed interest in idiolectal characteristics; such variation between speakers is explicitly addressed in the contributions by Cornips and Salzmann. Neef’s contribution, in contrast, focuses on variation within individual speakers, and attributes them to different stylistic levels. On the other hand, the contributions to this volume share a special interest in discovering intra-linguistic covariations and, ultimately, linguistic determinants of grammatical variables.An extreme case of grammatical determination is complementary distribution, where one variant occurs categorically in a precisely defined syntagmatic environment, and different variants must occur elsewhere. Besides such strict rules of co-occurrence, many cases of grammatical variation have been described in which aspects of context favor one variant over its competitors. Such weaker contextual determinants may arguably be found in all areas of linguistic structure. Indeed, in this volume, determinants are identified in prosody (De Sutter), morphology and syntax (Abraham, Berman, Cornips, De Sutter, Sahel), lexical choices (Cappelle, Chalcraft, Sahel), semantics (Cappelle, Klein, Mondorf), as well as in information and discourse structure (Klumpp). In the remainder of this section, four general topics, which are of central theoretical importance, will be briefly discussed. First, given the impressive range of extra-linguistic and language-internal determinants of grammatical variation, it is a natural question to ask whether a purported determinant of a variation phenomenon can be shown to be dependent on another determinant of the same variable. Of course, the interrelatedness between determinants of variation is of particular interest to multivariate analyses (cf. the contributions to this volume by De Sutter and Sahel). When extra-lingustic and languageinternal determinants are studied in combination, variation between different sociostylistic levels or between oral and written language use is sometimes claimed to be due to the higher overall grammatical complexity of ‘elevated’ registers or of literacy in general (cf. Neef, this volume). Note, however, that closer empirical investigation of non-standard varieties and unscripted speech has shown that at least in some cases, such correlations are not borne out by the data (cf. Auer 2002 on clause combining in informal spoken and in written German). Another line of research emphasizes discourse and processing needs as being fundamental for the grammar of spoken vernaculars (cf. Abraham, this volume, and references cited there). Conversely, written language tends to obey more closely to ‘canonical’ orderings and clause combining strategies. A second point which strikes us as fundamental for the general conception of grammatical variation is the extent to which all grammatical variables can be shown to depend on language-internal or extra-linguistic determinants. This issue has traditionally been explored in discussions about the existence

Introduction

5

of ‘free’ variation, i.e., variables which are not subject to either linguistic or extra-linguistic conditioning. The concept of free variation has first been formulated in phonology. As is well-known, a phoneme can have two or more ‘facultative’ allophones, i.e., realizational variants that are not conditioned by the syntagmatic environment. The choice between these variants may be conditioned by sociolinguistic and stylistic factors, but need not, as Trubetzkoy (1939: 43) already emphasizes. In the latter case, we are dealing with genuinely free variation, variation without any semiotic significance whatsoever – an observation that structuralists, as well as many modern theoretical linguists, might find hard to swallow. In fact, much variationist research is motivated precisely by the scientific goal of discovering determinants of variation even in cases where no such determinants seem to be available at first sight (cf. the optionality of certain relativizers in English, and the analyses by Fox and Thompson 2007 and Wasow, Jaeger and Orr in press). It remains to be seen whether all purported cases of free variation will eventually turn out to be subject to intricated patterns of determination, or whether we will have to acknowledge the existence of “some entropic remainder” (Cappelle, this volume), that is, of differences that really don’t make a difference. A third issue that has attracted much attention in recent linguistic scholarship is gradience (see Fanselow, F´ery, Vogel and Schlesewsky (eds.) 2006). Variants may not all enjoy the same degree of grammaticality, semantic acceptability or pragmatic felicity. While it has long been known that some grammatical variants are, in some contexts of occurrence or irrespective of context, judged better than others, the implications of gradience for variationist methodology and for grammatical description in general have only recently begun to be taken seriously. Some directions for dealing with gradience, and for complementing the study of linguistic variation by comparing production and perception, are indicated in the contributions by Abraham, Cornips, Salzmann, and Schiering. Finally, as a fourth controversial issue we should at least mention the status of token frequencies for grammatical description. Though few linguists would deny that information about differences in the frequencies of grammatical variants is highly valuable for descriptive reference grammars, no consensus seems as yet to have been reached about the role frequency should play in the description and modeling of linguistic competence. The contributions in this volume reflect this divergence of scholarly opinion among linguists. In the next section, these issues will turn up again in our survey of the role of variation for some recent theories of grammar.

6

Andreas Dufter, J¨urg Fleischer and Guido Seiler

3. Modeling variation in grammatical theory 3.1 Two approaches to the scientific study of language During the last decades, the development of grammatical theory has been dominated by generative grammar in the Chomskyan tradition, directly and, sometimes, indirectly (i.e., via differentiation from mainstream generativist linguistics). This line of investigation and the sophisticated formal apparatus it employs made it possible to gain a better understanding of the highly abstract internal organization of the language faculty, the units and limitations of cross-linguistic variation, the respective relevance of universal and language-specific properties, patterns of language acquisition, the relation of the language faculty to human cognition as a whole – or at least to think about all these issues in a nontrivial fashion. For quite a long period of time, variation phenomena were simply ignored in the development of formal linguistic theory. One important property of the generative enterprise was idealization. Under such a dichotomizing view of competence and performance, ‘trivial’ realities about speech communication such as working memory limitations, but also cultural factors such as politeness or prestige associated with certain linguistic forms, were considered irrelevant for the province of linguistics proper, conceived of as the study of linguistic competence. Intuitions about grammaticality (usually those of the linguists themselves) were the most important source of data for at least two reasons (Henry 2002): First, intuitions are assumed to be largely free of all the performance noise that is pervasive in actual speech, so that they provide access to the internalized grammar in a much more direct way. Second, a theory is supposed to make predictions, in our case, about the grammaticality of sentences. In order to test whether the theory’s predictions hold, it is mandatory to test whether a logically possible structure in a given language is impossible (i.e., ungrammatical) or just unattested. Intuition is the easiest way to obtain negative evidence, which is crucially lacking in historical data (Labov 1994: 11).1 As a perhaps unintended side-effect of the research interest and the methodology related to it, very little interest has been accorded to non-standard varieties, at least until the 1990s. The intuitive judgements made by most formal linguists just happened to be about sentences of standard languages. As a consequence of the lack of interest in variation, the formal apparatus has become unable to generate variable outputs. According to Cornips (2005) and Henry (2005), earlier versions of generative grammar were somewhat better equipped to generate variation. As long as the grammar consisted of a battery of rewriting rules, some of those rules could apply optionally, or be associated to some probability of application in the spirit of the variable rules approach, without much changing the overall

Introduction

7

architecture of the grammar. However, when the formal apparatus became more restrictive – which is, to be sure, desirable, since only a restrictive theory is a predictive theory –, the modeling of variation became increasingly difficult and ultimately impossible. If cross-linguistic variation is implemented in terms of different parameter settings, it follows that a single language does not allow for multiple parameter values at a time. Within the apparatus of Minimalism, cross-linguistic variation is generated by features being either weak or strong. These features simply cannot be both, or semi-weak. It follows that movement operations are triggered or fail to be so, but cannot be triggered to some extent. Not much longer after the advent of generative grammar, a very different line of research became established: sociolinguistics. Sociolinguistics did not come into being ex nihilo, but it tied up with traditions such as dialectology and historical linguistics. These traditions were tangential to the principal concerns of earlier generative theory, but of paramount importance for the methodology of sociolinguistics in that the collection, description and analysis of data is to be conducted in maximal closeness to actual usage, thus minimizing idealization (Henry 2002). Sociolinguistics is interested not only in structural variation within languages, but in particular in the ways variation is determined by extralinguistic, social reality. Thus, both generative grammar and sociolinguistics are interested in aspects of language structure, and, moreover, both are interested in interfaces. However, whereas the most important interface for generative grammar is the one between language structure and human cognition, sociolinguistics takes the interface of language structure and the structure of society as its object of investigation. It seems obvious to us that competence-oriented approaches and approaches focusing on social variability are not incompatible with each other, but rather two complementary subfields of a much more general enterprise, namely understanding how language works, in all its interdependent aspects (see Berruto 2004; Barbiers 2005; Cornips and Corrigan 2005, who argue for an integrative approach in a similar spirit). If it is right that theoretical approaches to the internal structure of grammar and research into variation are perfectly compatible, or even, as we would like to propose, in need of each other, one question immediately arises: Why has there been so little collaboration between the two traditions of research? From the point of view of variationist sociolinguistics, many of the highly abstract notions used in e.g. generativist syntactic theory have just been inapplicable. Mainstream generative frameworks were ill-equipped for the analysis of variation facts and their social conditioning so that there has not been much to gain by incorporating generative concepts and methods into research agendas focusing on the relationships of structure, social conditioning, and social function. From the point of view of formal theoretical linguistics, much work of

8

Andreas Dufter, J¨urg Fleischer and Guido Seiler

sociolinguistics has been considered irrelevant for research questions pertaining to the architecture of the language faculty. To make a very strong, perhaps controversial point, one could even say that sociolinguists have contributed to generativists’ ignorance in matters of variation since they have treated variation as a phenomenon that has very much to do with society but very little with grammar. Given the competence/performance dichotomy, the kinds of variation sociolinguists are interested in appeared to belong to the realm of performance, and thus lie outside the range of phenomena for which a theory of linguistic competence has to account. More specifically, as long as variation appears to be always socially or regionally stratified, it is irrelevant for a model of grammar. As for social stratification, it is unproblematic for the competence-oriented approach to assume that the grammar of the English language spoken by a male, middle-class teenager is different from the grammar of the English language spoken by a female, working-class senior. Similarly, the ‘Englishes’ of speakers from different regions can be argued to instantiate different grammatical systems. The varieties themselves are taken as languages in the sense of individual competences; the variation between those varieties can be taken as minimal differences between very similar grammars. In a similar vein, we would like to emphasize the ontological gap between intra-speaker and inter-speaker variation because it has sometimes been overlooked in the grammatical literature on dialects. Generative linguists, for instance, developed a keen interest in dialect syntax from the 1990s onwards. Of course, dialects deserve to be investigated on a theoretically sophisticated level since they are natural languages like any other, and have even been argued to provide a more direct window into ‘natural’ language acquisition and change than standard varieties, where normative pressure is always liable to interfere (Weiß 2001). However, research into dialects does not per se constitute research into variation. When linguists discuss ‘dialect variation’, they usually refer to differences between two or more regional dialects of a language, or between a regional dialect and the corresponding standard variety. Since this parlance, widespread as it may be, is liable to create confusion, we will refer to such differences as diatopic (Coseriu 1956; Seiler 2004). Crucially, diatopic differences involve different varieties and need to be distinguished as such from variation within one dialect. Given that many studies on ‘dialect variation’ turn out to be comparative diatopic studies, it is only to be expected that formal architectures of grammar that were designed to account for macro-variation do not need to be modified in any way for this particular subtype of micro-variation. However, the picture becomes more challenging for a formal model of grammar if variation is observed within individual speakers, in particular, if these variants cannot be plausibly related to a switching between or a mixing of different

Introduction

9

diatopic varieties or of registers. Indeed, it has been argued that such a competition of linguistic variants may not be an uncommon situation in dialects. The reason for this is straightforward: In contrast to standard languages, where prescriptive intervention and normative effects of literacy have ousted over time forms and constructions qualified as ‘incorrect’, variants have a much greater chance of survival in non-scripted varieties. In the next subsection, we will turn to some proposals to account for such ‘irreducible’ variation within formal models of grammatical competence.

3.2 Grammatical architectures for system-internal variation In a sense, the generation of variable outputs may run counter to our intuitions about grammar, understood as an algorithm mapping some underlying input to a well-formed structured expression. In cases of variation, two (or more) output options are assumed to be identical at some level of abstraction (Adger 2006: 504). If the grammar requires an expression to occur at a specific position within its word form, phrase, clause, or sentence, it cannot alternatively occur at some other position for no obvious reason. Although we are convinced that variation, and even totally unconditioned (‘free’) variation exists, and although this book is intended to make a strong case for variation, it needs to be acknowledged that variation is not something we expect a grammar to produce in the first place. Such a null hypothesis is also implicit in theoretical approaches to grammar. As mentioned earlier, more recent frameworks of the generative tradition, such as Principles and Parameters as well as Minimalism, do not assume grammar to generate variable outputs. This is why it is technically difficult in these frameworks to have grammar generate, alternatively or simultaneously, two or more variant forms or constructions. As already mentioned in 3.1, rule-based earlier versions of generative theory did not face such principal obstacles. In view of this, it has even been suggested that “adequately characterising the nature and limits of human language requires a return to a concept of rules and a research programme which seeks to identify and constrain the nature of the rules” (Henry 2005: 120). The need for taking into account system-internal variants becomes even more obvious in historical linguistics. Every linguistic change presupposes a state of variation (Weinreich, Labov and Herzog 1968: 188; Anttila 2002: 213). Historical change is gradual, and thus involves periods of variation between older and newer options. During such transition periods, a speaker must be able to cope with this reality. In his pioneering work on historical syntax from a generative perspective, Anthony Kroch proposed that competing variants belong

10

Andreas Dufter, J¨urg Fleischer and Guido Seiler

to different, competing grammars which are both present in a speaker’s mind (Kroch 1989, 2000). The point of this kind of analysis is that the grammars themselves remain free of variation, even if competing variants are observed in a speaker’s production. Rather, variation is a function of the interplay of the two grammars. This approach, which can also be referred to as ‘internalized diglossia’, invokes a couple of serious issues, although it seems to us technically feasible (Seiler 2004; Bresnan et al. 2007; Salzmann, this volume). First, variation, understood as system-internal optionality, is completely different in nature from what the term diglossia was originally intended to designate in sociolinguistics (Ferguson 1959). Second, the number of grammars involved increases exponentially with every variable linguistic element. This, in turn, raises serious problems in terms of computational complexity. Third, the pseudo-diglossic approach stipulates that the choice of variants correlates in all instances with extralinguistic factors such as different registers and/or speech styles since the speaker must somehow know when to make use of which grammar – unless one wanted to postulate a ‘meta-grammar’ regulating the power of the parallel grammars. However, in situations where both variants belong to the same level of style of an individual speaker, or where the choice of variants is biased by the influence of purely linguistic factors, a pseudo-diglossic analysis seems to be highly implausible. Such cases are, for instance, the English dative alternation (Bresnan et al. 2007), prepositional dative marking in Upper German (Seiler 2003, 2007), or variation in Finnish morphophonology (Anttila 2002). In this volume, more examples of this type are studied in detail, e.g. the English verbal particle placement (Cappelle), word order in Dutch verb clusters (De Sutter), agreement systems in English dialects (Haser and Kortmann), or Swiss German relative clause formation (Salzmann). M¨uller (2000: 193) critically discusses a second strategy of modeling variation which he calls pseudo-optionality. In contrast to parallel grammar theories, pseudo-optionality does not involve the generation of variable outputs for a given input. Instead, the variants are different already at the level of underlying representation, or feature content. Strictly speaking, the two expressions are not variants of each other, but featurally distinct. If this were always the case, the grammar would be free of variation, mapping underlying representations in a deterministic way to surface expressions. Although we believe that such a strong claim cannot be upheld, we think that from a methodological point of view, pseudo-optionality is a fruitful concept because it encourages the researcher to control for all possible functional differences between the two expressions. Barbiers (2005) proposes a novel way of organizing the division of labor between formal syntax and sociolinguistics. Starting out from the assumption that grammar must be able to generate multiple outputs, his case study proposes

Introduction

11

an implementation for variable orderings within verb clusters in Dutch dialects. According to Barbiers, grammar must provide the set of all variants and is constrained only by principles of Universal Grammar. Conversely, the actual choice of variants is regulated by extralinguistic factors such as regional and social preferences (in this respect the approach is reminiscent of Newmeyer 2003). In our opinion, this proposal faces at least one serious problem: Given that all dialects of Dutch are claimed to share the same underlying grammar, at least as far as verb clusters are concerned, the question arises as to how a speaker should be able to acquire this ‘Common Dutch’ grammar with all its variants in the first place. It must be noticed that this set of all verb cluster options inevitably includes variants that a speaker has never heard and will never produce. Therefore, it seems to us that too great a role is ascribed to sociolinguistics. More specifically, we do not consider it a purely sociolinguistic matter whether a sentence is grammatical or not in a given variety; yet this is exactly what follows from Barbiers’ approach (cf. Salzmann, this volume). Another Minimalist approach to variation is represented by Adger (2006). Adger traces back variation in subject–verb agreement (was vs. were) in varieties of English to the featural specification of the copula verb. Some values of uninterpretable (i.e., agreement) features are assumed to be underspecified. They can therefore match with different features provided by pronominal forms. This analysis remains relatively conservative as for the technical apparatus of Minimalism since no variable rules or probabilities are built into the grammar. However, the restricted set of devices (features are on or off or underspecified, feature combinations must match) nicely works out not only to generate variable outputs, but also to predict frequencies. Adger (2006: 505) compares this account with Mendelian genetics: the underlying mechanism (combinatorics of genes) is categorical in its substance, but the output that it generates is frequentistic. A major shortcoming of Adger’s approach is that it is, as far as we can see, not applicable beyond agreement phenomena. The advent of Optimality Theory opened new avenues of research into variation in grammar. In standard Optimality Theory, output candidates are evaluated by a set of violable but ranked well-formedness constraints which identify exactly one candidate as the most harmonic with regard to its constraint violation profile whereas all other candidates are ungrammatical. However, Optimality Theory’s overall architecture invites for slight adjustments that result in the generation of variable outputs. As Anttila (2002; 2007) points out, leaving conflicting constraints unranked leads to more than one grammatical output. That is, if candidate I is penalized by constraint A and candidate II by constraint B, we expect I to be grammatical given a ranking A  B and II in case of B  A, respectively, because I’s violation of A is more fatal than the

12

Andreas Dufter, J¨urg Fleischer and Guido Seiler

violation of B and II’s violation of B is more fatal than the violation of A, respectively. However, if A and B are unranked (tied), both candidates violate one constraint which leads to an identical violation profile. If no other candidate exists which is more harmonic, both I and II are predicted to be grammatical. The tied constraints approach predicts multiple outputs, but to some degree also probabilities. Assuming that all mutual rankings of tied constraints are equally probable at a concrete candidate evaluation, we expect both I and II to occur with equal probability in the case at hand. In more complex cases, more tied constraints may be involved, all rankings of which are equally probable. If a candidate is most harmonic in a majority of rankings in actual evaluations, it is predicted that the candidate will occur more frequently. A more radical extension of standard Optimality Theory is proposed by Boersma (1997) for phonology and Bresnan et al. (2001) for syntax (see Seiler 2004 for an application in dialect syntax). In the model of Stochastic Optimality Theory, constraints are assumed to be positioned on a scale of real numbers with different ranking distances. At each candidate evaluation, some temporal noise is added, possibly perturbating the order of constraints. The closer together the constraint positions are on the continuous hierarchy, the greater is the probability of reversed rankings. Stochastic Optimality Theory not only manages the generation of competing variants, but it also predicts frequency distributions of competing variants. Furthermore, factors influencing variant choice (such as, for the example of the English dative alternation, givenness, pronominality, animacy, definiteness, phonological weight; cf. Bresnan et al. 2007) are claimed to be an integrate part of the knowledge of grammar, a position radically different from Newmeyer (2003). Recent years have also seen vibrant grammatical research activities conducted within the family of Construction Grammar approaches (see Goldberg 2006, Jackendoff 2008, and literature cited therein). In Construction Grammar, any piece of linguistic structure, be it concrete (idioms) or highly abstract schemata, such as “subject noun phrase + verb”, can be part of the grammar. The overall organization of grammar is a structured set of constructions and thus bears more similarity to the lexicon component within other models. In Construction Grammar, it is relatively easy to account for the co-existence of variants, since nothing prevents two or more competing construction schemata for one and the same structural task to be stored side by side. While the potential of Construction Grammar to capture variation phenomena in a straightforward way is perhaps unrivaled among the grammatical architectures on the market, there are still a number of open questions about Construction Grammar that have arguably been addressed in a more satisfying way by its competitors. One such question is whether the factors that influence variant distribution should

Introduction

13

be an integral part of the grammar or not. A second theoretical issue that needs to be explored further is the nature of generalization. Can generalizations about structural properties within a language be formulated in parallel fashion to cross-linguistic generalizations, using the same types of inheritance networks? Shouldn’t there be a somewhat different status accorded to more general typological principles, constraints, or parameters? Is there really no difference between the modeling of micro-variation and macro-variation? It seems that construction schemata can be postulated rather ad hoc, such that the limits of what is possible in language do not follow from the theory (as it is claimed by models of the Chomskyan tradition). In sum, a division can be made between models of grammar which do not generally allow for variable outputs (Principles and Parameters, Minimalism, classical Optimality Theory) and those where variation is an integral aspect of grammatical architecture. Within the latter group, models differ in that they either integrate factors that predict variant frequencies into the grammar (Stochastic Optimality Theory) or disregard factors determining the choice between variants (Optimality Theory with constraint ties, Construction Grammar).3

4. On the cross-fertilization of formal and variationist linguistics In this concluding section, we would like to mention some reasons for taking variation seriously in grammatical theory, and vice versa. In fact, we are convinced that theoretically minded grammarians and descriptively oriented variationist linguistics could benefit a lot from taking each others’ work into account. Variationist research can take advantage of the formal grammatical tradition in several respects. First, and most generally, any description of a structural regularity in a variety can, and should, incorporate what theoretical linguistics has to say about that kind of phenomenon. Second, formal linguistics can provide us with precise tools and concepts to describe empirical observations. Third, when a linguistic variable is investigated, it is mandatory to assess the relative importance of all identifiable determinants of variation, both those internal and those external to the linguistic system. What this amounts to is that in order to estimate the relevance of various extralinguistic factors, the possible influence of linguistic factors must inevitably be taken into consideration as well. Fourth, formal linguistic theory can be of help in the development of hypotheses about co-occurrence relations between different structural options (see e.g.

14

Andreas Dufter, J¨urg Fleischer and Guido Seiler

Heap 2000). Fifth and finally, it is often the case that a conjunction of structural, social and stylistic factors can be shown to be the best predictor for the probability of occurrence of alternants (Podesva 2006). The benefits to be obtained from the study of variation for a competence model of grammar may seem even more obvious. Variation studies confront linguistic theories with new types of data, including frequentistic and gradient ones, and thereby open up a different perspective on grammar. Speakers turn out to possess an intimate knowledge of linguistic variation even within a single variety, a fact that needs to be reflected somehow in our model of linguistic competence. Of course, fundamental questions continue to be the object of scholarly debate. No consensus seems to have been reached yet as to the question of whether grammars need to be able to supply multiple outputs, even if a majority of linguists today would probably answer in the affirmative. Likewise, different linguists will locate differently the dividing line between grammatical and other types of linguistic competence, including knowledge of extra-linguistic determinants of variation. In any event, however, a theory of grammar that simply chooses to dismiss variation, for technical reasons or some notion of elegance, is of little help in describing and modeling linguistic reality. A final argument for taking into consideration variationist descriptions in grammatical theory is this: Given that a big task of linguistic theory is to provide a cognitively adequate characterization of speaker competence, it is of prime importance to distinguish carefully between intra- and inter-speaker variation. Variants from different dialects or sociolects may well turn out, upon closer scrutiny, not to be found within the linguistic repertoire of any single individual. If such variants are lumped together in grammatical descriptions, formal grammatical models based on such descriptions will inevitably fall short of modeling the linguistic competence of real speakers, and thus miss their goal. According to Chambers (2003: 26–29), the fact that a scientific discipline moves beyond the study of categoricity towards the study of variation can be taken as a sign that the discipline has become mature. The contributions in this volume bear witness to this maturity, and to the empirical scrutiny and theoretical sophistication of current research on variation in grammar.

Introduction

15

Notes ∗

We are grateful to Elvira Glaser, Elisabeth Stark and Richard Wiese for valuable comments on a previous version of this introduction. Needless to state, the usual disclaimers apply. 1. For that reason historical linguistics is forced to make “the best use of bad data” (Labov 1994: 11). It is therefore no surprise that most current variationist work that has theoretical ambitions is conducted with living languages. 2. To be fair, it must be added that such a bias towards standard languages can also be found in typological work. Cf. the EUROTYP project where vernacular varieties are clearly underrepresented compared to languages with a long written tradition – to which group most European languages happen to belong (see Siewierska (ed.) 1998, and subsequent volumes in the series). 3. Of course, these decisions have an impact beyond the internal architecture of the grammar itself. If variation is assumed to be an integral part of grammar, it is expected that variation is a natural part of grammar during the whole process of language acquisition (see Henry 2002, advocating this position). As for historical linguistics, stages of variation need not necessarily be expected to be unstable, pace Kroch (2000).

References Adger, David 2006 Anttila, Arto 2002 2007 Auer, Peter 2002

Barbiers, Sjef 2005

Combinatorial variability. Journal of Linguistics 42(2): 503–530. Variation and phonological theory. In Jack K. Chambers, PeterTrudgill and Natalie Schilling-Estes (eds.) 2002, 206–243. Variation and optionality. In The Cambridge Handbook of Phonology, Paul de Lacy (ed.), 519–536. Cambridge: Cambridge University Press. Schreiben in der Hypotaxe – Sprechen in der Parataxe? Kritische Bemerkungen zu einem Gemeinplatz. Deutsch als Fremdsprache 39(3): 131–137. Word order variation in three-verb clusters and the division of labour between generative linguistics and sociolinguistics. In Cornips and Corrigan (eds.) 2005, 233–264.

Berruto, Gaetano 2004 The problem of variation. The Linguistic Review 21(3/4): 293–322. Boersma, Paul 1997 How we learn variation, optionality, and probability. Institute of Phonetic Sciences, University of Amsterdam, Proceedings 21: 43–58.

16

Andreas Dufter, J¨urg Fleischer and Guido Seiler

Bresnan, Joan, Anna Cueni, Tatiana Nikitina and R. Harald Baayen 2007 Predicting the dative alternation. In Cognitive Foundations of Interpretation, Gerlof Boume, Irene Kraemer, and Joost Zwarts (eds.), 69–94. Amsterdam: Royal Netherlands Academy of Science. Bresnan, Joan, Shipra Dingare and Christopher D. Manning 2001 Soft constraints mirror hard constraints: Voice and person in Lummi and English. In On-line Proceedings of the LFG2001 Conference, Miriam Butt and Tracy Holloway King (eds.). Stanford: CSLI Publications. [http://csli-publications.stanford.edu/ LFG/6/lfg01-toc.html] Chambers, Jack K. 2003 Sociolinguistic Theory: Linguistic Variation and its Social Significance. Oxford: Blackwell. Chambers, Jack K., Peter Trudgill and Natalie Schilling-Estes (eds.) 2002 The Handbook of Language Variation and Change. Malden, Mass./ Oxford: Blackwell. Cheshire, Jenny 2005 Syntactic variation and spoken language. In Cornips and Corrigan (eds.) 2005, 81–106. Cornips, Leonie 2005 Variation and formal theories of syntax, Chomskyan. In Encyclopedia of Language & Linguistics, Keith Brown (ed.), vol. 13, 330–332. Oxford: Elsevier. Cornips, Leonie and Karen P. Corrigan 2005 Toward an integrated approach to syntactic variation: a retrospective and prospective synopsis. In Cornips and Corrigan (eds.) 2005, 1–27. Cornips, Leonie and Karen P. Corrigan (eds.) 2005 Syntax and Variation. Reconciling the Biological and the Social (Current Issues in Linguistic Theory, 265). Amsterdam: John Benjamins. Coseriu, Eugenio 1956 La geograf´ıa ling¨u´ıstica. Montevideo: Instituto de Filolog´ıa. Delbecque, Nicole, Johan van der Auwera and Dirk Geeraerts (eds.) 2005 Perspectives on Variation. Sociolinguistic, Historical, Comparative (Trends in Linguistics; Studies and Monographs, 163). Berlin/New York: Mouton de Gruyter. Fanselow, Gisbert, Caroline F´ery, Ralf Vogel and Matthias Schlesewsky (eds.) 2006 Gradience in Grammar. Generative Perspectives. Oxford: Oxford University Press. Ferguson, Charles A. 1959 Diglossia. Word 15 (2): 325–340. Fox, Barbara A. and Sandra A. Thompson 2007 Relative clauses in English conversation: relativizers, frequency, and the notion of construction. Studies in Language 31(2): 293–326.

Introduction

17

Goldberg, Adele 2006 Constructions at Work. The Nature of Generalization in Language. Oxford: Oxford University Press. Heap, David 2000 La variation grammaticale en g´eolinguistique: les pronoms sujet en roman central (Lincom Studies in Romance Languages, 11). Munich: Lincom Europa. Henry, Alison 2002 Variation and syntactic theory. In Chambers, Trudgill and SchillingEstes (eds.) 2002, 267–282. 2005 Idiolectal variation and syntactic theory. In Cornips and Corrigan (eds.) 2005, 109–122. Jackendoff, Ray 2008 Construction after construction and its theoretical challenges. Language 84(1): 8–28. Kortmann, Bernd (ed.) 2004 Dialectology meets Typology. Dialect Grammar from a CrossLinguistic Perspective (Trends in Linguistics; Studies and Monographs, 153). Berlin/New York: Mouton de Gruyter. Kroch, Antony S. 1989 Reflexes of grammar in patterns of language change. Language Variation and Change 1(3): 199–244. 2000 Syntactic change. In The Handbook of Contemporary Syntactic Theory, Mark Baltin and Chris Collins (eds.), 699–720. Oxford: Blackwell. Labov, William 1975 What is a Linguistic Fact? Lisse: Peter de Ridder Press. 1994 Principles of Linguistic Change. Volume 1: Internal Factors. Oxford: Blackwell. M¨uller, Gereon 2000 Elemente der optimalit¨atstheoretischen Syntax. T¨ubingen: Stauffenburg. Newmeyer, Frederick J. 2003 Grammar is grammar and usage is usage. Language 79(4): 682–707. Podesva, Robert 2006 Phonetic detail in sociolinguistic variation. Ph. D. diss., Department of Linguistics, Stanford University. Seiler, Guido 2003 Pr¨apositionale Dativmarkierung im Oberdeutschen (Zeitschrift f¨ur Dialektologie und Linguistik, Beiheft 124). Stuttgart: Steiner. 2004 On three types of dialect variation, and their implications for linguistic theory. Evidence from verb clusters in Swiss German dialects. In Kortmann (ed.) 2004, 367–399.

18

Andreas Dufter, J¨urg Fleischer and Guido Seiler 2007

Microvariation in LFG and OT. In Syntactic Variation: Rules, Hierarchies and Statistics. A Festschrift for Joan Bresnan, Annie Zaenen, Chris Manning, Jane Grimshaw, Jane Simpson and Joan Maling (eds.), 529–547. Stanford: CSLI. Siewierska, Anna (ed.) 1998 Constituent Order in the Languages of Europe (Empirical Approaches to Language Typology 20: EUROTYP, 1). Berlin/New York: Mouton de Gruyter. Trubetzkoy, Nikolai S. 1939 Grundz¨uge der Phonologie. Prague: Cercle Linguistique. [Reprint: G¨ottingen 1958: Vandenhoeck & Ruprecht.] Wasow, Thomas, T. Florian Jaeger and David M. Orr In press Lexical variation in relativizer frequency. To appear in Expecting the Unexpected: Exceptions in Grammar, Horst J. Simon and Heike Wiese (eds.). Berlin/New York: Mouton de Gruyter. Weinreich, Uriel, William Labov and Marvin I. Herzog 1968 Empirical foundations for a theory of language change. In Directions for Historical Linguistics. A Symposium, Winfred P. Lehmann and Yakov Malkiel (eds.), 95–195. Austin/London: University of Texas Press. Weiß, Helmut 2001 On two types of natural languages. Some consequences for linguistics. Theoretical Linguistics 27(1): 87–103.

Parameter-based and Minimalist approaches

Methodological considerations on grammar variation. The right periphery as an OV/VO deciding parameter more so than the left periphery: Gradience in the verb cluster Werner Abraham Abstract. If it is correct that the (nominal) left periphery in Germanic stems the tide of OV-to-VO a lot more solidly than the (verbal) right periphery, then we expect also the right periphery to show a structurally representable gradient weakening of strict OV3V2V1 . This is what this paper tries to show by applying Optimality-Theoretic principles to the variants in Southern German dialects and in Dutch. It will be seen that the motivation behind left periphery giving way to OV-diagnostics later, and more hesitantly, than the right periphery, has to do with the different discourse areas of the clausal structure of the late OVlanguages, German and Dutch. Processing limitations play a determining role in the gradient verb cluster differences between spoken (dialect) and standard varieties. Dutch gives way to variants of the spoken vernacular more easily than Standard German ever has.

1. Introduction This discussion takes the following premises as points of departure:1 (i)

The right periphery of the clause shows more gradient weakening of OV3V2V1 than any other clausal substructure. The typological change from OV to VO begins in the right periphery.

(ii)

Category dependent (verb vs. noun complemental) licensing switches as well as processing limitations play a determining role in the gradient verb cluster differences between spoken (dialect) and standard varieties. This is evidenced by North Italian substandard varieties of German (Pladen, Cimbrian and M`ocheno, the Fersen valley dialects) as well as Dutch.

(iii)

Dutch gives way to variants of the spoken vernacular more easily than Standard German ever has. Its typological change from OV to VO also begins in the right periphery.

22

(iv)

Werner Abraham

Left and right clausal peripheries are interconnected to the extent that Comp/Co, in the left periphery, and Vo , in the right periphery, are in complementary distribution (with Comp being filled verbally only, if not prefilled by a Comp category (embedding conjunction)). Yet, Comp is on the right edge of the left periphery, whereas Vo is the leftmost position in the right periphery. This relative difference accounts for the different behavior of left-peripheral and right-peripheral categories in diachronic change.

These premises will be empirically motivated in the ensuing four sections, viz. (1) microvariation (as opposed to macrovariation), (2) North Italian German dialects, (3) an Optimality-Theoretic approach to verb clustering (verb reconstruction) and (4) summary and generalization. Section 1 discusses the value of morpho-syntactic studies of oral dialects against a general typological background. The clausal structure of German is introduced with its wide middle field hosting scrambling sites as well as A’-sites for adverbials and, in addition, makes amenable that discourse functions such as Thema and Rhema and contrasting focus are computable solely on the basis of repositioning and refocusing. This sets the scene for Section (2), the discussion of data from dialectal Cimbrian2 and Pladen3, both German varieties (so-called “linguistic islands” or “enclaves”) spoken in Northern Italy. The two dialects are interesting because, notwithstanding the common German origin, they display typical differences of Germanic OV vs. Italian-Romance VO in the verb cluster. The order of the verbs in the right peripheral verb cluster is itself a strong indicator of typological OVvs. VO-categorisation. Section (3) takes up the right clausal periphery demonstrating the crucial differences between the OV-based right periphery (verbal cluster) and the VO one. The German-based dialect of Pladen German, more strongly influenced by the environing Italian both along social and linguistic criteria than its German sister dialect Cimbrian, displays far stronger traits of VO than Cimbrian. The different linearizations in the verb clusters are scrutinized against basic explanatory ordering principles such as a scoping (typological OV/VO-related, licensing) one, a processing one, and a topological agreementrelated one. An Optimality-Theoretic sketch of word order in German allows one to hierarchize these different ordering principles and correlate them to the general hierarchy of module-based linguistic processing. It will be seen that the Optimality Theory based solution to the gradient OV vs. VO leaning in the languages under discussion cannot explain the empirical divergences and will hence be replaced by a licensing split dependent upon complement category and licensing direction.

Gradience in the verb cluster

23

2. Microvariation (as opposed to macrovariation?) Micro-variation, or, more specifically, micro-comparative syntax, can be thought of as comparative syntax work done on a set of very closely related languages or dialects. As a first approximation, the degree of historical relatedness of languages or dialects can be taken as an informal guideline for the degree of syntactic closeness (Kayne 2005: 280). Beyond such plausible, hardly dismissible characteristics, Beninc`a (1994) sets the tune more distinctly by spelling out specifics. In this sense, the micro-variation observed between closely related dialects and dialects and their written standard is better manageable in certain respects, including that properties that both codes share are certain not to be arbitrary. Furthermore, and more importantly, Beninc`a regards the observable variations between closely related North Italian dialects as attesting to the limits of syntactic variation. Given the rich dialectal variations, what is not attested may be taken to be impossible. In other words, dialect investigation may set the boundary between possible and impossible human languages (Beninc`a 1994: 8) and the task of syntax would then be to account for these gaps. Kayne, in his review of Beninc`a (Kayne 2005: 59–64), drives the parameterization further by pointing out that, in the spirit of Greenberg’s word order universals, it is structural implications that turn out so much more probable and realistic in closely related languages and vernaculars. Results gained from comparison are easier to pin down, and the theoretical results have a higher rate of probability (than, e.g., similarities between German and Navajo, or German and Kashmiri). Kayne’s own interesting example (Kayne 2005: 283) is the link between the position of clitics in French infinitives (proclisis) vs. Italian infinitives (enclisis) and the observation of an equally categorical difference in certain control constructions, see (1) and (2): (1)

(2)

French *Jean ne sait pas si partir Jean neg know if leave.inf Lit. ‘Jean doesn’t know whether to leave.’ Italian Gianni non sa se partire Gianni neg know if leave.inf Lit. ‘Gianni doesn’t know whether to leave.’

Apart from what the specific real solution is, notice that dialectal (South) German might have a totally different solution;4 see (3) and (4):

24

(3)

(4)

Werner Abraham

German Wir wissen nicht wen suchen/wo(hin) gehen/ we know not whom seek/where (to) go/ wem trauen/wann kommen whom trust/when come Lit. ‘We do not know whom to seek/where to go/whom to trust/when to come.’ German *Wir wissen nicht ob gehen/wen sehen/ warum kommen we know not whether go/whom see/ why come Lit. ‘We do not know whether to go/whom to see/why to come.’

The postulate is that it is far-ranging links that help come closer to the distinction between possible and impossible linguistic codes (in the Greenbergian (Greenberg 1966) sense; see also Kayne 2005: 326–332). Let us, in this spirit, first pursue the question as to whether there are certain a priori structures that are expectably subject to change across close L-varieties (and, likewise, closer candidates to diachronic change).5 The goal of the present paper is to render an account of the variants in verb clusters in some dialects of German and of Dutch. I base my empirical findings on Abraham (1994) where an Optimality Theory-like explanation was attempted, which is repeated here in more modern Optimality Theory-calibrated terms. The gist of the present recast of my earlier observations (see, more recently, Wurmbrand 2004, who does not go for an explanation of the distributional data at all) is to explain the gradient differences of the empirical data, primarily the differences in the orders of the German vernaculars and those of Dutch. The claim, then and in consonance with much of the comparative literature, was that Dutch had reached a state closer than German to giving up OV in favor of VO. In what follows, I shall first sketch the prerequisites to answering the question why this typological change should emerge in verbal clusters before elsewhere in the clause. To this end, I shall take up arguments extended by Fuß and Trips (2002) and Grewendorf and Poletto (2005). As will be seen from comparative work on older states of English as well on originally German dialects in the Romania, it is indeed the right periphery which gives way to essential VO-traits before any other, and above all the left periphery, weaken and yield. This is in itself worthy of explanation. The ensuing discussion is closely linked to the German clausal structure both of the written standard and the orally represented dialects. Furthermore, the exclusively oral linguistic representation of the dialects make the discussion dependent in high measure on discourse parsing. This is why I introduce the

Gradience in the verb cluster

25

[‘Thema/rhema’ in the second column means that topicalized clausal members can be in both discourse functions simultaneously. Capitals mark focus/rhema: a focus constituent or its head may be marked by a grammatical accent if the focused element is in its base position; it is marked by a contrastive accent if it has been moved out of its base position. Base positions: inside the verb phrase for indefinites, outside for definites and pronominals. The definite subject is the only constituent which is unmarked in topic position. PART = (modal/discourse) particle. GA = grammatical (clause-default) accent, CA = contrastive clausal accent. Notice that Co and Vo are in complementary distribution: V moves to C(omp) unless C is filled by a Comp-category, i.e. an embedding subjunction such as wenn, dass etc.] Prefield 1st Vbracket

(i)

Coord [CP/IP SpecCP – –

(ii) Aber But (iii) – (iv) Aber But (v)

sie she Wieso why wieso why Wieso Why

Co Wann If gibt gives –

–

Middle Field thema WP (clitic AgrOP [VP pronouns) -ts (es) mir des – you you me that

’s ’m it him ’m ’s him it ’s it

– – –

2nd Vbracket [VP Vo ]]] – sie her –

denn after all denn sofort – after all quickly denn p¨unktlich ihm after all on time him

Postfield

A CCENT

rhema Extraposition

SAGT-S

tell-you –

heut today gerne with delight

ABGEBEN ?

GA

CA

GA

hand over heute? show today abgeben? – hand over ZEIGEN

GA

CA

Figure 1. The computable link between the clausal structure and discourse functions in German

main tenets of the clausal structure of German and the syntactic computability of its discourse-functions (Thema, Rhema, repositioning and consequent refocusing) in some detail in the following figures. They demonstrate that German, the dialects even more so than the written standard, is highly discourseprominent to the extent that pronominals are basically in the left (i.e., thematic) range of the middle field and that nominal reference (definiteness vs. indefiniteness) on all arguments is separated according to the discourse functions of Thema and Rhema. Since refocusing takes place when Thema-born and Rhema-generated clausal parts undergo movement for discourse-satisfying reasons, Thema/Old/Presupposed and Rhema/New/Unpresupposed positions are syntax-computable. No extra descriptive discourse level needs to be introduced. German dialects realize all pronominals as clitics as well as definite determiner phrases in the left-most section of the middle field, WP. The (potentially wide) middle field is sharply organized on the basis of the discourse functions Thema vs. Rhema. German dependent clauses reach to IP only.6 Counter to English, independent clauses are invariably CP, without expansions beyond CP

26

(i)

Werner Abraham –

–

–

–

Coord –

[CPSpec,CP –

Co –

–

–

[IP Spec,IP Es Expl-it es Expl-it

(ii) Aber But

⇐ ⇒ Infl u¨ bert¨olpeltei fool u¨ bert¨olpelni fool

Rhema i [VP prinzipiell in principle jedesmal every time

[VP der Igel den Hasen the hedgehog the hare Igel Hasen hedgehogs hares

Vo ]]]] ti ti

Figure 2. All-rhematic presentational/existential clauses in German (dialects)

in Rizzi’s sense (Rizzi 1997). For a middle field upgraded for further discourse functions and definiteness vs. indefiniteness see Figure 4 below. Adverbs are placed according to scopal reach over different structural periods in the prefield (Spec,CP) and the middle field (from C to Vo ). English German

Default focus placement Thema before Rhema VO/OV V – focus – adverb/Thema – VO Thema/adverb – focus – V + OV

Figure 3. Adverb placement and the generalized information flow: “Thema before Rhema”

The German sentence reaches no higher than CP. Counter to Italian or English, there is no expansion beyond that – topics and foci are located in the clausal prefield or the middle field according to their discourse functional status; see Figure 4 (from Abraham 2007). This concludes the sketch of the clause structure and the syntactic computability of discourse functions in German. Our ensuing discussion will be undertaken against this background. It will turn out that the discussion of the order of elements within the verb cluster is contingent, among other criteria, on the clausal default focus and questions of early subject identification in the parsing process.

3. North Italian German dialects As Fuß and Trips (2002) have analyzed, change from Old English to Middle and Modern English starts on the position of the finite auxiliaries. Finite auxiliaries raise in the verb cluster from the original (OV-determined) V2–V1 to (VO-determined) V1–V2. The more OV-to-VO-resistant characteristics comprise the position of the separable verb particles (We put them up) and the position of quantified objects (German Ich hab jemand gesehen ‘I-have-someoneseen’/*Ich habe gesehen jemand). Grewendorf and Poletto (2005) extend this

Gradience in the verb cluster

27

CP 3

C’ 3 o

C 3 Tophigh SentAdv er/(e)s/(ih)n 3 vermutlich FocPcl-ext ‘presumably’ 3 Foccl-ext TopPlow 3 AgrSP Toplow 3 AgrS NegP 3 Spec AgrOP 3 AgrO VP 6 Foccl-int V

Figure 4. Discourse grammatical nodes in German clause structure

intra-clausal comparison to the distinction between the left, and more conservative, periphery and the right periphery, which yields more easily. Notice that, when speaking about gradient diachronic changes, this is in itself a valuable indicator and an explanans of gradience (see, most illuminatingly, Fanselow et al. (eds.) 2006). Comparison between two German dialects in the Romance diaspora – North Italian Lusern, the Cimbrian German-speaking islands, and Pladen German in the mountains between Trento and Vicenza; the villages are part of the “thirteen communities” and the “seven communities”, German speaking islands in a purely Italo-Romance environment – reveals in an illuminating way where in the structure of the (German) OV clause the advent of VO is manifested first and where, by contrast, the original OV-order is more, or the most, resistant. Needless to say, the German in these enclaves has been under continuous contact pressure from the Romance neighbor languages, which are all VO. See (iv) in Section 1. The following characteristics attest to the still existing OV, the original order type of Old High German and Old Italian as well as Modern Standard German. Their occurrence is restricted to the left periphery. The data in (5)–(7) illustrating “Germanic” subject inversion in verb second declaratives, with non-subjects in the Spec,CP prefield, are from Grewendorf and Poletto (2005).

28

(5)

Werner Abraham

Pladen German Hainte tuit de Mame de Kinder ins Pette. today does the mother the kids to-the bed ‘Today mother puts the kids to bed.’

By contrast, Lusern has lost subject inversion if nominal subjects are present, cf. (6).Yet, with a clitic subject, inversion is possible, as shown by (7). (6)

Lusern M`ocheno *Ha¨ute geat dar Gianni vort. today goes the Gianni away ‘Today Gianni goes away.’

(7)

Lusern M`ocheno Ha¨ute geat-a vort niamat. today goes-he away nobody ‘Today nobody goes away.’

This is in itself not waterproof evidence for the loss of verb second, but points at an independent factor in this syntactic variation (see our assumption of a contact induced change at the end of Section 4). Typological VO vs. OV is decided, among other criteria, by the following characteristics (the first three, (8a–c), are due to Fuß and Trips 2002; for a more detailed survey as between OV-German and VO-Icelandic cf. Haider 2000a, 2003, 2005): (8)

a. Abstracting away from dislocation effects which may blur the picture, postverbal verb particles in independent clauses determine OV.7 b. Postverbal light elements such as pronouns or light adverbs, which cannot be the result of rightward movement, indicate OV. c. The order “finite verb – object – main verb” determines OV. By contrast, the linear sequence “verb – participle – object” is a strong indicator of VO.8 d. The middle field (between C/I and the finite lexical verb or auxiliary in clause-last position) is subject to discourse-determined focus vs. Thema distribution, with focus placed on the right edge (Abraham 1995, 1997); compare Figures 1–3 above. This constraint, however, is of little avail for our discussion of the right periphery. e. The finite predicate component in the dependent clause is in absolute verb-last position (with limited parsing exceptions; Abraham 1995, 1997).

Gradience in the verb cluster

29

f.

Verb projection raising (as claimed for Zurich German; cf. Haegeman and van Riemsdijk 1986, Sch¨onenberger 1995), which is sensitive to discourse functions (like Thema and Rhema), is indicative of OV (Fuß 2002, 2004; Pintzuk 1999). Schmid and Trips (2003: 13) show that as soon as the language in question turns into the VO type verb projection raising is lost.9 g. Inversion of finite auxiliaries or modal verbs across the non-finite full verb in the verb cluster (usually concomitant with infinitivus pro participio-IPP/German Ersatzinfinitiv) need not indicate VO. More so, in three-elemental verb clusters of OV only verb first inverts (all the way or partly), but never verb second and verb third (Abraham 1995, 1997). Notice that this is similar to the structure condition in (8c) above, the exception being that it works inside the verbal complex.10 h. OV-based verb projection raising in a dependent clause excludes a (raised) finite modal verb before an indefinite object or an unstressed adverb. This predicts the ungrammaticality of “finite modal verb – (*indefinite object/adverb) – V0 ” for the OV-type.11 i. The adjacency of finite auxiliary and participle typical of VO cannot be suspended by any type of adverb. This predicts the ungrammaticality of “finite auxiliary – (*indefinite object/adverb) – V0 ” for the OV-type, while it is grammatical for VO, as illustrated by English (that) he has already solved the problem. (8h) and (8i) may be taken as sub-constraints to (8f). To substantiate (8d), see Pladen (illustrations from Grewendorf and Poletto 2005). Pladen German (9b) is out unless the moved accusative object carries contrastive accent and thereby indicates movement under refocusing as in (9c). (9)

Pladen German a. Geischter hot der Hons de Mame me Mork yesterday has the Hans the mother on-the market pagaignet. ran-into ‘Yesterday Hans ran into the mother at the market.’ b. *Geischter hot der Hons me Mork de Mame yesterday has the Hans on-the market the mother pagaignet. ran-into ‘Yesterday Hans ran into the mother at the market.’

30

Werner Abraham

c.

(10)

Geischter hot der Hons me Mork DE MAME yesterday has the Hans on-the market the mother pagaignet ober et de Schweischter. ran-into but not the sister ‘Yesterday Hans ran into the MOTHER at the market, not into the sister.’

Pladen German I waas as de Mame de Alan me Morke gekaaft hot. I know that the mother the eggs on-the market bought has ‘I know that mother bought the eggs at the market.’

The sequence “finite verb – past participle – object” should not be acceptable in an OV-language – which bears out for Pladen German (see (8c) above): (11)

Pladen German *Sie hont mer gesouk, as ana Kronkhet hot geteten de they have me told that an illness has killed the Kie. cows ‘They have told me that an illness killed the cows.’

Both the finite auxiliary and the non-finite modal verb can precede the main verb in dependent clauses – which is more than a true OV-language usually does in terms of inversion within the verb cluster: (12)

Pladen German Si hot gesouk, as de Mame de Teller hot gebellt she has said that the mother the plates has wanted aufspieln. rinse ‘She has said that mother wanted to rinse the plates.’

Unlike in OV-based verb projection raising, any modal verb may precede an indefinite object or unstressed adverb in dependent clauses – something that makes (13) typical of a VO-derivation (characteristic of the linguistic Italian environment, see (8h) above). (13)

Pladen German as der Hons hot gekeint gut schlofn / schnell lafn that the Hans hat could well sleep / fast run ‘that Hans could sleep well / run fast’

Gradience in the verb cluster

31

It may not be so clear whether this is a special case of verb projection raising or a genuine VO-structure. Yet, summarizing, Pladen German can be generalized to follow the following ordering restriction: (14)

(Aux) – Pronoun – (Aux) – DativeCase – (Aux) –Accusative – DativePrep – (Aux) – V – (Aux)

(14) embodies the generalization that in Pladen German auxiliaries and modal verbs, but not main verbs, indicate VO. The fact that a main verb is compatible with strict OV is corroborated by the clause final position of the separable verb particles: (15)

Pladen German a. Hainte de Mame spielt de Teller auf. today the mother rinses the plates off b. *Hainte de Mame spielt auf de Teller. today the mother rinses off the plates ‘Today, mother rinses the plates.’

What counts in this context is the fact that Pladen German is in a transient state from OV to VO. However, VO order is limited to the verbal grammatical categories of auxiliary and modal verbs, not to the full verb. In other words, the transient state reconstructs in the categorial realm of verb clustering auxiliaries and modal verbs. This mirrors precisely the development from Old English OV to Middle and Modern English VO, where the new order also embraces auxiliaries and modal verbs in the verb cluster first: from OV-Old English (like German) “finite verb – object – non-finite verb” to VO-Late Middle English “finite verb – non-finite verb – (light) object”. Cimbrian, the other Germanic-based North Italian dialect, clearly embraces the VO-type as is demonstrated by the typical order “finite verb – participle – object” (illustrations from Grewendorf and Poletto 2005): (16)

Cimbrian German ¨ Ha¨ute die Mome hat gekoaft die Oala. today the mother has bought the eggs ‘Today mother has bought the eggs.’

32

Werner Abraham

This typical VO order prevails even in dependent clauses: (17)

Cimbrian German Dar issese darz¨urnt obrom gestarn honne i get there is-she-refl maddened because yesterday have I given an Libar in Peatar. a book him Peter ‘She has got mad because I gave a book to Peter yesterday.’

The adjacency of finite auxiliary and participle typical of VO cannot be suspended by any type of adverb (see (8i) above), cf. (18): (18)

Cimbrian German a. *Dar issese darz¨urnt obrom i hon gestarn get here is-she-refl maddened because I have yesterday given an Libar in Peatar. a book him Peter b. Dar issese darz¨urnt obrom i hon get there is-she-refl maddened because I have given gestarn an Libar in Peatar. yesterday a book him Peter ‘She has got mad because I gave a book to Peter yesterday.’

However, the VO-pattern is not realized throughout. As far as indefinite quantifiers are concerned, the linear pattern “finite verb – quantified pronominal object – participle”, which is typical of OV and which is ungrammatical in pure VO-languages, leads to the conclusion that Cimbrian, which is clearly en route to VO, has retained original traits of OV. It is crucial to see that this conservative characteristic is part of the middle field and not of the right periphery. Trips (2002: 203) points at exactly the same phenomenon in Middle English – a period that has almost finished the transition from Old English OV to VO. (19)

Middle English (Trips 2002: 203) He haþ on us mercy, for he may al þynge do. he has on us mercy since he may all things do ‘He has mercy upon us since he may do all things.’

Note that Middle English (19) contradicts (8h) outright, (8h) holding for OV. One of the least ambiguous indicators of conservative OV is the verb-last position of separable verb particles. Cimbrian obeys this positional generalization, yet it also allows VO-indicative “verb particle – predicative participle”.

Gradience in the verb cluster

33

Grewendorf and Poletto (2005: 18) point out that, at the same time, even with the preverbal verb particle, the (OV-derived) order “finite auxiliary – *object – predicative participle” is ungrammatical (cf. (20c)) – a lot less even in embedded relative clauses. (20)

Lusern M`ocheno a. I hon au-gehort die Arbat ka Tria. I have up-stopped the work in Trento ‘I have stopped working in Trento.’ b. I hon gehort-au di Arbat. I have stopped-up the work ‘I have stopped working.’ c. *I hon die Arbat au-gehort. I have the work up-stopped ‘I have stopped working.’

(21)

Lusern M`ocheno a. dar Mann bo da o hat geheft a na¨uga Arbat. the man who there on has started a new work ‘the man who has started a new work’ b. *I gloabe che dar Hons vort is gont. I believe that the Hans away is gone ‘I believe that Hans has gone away.’

Clearly, this shows that Cimbrian M`ocheno has not brought to a close the transition from OV to VO. Matrix clauses behave the OV-way with respect to verb particles. Embedded clauses, on the other hand, give way to VO early on by allowing objects after the predicative participle. In sum, the gradience with which the OV-property of the two Ladinian Germanic dialects give way is the following, with 3 as the most conservative, i.e. least modern and most resistant, and 1 as the weakest, least resistant, OV-trait. Compare again the constraints in (8a) to (8h) above: (22)

Hierarchy of OV-traits 1 Auxiliary and modal verb raising < 2 quantified pronominal objects < 3 separable verb particles (< 4 non-subjects in the prefield, Spec,CP, i.e. “Germanic subject inversion in verb second declaratives”)

34

Werner Abraham

Trait 1, auxiliary and modal verb raising, in particular, allows for inverted orders of the finite and the infinite verbal components in OV. Compare Standard German dass er den Teppich will wegwerfen, lit. ‘that he the carpet wants to throw away’. From the two different developmental states in Cimbrian and Pladen German, one would like to conclude that, given identical exposure to Romance contact, the linguistic change from OV to VO follows an interior, contact-independent sequence determined by the clausal structure and explicable in terms of control over different structural domains – in particular, the right (verb/Rhemacontrolled) vs. left (complementizer/noun/Thema-controlled) peripheries. However, admittedly, the left periphery, 4, in particular, has not been compared as to co-temporal gradience with the rest of the word order criteria in (22). Given that, our surmise of a change induced by contact to Romance VO languages remains a viable alternative.12 We shall exploit this insight in investigating more closely the variation in the verb cluster in Modern Dutch and Alemannic, Austrian and Bavarian (based primarily on Abraham 1995; for a data collection of West Germanic, see also Wurmbrand 2004). The intent is to develop a generalization over the gradience with which the logical OV-order gives way to VO.

4. An Optimality-Theoretic approach to verb clustering (‘verb reconstruction’) The following has become common parlance in modern linguistic research: (23)

Word order is not only dependent on or determined by (the hierarchical relations in) a single syntactic representation, but arises also as a result of linear precedence rules active in particular word order domains or as a result of the interaction between different levels of syntactic representation.

(24)

Word order often is the result of the interaction of a limited number of violable output oriented constraints.

This is behind the attempt in this essay to determine, as closely as possible, the variance of the members of verb clusters in German (spoken variants) and in Dutch. It is assumed that both triggers, and thereby criteria, of change and processing economy play a determining role (see, in general terms, for the spoken codes of German, Dutch, and Afrikaans Abraham 1999a, 1999b, 1999c, 2002a, 2004; Abraham and Conradi 2001; for the diachrony in English see van Gelderen 2004).13

Gradience in the verb cluster

35

4.1 Optimality-Theoretically described and explained word order in German According to Engel (2004: 95), the basic order of the finite verb in the German dependent clause is explained as in (25) to (27) below. (25) Morgen reist Peter ab. tomorrow leaves Peter off ‘Peter will leave tomorrow.’

Prefield subject ObHd Stay-X *Lx Mv Stay-XP

a. [CP Morgen reisti [IP Peterj [VP tj ab-ti ] ti ]] b. [CP Morgen e [IP Peterj [VP tj ab-ti ] reisti ]]

*!

c. [CP Morgen reisti [VP Peter ab-ti ]]

*!

**

**

*

*

*

*

*

*

d. [CP e Reisti [IP morgen [IP Peterj [VP tj ab-ti ] ti ]]] *!

**

**

*

e. [IP Morgen [IP Peterj [VPtj ab-ti ] reisti ]]

*

*

*

*!

(26) Prefield *X’-Adj *Cp-Adj Stay-XP

Den Karl hat wahrscheinlich der Fritz verpr¨ugelt. the Karl has probably the Fritz beaten.up ‘Fritz probably beat up Karl.’ a. [CP Den K.k hati [IP wahrscheinlich [IP der Fritzj [AuxP [VP tj verpr¨ugelt] ti ] ti ]]]

**

b. [CP Den K.k hati ti ] [IP der Fritzj [AuxPwahrscheinlich [AuxP [VP tj verpr¨ugelt] ti ]]] c.

[CP Den K.k wahrscheinlich [C hati [IP der Fritzj [AuxP [VP tj verpr¨ugelt] ti ] ti ]]]

d.

[CP Wahrscheinlich [CPden K.k hati [IP der Fritzj [AuxP [VP tj verpr¨ugelt] ti ] ti ]]]

** *!

** *!

**

In the absence of a finite verb, Co is filled by a complementizer (Engel 2004: 98): (27) weil Hans Karla gek¨usst haben soll. because Hans Karla kissed have should ‘because Hans has allegedly kissed Karla.’

PureEp Subject ObHd Stay-X Stay-XP

a. [CP weil [IP Hansj [AuxP1 [AuxP2 [VP tj Karla gek¨usst] haben] ti ] solli ]] b.

[CP weil [IP Hansj [AuxP1 [AuxP2 [VP tj Karla gek¨usst] haben] soll] e]]

c.

[CP weil [AuxP1 [AuxP2 [VP Hans Karla gek¨usst] haben] soll]]

d.

[CP solli weil [IP Hansj [AuxP1 [AuxP2 [VP tj Karla gek¨usst] haben] ti ] ti ]] *!

*

* *

*! *! **

*

36

Werner Abraham

(i) Appendix: Constraints and rankings as used in (25)–(27) (Engel 2004: 299) *CP-ADJ: INVERSION:

X’-ADJ: LXMV: OBHD: PREFIELD: PUREEP:

STAY-X: STAY-XP: SUBJECT: X’-ADJ:

There is no phrase adjoined to CP. X0 and Y0 cannot be co-indexed if there is an overt phrase ZP, ZP c-commanding a segment of IP, and X0 c-commands ZP and ZP ccommands Y0 . There is no adjunction to the X’-level. A lexical head cannot move. A projection has a head. The specifier position of matrix CP is filled. No adjunction takes place to the highest node in a subordinate extended projection; and no movement takes place into the highest head of a subordinate extended projection. X0 -trace is not allowed. XP-trace is not allowed. The highest A-specifier is structurally realized. There is no adjunction.

(ii) Constraints and rankings not used in (25)–(27), but belonging to the linearization constraints in German (Engel 2004: 299) ADJCON: ADV3} {0–1, 2–3, >3}

{separable, non-separable} {complement of the main verb, complement of a preverbal head, adjunct/no extraposition} 5 Length of the middle field {0–2, 3–5, 6–8, 9–11, 12–14, >14} words 6 Information value of the last preverbal word {highly informational, intermediately informational, low informational} 7 Inherence of the last preverbal content word {inherent, non-inherent} 8 Type of finite verb/categorial status of the {copular verb, temporal auxiliary, participle passive auxiliary zijn, passive auxiliary worden, unclassifiable} 9 Syntactic persistence {previous participle-first word order, previous participle-final word order, no previous verb cluster} 10 Frequency of the participle {0–∞} 3 4

order, whereas an interval of more than three unaccented syllables stimulates participle-final word order (same reference point). The operationalization of the variable morphology of the participle (explanatory variable 3) was unproblematic since both values are formally identifiable: separable participles contain a particle which can be detached from the verb in certain contexts (e.g., when the verb is used in the simple present; afgemaakt ‘finished’) whereas non-separable participles cannot be split (gemaakt ‘made’). Based on previous research (cf. Section 2), we hypothesize that separable participles, compared to non-separable participles, stimulate the use of participle-final word order. The original variable presence vs. absence of an extraposed constituent (cf. Table 1) was recoded as grammatical relationship between the postverbal constituent and its head (explanatory variable 4). This variable not only incorporates the presence vs. absence of an extraposed constituent, but also its syn-

Towards a multivariate model of grammar

239

tactic function in the clause and the head to which it relates. As a consequence, the assocation between the extraposed constituent and the choice of word order can be studied more in depth. A distinction is made between clauses without an extraposed constituent (cf. (13)), clauses with an extraposed constituent that functions as the complement of a preverbal head (a noun, adjective or adverb; cf. (14)), clauses containing an extraposed constituent that functions as the complement of the main verb (cf. (7b), repeated as (15) below) and clauses with an extraposed constituent that functions as an adjunct (16). In order to distinguish between the different values, a set of criteria was developed based on the Dutch reference grammar ANS (Haeseryn et al. 1997; cf. De Sutter, Speelman and Geeraerts 2008 for a complete overview of the criteria). (13)

dat ze voor het vaderland worden gedaan (DS) that they for the father-country are done ‘that they are done for the country’

(14)

dat precies hier het verschroeide [lijk]head werd gevonden that exactly here the burned body was found van een Afrikaan die al dood geweest moest zijn of an African who already dead been must be toen het vuur uitbrak (DS) when the fire begun ‘that the burned body of an African – who must have been dead by the time the fire begun – was found exactly here’

(15)

dat ze nooit [weggestoken]head heeft dat ze elke that she never hidden has that she every dag een glas gin drinkt (DS) day a glass of gin drinks ‘that she has never hidden that she drinks a glas of gin every day’

(16)

dat de Vlaamse regering heeft [geknoeid]head bij de that the Flemish government has tampered with the aanstelling van Marc Cl´emeur tot intendant van de appointment of Marc Cl´emeur to manager of the Vlopera (DS) Vlopera ‘that the Flemish government has tampered with the appointment of Marc Cl´emeur as manager of the Vlopera’

240

Gert De Sutter

In a preliminary analysis, it turned out that there is no significant difference between the values adjunct and no extraposed constituent, as a consequence of which these two values are lumped together in the remainder of this paper. Based on previous research (cf. Section 2), we hypothesize that the presence of a postverbal constituent, compared to the absence of a postverbal constitent, stimulates the use of participle-first word order. There are no hypotheses concerning the grammatical relationship and type of head since both are new to the research field. The variable length of the middle field (explanatory variable 5) has a straightforward operationalization, viz. the number of words between the complementizer dat ‘that’ and the verb cluster. Based on previous research (cf. Section 2), we hypothesize that the length of the middle field correlates positively with the use of participle-final word order. The variable information value of the last preverbal word (explanatory variable 6) is based on the original variable definiteness of the last preverbal constituent (cf. Table 1). The latter variable only includes information about nouns, whereas the new variable includes information on other word classes too. The variable takes three easily identifiable values: highly informational (i.e. nouns/verbs/numerals), intermediatly informational (i.e. adjectives and adverbs) and low informational (i.e. pronouns, conjunctions and prepositions). The arrangement in three groups has both a technical and linguistic reason. The technical reason is that not all differences between the individual word classes were statistically significant. A partitioned chi-square test showed that nouns, verbs and numerals do not differ significantly in their effect on the choice of word order and can therefore be grouped together. In the same way, it was found that adjectives and adverbs constitute a group, and, finally, that pronouns, conjunctions and prepositions can be merged in one group. From a linguistic perspective, each of these groups are internally coherent, as their members display a similar information value: the group with nouns and verbs is highly informational (they contribute the most to the referential content of the clause), adjectives and adverbs have an intermediary information value (as they are mostly used to modify nouns and verbs), whereas prepositions, conjunctions and pronouns have a low information value (they mainly have a mere syntactic or discursive function, without contributing to or modifying the referential content). In previous research (cf. Section 2), it was found that indefinite nouns (vs. definite nouns) stimulate the use of participle-final word order. If one assumes that indefinite nouns have a higher information value than definite nouns (since indefinite nouns introduce new referents), one could hypothesize that the degree of information value correlates positively with the use of participle-final word order.

Towards a multivariate model of grammar

241

The operationalization of the variable inherence of the last preverbal constituent (explanatory variable 7) is not straightforward. First, the variable comprises rather different types of constructions (such as complement parts of a copular verb, predicative adjuncts and fixed expressions; Haeseryn et al. 1997: 1350–1352); therefore, we chose to narrow the variable down to only one specific type, viz. fixed verbal expressions (e.g., in staat zijn ‘to be able to’). Second, there is no general consensus on the definition of a fixed expression so that an operationalization is not self-evident; therefore, we chose to define a fixed verbal expression as a fixed pattern consisting of a participle and at least one preverbal word. Since a fixed pattern implies that the pattern is recurrent, we identified the fixed verbal expressions in our data set by computing the statistical attraction between the participle and the left collocate of the verb cluster (i.e. the word preceding the verb cluster).6 More particularly, we used the log likelihood ratio (LLR), which is well-known in statistical collocation research (e.g., Dunning 1993) to compute whether two words co-occur significantly. Each combination of a participle and a left collocate in our data set is assigned an LLR score on the basis of which the statistical significance can be determined. This yielded a binary variable: fixed expressions (significant attraction) vs. nonfixed expressions (non-significant attraction). Based on previous research (cf. Section 2), we hypothesize that fixed verbal expressions, compared to non-fixed verbal constructions, stimulate the use of participle-final word order. As indicated above, the operationalization of the variable type of finite verb/categorial status of the participle (explanatory variable 8) in previous research was corrupted by vague semantic notions and morphosyntactic addition tests that are able to detect only the adjective potentiality of a certain participle, but not the adjective reality. In order to get a firmer grip on the status of the participle and the finite verb, we developed an algorithm that aims at disambiguating the status of the finite verb and participle. In the first module used to check the actual status of the participle, the categorial membership is determined on the basis of intersubjectively validated contextual cues, such as the presence of a specifying temporal or locative adjunct, or an agent introduced by door ‘by’. If there are no contextual cues present, the participle is sent to the second module, where the categorial membership is determined on the basis of several types of probability measures. These measures quantify the potential of the participle to be used as a verb or an adjective. One of these measures, for instance, quantifies the degree of adjectivity by dividing the token frequency of a participle by the type frequency of the corresponding verb. The idea that the relative importance of the participle in the verbal paradigm is a good indicator for the degree of adjectivity of a participle or, in other words, for the extent to which a participle has developed a specialized, divergent, adjectival meaning

242

Gert De Sutter

next to the verbal meaning is based on the traditional view that obsolete participles, i.e. participles without corresponding verb forms, have to be categorized as adjectives. If the probability measures are not reliable, e.g. in cases of low frequency, it is sent to the third module. Here, the participle is disambiguated on the basis of morphosyntactic addition tests to find out whether it is at all possible to use the participle as an adjective. Examples of these tests are prefixing by on- ‘un-’, grading by zeer, heel, hoogst, te ‘very’ and pronominalization. The algorithm is discussed at large in De Sutter (2005). For the analysis in Section 4, the following categories are discerned: copulative verb (17), zijn ‘to be’ as passive auxiliary (18), worden ‘to be’ (19), temporal auxiliary (20) and unclassifiable (i.e., participles that could not be disambiguated by the algorithm; cf. (21)). Based on previous research (cf. Section 2), we hypothesize that the copulative use of the auxiliary verb zijn ‘to be’ only allows participle-first word order (compared to the other categories). (17)

dat polio of kinderverlamming sinds 1960 uitgeroeid is in that polio or infantile-paralysis since 1960 exterminated is in Joegoslavi¨e (DS) Yugoslavia ‘that polio or infantile paralysis is exterminated inYugoslavia since 1960’

(18)

dat zijn bezittingen in de VS door Washington bevroren that his possessions in the US by Washington frozen waren (DS) were ‘that his possessions in the US were frozen by Washington’

(19)

dat ze in Bhutan zwaar werden gediscrimineerd that they in Bhutan severely were discriminated-against ‘that they were severely discriminated against in Bhutan’

(20)

dat de EU haar ultimatum had verlengd (DS) that the EU her ultimatum had extended ‘that the EU has extended its ultimatum’

(21)

dat de toeschouwers de indruk krijgen dat het stuk op that the spectators the impression get that the piece on waargebeurde feiten gebaseerd is (DS) historical facts based is ‘that the audience gets the impression that the piece is based on historical facts’

Towards a multivariate model of grammar

243

The variable syntactic persistence (explanatory variable 9), which has not yet been scrutinized in a corpus-based study, was operationalized as follows: First, we checked whether there was a verb cluster in the preceding context of the target clause (span: ten preceding clauses). Second, if there was a preceding verb cluster, its structure was determined.7 Based on previous research (cf. Section 2), we hypothesize that the structure of a preceding verb cluster affects the structure of the verb cluster in the target clause. Thus, a preceding participlefirst word order cluster stimulates the use of participle-first word order in the target clause and a preceding participle-final word order cluster stimulates the use of participle-final word order. The only completely new variable, frequency of the participle (explanatory variable 10), was introduced to the research field because previous (psycholinguistic) research has shown that the frequency of a certain lexical item influences language production and comprehension and, consequently, the result of language processes: “Higher frequency of a unit or pattern results in a greater degree of what Langacker terms entrenchment, i.e. cognitive routinization, which affects the processing of a unit” (Kemmer and Barlow 2000: x). We used the frequency information in the Celex database (Baayen, Piepenbrock and van Rijn 1993). Since this is a completely new variable, no hypothesis is formulated.

3.4 Statistical analysis The statistical package R 2.0.1 (2003) is used to conduct the logistic regression analysis. It is important to note that the focus will predominantly lie on the interpretation of these analyses, not on the technical details (cf. Agresti 1996).

4. Logistic regression analysis The present section presents and discusses the results of a logistic regression analysis. This technique enables us to formulate an answer to the following, still unanswered questions: (i) What is the relative impact of each of the ten most cited explanatory variables that were selected on the basis of previous research; which variable has the largest contribution to the determination of the word order; are there any variables which do not significantly contribute to the choice of word order given the effect of all other variables? (ii) What is the explanatory power of the statistical model (i.e. the set of ten explanatory variables)? (iii) To what extent can we predict the word order on the basis of the ten selected explanatory variables?

244

Gert De Sutter

4.1

Results

Table 3 displays the results of the logistic regression analysis in which the choice for participle-final word order (vs. participle-first order) was modeled as a function of the explanatory variables mentioned in Section 3.3 (to avoid multicollinearity, interactions were not included; see the variance inflation factors (VIFs) at the bottom of Table 3). The direction and strength of the relative impact of each of the explanatory variables on the choice of word order in Dutch clause final verb clusters is indicated in terms of odds ratios. An odds ratio > 1 signifies that the given explanatory variable stimulates participle-final word order; an odds ratio < 1 signifies that the explanatory variable stimulates participle-first word order. An odds ratio value farther from 1 (in each direction) indicates a stronger effect. For instance, auxiliary of time (odds ratio = 18.30) has a stronger effect on the choice of word order than passive zijn (odds ratio = 7.82), as 18.30 is farther away from 1 than 7.82. The asterisks behind the odds ratios indicate the level of significance (* significant at p < .05, ** significant at p < .01, *** significant at p < .001). Before we start discussing the results in Table 3, please note that all categorical variables (i.e. all variables except frequency of the participle, which is a continuous variable) are split up according to the different values they can take. One of the values of each variable functions as reference value (the italicized values in Table 3), which means that its effect equals 0 so that it can function as a standard for the evaluation of the other values of that categorical variable. Thus, the non-reference values always have to be evaluated in comparison with the reference value. The categorical variable inherence, for instance, has two values: no fixed expression and fixed expression. The former value is the reference value for this categorical variable (its effect equals 0); the effect of the latter value (2.26) signifies that the odds for participle-final word order is 2.26 times higher when the verb cluster is part of a fixed expression as compared to verb clusters that are not.8 Table 3 reveals that eight out of ten explanatory variables have a statistically significant effect on the choice of word order in Dutch clause final verb clusters. These variables thus have a unique effect which cannot be traced back to the effect of one of the other variables in the statistical model. The significant variables include type of finite verb/categorial status of the participle, morphological structure of the participle, information value of the last preverbal word, inherence of the last preverbal word, length of the middle field, grammatical relationship between the extraposed constituent and its head, syntactic persistence and frequency of the participle (since this is a continuous variable, the odds ratio cannot be computed; instead, the estimate β is given together with

Towards a multivariate model of grammar

245

Table 3. Results of the logistic regression analysis (with reference values in italics) Explanatory variable TYPE OF FINITE VERB / CATEGORIAL STATUS OF THE PARTICIPLE

MORPHOLOGICAL STRUCTURE OF THE PARTICIPLE

LENGTH OF THE MIDDLE FIELD

INFORMATION VALUE

SYNTACTIC PERSISTENCE

INHERENCE GRAMMATICAL RELATION OF EXTRAPOSITION TO HEAD

DISTANCE TO PREVIOUS ACCENTED SYLLABLE

DISTANCE TO FOLLOWING ACCENTED SYLLABLE

Copular zijn Auxiliary of time passive zijn worden Unclassifiable auxiliary Non-separable Separable 0–2 words 3–5 words 6–8 words 9–11 words 12–14 words >14 words Low Intermediate High Participle-first order No previous cluster Participle-final order No fixed expression Fixed expression Adjunct/no extraposition Complement of main verb Complement of preverbal noun > 3 syllables 0–1 syllable 2–3 syllables > 3 syllables 0–1 syllable 2–3 syllables

FREQUENCY OF THE PARTICIPLE N = 2390 Goodness-of-fit χ²: df = 2367, p > .05 Model χ²: df = 22, p < .0001 c = 0.803 VIF < 10

Odds ratio Exp(β β) 18.30 *** 7.82 *** 11.73 *** 4.34 *** 3.87 *** 2.03 *** 2.29 *** 2.29 *** 2.57 ** 1.98 1.41 1.94 *** 1.72 *** 3.28 *** 2.26 *** 0.47 *** 1.21

0.86 1.11 1.00 0.97 β = 2.44E-06 (ASE = 7.74 E-07 ) **

246

Gert De Sutter

theASE). Note, however, that not all values within these variables are significant: complement of a preverbal noun, length of the middle field < 14 words and intermediate information value of the preceding word turn out to be insignificant in the present model. The only two variables that are not significant as a whole are both prosodic variables distance between participial accent and last preverbal accent and distance between participial accent and first postverbal word accent. It might not come as a surprise that prosodic variables do not have an effect on the response variable in a data set that consists of written data only. Nevertheless, this finding refutes the assumption that prosodic characteristics can influence word order in Dutch clause final verb clusters (De Schutter 1996), at least when it comes to verb clusters in Belgian, journalistic Dutch. Future research will have to investigate whether our finding also applies for other types of written material (Netherlandic, non-journalistic). When doing so, researchers must avoid studying the effect of the prosodic variables in isolation, as previous research has already shown that they do have an effect when no other explanatory variables are taken into account (cf. De Sutter, Speelman and Geeraerts 2008). Thus, future research will also need to use multivariate statistics in order to find whether prosody plays a more prominent role in other types of language registers. The explanatory variables that are significant all have the hypothesized effect (cf. Section 3.3): Clusters containing an auxiliary verb of time, passive zijn, worden or an unclassifiable auxiliary verb (compared to zijn as copulative verb) stimulate participle-final word order, just as clusters containing a separable participle (compared to a non-separable participle). Verb clusters that are preceded by a highly informational preverbal word (compared to a low informational word) or by another participle-final word order cluster all make participle-final word order more likely. Also, if the target cluster is not preceded by any other verb cluster, participle-final word order is favored. If the participle is part of a fixed verbal expression (vs. no fixed expression), participle-final word order is chosen more often. Finally, increasing length of the middle field influences the choice of participle-final word order significantly. Only postverbal complements of the main verb (vs. adjunct/no postverbal constituent) stimulate the opposite order. The newly introduced variable frequency of the participle has a positive effect on the choice of participle-final word order, i.e. the higher the frequency of the participle, the more participle-final word order clusters are chosen. The results of the logistic regression analysis lead us to conclude that most of the variables that were introduced and tested in previous monovariate analyses are also relevant in a multivariate analysis, and, hence, that the choice of word order in Dutch clause final verb clusters is simultaneously affected by eight different language-internal variables. More particularly, logistic regression analysis

Towards a multivariate model of grammar

247

indicates that different types of language-internal variables (morphosyntactic, semantic, discursive) simultaneously influence the choice of word order. However, these variables do not affect the choice of word order equally, as the size of the impact they have varies substantially: From the odds ratios in Table 3, it can be derived that the predictor auxiliary of time has the largest impact on the choice of word order, followed by the auxiliary verb worden. More particularly, the odds for participle-final word order is more than eighteen times and eleven times higher when the verb cluster consists of an auxiliary of time and the auxiliary worden, compared to clusters with copular zijn. The following overview provides a complete classification of the relative impact of all significant explanatory variables (in descending order): 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14.

auxiliary of time (o.r. = 18.30) auxiliary worden (o.r. = 11.73) auxiliary zijn (passive) (o.r. = 7.82) unclassifiable auxiliary verb (o.r. = 4.34) separable participle (o.r. = 3.87) syntactic persistence: participle-final order (o.r. = 3.28) length of the middle field: 12–14 words (o.r. = 2.57) length of the middle field: 6–8 words (o.r. = 2.29) length of the middle field: 9–11 words (o.r. = 2.29) inherence: fixed expression (o.r. = 2.26) postverbal constituent: complement of the verb (o.r. = 0.47) length of the middle field: 3–5 words (o.r. = 2.03) informationality: high (o.r. = 1.94) syntactic persistence: none (o.r. = 1.72)

Besides the evaluation of the individual explanatory variables, we can also evaluate the success of the statistical model as a whole, i.e., to what extent can the variation at hand adequately be described and predicted by the set of eight explanatory variables? The model statistics at the bottom of Table 3 show that (i) the statistical model provides a good description of the data (p > .05), (ii) it reduces the degree of unexplained variance significantly (p < .0001), and (iii) it can be used to predict the choice of word order in unseen data (c = 0.803 after 100 bootstrap repetitions).

4.2 Discussion The logistic regression analysis has shown that the choice of word order in Dutch clause final verb clusters is multivariate in nature, as it is simultaneously influ-

248

Gert De Sutter

enced by eight language-internal (morphosyntactic, semantic and discursive) variables. Since most of these variables were already mentioned and tested in previous monovariate analyses (i.e., independently of each other), we are now able to compare and evaluate both approaches (monovariate vs. multivariate). Generally speaking, the majority of the variables that were already introduced in previous research appeared to have a similar effect in both approaches. This does not mean, however, that both approaches are perfectly interchangeable. First of all, what makes the multivariate approach more powerful than the monovariate analysis is that the effects of the variables can be compared directly (the same methodology and analytical techniques were used); as a result, we now know how the variables relate to each other in their effect on the choice of word order (for instance, what is the most powerful variable?). Second, the multivariate analysis has pointed out that not all variables are equally relevant to the explanation and prediction of the word order variation in Dutch verb clusters: Both prosodic variables assumed to be key determinants of word order variation in Dutch verb clusters (cf. De Schutter 1996), even in written language, did not survive the multivariate analysis (i.e., they are redundant). Obviously, a statistical analysis of word order variation can never be the endpoint of a linguistic enquiry. Therefore, one has to take the additional step to interpret the results of the logistic regression analysis in a linguistically relevant way. Why, for instance, do high frequency participles increase the odds for participle-final order? Or why do extraposed complements increase the odds for participle-first order? Given limitations of space, we are not able to fully elaborate on this matter, but we will briefly touch on a possible interpretation which we will illustrate on the basis of the effect of participial frequency (for more information, cf. De Sutter 2005). Recall that the odds for participle-final word order increased with increasing frequency of the participle. If one accepts that the frequency of a given linguistic item is positively correlated with cognitive accessibility and entrenchment in language production and language perception – “Higher frequency of a unit or pattern results in a greater degree of what Langacker terms entrenchment, i.e. cognitive routinization, which affects the processing of a unit” (Kemmer and Barlow 2000: x) – one can hypothesize that participle-final word order is chosen in circumstances where selection and identification of the participle is (relatively) easy. Mutatis mutandis, participle-first word order is chosen in circumstances where selection and identification of the participle is (relatively) difficult. This cognitive interpretation, which obviously has to be considered a hypothesis, can be applied to some of the other variables in the model (e.g., the type of postverbal constituent) so that the following overarching interpretation/ hypothesis can be derived from the results of the logistic regression analysis:

Towards a multivariate model of grammar

249

Participle-first word order is the basic word order that is chosen in circumstances where language users experience processing difficulties in processing and perceiving linguistic utterances; participle-final word order is a sociostylistic option (prestige) that is chosen in circumstances where processing and perception is relatively easy (cf. De Sutter 2005).

5. Conclusions and outlook For the first time in over fifty years of empirical research into word order variation in Dutch clause final verb clusters, the present paper has formulated a linguistically-based and statistically-validated answer to three interrelated questions: (i) Which language-internal variables determine the choice of word order in Dutch clause final verb clusters; (ii) what is the relative impact of each of these variables; and (iii) what is the explanatory and predictive power of this complex of variables? While previous research exclusively conducted monovariate analyses, i.e. testing the effect of one variable on the choice of word order without taking into account the effect of other, possibly interacting variables, the present paper collected the variables that were introduced in previous research, operationalized them so that they could be used in a systematic, objective and generalizable way, introduced a couple of new variables, and entered them in a binary logistic regression model. These analyses have pointed out that the word order variation at hand is simultaneously affected by eight variables, the semantic variable being the most important one (i.e. the status of the participle/auxiliary verb), and that the most specific prosodic variables (i.e. the distance between word accents) which were claimed to be key determinants in previous research do not contribute to the explanation and prediction of the word order variation in formal written Belgian Dutch clause final verb clusters. Moreover, the statistical model with eight explanatory variables appeared to be able to explain and predict the bulk of the variation in the data set. On a methodological level, we have shown that syntactic variation research needs a rigorous quantitative, corpus-based approach in order to disentangle the variables that guide the choice between syntactic alternatives. As has become clear, such an approach does not only involve getting the right data or starting from articulate linguistic hypotheses, but it crucially involves linking data to hypotheses, i.e., operationalizing linguistic hypotheses in ways that make them empirically testable on the available data and statistically analyzing the data by means of sophisticated techniques.

250

Gert De Sutter

On a theoretical level, we have pointed out that a syntactic variation phenomenon is determined by several types of influences (cf. also Grondelaers 2000; Gries 2003; Wulff 2003 and several others) so that our traditional conception of models of grammar must be broadened in order to include lexical, pragmatic, psycholinguistic, sociolinguistic and morphosyntactic variation. What’s more, the influencing variables that emerged from the multivariate analysis are generally in line with results from other multivariate analyses into intraconstituent variation (e.g., Wulff 2003) in that sense that they seem to support the idea that syntactic variation phenomena of the intraconstituent type function in a completely different way than types of interconstituent variation. What binds intraconstituent variation phenomena is that they do not seem to be affected by the typical discursive variables (e.g., topic/focus), whereas most of the interconstituent types do (Wasow 2002; Gries 2003). Obviously, more research is needed in order to clarify this issue and to specify what the common characteristics of each of these types are. Several issues remain to be addressed in future research. In order to increase our understanding of the type of word order variation, the logistic regression model presented here must be complemented in many different ways. First, by including interaction terms and/or, when possible, by including new explanatory variables. Second, by looking at the word order variation from other (statistical) perspectives: loglinear analysis, cluster analysis and correspondence analysis.

Notes ∗

This research was conducted at the University of Leuven and supported by grant G.0086.03 from the Research Foundation Flanders (FWO-Vlaanderen). I would like to thank my supervisors, Dirk Geeraerts and Dirk Speelman (University of Leuven), and one anonymous reviewer for their comments and suggestions. The usual disclaimers apply. 1. Attested examples are followed by the reference to the location in the CONDIV corpus (cf. Section 3; DS = De Standaard). Paradigmatic variants of these corpus examples, such as (1b), are exclusively dealt with in the context of the attested variant. These variants differ from their original in that word order has changed (for illustrative reasons). 2. One notable exception is Swerts and van Wijk (2005). However, since these authors are mainly interested in the effect of prosody on word order, they only explored the relationship between their prosodic variables and added a couple of other nonprosodic variables. Still, a full-scale, systematic multivariate analysis ofall (languageinternal) variables previously mentioned in the literature has not yet been undertaken.

Towards a multivariate model of grammar

251

3. However, note that some researchers argue that the word order variation correlates with a semantic-pragmatic difference (cf. Pardoen 1991). 4. Examples (2), (3b), (4a) and (4b) are extracted from De Schutter (1996: 215); examples (5a) and (6a) are corpus examples (DS), the other examples are paradigmatic variants. 5. However, the identification of the relevant contexts was not always straightforward (e.g., do all words that have the formal architecture of a participle have to be considered as a participle?). Cf. De Sutter (2005) for a discussion of these cases and how they are dealt with in this study. 6. Actually, we only computed statistical attraction if the verbs were preceded by nouns since this yielded the best results (cf. De Sutter 2005). If the verb cluster was preceded by a word from another word class, statistical attraction was considered to be 0 (i.e. no fixed verbal expression). 7. The preceding verb clusters must apply to the following selection criteria: (i) they must at least be bipartite and (ii) the order of the elements is variable. Consequently, verb clusters consisting of three or more verbal elements, verb clusters consisting of infinitives instead of participles and verb clusters in main clauses are allowed. A partitioned chi-square test revealed that all types of preceding verb clusters can be traced back to the two basic types, viz. participle-first and participle-final word order (the classification in two types was done by looking at the position of that participle in relation to the position of the auxiliary verb). 8. This study did not presume a priori that verb clusters containing a copular finite verb can only occur in participle-first word order. If one takes into account that participle-final word order is the prestigious variant (cf. Haeseryn 1990), it is not inconceivable that copular verbs are used in participle-final word order too. The output of the algorithm that is used to disambiguate the status of the finite verb confirms this: 15% of the copular verbs occur in participle-final word order. Examples: (dat) ze erin was geslaagd de rust enigszins te herstellen in Karachi (‘that she succeeded in restoring the quiet in Karachi somewhat’); (dat) opnieuw een meisje was vermist (‘that again a girl was missing’) (DS). Because of this variation, found in constructions with copular verbs, there can be no objection against the use of the value copular verb as the reference value in the regression analysis (thus, the use of this reference value does not distort the results for any of the other values or variables).

References Agresti, Alan 1996

An Introduction to Categorical Data Analysis. New York/Chichester/ Brisbane/Toronto/Singapore: Wiley. Arnold, Jennifer E., Thomas Wasow, Anthony Losongco, and Ryan Ginstrom 2000 Heaviness vs. newness: the effects of structural complexity and discourse status on constituent ordering. Language 76(1): 28–55.

252

Gert De Sutter

Baayen, R. Harald, Richard Piepenbrock, and Hedderik van Rijn 1993 The CELEX Lexical Database. Philadelphia: Linguistic Data Consortium. De Schutter, Georges 1996 De volgorde in tweeledige werkwoordelijke eindgroepen met voltooid deelwoord in spreek- en schrijftaal. Nederlandse taalkunde 1(3): 207– 220. De Sutter, Gert 2005 Rood, groen, corpus! Een taalgebruiksgebaseerde analyse van woordvolgordevariatie in tweeledige werkwoordelijke eindgroepen. Unpublished doctoral dissertation, University of Leuven (http://hdl.handle.net/1979/98). 2007 Naar een corpusgebaseerde, cognitief-functionele verklaring van de woordvolgordevariatie in tweeledige werkwoordelijke eindgroepen. Nederlandse taalkunde 12(4): 302–330. De Sutter, Gert, Dirk Speelman, and Dirk Geeraerts 2005 Regionale en stilistische effecten op de woordvolgorde in werkwoordelijke eindgroepen. Nederlandse taalkunde 10(2): 97–128. 2007 Luisteren schrijvers naar hun innerlijke stem? De invloed van ritmische factoren op de woordvolgorde in werkwoordelijke eindgroepen. Neerlandistiek.nl 07(01 (online publication:) http://www.neerlandistiek.nl/). 2008 Prosodic and syntactic-pragmatic mechanisms of grammatical variation: the impact of a postverbal constituent on the word order in Dutch clause final verb clusters. International Journal of Corpus Linguistics 13(2): 194–224. Dunning, Ted 1993 Accurate methodes for the statistics of surprise and coincidence. Computational Linguistics 19(1): 61–74. Giv´on, Talmy 1988 The pragmatics of word-order: predictability, importance and attention. In Studies in Syntactic Typology, Michael Hammond, Edith A. Moravcsik and Jessica R. Wirth (eds.), 243–284. Amsterdam: John Benjamins. Gries, Stefan Thomas 2003 Multifactorial Analysis in Corpus Linguistics: A Study of Particle Placement. London/New York: Continuum Press. Grondelaers, Stefan 2000 De distributie van niet-anaforisch er buiten de eerste zinsplaats. Sociolexicologische, functionele en psycholingu¨ıstische aspecten van er’s status als presentatief signaal. Unpublished doctoral dissertation, University of Leuven.

Towards a multivariate model of grammar

253

Grondelaers, Stefan, Katrien Deygers, Hilde van Aken, Vicky van den Heede, and Dirk Speelman 2000 Het CONDIV corpus geschreven Nederlands. Nederlandse taalkunde 5(4): 356–363. Haeseryn, Walter J.M. 1990 Syntactische normen in het Nederlands. Een empirisch onderzoek naar volgordevariatie in de werkwoordelijke eindgroep. Doctoral dissertation, University of Nijmegen. Haeseryn, Walter J.M., K. Romijn, G. Geerts, J. de Rooij, and M.C. van den Toorn 1997 Algemene Nederlandse Spraakkunst, 2nd ed. Groningen/Leuven: Wolters/Noordhoff. Hartsuiker, Robert J., and Casper Westenberg 2000 Word order priming in written and spoken sentence production. Cognition 75(2): B27–B39. Hawkins, John A. 1994 A PerformanceTheory of Order and Constituency. (Cambridge Studies in Linguistics.) Cambridge: Cambridge University Press. Kemmer, Suzanne, and Michael Barlow 2000 Introduction. In Usage-based Models of Language, Michael Barlow and Suzanne Kemmer (eds.), vii–xxviii. Stanford: CSLI. Lenz, Barbara 1993 Probleme der Kategorisierung deutscher Partizipien. Zeitschrift f¨ur Sprachwissenschaft 12(1): 39–76. Pardoen, Justine 1991 De interpretatie van zinnen met de rode en de groene volgorde. Forum der letteren 32(1): 1–22. Pauwels, Anita 1953 De plaats van hulpwerkwoord verleden deelwoord en infinitief in de Nederlandse bijzin. Leuven: Symons. R Development Core Team 2003 R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing. [http://www.R-project.org] Speelman, Dirk 1997 Abundantia Verborum. A computer tool for carrying out corpus-based linguistic case studies. Unpublished doctoral dissertation, University of Leuven. [http://wwwling.arts.kuleuven.ac.be/genling/abundant/] Swerts, Marc 1998 Ritme als verklarende variabele voor de keuze tussen groene en rode werkwoordvolgorde in het Nederlands. Nederlandse taalkunde 3(4): 299–308. Swerts, Marc, and Carel van Wijk 2005 Prosodic, lexico-syntactic and regional influences on word order in Dutch verbal endgroups. Journal of Phonetics 33(2): 243–262.

254

Gert De Sutter

Van Dale Lexicografie 2002 Van Dale Groot Woordenboek der Nederlandse Taal. Utrecht/ Antwerpen: Van Dale Lexicografie. [CD-ROM] Wasow, Thomas 2002 Postverbal Behavior. (CSLI Lecture Notes 145.) Stanford, Cal.: CSLI. Wulff, Stefanie 2003 A multifactorial corpus analysis of adjective order in English. International Journal of Corpus Linguistics 8(2): 245–282.

Synchronic variation in diachronic perspective: Question formation in Qu´ebec French∗ Martin Elsig and Shana Poplack Abstract. This paper traces the evolution of French interrogative structure after it was transplanted to Canada, by analyzing the variable expression of yes/no questions over a century and a half of real-time speech. In a radical departure from the current one-variant system of European French, the four original variants continue to divide the labor – linguistic and social – of question formation in Qu´ebec French. Moreover, despite rate changes, the conditioning of the variability has remained stable over the duration. We argue that the variable structure of Qu´ebec French yes/no question formation is a retention and that European French has innovated.

1. Introduction Variability in question formation is a well-documented feature of French syntax. In yes/no questions, five distinct variant forms have been competing for centuries: inversion of (clitic) subject and verb (INV), as in (1), complex inversion (C-INV) (2), rising intonation (INT) (3), phrase-initial interrogative particle est-ce-que (ECQ) (4), and its post-verbal counterpart tu (TU) (5). (1)

As-tu (INV) d´ej`a parl´e avec un vrai Fran¸cais de France l`a? ‘Have you ever spoken to a real Frenchman from France?’ (XX.105.2768)1

(2)

Et le roi est-il (C-INV) icitte? ‘And the king, is he here?’ (XIX.036.3932)

(3)

Ah, toi tu restes pas (INT) avec tes parents? (XX.112.1819) ‘Oh, you don’t live with your parents?’

(4)

Mes bombes est-ce que (ECQ) je les largue ici? ‘My bombs, do I throw them here?’ (XX.078.1502)

256

Martin Elsig and Shana Poplack

(5)

Tu vas-tu (TU) eˆ tre plus mari´e oubedonc moins mari´e? ‘Are you gonna be more married or less married?’ (XX.079.1471)

Empirical studies of European varieties report that the variability illustrated in (1) to (5) has resolved itself in favor of INT (3), with ECQ persisting as a minor contender. INV, once the quintessential interrogative marker, is now restricted to literary use. The spread of TU is said to have been blocked by stigma (Behnstedt 1973: 32; Foulet 1921: 271–272), while C-INV has disappeared altogether from speech. In Qu´ebec French, on the other hand, not only are the “extinct” variants thriving, but as we shall see, they each fulfill a well-defined function. As a result, the Canadian system of question formation appears structurally more complex than that of its source, the opposite of what is expected of transplanted dialects (e.g. Britain 2004). In this paper, we investigate the continuing evolution of question formation by tracing the variable expression of yes/no questions before and after the French settlement of Qu´ebec during the seventeenth century.

2. Data and Method The corpora on which our analyses are based, displayed in Table 1, are particularly well-suited to this endeavor. Table 1. Data sources Century Source

Time span

20

Corpus du fran¸cais parl´e a` Ottawa-Hull (Poplack 1989)

speakers born 1898–1965

19

R´ecits du fran¸cais qu´eb´ecois d’autrefois (Poplack and St-Amand 2007)

speakers born 1846–1895

17

Seventeenth-century popular French plays

1629–1663

16–20

R´epertoire historique des grammaires du fran¸cais (Poplack, Jarmasz, Dion and Rosen ms.)

1530–1998

Two represent vernaculars spoken in Qu´ebec in the nineteenth and twentieth centuries. A corpus of seventeenth-century popular plays and a compilation of normative grammars dating from 1530 to the present are complementary diachronic sources representing a benchmark before the language was trans-

Synchronic variation in diachronic perspective

257

planted. We make use of the prescriptive tradition to date the variant forms, assess their institutional acceptance, and most important, ascertain the factors conditioning their selection (Poplack, Jarmasz, Dion and Rosen 2002; Poplack and Dion 2004). Real-time analysis spanning several centuries will help pinpoint the locus and time of change, if any.

3. Variable Context From each of the usage corpora we extracted every non-rhetorical question requiring a yes/no answer, noting the variant selected for each, excluding all others (e.g. wh-questions, as in (6), fixed expressions (7), non-sentential questions (8), imperatives (9), rhetorical questions (10), echo questions (11), and interrogative tags (12)). (6)

Il dit “Pourquoi tu as pas tu´e l’ours blanc?” ‘He says “Why didn’t you kill the white bear?” ’ (XIX.036.2670)

(7)

Hey, c¸ a fait longtemps, tu sais? ‘Hey, it’s been a long time, you know?’ (XX.096.79)

(8)

Les pattes rondes c¸ a? ‘The round paws, there?’ (XIX.043.2452)

(9)

Allons- allons, disons au roi que le prince est pas mort? ‘Let’s go, shall we tell the king that the prince isn’t dead?’ (XIX.038.2478)

(10)

Tu as peut-ˆetre d´ej`a vu c¸ a ces charti`eres l`a ou entendu parler[. . . ]? ‘You may have already seen those window bars, or heard of them [. . . ]?’ (XIX.018.820)

(11)

“Ah moi j’aime pas c¸ a”. “Tu aimes pas c¸ a?” ‘ “Ah, I don’t like that”. “You don’t like that?” ’ (XX.103.398)

(12)

Ils grasseyent eux-autres a` Montr´eal, hein? ‘They have a guttural R in Montreal, eh?’ (XX.089.1725)

Table 2 compares variant distribution in contemporary Qu´ebec and European French.

258

Martin Elsig and Shana Poplack

Table 2. Distribution of variants in Qu´ebec French and European French

Intonation -tu Inversion Est-ce que Complexinversion Total

Qu´ebec French European French Pohl Terry This Fox 1965 1970 study 1989 % % % % 86 86 35 36 0 – 33 34 0 11 26 29 14 3 6 1 – – – –

Ashby 1977 % 80 – 9 11 –

S¨oll 1982 % 91 – 1 8 –

Coveney 2002 % 79 – – 21 –

776

130

452

180

871

816

3016

INT is basically the only viable variant in Europe; in Canada, there are three. Why should speakers select one rather than another? Most scholars who have addressed this issue believe the variants differ pragmatically, conveying nuances like emphasis, doubt, astonishment, weak curiosity, etc. (e.g. Pohl 1965; Behnstedt 1973; Coveney 2002). We attempted to operationalize and test these claims on our data. Reliability tests showed that inter-judge consensus for most interrogative tokens could not be achieved. It was often impossible to determine, for example, whether a given question was a request for information, a request for opinion, or a check on the knowledge of the addressee (Coveney 2002: 128), let alone to identify some of the more subtle readings listed above.2 Such inaccessibility to speakers’ intentions is antithetical to accountable quantitative methodology, since this requires that the set of alternative explanatory factors be exhaustive and consistently identifiable for all or almost all tokens. We have no doubt that questions may admit the variety of readings scholars have attributed to them, and that these pragmatic effects can be both conveyed and interpreted by speakers. However, no one has yet succeeded in demonstrating empirically that any of the distinct readings are correlated with variant choice in discourse. Accordingly, in this study we focus on the role of linguistic factors (e.g. subject type, and frequency, form, semantics, syllable length and lexical identity of the verb, as well as polarity). Most of these, along with speech style, have long been invoked as explanatory of choice of interrogative variant (e.g. Pohl 1965; Terry 1970; Behnstedt 1973; Ashby 1977; S¨oll 1982; Dewaele 1999; Coveney 2002). We analyze their combined effect using Goldvarb 2001 (Rand and Sankoff 1990), which enables us to contextualize the role of the variants within the system, with a view to elucidating its evolution over time.

Synchronic variation in diachronic perspective

259

4. Results

%

Turning first to contemporary Canadian French, a first important finding (Figure 1) is that negative questions are virtually only expressed by INT, as in (3). Ensuing analyses thus deal only with affirmative questions. 10 0 90 80 70 60 50 40 30 20 10 0

Intonation

Inversion

Negative

TU

Est-ce que

Affirmative

Figure 1. Distribution of variants according to polarity

Table 3 displays four independent variable rule analyses of the other factors selected as significant to variant choice.3 The table shows that the major task of INV is to form questions directed towards an interlocutor and hence featuring a second person pronoun, as in (13), while questions involving other subjects tend to be formed with the interrogative particle TU, as in (14). (13)

Bien j’ai dit, es-tu (INV) fou toi? ‘So I said: “Are you nuts?” ’ (XX.112.1980)

(14)

Bien j’avais-tu (TU) de l’air niaiseuse? ‘Well, did I look silly?’ (XX.117.2122)

Other contexts favoring INV, albeit to a lesser degree, include verbs of cognition, as in (15), and polysyllabic verbs (16). Here INV is complementary to TU, which in turn behaves like a default variant (cf. also Elsig 2009). We also note that verb frequency, claimed by many (e.g. Pohl 1965; Behnstedt 1973; Ashby 1977; Dewaele 1999) to explain why INV persists at all, has no effect. (Nor does lexical identity, though not shown here).

260

Martin Elsig and Shana Poplack

Table 3. Variable rule analyses of factors selected as significant to variant choice: twentieth century (affirmative tokens only; brackets indicate nonsignificant factors) Variant Total N Corrected mean: Subject type 2nd person Others Verb form Synthetic Periphrastic Verb semantics “Cognitive” Others Verb syllables Polysyllabic Monosyllabic Verb frequency Frequent (21+) Mid (6–20) Rare (1–5) Style Careful Casual

Inversion 205 .30

-tu 247 .36

Est-ce que 50 .07

Intonation 186 .27

100% 0%

.36 .70

.37 .69

[] []

.47 .64

[] []

[] []

[] []

.77 .36

.43 .54

– 100%

[] []

.64 .45

.33 .57

.63 .45

[] []

[] [] []

[] [] []

[] [] []

[] [] []

.44 .61

.46 .58

.65 .24

.52 .45

(15)

Maintenant, comprenez-vous (INV) que pour le min´eral je peux le localiser maintenant? ‘Now, do you understand that for the mineral, I can locate it now?’ (XX.082.2979)

(16)

Mangeriez-vous (INV) votre p`ere? ‘Would you eat your father?’ (XIX.004.1654)

The variants also have strong stylistic connotations, and these too are the opposite of those reported for Europe: In Canada, the rare ECQ (17), and to a lesser extent, INT (18), denote formality, while the other variants are relegated to casual speech, cf. (19) and (20). (17)

Excusez, est-ce que (ECQ) je peux le regarder? ‘Excuse me, may I see it?’ (XX.091.270)

Synchronic variation in diachronic perspective

(18)

Vous eˆ tes correcte comme c¸ a? (INT) ‘Is everything all right this way?’ (XX.119.2053)

(19)

Penses-tu (INV) que j’´etais faite comme un bœuf? ‘Do you think I was built like an ox?’ (XIX.009.1478)

(20)

Ouais, c¸ a a-tu (TU) du sacre bon sens? ‘Yeah, does that make any goddamn sense?’ (XX.84.1867)

261

We may summarize the main functions of the interrogative markers in twentieth century Qu´ebec French as follows: (i) negative polarity questions are expressed with INT, (ii) INV is specialized for direct questions, (iii) ECQ is a hyperstyle marker, and (iv) TU assumes most of the remaining (nonspecialized) work of question formation. This pattern is substantively different from what is reported for European French (Table 2), where INT is the default variant, with only a little support from ECQ, while TU, a frontrunner in Qu´ebec French, is practically nonexistent. How did the Canadian system arise?

4.1 Development of the French interrogative system A brief review of the historical record reveals that all of the variants have a long and venerable history, linked to the Middle French shift from verb–subject to subject–verb word order and ensuing efforts to rout out remaining inversions. This paved the way for the incursion of other forms, all of which had the virtue of reestablishing the desired subject–verb order. As far back as Middle French, INV co-existed with C-INV (Foulet 1921; Roberts 1993), which provided a solution to the problem of question formation with nominal subjects, since it contains both subject–verb and verb–subject word order. TU is widely believed to have originated through reanalysis of C-INV (cf. Picard 1992). But for reasons that are still not entirely clear, it was never accepted in polite discourse. Eventually, on analogy with the qu’est-ce que ‘what is it that’ paradigm for wh-questions, the construction ECQ emerged, grammaticizing to interrogative particle by the sixteenth century. In contrast to TU, this variant came to be explicitly ratified by the Acad´emie Fran¸caise, especially in contexts involving first conjugation and monosyllabic verbs where inversion was deplored (Vaugelas 1647). INT, though perhaps the oldest (and currently the majority) variant in France, was never viewed as a serious contender.

262

Martin Elsig and Shana Poplack

Summarizing, according to historical accounts, the prototypical interrogative variant, INV, was ousted due to loss of the verb–subject word order it instantiates. It is said to have disappeared first from the most salient or problematic contexts (e.g. subject noun phrases, first person pronouns, first conjugation monosyllabic verbs), and persisted where it was most entrenched: frequent verbs and collocations (Pohl 1965; Behnstedt 1973; Dewaele 1999; Coveney 2002). Ensuing analyses test these hypotheses. INV’s competitors, all of which reinstated the desirable subject–verb order, infiltrated the system to varying degrees, due, at least in part, to differences in institutional and social acceptance.

4.2

Question formation in seventeenth-century French

Having reviewed the reported state of the language at the time of the French colonization of Qu´ebec in the seventeenth century, we now investigate the extent to which the scenario outlined above captured contemporaneous usage, as instantiated in the works of Corneille, Moli`ere and Richer, popular playwrights of the time. We make no claims about the extent to which rates of variant use reflect the speech of either actors or audience. But the structure of their variable selection, as revealed by the constraint hierarchies associated with each, can yield valuable information on their trajectory of development. This will be our focus in this section. Table 4 displays two independent variable rule analyses of the contribution of factors to variant choice in seventeenth-century plays. The corrected means indicate that INV, mostly pronominal, was still by far the majority variant of the time; the rest is basically made up of INT. We first note that subject type, the most important predictor of INV in twentieth-century French, was not significant in the seventeenth century. This is because at the time, INV still occurred with all pronominal subjects, including the prescriptively undesirable je ‘I’, a full 82% of which were inverted. This is illustrated in (20). (20)

Vous offens´e-je (INV) en parlant de la sorte? ‘Do I offend you by speaking this way?’ (XVII.M.154.781)

Second person questions favored INT, and questions involving subject noun phrases were formed with C-INV or, to a lesser extent, INT. At this stage INV in yes/no questions was already vanishingly rare with noun phrases, but still occurred freely with all pronominal subjects. This is consistent with the observed frequency effect: INV was already favored in frequent verbs, foreshadowing

Synchronic variation in diachronic perspective

263

Table 4. Variable rule analyses of factors selected as significant to variant choice: seventeenth century (affirmative tokens only) Variant Total N Corrected mean: Subject type 2nd person Others Verb frequency Frequent (21+) Mid (6–20) Rare (1–5) Verb form Synthetic Periphrastic Verb semantics “Cognitive” Others Verb syllables Polysyllabic Monosyllabic Style Careful Casual

Inversion

Intonation

Pronominal 502 .65

Complex 57 .07

205 .27

[] []

– 100%

.59 .42

.56 .42 .38

[] [] []

.46 .52 .60

.51 .40

[] []

.48 .64

.58 .45

.34 .61

.43 .55

[] []

.39 .58

[] []

[] []

[] []

[] []

its eventual recession. The twentieth-century tendency for INV to occur with “cognitive” verbs was already in place. Although the Acad´emie prescribed ECQ to replace INV in first person subjects, especially with monosyllabic verbs, the form was considered too colloquial throughout the seventeenth century to be admitted to writing (Foulet 1921). This may explain why we found so few of them (N = 14) in the plays. In this context, it is perhaps not surprising that number of syllables was not a significant predictor of variant choice. But we can already detect a clear preference for questions involving monosyllabic verbs to occur with another variant: C-INV. Summarizing, in seventeenth-century (approximations of) speech, INV was still used productively with a variety of subject pronouns and verbs, though there were already harbingers of the current situation, since questions involving noun phrases, monosyllabic and infrequent verbs tended not to be inverted, all testifying to the gradual restriction of this variant.

264

Martin Elsig and Shana Poplack

4.3 The trajectory of variant choice over three centuries Table 5 summarizes the trajectory of variant choice over the three centuries we studied, as illustrated by the relative contributions to their selection over the duration. Despite rate changes (observed in the corrected means), in all cases but one, the hierarchy of linguistic constraints is the same from the nineteenth to the twentieth centuries. This suggests that any changes to the system must have predated this time. Even the clear stylistic associations noted earlier for contemporary French were firmly in place by the nineteenth century: INV – literary for Europeans – was already favored in casual contexts, as was, more predictably, TU. INT, and especially ECQ (both colloquial in European French) had become formal variants. But from a linguistic perspective, the system has remained essentially unchanged. For example, although C-INV disappeared, its associated conditioning was transferred to TU. Constraints on TU are now largely mirror-images of those on INV, consistent with its emerging role as majority variant: Where INV is favored, TU is disfavored. Even the differences in the contributions of subject type and verb frequency seem to be apparent only. Recall that in seventeenth-century France, INV could still occur freely across the entire pronominal paradigm, explaining why subject type was not selected as significant. Two centuries later, INV was highly promoted by second person pronouns, a constraint which has now become categorical. But despite its productivity with pronominal subjects, INV was receding from rarer contexts. So even though frequency appears from Table 5 to have had no effect on variant selection since the nineteenth century, we suggest that this is what in fact underlies the retention of INV. But why should INV be restricted to second person subjects? Figure 2 shows that these have always accounted for more questions than any other grammatical person. We can assume that the French brought to Canada also contained a disproportionate number of such contexts, explaining the current persistence of INV here. The only real departure from seventeenth-century French involves negative polarity: it was strongly associated with INT by the nineteenth century in Qu´ebec. There is no evidence of this effect in the earlier materials.

5. Discussion We may now return to the initially puzzling distributions displayed in Table 1. The Qu´ebec interrogative system, for all its differences from its contempo-

Synchronic variation in diachronic perspective

265

Table 5. Variable rule analyses of factors selected as significant to variant choice: seventeenth, nineteenth and twentieth centuries (affirmative tokens only) Variant Century Total N Corrected mean Subject type 2nd person Others Verb frequency Frequent (21+) Mid (6–20) Rare (1–5) Verb form Synthetic Periphrastic Verb semantics “Cognitive” Others Verb syllables Polysyllabic Monosyllabic Style Careful Casual

Inversion Pronominal Complex 17 19 20 17 19 502 289 205 57 2 .65 .47 .30 .07 –

-tu

Est-ce que

Intonation

20 0 –

17 0 –

19 20 17 157 247 10 .25 .36 .01

19 3 –

20 50 .07

17 19 20 205 171 186 .27 .28 .27

[] []

.77 .06

100% – – 0% 100% –

– –

– –

.31 .88

.36 .70

– –

– –

.37 .69

.59 .42

[] []

[] []

.56 .42 .38

[] [] []

[] [] []

[] [] []

– – –

– – –

– – –

[] [] []

[] [] []

– – –

– – –

[] [] []

.46 .52 .60

[] [] []

[] [] []

.51 .40

.53 .37

.47 .64

[] []

– –

– –

– –

[] []

[] []

– –

– –

[] []

.48 .64

[] []

[] []

.58 .45

.64 .44

.77 .36

.34 .61

– –

– –

– –

[] []

.43 .54

– –

– –

– .43 100% .55

.38 .56

[] []

[] []

[] []

.64 .45

.39 .58

– –

– –

– –

[] []

.33 .57

– –

– –

.63 .45

[] []

.42 .54

[] []

[] []

.27 .52

.44 .61

[] []

– –

– –

– –

.19 .54

.46 .58

– –

– –

.65 .24

[] []

.80 .47

.52 .45

%

100 90 80 70 60 50 40 30 20 10 0 17th

19th 2nd person Others

20th

Figure 2. Distribution of questions according to grammatical person of addressee

rary European counterparts, appears to be a direct descendant of the one in use in seventeenth-century France at the time of colonization. That system still prominently featured INV. Its total demise was blocked, not by entrenchment in frequent verbs or collocations, but rather, by the preponderance of its favored contexts of occurrence: direct questions. C-INV, already moribund in the seventeenth century, disappeared altogether, but not before it transferred its functions

266

Martin Elsig and Shana Poplack

to its alter ego TU. We know that TU (in its variant form [ti]) was (and is) widespread in many regions of France, though heavy institutional stigma curtailed its use. That stigma either failed to accompany the form to Canada, or disappeared thereafter. Table 6 confirms that not only is TU clearly on the increase among the young, but it is also favored by women. ECQ, still rare in our seventeenth-century materials, failed to gain a true foothold in Qu´ebec. This paved the way for its eventual conversion into a hyper-formal, upper-class variant, never used by speakers under 35. INT, interrogative marker par excellence in European French, somehow associated itself with negative questions. This is the only development for which we have found no precursor in the history of the language, though it is also attested in contemporary European varieties (Terry 1970; S¨oll 1971; Hansen 2001; Coveney 2002: 212–213 citing Borillo 1979). Table 6. Variable rule analysis of extralinguistic factors selected as significant to variant choice: twentieth century (affirmative questions only) Variant Total N Corrected mean: Sex Female Male Age 35+ 15–34 Socioeconomic class Upper middle Working

Inversion 205 .30

-tu 247 .36

Intonation 186 .27

Est-ce que 50 .07

[] []

.55 .45

[] []

[] []

.56 .37

.42 .60

[] []

100% –

[] []

[] []

[] []

.85 .16

The distribution of these same interrogative variants in Europe is considered so compartmentalized that their variable selection is said to be achieved through code-switching. INV – pronominal and complex – is a feature of Standard French grammar, while INT and ECQ are restricted to “fran¸cais populaire” (de Wind 1995; Kaiser 1996; Bonnesen and Meisel 2005). For Canadian French, no such analysis is required, since all but one of the variants continue to be implicated in question formation. Moreover, our analysis suggests that the contemporary Canadian interrogative system is a rather faithful representation of the system brought over from France. Aside from some rate differences involving the two major contenders, little has changed since the seventeenth century. The spectacular changes have taken place in France, for reasons no one has yet elucidated.

Synchronic variation in diachronic perspective

267

Given the remarkable continuity of the linguistic conditioning of question formation, we may surmise that, as in Canada, it is the social embedding which has driven the change.

Notes ∗

The research reported here is part of a larger project entitled ‘Confronting prescription and praxis in the evolution of grammar’, supported by Social Sciences and Humanities Research Council and Killam Foundation grants to Poplack. Poplack holds a Canada Research Chair in Linguistics. Elsig’s participation in this project was generously funded by the Deutsche Forschungsgemeinschaft as part of the research project ‘Multilingualism as cause and effect of language change: Historical syntax of Romance languages’, directed by J¨urgen M. Meisel and Esther Rinke. It is one of currently nineteen projects within the Collaborative Research Center on Multilingualism established at the University of Hamburg. Parts of this paper appeared as: Elsig, Martin, and Shana Poplack. 2006. Transplanted dialects and language change: question formation in Qu´ebec. University of Pennsylvania Working Papers in Linguistics 12 (2): 77–90. 1. Codes refer to corpus (XIX =R´ecits du fran¸cais qu´eb´ecois d’autrefois (Poplack and St-Amand 2007); XX =Corpus du fran¸cais parl´e a` Ottawa-Hull (Poplack 1989); XVII =Corpus of 17th-century popular French plays), speaker and line number. Examples are reproduced verbatim from audio recordings or plays. 2. This problem is compounded by the fact that a single token may assume more than one function (e.g. Hansen 2001: 480). 3. Potential interactions (amongst e.g. grammatical person, number of syllables, frequency) have been controlled for; factor effects reported here and in ensuing tables are independent.

References Ashby, William 1977 Interrogative forms in Parisian French. Semasia 4: 35–52. Behnstedt, Peter 1973 Viens-tu? Est-ce que tu viens? Tu viens? Formen und Strukturen des direkten Fragesatzes im Franz¨osischen. T¨ubingen: Narr. Bonnesen, Matthias, and J¨urgen M. Meisel 2005 Die “Subjekt–Verb Inversion” in Interrogativkonstruktionen des gesprochenen Franz¨osischen: Zum Problem der syntaktischen Variation. In Deutsche Romanistik – generativ, Georg A. Kaiser (ed.), 31–48. T¨ubingen: Narr.

268

Martin Elsig and Shana Poplack

Borillo, Andr´ee 1979 Britain, David 2004

Coveney, Aidan 2002

La n´egation et l’orientation de la demande de confirmation. Langue Fran¸caise 44: 27–41. Geolinguistics – diffusion of language. In Sociolinguistics/Soziolinguistik. An International Handbook of the Science of Language and Society/Ein internationales Handbuch zur Wissenschaft von Sprache und Gesellschaft, 2nd ed., Ulrich Ammon, Norbert Dittmar, Klaus J. Mattheier, and Peter Trudgill (eds.), 34–48. Berlin: Mouton de Gruyter.

Variability in Spoken French. A Sociolinguistic Study of Interrogation and Negation. Bristol/Portland: Elm Bank. De Wind, Maarten 1995 Inversion in French. Ph.D. dissertation, University of Groningen. Dewaele, Jean-Marc 1999 Word order variation in French interrogative structures. ITL Review of Applied Linguistics 125–126: 161–180. Elsig, Martin 2009 Grammatical Variation across Space and Time: the French Interrogative System. Amsterdam/Philadelphia: John Benjamins. Foulet, Lucien 1921 Comment ont e´ volu´e les formes de l’interrogation. Romania 47: 243– 348. Fox, Cynthia 1989 Syntactic variation and interrogative structures in Qu´eb´ecois. Ph.D. dissertation, Indiana University at Bloomington. Hansen, Maj-Britt Mosegaard 2001 Syntax in interaction. Form and function of yes/no interrogatives in spoken standard French. Studies in Language 25(3): 463–520. Kaiser, Georg A. 1996 V2 or not V2? Subject–verb inversion in Old and Modern French interrogatives. In Language Change and Generative Grammar, Ellen Brandner and Gisella Ferraresi (eds.), 168–190. Opladen: Westdeutscher Verlag. Picard, Marc 1992 Aspects synchroniques et diachroniques du tu interrogatif en qu´eb´ecois. Revue qu´eb´ecoise de linguistique 21(2): 65–74. Pohl, Jacques 1965 Observations sur les formes d’interrogation dans la langue parl´ee et dans la langue e´ crite non-litt´eraire. In Linguistique et philologie romanes. Actes du Xe Congr`es International de Linguistique et Philologie Romanes, Georges Straka (ed.), vol. 1: 501–513. Paris: Klincksieck.

Synchronic variation in diachronic perspective Poplack, Shana 1989

269

The care and handling of a mega-corpus: the Ottawa-Hull French project. In Language Change andVariation, Ralph W. Fasold and Deborah Schiffrin (eds.), 411–451. (Current Issues in Linguistic Theory 52.) Amsterdam: John Benjamins. Poplack, Shana, and Nathalie Dion 2004 The French future in grammar and speech. Presentation at NWAV 33. Ann Arbor: University of Michigan. Poplack, Shana, Lidia-Gabriela Jarmasz, Nathalie Dion, and Nicole Rosen 2002 The evolution of French prescriptive discourse: constructing the R´epertoire historique des grammaires du fran¸cais. Technical Report. Sociolinguistics Laboratory. University of Ottawa. Poplack, Shana, and Anne St-Amand 2007 A real-time window on 19th century vernacular French: the R´ecits du fran¸cais qu´eb´ecois d’autrefois. Language in Society 36(5): 707–734. Rand, David, and David Sankoff 1990 GoldVarb. Version 2. A variable rule application for the Macintosh. Centre de recherches math´ematiques. Montr´eal: Universit´e de Montr´eal. Roberts, Ian 1993 Verbs and Diachronic Syntax. A Comparative History of English and French. (Studies in Natural Language and Linguistic Theory 28.) Dordrecht: Kluwer. S¨oll, Ludwig ´ 1982 L’interrogation directe dans un corpus de langage enfantin. In Etudes de grammaire fran¸caise descriptive, Franz Josef Hausmann (ed.), 45– 54. Heidelberg: Groos. Terry, Robert ´ 1970 Contemporary French Interrogative Structures. Montr´eal: Editions Cosmos. Vaugelas, Claude Favre de 1647 Remarques sur la Langue Fran¸coise. (Nouvelle e´ dition en deux volumes de 1880 comprenant le texte de l’´edition originale, des remarques in´edites, une clef in´edite de Conrart, tous les commentaires du XVIIe si`ecle, des notes nouvelles, une introduction et une table analytique des mati`eres.) Paris: A. Chassang.

Agreement in English dialects Verena Haser and Bernd Kortmann Abstract. This paper deals with variation in agreement systems in English dialects. We will argue that a usage-based constructionist approach offers a compelling explanation why certain theoretically possible agreement systems noted in previous resarch are not attested. Our corpus study attempts to apply insights of usage-based constructionist models to the study of “pronoun + past BE” combinations in several British dialects, investigating frequency distributions for was- and were-generalization for different pronouns. The regular distributions patterns observed can be easily accounted for within a usage-based framework. But even the occasional (from a usage-based view) unexpected frequency distributions seem to lend themselves to a constructionist account, though not necessarily of the usage-based type. From a methodological point of view, we will make a case for neatly distinguishing between “he + past BE” and “she + past BE” as separately stored construction schemata. Despite the fact that both he and she belong to the same syntactic category of third person singular pronouns, they do not constitute a uniform context for were-generalization.

1. Outline of the present study Having suffered a long period of relative neglect, the study of morphosyntactic variation has come into its own in recent years, due to a growing awareness among both functionalists and formalists of its relevance to syntactic theorizing. There are essentially two ways in which the study of morphosyntactic variation can pose a challenge to syntactic theories. First, specific constructions found in dialects but absent in standard languages are frequently left unaccounted for in mainstream frameworks. Second, the fact that a single speaker may possess and use a rich array of alternative constructions expressing the same meaning presents problems to some well-known syntactic models. Agreement constructions in different varieties of English present a challenge to syntactic theories in both of these respects. We hope to show that a constructionist approach is well suited to modeling syntactic variation, both explaining apparently elusive agreement structures occurring in dialects and accounting for intra-speaker/intra-dialect variation. Our main emphasis will be on performance- (or usage-)based models of construction grammar. The first

272

Verena Haser and Bernd Kortmann

part of our paper will review some previous work on this topic, tracing its major implications for syntactic theories. We will have a look at the so-called Northern Subject Rule, an agreement pattern attested in Northern varieties of Britain and Ireland that lends itself to a constructionist account. Subsequent sections will apply basic principles of construction grammar to other kinds of agreement patterns found in British dialects. At the heart of this study is an investigation of combinations of pronouns and past tense forms of BE in English dialects.

2. Usage-based models of construction grammar Construction grammar is one of the most important syntactic theories to have emerged in relatively recent years. The general theoretical background to constructionist approaches is expounded in several recent works, including Goldberg (1995; 2006) and Croft (2001), and will not be rehearsed in detail here. The key assumption shared by all construction grammarians is that linguistic knowledge is organized in terms of constructions, with constructions being defined as follows: “Any linguistic pattern is recognized as a construction as long as some aspect of its form or function is not strictly predictable from its component parts or from other constructions recognized to exist” (Goldberg 2006: 6). As will be seen below, some construction grammarians assume that even constructions that are predictable from a more general construction schema may be represented independently in speakers’ minds. Another pivotal constructionist assumption is that constructions may differ significantly not only in terms of their size – for example, both simple words and idioms are recognized as constructions – but also in terms of their level of abstraction. Constructions may be as abstract as the rule for passivization in English [Subj aux VPpp (PPby ); e.g., My cat was chased by John’s dog]; but we also find much more specific constructions (e.g., specific idioms). Constructions can thus be arranged in construction hierarchies, with very general constructions at the top, followed by several levels of increasingly specific constructions that can be derived from the respective abstract construction schemata. An example of such a hierarchy for Standard English might include, among other things, the Subject-Verb schema as the most abstract schema, which subsumes the two more specific schemata “Third person singular subject + verb + -s” and “Non-third person singular subject + verb + zero-inflection”. He goes and we go, respectively, represent two instances of the latter two schemata, which in turn function as mother nodes of even more specific schemata such as “he + verb + -s”, “I + verb + zero-inflection”, etc. (instantiated respectively by he

Agreement in English dialects

273

drinks and I think, for example). As will be seen shortly, such a hierarchy might at the lowest level include highly specific constructions like I was or he has. One major issue that divides different versions of construction grammar relates to the question whether a constructionist approach should take into consideration performance factors such as discourse frequency. We will see that a framework that does take these factors into account may indeed explain facts which are left unexplained – or which are at least not as easily explained – in a competence-based model. Competence-based versions of construction grammar have been developed by Fillmore and Kay (1993), while Croft (2001) or Goldberg (2006) are important proponents of usage-based models. As the term “usage-based” already implies, a usage-based model of syntactic representation assumes that the way a construction is used in everyday language has implications for how this construction is mentally represented. Below we will give examples how such a model can be applied to the so-called Northern Subject Rule. In a usage-based model, a pivotal feature of constructions is their frequency of use in discourse, also called token frequency. A construction’s token frequency has an effect on whether this form is independently stored in a speaker’s mind. The hypothesis is that each time a word (or construction) is used, it activates a node or pattern of nodes in the mind, and frequency of activation affects the storage of that information, leading to its ultimate storage as a conventional grammatical unit. (Croft and Cruse 2004: 293)

An item which is activated frequently enough to be stored as an independent unit in speakers’ minds is said to be entrenched. A crucial feature of usagebased approaches is that specific units can be stored independently even if they can be subsumed under a general schema which is already present in speakers’ minds (cf. Croft and Cruse 2004: 293–294). For example, the Standard English construction schema “I + verb + zero-inflection” (e.g., I sing) is likely to be stored separately from the more general construction schema “subject + verb”, even though the more specific schema can be derived from the more general schema. The reason for postulating a separate mental representation for the more specific schema lies in the high discourse frequency of pronouns – and hence of pronoun–verb combinations. As a result, these constructions are strongly entrenched and thus represented separately from the more general construction schemata under which they can be subsumed. A separate mental representation will also be postulated for even more specific constructions that occur extremely frequently in discourse, such as I was or they were.

274

Verena Haser and Bernd Kortmann

3. Applying the usage-based model: The Northern Subject Rule Pietsch (2005) demonstrates the advantages of a constructionist approach for the study of specific agreement patterns in Northern varieties of Britain and Ireland.1 More specifically, the author investigates the so-called “Northern Subject Rule”2 , which can roughly be stated as follows (cf. Pietsch 2005: chapter 1): All present tense verbs may have the ending -s, unless their subjects are personal pronouns which are immediately adjacent to the verbs (third person singular verbs always have final -s, as in Standard English). (cf. Pietsch 2005: 5–6)

Within the Northern Subject Rule, a distinction is usually made between a typeof-subject constraint and a position-of-subject constraint. The type-of-subject constraint basically says that the verbal ending -s does not occur after a particular type of subject, viz. pronouns (except the third-person singular pronouns). The position-of-subject-constraint specifies that verbal -s does not occur in certain positions, specifically it is absent in those cases where the subject and verb are directly adjacent. Examples illustrating the position-of-subject constraint are We cry and laughs; We always drives. As Pietsch (2005: chapter 2) argues at length, pertinent studies by scholars working in formalist traditions have not offered a satisfactory and unified account of the two constraints involved in the Northern Subject Rule. We will see that a usage-based constructionist approach seems more promising when it comes to capturing the complexity of this rule as well as the variation patterns observed in actual discourse. The greatest challenge faced by scholars investigating the Northern Subject rule is to explain the absence of the verbal inflection -s in the cases specified by the type-of-subject constraint and the position-of-subject constraint. Furthermore, there is considerable variation with regard to the occurrence of -s in cases that are not covered by these constraints. How should one explain, for example, why the likelihood of verbal -s in constructions involving pronoun plus verb increases in proportion to the increase in distance between subject pronoun and verb? Thus, a construction such as We always drive/s is more likely to occur with verbal -s than We drive, but it is less likely to occur with verbal -s than We almost never drive/s. The key to explaining this fact lies in the notion of entrenchment. As already noted, this concept relates to the fact that a mental representation of an item or a construction is reinforced in proportion to the frequency with which this representation is activated. As a result of entrenchment, constructions that are

Agreement in English dialects

275

very frequently used become established as separate mental representations that are to some extent independent of the more general schemata which they exemplify. How can the notion of entrenchment be invoked as a crucial factor accounting for the above-noted facts? The answer is as follows. Pietsch (2005) assumes the existence of two construction schemata relevant to explaining the variation observed. First, there is an abstract schema relating to combinations of subject and verbs in general. This general schema is associated with verbal -s (“subject + verb + -s”). In addition to the general schema, however, more specific schemata involving combinations of subject pronouns and verbs have to be posited, since such combinations are very frequently used in discourse and therefore strongly entrenched. Importantly, these more specific schemata of the type I/we/you/they + verb are not associated with verbal -s. Rather, they are associated with -∅, as in Standard English. This provides us with an explanation of the type-of-subject constraint: Constructions such as I (we/you/they) + verb are stored as mental representations that are highly entrenched. As a consequence of entrenchment, they are represented separately from the more abstract schema “subject + verb + -s” and thus need not (and do not) feature the inflectional ending -s. In a similar vein, the position-of-subject constraint can be accounted for by appealing to basic notions of a performance-based model of construction grammar. Pietsch’s (2005) general idea is that the Northern Subject Rule involves a competition between the more abstract and the more concrete construction schemata. The specific schemata I/we/you/they + verb are most likely to be activated in those cases where pronoun and verb are directly adjacent, because the prototypical and most frequently encountered instances of this schema comprise a subject pronoun plus an adjacent verb. The greater the distance between pronoun and verb, i.e., the more intervening material occurs between the two, the smaller the likelihood that the specific schema is accessed: Constructions characterized by increasing distance between pronoun and verb are increasingly dissimilar to the prototypical construction schema “subject pronoun + contiguous verb”. This account contains the gist of a viable explanation of variability in the choice of concord patterns: Which variant a speaker opts for is determined or at least strongly influenced by various (general and specific) construction schemata, which are in competition with each other. Which of these construction schemata wins out and is accessed is ultimately a matter of entrenchment. Speakers tend to opt for that variant of a variable which most closely matches a strongly entrenched construction schema. For a detailed schematic representation of the construction hierarchies that have to be posited for the various pos-

276

Verena Haser and Bernd Kortmann

sible agreement systems in Northern varieties of English the reader is referred to Pietsch (2005: chapter 2.3).

4. Variation patterns as part of linguistic competence? Phenomena such as the Northern Subject Rule raise the question whether we should consider variation patterns to be part of competence, or whether one should espouse the view defended, for example, by Newmeyer (1998) that choosing one particular form over a variant is solely determined by performance principles (such as “old information precedes new information”; “heavy constituents are placed at the end of sentences”). Several authors have offered compelling arguments in favour of the former view. For instance, Henry (2002) observes that very young children already acquire not only the different variants of a variable used in their environment, they also use them with roughly the frequency with which they are used by adult speakers. This observation, along with several others adduced in her paper, suggests that statements to the effect that a particular variant occurs with a particular likelihood should be viewed as part of competence: “The children have learned the statistical distribution of forms at an early stage, apparently indicating that it is possible to acquire the statistical properties of syntactic structures as part of the acquisition process” (Henry 2002: 279). Pietsch (2005) also assumes that statements of probability constitute part of speaker competence. He approaches the topic from a somewhat different angle than Henry (2002), arguing that potential alternative views to his position cannot possibly be correct: It cannot be true that the choice of a variant is merely a matter of performance principles. Pietsch’s line of argument will be summarized in the following. Pietsch rejects Newmeyer’s (1998) view that pragmatic factors may explain why one variant is preferred over another in a particular context. At least in some cases, such pragmatic explanations fail. Granted, patterns of variation do exist which can be explained in terms of general pragmatic principles. A paradigm example would be the alternation between Chris saw Bryce and It was Bryce whom Chris saw. Here, the second variant is used in cases where the knowledge is presupposed that Chris saw someone. Pragmatic explanations of this type are related to “universally available, functionally grounded mechanisms of performance” (Pietsch 2005: 30). However, the structured variation in agreement patterns exemplified by the examples cited below (from Pietsch 2005: 30) cannot plausibly be motivated by universal pragmatic mechanisms.

Agreement in English dialects

(1)

a. b. c.

277

Many was going. Lots of them was going. Them ones was going.

The use of was is more frequent in examples of the type (1b) and (1c) than in (1a). These context-dependent differences in the likelihood that verbal -s occurs have to be accounted for in some way. They cannot be explained by appealing to performance principles. A plausible alternative is to assume that knowledge about such quantifiable differences in the distribution of variants is part of speaker competence. Pietsch also rejects the position espoused by Kroch (1994) that some sort of psycholinguistic processing mechanisms account for such examples of systematic variation. Positing universal processing mechanisms as an explanation of systematic patterns of variation fails in cases where this variation is determined by factors that are distinctly dialect-specific.The Northern Subject Rule can once again serve as an example: Constructions involving them as subject display a tendency to feature verbal -s; on the other hand, constructions involving they as subject tend to combine with verbs without overt inflectional ending. According to Pietsch (2005: 31), such patterns of variation “cannot be predicted from any more general principles but must be learned as such”. As a consequence, it seems plausible to assume that they are a matter of speaker competence. One might object that Pietsch’s examples are not ideally suited to make his point. For instance, the last example may seem problematic if one subscribes to the competing grammars model advanced by Kroch (1994). From the perspective of Kroch (1994) one might observe that the subject form them is typical of non-standard varieties of English – as opposed to they, which is the form that is characteristic of Standard English. Similarly, the use of verbal -s (as opposed to zero) as verbal inflection in the context of plural subjects is typical of nonstandard varieties. One might thus argue that differences in frequency between constructions such as They sings vs. Them sings (or possibly also They was vs. Them was) can be attributed to the fact that the non-standard form them is likely to trigger and combine with another non-standard form (i.e., with “verb + final -s” – or in the case of the verb BE, with was; these are the non-standard forms occurring after plural subjects). Conversely, the standard form they is likely to trigger a corresponding standard form of the verb (zero-inflection or were). Such a state of affairs is very much in line with Kroch’s competing grammar model of variation: In the example at hand, two grammars seem to be involved; the form belonging to the standard grammar tends to preferentially combine with another form belonging to the standard grammar (rather than to the dialect grammar); and vice versa, forms relating to a non-standard grammar combine

278

Verena Haser and Bernd Kortmann

with other forms relating to the non-standard grammar. Kroch’s idea would be that only one of these grammars is acquired by the child, the other is acquired at a later stage. So only one of these grammars is properly speaking part of speaker competence. Possibly, a similar explanation might be found for the differences in frequency distribution between the contexts in (1) above. It should not go unnoticed, though, that Kroch’s model by itself does not really offer a convincing explanation for the observed patterns of variation (cf. Pietsch 2005: 31). In any case, we will encounter further data below which do seem to confirm the view that statements of probability are indeed – at least in certain cases – part of speaker competence. Note also that Kroch’s overall claims appear to be incompatible with Henry’s (2002) above-mentioned observations that very young children can already produce variants that Kroch might attribute to different grammars and that the frequency distribution of these variants in children’s speech is very much comparable to the distribution of the variants in adult language.

5. Agreement patterns involving pronoun plus BE: Anderwald (2002) We will now take a closer look at data from FRED, the Freiburg English Dialect Corpus compiled by the Freiburg research group (cf. Hernand´ez 2006). More specifically, the focus will be on past tense forms of BE when the verb is combined with personal pronouns. Past tense BE is the only English past tense verb which is irregular. This fact can be explained by appealing to the notion of entrenchment. It is the highly frequent forms which are most likely to develop and maintain irregularities (cf. Croft and Cruse 2004). Needless to say, BE is the most frequently used verb in English. Forms that are less entrenched are far more likely to be regularized. Even past tense BE, however, is affected by regularization tendencies, at least in dialects (cf., e.g., Trudgill 1990: 98; Anderwald 2002: chapter 8). Some dialects have generalized was, i.e., was occurs in all persons in the past tense (I was, you was, he/she/it was, we was, they was). Others have generalized were (I were, you were, he/she/it were, etc.). Finally, there are also mixed systems in which was is used in positive contexts (I/you/he/we/they was happy), while were tends to be restricted to negative contexts (I/you/he/we/they weren’t happy). In recent years, several studies have been published on the subject (cf. inter alia Anderwald 2002; Christian et al. 1988; Tagliamonte 1998; Tagliamonte and Smith 1999; Wolfram and Sellers 1999).

Agreement in English dialects

279

Anderwald (2002) is a recent study of past tense BE in the British National Corpus. We will try to reinterpret some of her major findings from a constructionist perspective. Anderwald investigates personal pronouns and existential there combined with was/wasn’t/were/weren’t. Restricting attention to combinations of personal pronouns and past BE was necessary for practical reasons, since the figures for combinations of full noun phrases and pronouns were too large to be manageable. The same is true for our case study presented in Section 7. However, considering pronoun-verb combinations on their own terms makes sense from a constructionist perspective, since full noun phrases show markedly different frequencies for were-generalization than do third person pronouns (e.g., Tagliamonte 1998). Since the frequency distributions for combinations of pronouns and past BE are significantly different from those for full noun phrases plus past BE, it seems vital to focus on overall constructions such as “pronoun + past BE”, distinguishing neatly between different types of subjects. This procedure will be followed throughout this paper. One of the results of Anderwald’s investigation is that generalized was with plural subjects (e.g., they was) occurs very frequently in British dialects. Considerably less frequent is the corresponding generalization involving were with singular subjects (e.g., I were/He were): Generalized were occurs roughly half as frequently as generalized was. These observations apply in positive contexts only. In negative contexts, generalized were is considerably more frequent than generalized was (cf. Anderwald 2002: 178). From a usage-based construction grammar perspective, the prevalence of was is easily explained with the help of the notion of entrenchment. Since combinations of pronouns and was are far more frequent than combinations of pronouns + were (“was occurs much more frequently [. . . ] than were”; Anderwald 2002: 183), individual constructions such as He was or I was – and possibly also the superordinate construction schema “pronoun + was” – are more entrenched than the corresponding constructions involving pronouns combined with were (cf. also Tagliamonte and Smith 1999). It is in line with basic principles of a usage-based framework to assume that the more entrenched construction schema is less easily given up. In other words, a less entrenched construction like they were is relatively likely to be given up in favour of they was; by comparison, a more entrenched construction like I was is less likely to be given up in favour of I were. Another intriguing finding is that generalization occurs much more frequently in negative clauses than in positive clauses (cf. Anderwald 2002: 180– 182).

280

Verena Haser and Bernd Kortmann

The striking difference between generalization in positive and negative contexts in most dialect areas points to the fact that any generalization strategy is strongly preferred in a negative context over positive contexts. This tendency accords well with a strong cross-linguistic trend that many grammatical distinctions which hold in positive clauses are levelled under negation. Negative clauses in general therefore tend to be less differentiated than their positive counterparts (Anderwald 2002: 182; emphasis added).

The fact that generalization is most often found in negative contexts can be easily accounted for in usage-based approaches to construction grammar. Negated combinations of pronouns and past tense BE are far less frequent – and hence less entrenched from a cognitive perspective – than the corresponding positive constructions. But even though the respective constructions vary with respect to their degree of entrenchment, both the negated constructions and the positive constructions occur frequently enough to be cognitively entrenched to such a degree that they are represented as separate units in speakers’ minds. Thus we will assume that constructions of the type I (he/she/it) wasn’t and I (he/she/it) weren’t are indeed stored separately from the corresponding negated constructions I (he/she/it) was and I (he/she/it) were, with the negative constructions being less entrenched than their positive counterparts. Research on entrenchment (e.g., Pietsch 2005) suggests that strongly entrenched (high-frequency) constructions will be far less likely to be given up in favour of competing construction schemata than weakly entrenched constructions. Applied to our case study, this means that instances of generalization – i.e., changes of form from was(n’t) to were(n’t) or vice versa – will tend to predominantly affect the weakly entrenched constructions I (he/she/it) wasn’t and I (he/she/it) weren’t, rather than their positive counterparts. The more entrenched non-negated construction types are more likely to resist being superseded by the competing generalized construction schema. Hence, speakers will produce far more instances of generalized forms for negative contexts than for positive contexts. This in turn implies that generalization tendencies will – at least in the early stages of levelling – be able to firmly establish themselves only in negative contexts. Of course, in actual discourse individual instances of generalization may indeed occur in positive contexts as well. However, these cases will typically not be frequent enough to catch on and initiate a permanent change in the paradigm, since strongly entrenched constructions will in most cases succesfully resist being superseded by a competing construction schema. Even if a whole paradigm will eventually undergo a levelling process, the non-negated/assertive forms will be able to resist the tendency towards levelling for a longer time than the negated forms.

Agreement in English dialects

281

Another intriguing finding from Anderwald (2002) relates to possible combinations of was- and were-generalization. The next section will be devoted to this topic.

6. Possible combinations of was- and were-generalization Anderwald (2002: chapter 8) notes that there are four possible combinations of was- and were-generalization: (i) was-generalization in positive and negative contexts (ii) were-generalization in positive and negative contexts (iii) was-generalization in positive contexts and were-generalization in negative contexts (iv) were-generalization in positive and was-generalization in negative contexts Of these four possibilities, the first option is frequently found, especially in several non-British dialects. The last one, however, is not attested. Anderwald (2002: 184) therefore suggests the following implicational tendency for nonstandard past tense BE: Non-standard were → non-standard weren’t. Why is system (iv) not attested? At first sight, this generalization is surprising. After all, a contrast between were in positive contexts and wasn’t in negative contexts would make considerable sense from a functionalist perspective. Consider the parallel system (iii) which features was-generalization in positive and weren’t-generalization in negative contexts, and which is attested. As Anderwald has shown for this system, the contrast between positive was and negative weren’t, which was brought about by a “remorphologization” (Wolfram and Schilling-Estes 1996) of was/were forms, makes sense from a cognitive perspective: [. . . ] a singular-plural distinction is clearly not necessary cognitively for past tense verbs in English: the past tense forms of BE constitute the only verb paradigm where this distinction is still in place. A positive-negative distinction on the other hand is extremely valuable, as the information carried by the negator is very important [. . . ] (Anderwald 2002: 191)

This positive-negative distinction is clearly indicated with the help of the marked acoustic contrast between was and weren’t – the acoustic differences between was and weren’t being more pronounced than those between was and wasn’t (or between were and weren’t). In principle, a very similar motivation could be given for the non-existent or at least very rare system (iv) which builds on a

282

Verena Haser and Bernd Kortmann

contrast between were in positive and wasn’t in negative contexts. Why then is such a system hardly ever found? This section proposes a usage-based constructionist explanation of the impossibility of system (iv), capitalizing on two assumptions expounded above. First, highly specific and closely related constructions can be stored separately in speakers’ minds; e.g., both the positive construction I (you, he, etc.) were and the negated construction I (you, he, etc.) weren’t are stored separately. Secondly, less entrenched constructions like I (you, he, etc.) weren’t are typically affected by generalization to a greater degree – and hence superseded by the generalized forms at an earlier stage – than the corresponding positive constructions I (you, he, etc.) were. The following account – as well as the corresponding Figure 1 – are necessarily in some respects idealized. Thus, they assume ideal systems in which changes in specific constructions become categorical, i.e., either were- or wasgeneralization come to be used throughout (part of) the agreement system. This simplification is hoped to yield a clearer picture of the central forces underlying the developments involved. Actual agreement systems may involve long-lasting competition between the relevant generalization types; as a result, was- and were-forms may occur in the same range of contexts, albeit with different frequencies. In principle, however, only small modifications to our account are necessary if we want to do justice to the non-categorical nature of actual generalization processes; some of these modifications will be sketched below. The following exposition makes crucial reference to Figure 1, which gives a simplified sketch of the developments that lead to the three attested systems of generalization types. Boxes indicate sets of constructions that differ in degree of entrenchment. Entrenchment is indicated by the boxes’ different degrees of thickness. Stage 1 in Figure 1 sketches the standard system, which acts as the starting point of the developments under discussion. The we/you/they weren’t group of constructions represent the least entrenched combinations of personal pronouns and past BE in stage (1). These constructions are combinations of the comparatively3 weakly entrenched negated verb forms and the comparatively4 weakly entrenched plural pronouns. Consequently, the overall constructions are almost bound to be comparatively weakly entrenched as well. It follows from the preceding observations that the “mother node” of the we (you/they) weren’t constructions, viz. the superordinate construction schema “pronoun + weren’t”, is the least entrenched superordinate schema depicted in stage (1). The instances of the “pronoun + wasn’t” schema form another set of constructions which is relatively weakly entrenched (compared to the instances of the assertive schema “pronoun + was + positive5 ”).

Agreement in English dialects

283

Stage (1): Point of departure (standard system) Pron + was + positive

Pron + wasn’t

I was

I wasn’t he wasn’t

he was

…

…

Pronoun + were + positive

Pronoun + weren’t

we were

we weren’t they weren’t

they were

…

First Scenario: stage (2): were-generalization

Pron + weren’t

Pron+was+pos. I was

he was

…

Pron+were+pos. we were

Second scenario: stage (2*): was-generalization

they were

stage (3a): were comes to be established as the general past BE form used for all persons in positive contexts

I weren’t he weren’t we weren’t they weren’t

…

Pron+wasn’t

Pron+was+pos I was

he was

…

Pron+ were+pos we were

they were

I wasn’t he wasn’t we wasn’t they wasn’t …

stage (3b):

stage (3*a):

stage (3*b):

establishment of was as the form used for all persons in positive contexts

“pronoun + was” schema comes to be used throughout positive contexts as well

impossible system: were comes to be established as the form generally used in positive contexts

Figure 1. Developments leading to the four possible combinations of was and weregeneralization

For reasons expounded in Section 5, generalization will ceteris paribus tend to first affect (instances of) the two weakest schemata (“pronoun + wasn’t” or “pronoun + weren’t”). Stages (2) and (2*) represent the results of two possible developments proceeding from stage (1). The first scenario – stage (2) – is that were-generalization occurs, starting off predominantly in negative contexts. This leads to a change in the “pronoun + wasn’t” constructions, which are increasingly superseded by “pronoun + weren’t” constructions. Thus by stage (2), the “pronoun + weren’t” schema has become more entrenched, ousting in idealised cases the “pronoun + wasn’t” schema. Crucially, this development also results in increased entrenchment of the more abstract superordinate schema “pronoun + were”. The “pronoun + were” schema covers all pronoun + were combinations, including both positive and negative contexts. In other words, the “pronoun + were” schema encompasses

284

Verena Haser and Bernd Kortmann

constructions of the type “pronoun + were + positive” and constructions of the type “pronoun + weren’t”. In our diagram, the “pronoun + were” schema is given by the outlines of the geometric figure encompassing the “pronoun + weren’t” schema and the “pronoun + were + positive” schema. The “pronoun + were” schema is thus not to be confused with the “pronoun + were + positive” schema, which relates only to those “pronoun + were” combinations that are used in positive (i.e., non-negated) contexts. Positing an abstract schema “pronoun + were” and a more specific “pronoun + were + positive” schema is in line with the logic of construction hierarchies posited in construction grammar and as we will see appears to be essential to providing an account of the regularity found by Anderwald. There are two paths of developments that can be subsequently followed by a system that has reached stage (2). One possibility – stage (3a) – is that the “pronoun + were” schema spreads and comes to prevail in positive contexts as well. This development is motivated by analogy and made possible due to the increase in entrenchment of the “pronoun + were” schema that has resulted from the previous developments. Another possible development is for stage (2) to be followed by a stage where was comes to be used in combination with all personal pronouns in positive contexts. This development results in stage (3b) in Figure 1. Here, the prevalence of was-forms in positive contexts leads to a loss (or, in less ideal systems, to a drastically reduced frequency) of the were-forms that in the standard system were used in combination with we, they, and you. The introduction of wasforms throughout assertive (non-negated) contexts is motivated by iconicity, i.e., maximal acoustic differentiation between asserted sentences (which feature was-constructions throughout) and negated sentences (which feature weren’tconstructions). The spread of was in positive contexts is made possible by the intrinsically strong entrenchment of “pronoun + was” combinations. The foregoing considerations relate to a scenario where were-generalization comes to predominate the early stages of paradigm levelling.The second possible scenario is for was-generalization to become the predominant generalizationtype. This scenario is sketched as stage (2*) in Figure 1. Again, it is plausible to assume that generalization tends to be successfully established in negative contexts first. As a result of was-generalization in negative contexts, the “pronoun + wasn’t” schema comes to be more strongly entrenched, ousting the “pronoun + weren’t” schema. The loss (or, in less idealised systems, decrease in frequency) of the “pronoun + weren’t” schema also implies an increase in entrenchment of the abstract schema “pronoun + was”, which covers both asserted and negated contexts. This schema is symbolised by the outlines of the geometric figure encompassing the “pronoun + wasn’t” and the “pronoun + was + positive” schema.

Agreement in English dialects

285

Subsequent to was-generalization in negative contexts (i.e., stage 2*), two potential developments might occur. The first possibility is sketched as stage (3*a) in Figure 1: was-levelling spreads to positive contexts as well; as a result, was comes to be used throughout the paradigm. This spread is motivated by analogy. The factor ultimately enabling this development is the overwhelming predominance of the “was + pronoun” schema. This schema has been strongly entrenched from the outset; its degree of entrenchment was further increased due to the increase in frequency of (instances of) its “daughter” schema “pronoun + wasn’t”. On the other hand, another possible development (sketched as 3*b) is predicted not to occur if Anderwald’s implicational tendency is correct: weregeneralization in positive contexts does not occur once was-generalization has spread in negative contexts. Initially, this might seem surprising: After all, stage (3a) (see Figure 1) in our first scenario leads to a similar system, in which weregeneralization also comes to be established throughout positive contexts. Why then is (3a) possible, while (3*b) isn’t – even though both systems could be motivated by iconicity (maximal differentiation between negative and positive contexts)? The solution to our problem seems to be this.As a result of was-generalization in negative contexts, the abstract construction schema “pronoun + was” has become too entrenched by stage (2*) to be superseded by the “pronoun + were” schema. The latter schema in turn has been relatively weakly entrenched from the outset, and has become even less entrenched as a result of the loss (or at least decreased frequency) of the “pronoun + weren’t” constructions. Thus, the “pronoun + was” schema is much more likely to be accessed in actual discourse than the “pronoun + were” schema. The latter schema is increasingly pushed to the sidelines and eventually falls out of use, resulting in was-generalization throughout the paradigm. That this explanation – which makes crucial reference to abstract construction schemata (“pronoun + were” and “pronoun + was”) – is on the right track is suggested by the following observation: The more specific schema “pronoun + was + positive” displays the same degree of entrenchment in (3a) and in (3*b); similarly, the more specific “pronoun + were + positive” schema also displays the same degree of entrenchment in (3a) and (3*b). Therefore, the reason for the possibility of (3a) and the impossibility of (3*b) is not a matter of these two more specific construction schemata; rather, the impossibility of (3*b) as opposed to (3a) must relate to differences in degree of entrenchment between the superordinate schemata “pronoun + were” and “pronoun + was”. To summarize, we have shown that a usage-based constructionist approach can explain the – at first sight mysterious – implicational tendency noted in Anderwald (2002), which stakes out the limits of variation for pronoun + BE

286

Verena Haser and Bernd Kortmann

combinations in the past tense paradigm. Such a constructionist account differs from potential generative models in at least one crucial respect: Our account presupposes that negated forms are stored separately from the non-negated verb forms (since they are cognitively entrenched). In other words, negated constructions are stored as units and not created anew each time by means of movement operations which adjoin two items (a verb and n’t) as conceived of in generative grammar accounts (e.g., Radford 1997). Note that the above account could be re-cast as modelling the kinds of systems that are likely to occur – and hence the kinds of systems which are expected to be most frequently found – across English varieties. These systems are represented by the individual stages in Figure 1.

7. Pronouns + past tense BE in the Freiburg English Dialect Corpus We will now focus on traditional dialects, more precisely on data from the Freiburg English Dialect Corpus FRED. In particular, the following areas were investigated: Lancashire, London, Kent, Suffolk, Nottinghamshire, Devon, Wiltshire, Somerset, and Cornwall. Other areas covered by the FRED corpus were also studied, but they will not be discussed in the following, because the data for these regions did not yield any findings pertinent to the present study. As observed above, we had a look at combinations of personal pronouns plus past tense BE, searching specifically for pronouns and adjacent BE. Results are given in the tables in the appendix. It should be noted that hypothetical contexts (defined as in Anderwald 2002 as occurrences of pronouns after if ) were excluded from consideration. A major methodological problem is that the token occurrences of pronoun plus BE constructions in negative contexts are often too small to allow reliable statistical testing. This problem cannot be solved in a completely satisfactory way, although occasionally the differences in frequencies between different constructions are very suggestive and do seem to allow some (admittedly preliminary) conclusions. A conspicuous result of our study, which accords well with findings by other authors (e.g., Tagliamonte 1998) is that generalization may progress at different rates for different grammatical persons. In line with other studies on the topic, the third person pronoun they emerges as an item which is less likely to be combined with was than all other pronouns. Most interestingly, they is less likely to undergo was-generalization than we. This is true for five of the nine

Agreement in English dialects

287

areas covered in this study; differences are significant for three of them (Suffolk, Lancashire, county London). Devon, however, is a conspicuous exception to this general trend. Here, was-generalization has progressed much further in the third person plural than in the first person plural – at least in positive contexts (cf. the relevant tables in the appendix). The difference is statistically significant6 (p ≤ 0.001) in this case, whereas it is not significant for the other two areas where the percentage of was-generalization is higher for they than for we. Another interesting example of how generalization may progress at different rates in different contexts is provided by data from Wiltshire, viz. I were vs. He were: In positive contexts He were is relatively more frequent than I were, i.e., were-generalization occurs more frequently for he than for I – the difference is quite striking (and statistically significant; p ≤ 0.001). This is not a general pattern in our data, however. For Lancashire, the relationship is reversed. Again, the data are statistically significant (p ≤ 0.05). In other dialect regions, there is no significant difference in frequency distribution to be found between these contexts (e.g., Nottinghamshire). The above examples, which involve dialect-specific and idiosyncratic departures from general trends, or unpredictable and dialect-specific differences in frequency distributions between different contexts, represent patterns of variation that arguably cannot be explained with the help of universal psycholinguistic or pragmatic mechanisms along the lines of Kroch (1994) or Newmeyer (1998) (cf. Section 4 above). Following Pietsch’s line of reasoning, these patterns of variation should therefore be part of speaker competence. Frequency distributions which are not in line with the pattern typically observed in most other dialects, or which seem to vary randomly from dialect to dialect, are most unlikely to be due to pragmatic or psycholinguistic processes. Since the type of subject (including grammatical person and singular vs. plural) has in several studies emerged as an important factor influencing the choice of was vs. were, each possible combination of individual personal pronouns and adjacent past tense BE should be investigated separately. Indeed, many previous studies do distinguish between different combinations of pronouns and verbs in order to assess which pronouns are more likely to occur with generalized was or generalized were (to cite but two examples, see Tagliamonte 1998 and Wolfram and Sellers 1999). A distinction is usually made between the different persons of a paradigm. Yet, we may even make a case for a more fine-grained analysis than the one advanced in such papers. In our data, the figures for weregeneralization are at times significantly different for third person masculine vs. third person feminine pronouns. Let us therefore contrast the frequencies of were-generalization for he vs. she in positive contexts for Lancashire, Wiltshire, and Somerset.

288

Verena Haser and Bernd Kortmann

In the data from Lancashire, the relative frequencies for were-generalization in the third person singular masculine and feminine (in positive contexts) are significantly different: He were occurs in 12% of all cases of he plus past BE, while she were occurs in 18% of all cases of she plus past BE. Note that for the first person singular I, were-generalization occurs in 17% of all cases (i.e., I were occurs in 17% of all cases where some instance of “I plus past BE” occurs; I was in 83% of all cases). Taking into account differences in grammatical gender for third person singular pronouns thus requires us to reconsider the hierarchy of grammatical contexts that influence the incidence of were-generalization for Lancashire. If we tease apart occurrences of she were from occurrences of he were, the personal pronoun she turns out to be the most favourable context for were-generalization in the Lancashire paradigm, followed by first person I, which is in turn followed by he. The frequency distributions for I were vs. he were differ from each other in a statistically significant way; by contrast, she were patterns with I were (i.e., the difference in relative frequencies between she were and I were is not statistically significant). By contrast, if we do not distinguish between masculine and feminine pronouns, the differences in frequencies for were-generalization are not statistically significant. Taken together, combinations of third person singular pronouns plus were occur in 15% of all cases; a result which is not significantly different from the frequencies found for I were. In this way, distinguishing between he and she may well make a difference to the hierarchy of grammatical contexts that are found to favour a generalization strategy. Similar observations hold for Wiltshire. In Wiltshire, she again patterns with I rather than he as far as the probability for were-generalization is concerned: Differences between relative frequencies of I were and she were are not statistically significant; while there are significant differences between the relative frequencies of she were vs. he were and I were vs. he were. In this case, however, were-generalization occurs much more frequently with he (in 34% of the cases), than with I (18%) or she (23%). Finally, for Somerset there is again a significant difference between the relative frequencies for were-generalization in the contexts of he and she. However, in this case it is he which patterns with the first person pronoun I : The differences in relative frequencies between he were and I were are not statistically significant (were-generalization occurs in 15% of the cases for the third person singular masculine; and in 17% of the cases for the first person singular). She were, by contrast, occurs in 37% of the possible occurrences and is thus considerably more frequent than both he were and I were.

Agreement in English dialects

289

The above findings argue against collapsing together constructions involving he and she as subjects when considering factors influencing were-generalization, despite the fact that both he and she belong to the same grammatical category of third person singular pronouns. Our results are very much in line with a constructionist model, where even very specific and highly similar constructions like he was and she was can be stored as separate mental units and may therefore be affected by syntactic changes to markedly different degrees. The above examples of different relative frequencies for were-generalization in the contexts of he and she are once again not attributable to universal cognitive or pragmatic factors of the type envisaged by Kroch and Newmeyer (cf. Section 4). Thus, there seems to be no universal principle which could explain why in Wiltshire, he is preferred over she as the more frequent context for weregeneralization; while matters are different for Somerset. Again, the conclusion suggests itself that knowledge about such frequency distributions must be part of speaker competence.

8. Discussion In some cases, the above results may appear to be at odds with a usage-based model of construction grammar. Notably, she were should be expected to be generally more frequent than he were: Due to the considerably higher overall frequency of he in discourse (compared to she7 ), he was is more entrenched than she was – and therefore should be generalized less readily than the latter construction. But as we have seen, this is not the case for Wiltshire – even though it is true for Somerset and Lancashire. For other FRED-regions investigated, were-generalization is either practically non-existent in these contexts; or there is no statistically significant difference between were-generalization for he and she. Further studies are needed to determine whether other regions follow the – from a constructionist perspective – expected pattern for generalization to be more frequent in the context of the third person female pronouns. A study of further regions might reveal that Wiltshire is indeed the odd man out, patterning differently from most other dialect areas. Though matters are not altogether clear for the different relative frequencies of he were as opposed to she were, there is a related pattern found in pertinent literature (e.g., Tagliamonte 1998) that lends itself straightforwardly to an explanation based on the notion of entrenchment: As already observed, they is in general less likely to combine with was than we. This finding can be motivated by the higher discourse frequency of they in comparison to we (cf. Tagliamonte and Smith 1999; compare also the figures from a very large corpus like the

290

Verena Haser and Bernd Kortmann

British National Corpus, where they occurs considerably more often than we). As a consequence, they were is more entrenched than we were, and thus less likely to be superseded by the generalized (was-)form. Even if a usage-based account of the statistical differences between generalization in the contexts of she and he should eventually turn out to be problematic, a closer look at the data for were-generalization in these contexts reveals that some sort of construction grammar approach – not necessarily of the usagebased type – may indeed be indispensable if we want to give a satisfactory account of the frequency distributions observed. One caveat is necessary, however. The figures on which the following generalizations were based were typically too low to allow statistical testing. This is a general problem which can only be remedied by compiling larger corpora. However, the observed statistical tendencies in different dialects are consistent enough, at least when taken together, to support the following observations. Let us first have another look at the Wiltshire data. Here, were-generalization is relatively more frequent for he than for she – at least for positive contexts. In other words, in positive contexts, he were is significantly more frequent than she were. In negative contexts, however, matters are different: he weren’t is relatively less frequent than she weren’t. In fact, he weren’t does not occur at all. Now assume that Henry’s (2002) and Pietsch’s (2005) position is correct that patterns of variation and statements concerning a variant’s likelihood of occurrence should be considered part of speaker competence. In that case, Wiltshire speakers should have internalised knowledge (competence) of the following type: The combination she + was is relatively more frequent – by some specifiable amount – than he + was. Moreover, speakers must know that for negative contexts, he + was + n’t is relatively more frequent than she + was + n’t – again by some definite amount. Finally, they must know that he weren’t is – by some specifiable amount – significantly less frequent than he were. It seems highly implausible that these differences in relative frequencies are a matter of the individual lexical items involved. Rather, they are a matter of the overall complex constructions such as “she (or he) combined with was” / “She (or he) combined with was and negation operator”, etc. A close look at the data shows why this is the case. For example, it would be mistaken to simply assert that in Wiltshire, he displays a quantifiable preference for occurring with were. This is true for positive contexts, but seems false for negative contexts. So presence or absence of n’t also figures into the equation. Similarly, it would be mistaken to assert that were prefers he over she as the item it combines with; because in negative contexts, matters again seem to be different. Finally, differences in relative frequencies are not simply a matter of

Agreement in English dialects

291

the negation operator as such. One cannot plausibly claim, for example, that due to some (possibly dialect-specific) feature of the negation marker, negation tends to be a preferred environment for was rather than were.As is amply demonstrated by the data in general and the figures for Wiltshire in particular, this is not true. The frequency distributions do not only depend on the negation marker, but also on the grammatical person involved. Quite independently of the impossibility of relating specific quantifiable frequency distributions to individual items, it would in some cases even be mistaken to stipulate that particular items have an overall preference for occurring with certain other items. One cannot claim, for instance, that we has a preference for combining with was in London county; for this is only true for positive contexts. In negative contexts, there seems to be a general preference for combining we and were. Consider finally Lancashire. Here it would be mistaken to claim that were has a preference for combining with she rather than he. For in negative contexts, this tendency seems to be reversed (though again, differences are statistically significant only for positive contexts). These data suggest that the differences in relative frequencies between the various combinations of pronoun plus verb (plus negation) cannot be ascribed to features of individual items; rather they depend on the overall constructions involved. Ultimately, then, the assumption that statistical properties of linguistic forms are part of speaker competence implies (at least in these cases) that we should adopt a constructionist approach to the data. For at least in the examples covered above, the statistical properties of the two variants of our variable (past tense BE) depend on the overall constructions in which they occur, rather than on individual lexical items. Thus, language users associate statistical properties with certain complex combinations of items, i.e., constructions – which implies that the constructions are mentally represented in the first place.

9. Conclusion The constructionist approach has turned out to offer a satisfactory explanation of an agreement system that has proved difficult to model within formalist frameworks, viz. the Northern Subject Rule. We have also seen that other robust generalizations that have emerged from previous studies on agreement – some of which were confirmed in this study – can be easily explained within a usagebased constructionist model. But even the more idiosyncratic and highly dialectspecific variation patterns discussed above seem to require a constructionist approach.

292

Verena Haser and Bernd Kortmann

In a constructionist model, changes may primarily affect constructions, rather than the respective system in general. If we assume that syntactic changes (here generalization) primarily affect specific constructions rather than abstract paradigms, it will not be particularly surprising that the changes do not proceed in sync, that different constructions may undergo generalization at very different rates, and even that the observed frequencies may occasionally be unexpected from a cognitive (usage-based) perspective. If a change occurs in a construction, this construction may gain momentum and change in partial separation from related constructions. In most cases, a construction will change at a rate which is predicted by its degree of entrenchment; but since each construction constitutes a separately stored unit (if it is sufficiently frequent), it may occasionally be affected by change to an unexpectedly low (or high) degree. This account does not absolve us from identifying, if possible, the factors responsible for unexpected developments. Still, a model that assumes a partial independence of constructions goes some way toward explaining the possibility of contextdependent and perhaps even idiosyncratic frequency distributions. Conversely, the complementary constructionist idea that specific constructions are united by being members of the same superordinate schemata has a potential for modeling a number of recurrent patterns in agreement systems, such as the possibility and impossibility of specific combinations of was- and were-generalization noted by Anderwald (2002).

Appendix: Results for individual areas Table 1. Lancashire: Absolute figures and percentages of generalized forms I was I were You was You were He was He were She was She were We was We were They was They were

516 103 20 102 380 53 216 49 37 229 56 544

83% 17% 16.4% 83.6% 88% 12% 82% 18% 14% 86% 9% 91%

I wasn’t I weren’t You wasn’t You weren’t He wasn’t He weren’t She wasn’t She weren’t We wasn’t We weren’t They wasn’t They weren’t

20 5 4 7 16 4 8 0 2 9 6 48

80% 20% 36% 64% 80% 20% 100% 0% 18% 82% 11% 89%

Agreement in English dialects

293

Table 2. Nottinghamshire:Absolute figures and percentages of (non-) generalized forms I was I were You was You were He was He were She was She were We was We were They was They were

412 66 80 23 287 47 113 20 91 52 185 113

86% 14% 78% 22% 86% 14% 85% 15% 64% 36% 62.1% 37.9%

I wasn’t I weren’t You wasn’t You weren’t He wasn’t He weren’t She wasn’t She weren’t We wasn’t We weren’t They wasn’t They weren’t

17 3 3 3 7 0 1 0 6 1 6 8

85% 15% 50% 50% 100% 0% 100% 0% 86% 14% 42.9% 57.1%

Table 3. County London: Absolute figures and percentages of (non-)generalized forms I was I were You was You were He was He were She was She were We was We were They was They were

324 0 30 31 273 0 173 0 80 49 61 97

100% 0% 49% 51% 100% 0% 100% 0% 62% 38% 39% 61%

I wasn’t I weren’t You wasn’t You weren’t He wasn’t He weren’t She wasn’t She weren’t We wasn’t We weren’t They wasn’t They weren’t

22 0 4 9 15 0 6 0 4 8 2 9

100% 0% 31% 69% 100% 0% 100% 0% 33% 67% 18% 82%

Table 4. Suffolk: Absolute figures and percentages of (non-)generalized forms I was I were You was You were He was He were She was She were We was We were They was They were

540 0 19 287 546 0 238 0 18 427 10 955

100% 0% 6% 94% 100% 0% 100% 0% 4% 96% 1% 99%

I wasn’t I weren’t You wasn’t You weren’t He wasn’t He weren’t She wasn’t She weren’t We wasn’t We weren’t They wasn’t They weren’t

1 3 0 8 3 4 0 2 0 6 0 22

25% 75% 0% 100% 43% 57% 0% 100% 0% 100% 0% 100%

294

Verena Haser and Bernd Kortmann

Table 5. Devon: Absolute figures and percentages of (non-)generalized forms I was I were You was You were He was He were She was She were We was We were They was They were

321 0 24 13 215 0 79 0 4 64 68 137

100% 0% 64.9% 35.1% 100% 0% 100% 0% 5.9% 94.1% 33.2% 66.8%

I wasn’t I weren’t You wasn’t You weren’t He wasn’t He weren’t She wasn’t She weren’t We wasn’t We weren’t They wasn’t They weren’t

11 0 0 3 7 0 1 0 1 6 1 14

100% 0% 0% 100% 100% 0% 100% 0% 14.3% 85.7% 6.7% 93.3%

Table 6. Wiltshire: Absolute figures and percentages of (non-)generalized forms I was I were You was You were He was He were She was She were We was We were They was They were

457 100 20 70 192 100 85 26 12 156 39 426

82% 18% 22.2% 77.8% 65.8% 34.2% 76.6% 23.4% 7.1% 92.9% 8.4% 91.6%

I wasn’t I weren’t You wasn’t You weren’t He wasn’t He weren’t She wasn’t She weren’t We wasn’t We weren’t They wasn’t They weren’t

17 11 4 20 8 0 3 2 2 11 0 21

60.7% 39.3% 16.7% 83.3% 100% 0% 60% 40% 15.4% 84.6% 0% 100%

Table 7. Somerset: Absolute figures and percentages of (non-)generalized forms I was I were You was You were He was He were She was She were We was We were They was They were

340 70 10 42 336 60 66 38 51 157 75 285

82.9% 17.1% 19.2% 80.8% 84.8% 15.2% 63.5% 36.5% 24.5% 75.5% 20.8% 79.2%

I wasn’t I weren’t You wasn’t You weren’t He wasn’t He weren’t She wasn’t She weren’t We wasn’t We weren’t They wasn’t They weren’t

6 3 1 3 8 2 3 2 2 9 3 11

66.7% 33.3% 25% 75% 80% 20% 60% 40% 18.2% 81.8% 21.4% 78.6%

Agreement in English dialects

295

Notes 1. We wish to thank an anonymous reviewer for very helpful comments on our article. 2. As observed by an anonymous reviewer, there are intriguing parallels between the Brittonic split agreement system and the Northern Subject Rule. Regardless of whether or not Brittonic exerted a causal influence on agreement systems in Northern English dialects, data from Welsh lend themselves to analysis in terms of entrenchment along the lines discussed in this paper. Our reviewer also points out that certain facts argue against Pietsch’s assumption that the peculiar agreement patterns occurring in Northern English dialects and British Celtic are a “typological fluke” (see contributions in Filppula, Klemola and Pitk¨anen (eds.) 2002). Our data suggest that the original Northern Subject Rule pattern has been replaced by a less systematic pattern, which can be analysed in terms of entrenchment. 3. “Comparatively weakly entrenched” in this case means that the verb forms involved are less entrenched than the corresponding non-negated forms. 4. As already noted, the plural pronouns are comparatively weakly entrenched (frequent) compared to the singular pronouns I and he/she/it. 5. “Positive” or in abbreviated form “pos” refers to constructions used in positive (nonnegated) contexts. 6. The statistical results were arrived at by using a chi-square test of significance. 7. In the British National Corpus, for example, he occurs considerably more frequently than she; this is presumably a general trend in most varieties of English, since he often functions as a default form that can also refer to females, while she does not usually function as a default form.

References Anderwald, Lieselotte 2002 Negation in Non-Standard British English: Gaps, Regularizations and Asymmetries. London: Routledge. Christian, Donna, Walt Wolfram, and Nanjo Dube 1988 Variation and Change in Geographically Isolated Communities: Appalachian English and Ozark English. Tuscaloosa: University of Alabama Press. Croft, William 2001 Radical Construction Grammar. Oxford: Oxford University Press. Croft, William, and D. Alan Cruse 2004 Cognitive Linguistics. Cambridge: Cambridge University Press. Fillmore, Charles J., and Paul Kay 1993 Construction Grammar Coursebook, Chapters 1 throu 11. Manuscript, University of California at Berkeley. Filppula, Markku, Juhani Klemola, and Hili Pitk¨anen (eds.) 2002 The Celtic Roots of English. Joensuu: University Press.

296

Verena Haser and Bernd Kortmann

Goldberg, Adele E. 1995 Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. 2006 Constructions at Work: The Nature of Generalization in Language. Oxford: Oxford University Press. Henry, Alison 1994 Singular concord in Belfast English. Belfast Working Papers on Language and Linguistics 12: 134–176. 2002 Variation and syntactic theory. InThe Handbook of LanguageVariation and Change, Jack K. Chambers, Peter Trudgill, and Natalie SchillingEstes (eds.), 267–282. Malden, MA: Blackwell. Hern´andez, Nuria 2006 User’s Guide to FRED: Freiburg Corpus of English Dialects. Freiburg: Albert-Ludwigs-University. Kroch, Anthony 1994 Morphosyntactic variation. In Papers from the 30th Regional Meeting of the Chicago Linguistic Society: Vol 2: The Parasession on Variation in Linguistic Theory, K. Beals, J. Denton, R. Knippen, L. Melnar, H. Suzuki, and E. Zeinfeld. (eds.), 180–201. Chicago: Chicago Linguistic Society. Newmeyer, Frederick 1998 Language Form and Language Function. Cambridge, Mass.: The MIT Press. Pietsch, Lukas 2005 Variable Grammars: Verbal Agreement in Northern Dialects of English. T¨ubingen: Niemeyer. Radford, Andrew 1997 Syntactic Theory and the Structure of English:A Minimalist Approach. Cambridge: Cambridge University Press. Tagliamonte, Sali 1998 Was/were variation across the generations: View from the city of York. Language Variation and Change 10: 153–191. Tagliamonte, Sali, and Jennifer Smith 1999 Analogical leveling in Saman´a English: The case of was and were. Journal of English Linguistics 27: 8–26. Trudgill, Peter 1990 The Dialects of England. Oxford: Blackwell. Wolfram, Walt, and Natalie Schilling-Estes 1996 Dialect change and maintenance in a post-insular island community. In Focus on the USA, Edgar W. Schneider (ed.), 103–148. Amsterdam: John Benjamins. Wolfram, Walt, and Jason Sellers 1999 Ethnolinguistic marking of past be in Lumbee Vernacular English. Journal of English Linguistics 27: 94–114.

Semi-modal variation∗ Katarina Klein Abstract. A number of sources mention a very small class of German verbs called “semi-modals”. This class comprises only four verbs: drohen ‘threaten’, pflegen ‘usually do/happen’, scheinen ‘seem’, and versprechen ‘promise’. Semimodal verbs are characterized by the following properties: They govern an infinitive marked by zu ‘to’, they are raising-to-subject verbs, and they only license coherent constructions with their infinitival complements. Semantically, semimodals express some kind of modality. It is well known that scheinen may alternatively be realized with a complement clause. Though corpus analyses have shown that drohen and versprechen are frequently realized without an infinitival argument (Kiss 2005), these variants have not yet been investigated in detail. This article focuses on the semantic and syntactic properties of these constructions. The central claim of this paper is that infinitive-less drohen and versprechen are constructional variants of the semi-modal raising verbs. The current definition of semi-modals excludes these variants, hence it must be revised – or perhaps abandoned altogether.

1. Introduction Among the classes of German verbs that select for infinitival complements, there is a class called “semi-modal verbs” which consists of the four verbs scheinen ‘to seem’, drohen ‘to threaten’, pflegen ‘to be in the habit of/usually do’, and versprechen ‘to promise’.1 (1)

a. b. c. d.

Hans scheint zu gewinnen. ‘Hans seems to be winning.’ Hans droht zu gewinnen. ‘Hans is about to win (and this is not welcome).’ Hans pflegt zu gewinnen. ‘Hans usually wins.’ Hans verspricht zu gewinnen. ‘Hans will probably win.’

298

Katarina Klein

Semi-modals are characterized by three syntactic properties (Eisenberg 1999: 352–353, cf. also Zifonun, Hoffmann and Strecker 1997: 1282–12852 ): First, they govern an infinitive marked with zu ‘to’, i.e. the second status in Bech’s (1955) terminology. Second, they are so-called raising-to-subject verbs, i.e. their syntactic subject does not relate semantically to the semi-modal verb, but is analyzed as the overtly missing subject of their infinitival complement. Third, they form obligatorily coherent constructions with their infinitival complement such that the infinitival complement cannot be extraposed. Semantically, semimodals denote modal aspects of various kinds. There are several problems with this definition of semi-modal verbs. One of them is that semi-modals allow for constructional variants without infinitival complements. It is well known that scheinen may be realized with an expletive subject and a finite complement clause as propositional complement as illustrated by example (2).3 (2)

Es scheint, dass Hans gewinnt. ‘It seems that Hans will win.’

Semi-modal drohen and versprechen also provide constructional variants without infinitival complements. The following examples are taken from a newspaper corpus comprising the 1993 editions of the daily newspaper Neue Z¨urcher Zeitung (NZZ). (3)

a.

b.

(4)

a.

Dem Meister droht gar das Ausscheiden the.dat champion threatens even the.nom elimination in der Hallenmeisterschaft. in the indoor championship ‘The champion might even be facing elimination from the indoor championship.’ Im neuen Verfahren droht ihnen eine weitere In-the new trial threatens them a.nom further.nom Verurteilung. conviction ‘Yet another conviction threatens them in a new trial.’ Steigende Nachfrage verspricht zwar erh¨ohten increasing.nom demand promises though increased.acc Absatz, . . . sale ‘Though increasing demands promise higher sales, . . . ’

Semi-modal variation

b.

299

. . . zumal neue Techniken und Behandlungsarten particularly new.nom technologies and treatments immer bessere Aussichten auf eine v¨ollige Genesung always better.acc prospects to a complete recovery versprachen promised ‘. . . particularly as new technologies and medical treatments promised better and better prospects of complete recovery.’

These constructions cannot be dismissed easily. Kiss’ (2005) corpus analysis demonstrates that semi-modal versprechen is realized about three times as often without an infinitival argument as it is with an infinitive form. The nominal construction with drohen is not uncommon either: In an NZZ 1999 sample (comprising roughly 3.18 million tokens and 230,000 types), I found 89 out of 193 uses of semi-modal drohen realized with a nominal complement. This paper focuses on the latter variants of drohen and versprechen. The central claim of this paper is that infinitive-less drohen and versprechen are constructional variants of the semi-modal raising verbs drohen and versprechen. It will be argued that these verbs belong to the same semantic class and that they share further crucial properties. These variants are explicitly excluded by the current syntactic definition, as noted above. Hence, the current idea of a natural semi-modal class must be revised. The remainder of the paper is structured as follows. The next sections are concerned with semi-modal raising verbs. Section 2 briefly touches on the semantics of semi-modal verbs. Section 3 discusses the syntax of semi-modal raising verbs. Section 4 points out some problems with the definition given above. Section 5 will be about semi-modals without infinitival complements. Subsection 5.1 illustrates the usage of these constructions with selected examples taken from a newspaper corpus. Subsection 5.2 is concerned with the subject of semi-modal, infinitive-less versprechen. Subsection 5.3 deals with optional datives and with other properties of constructions without infinitival complement and compares them to raising constructions. A summary concludes the paper.

2. Semantic properties of semi-modals Most linguistic accounts define modality as a semantic category expressing necessity and possibility (e.g. Kratzer 1991, von Fintel 2006). There are some, however, which incorporate aspect, tense, and further kinds of event modifi-

300

Katarina Klein

cation to a certain degree (e.g. Dietrich 1992, Engel 2004). I subscribe to the former view. The category of modality comprises several subcategories (there is considerable dispute on the partitioning and hierarchical order of the domain, cf. Nuyts 2005 for an overview on the literature). Perhaps the most prominent categories are epistemic modality and evidentiality (the latter is not always considered to be a modal class), deontic modality, and circumstantial or dynamic modality. Epistemic and evidential modality are commonly understood as being opposed to the other classes of modality. Palmer (2001) subsumes them under the term “propositional modality” and opposes this to “event modality”, which corresponds roughly to “root modality” and comprises everything else. Putting aside these questions of nomenclature, we may view the modal component of scheinen, drohen and versprechen as belonging to the field of propositional modality. I will give some examples to briefly illustrate propositional modality: The sentences in (5) display epistemic uses of m¨ussen ‘must’ and k¨onnen ‘can’. The German modal verbs sollen ‘should’ and wollen ‘want’ may be used as quotative evidentials, as in (6). Epistemic modal operators evaluate the necessity or possibility with respect to the informational or doxastic state of the utterer (cf. Kratzer 1991). Evidential modal operators relate the modalized proposition to the source of the information of the statement (the evidence). In the case of sollen, this source is by hearsay. In the case of wollen, the referent of the subject is the quoted source. (5)

(Context: Somebody has been in this room.) a. Das muss Jan gewesen sein. ‘That must have been Jan.’ b. Das kann Jan gewesen sein. ‘That may have been Jan.’

(6)

a. b.

Er soll nichts davon gewusst haben. ‘He is supposed to have known nothing about it.’ Er will nichts davon gewusst haben. ‘He claims that he didn’t know anything about it.’

The specification of a source is neither defined in terms of necessity or possibility, nor can it be properly described by entailment scales. Thus, some authors (as, for instance, de Haan 2001) suggest that evidentiality should not be treated as a modal category. On the other hand, evidentials and epistemics have properties in common in that both lack an affirmative component of factuality, or, following Diewald’s terminology, assign the unspecified value [± nonfactual] to an utterance (cf. Diewald 2000).4

Semi-modal variation

301

Diewald relates semi-modal scheinen to evidential uses. She writes: “In contemporary German, scheinen is an unspecific marker of evidentiality. It indicates that the speaker has judged the factuality of the proposition based on some unspecified evidence” (Diewald 2000: 337, my translation). Hence scheinen is an inferential evidential. This type of evidential is closely related to epistemic modals: Inferential evidentials express an evaluation based on evidence, much like epistemic modals, which are based on the knowledge of the speaker (which obviously also includes perceived information). According to Diewald (2000), this meaning is derived from the main verb use of scheinen ‘shine’ (cf. (7a)). In the semi-modal use, scheinen has received a visual-evidential component (dem Anschein nach ‘according to appearance’, cf. (7b)) and has further been extended to non-visual uses. (7)

a. b.

Das Licht scheint durch den Stoff. ‘The light shines through the cloth.’ Der Wind scheint sich zu drehen. ‘The wind seems to be turning.’

Essentially, every semi-modal verb is a homonym of a main verb it is supposedly derived from. There is a considerable amount of literature on how epistemic uses develop by certain grammaticalization paths from other uses of modal verbs (e.g. Anderson 1986 or Bybee, Perkins and Pagliuca 1994; see Traugott 2005 for an overview). While Diewald (2000, 2001, 2004) proposes a grammaticalization of scheinen along this line, Heine and Miyahita (2004) present a comparable hypothesis on the grammaticalization path of semi-modal drohen and versprechen.5 The following examples (8) to (10) illustrate the main verb readings of drohen, pflegen, and versprechen (a) alongside the semi-modal readings (b). (8)

a. b.

(9)

a. b.

(10)

a. b.

Der Besitzer droht (dem Dieb), einen Strafantrag zu stellen. ‘The owner is threatening (the thief) to report him.’ Der Baum droht umzust¨urzen. ‘The tree threatens to topple over.’ Sie pflegt ihre Großmutter. ‘She takes care of her Grandmother.’ Sie pflegt samstags zu waschen. ‘She usually does the laundry on Saturdays.’ Die Großmutter verspricht (ihm), ihm ein M¨archen zu erz¨ahlen. ‘The grandmother promises (him) to tell him a fairy tale.’ Nachschub verspricht in den n¨achsten Minuten einzutreffen. ‘Fresh supplies promise to arrive in the next minutes.’

302

Katarina Klein

Most authors attribute epistemic modality to semi-modal drohen and versprechen, roughly ‘probably p’, where p is the propositional argument of the semi-modal (see Reis 2005: 127 for a survey on characterization of meanings). However, drohen and versprechen have some additional content. First, while versprechen indicates that the speaker has a neutral or positive attitude towards the proposition, drohen indicates a negative attitude. Obviously, these components of meaning relate to the felicity conditions of the speech act readings. The semi-modal meanings can be paraphrased as ‘there is some evidence that p will happen’, where p is an expectation or apprehension. In addition, the modal force of versprechen is stronger than that of drohen: versprechen can be paraphrased as ‘there is strong evidence that p will happen’. This could be due to a pragmatic effect: Sentences with drohen are typically used as a warning, triggering a Gricean implicature (cf. Grice 1967) ‘if nobody does anything about it’ – thus it also presupposes that p may still be prevented. Second, drohen and versprechen have an aspectual component of imminence, roughly ‘to be about to’. Drohen and versprechen prefer telic arguments and hardly ever occur with states (cf. (11)), while scheinen can combine with all sorts of events (cf. (12)). (11)

a.

Er verspricht/droht ein guter/schlechter Arzt zu werden. ‘He is likely to turn out a good/bad physician.’ b. *Er verspricht/droht ein guter/schlechter Arzt zu sein. ‘He is probably a good/bad physician.’

(12)

Er scheint ein guter Arzt zu werden/zu sein. ‘He seems to become/to be a good physician.’

Reis (2005: 130) points out the fact that epistemic modals are opposed to drohen and versprechen in this respect (cf. (13)). (13)

Er muss wohl ein guter Arzt ?? werden/sein. ‘He must(epist) ?? become/be a good physician.’

Consequently, Reis (2005) argues that drohen and versprechen are temporal and aspectual expressions rather than evidential modals. With respect to the temporal component, the proposition p is located in the future. Reis argues further that the verbs contain an aspectual component, namely that drohen and versprechen imply that the state of affairs at reference time is a kind of state that typically results in a change of state denoted by the proposition p. See the following example:

Semi-modal variation

(14)

303

a.

Nach dem Elfmeter drohte das Spiel zu kippen. ‘Nach dem Elfmeter befand sich das Spiel in einem f¨ur Kippen charakteristischen Vor- bzw. Anfangszustand.’ (Reis 2005: 129) ‘After the penalty kick the game was in a state that is a characteristic pre-state of a turn.’ b. *Nach dem Elfmeter musste (epist) das Spiel kippen. (Reis 2005: 129) ‘After the penalty kick the game had to(epist) turn.’

Reis (2005) provides example (14b) as further evidence for the difference between drohen and versprechen on the one hand and epistemic modals on the other. She argues that the latter cannot be used if the reference time is in the past.6 A third argument put forward by Reis is that epistemic modals cannot occur in questions (2005: 130): (15)

a.

Wo noch droht es heute zu regnen? ‘Where else might it be going to rain today?’ b. *Wo noch muss(epist) es heute regnen? ‘Where else must(epist) it rain today?’

It remains unclear whether Reis (2005) differentiates between evidential and epistemic modality for she points out that utterances with drohen and versprechen must be based on non-inferential evidence (she also refers to scheinen as a quasi-epistemic semi-modal). With respect to the evidence, using either of the verbs presupposes that the speaker-origo has direct access to this evidence. Reis (2005: 131) concludes that temporal-aspectual drohen and versprechen differ significantly from epistemic scheinen. It may be important to note that drohen and versprechen are fully acceptable with (atelic) event complements, especially if combined with comparative constructions as in (16). In these cases, it is the lack of a change of state that makes the complement acceptable. (16)

Die Auswertung droht (noch einige Stunden l¨anger) anzudauern. ‘The process of evaluation is about to last (a few hours longer).’

Here, we have a state that is typical for an extension (which is a change of state). There will probably be a change from a situation where the evaluation was going to be finished in a moment to another situation where the evaluation will be taking some more hours. This is in line with Reis’ observation that drohen and versprechen are mainly used in combination with situations that are analyzed as a change of state to-be:

304

Katarina Klein

Characteristics of a change of state may be taken as evidence for this change of state, but the change need not materialize in the real world. If the speaker expected a change of state (such as the termination of the process of evaluation in (16)) and then discovers that this change will not take place, drohen and versprechen are well chosen expressions. Yet, it is not the change of expectation which is crucial, since unexpected states are not acceptable with drohen and versprechen: (17)

a.

Ich dachte, sie wohnt noch in Amsterdam, aber nun scheint sie auf einmal in Berlin zu wohnen. ‘I thought she still lived in Amsterdam, but now it seems she lives in Berlin.’ b. *. . . aber nun droht/verspricht sie auf einmal in Berlin zu wohnen. ‘. . . but now I learn she probably lives in Berlin.’

Based on these findings, I propose to analyze drohen and versprechen as unspecified evidentials, comparable to scheinen, but complemented by two meaning components as mentioned above and recapitulated below. The differences in (14) and (15) are then accounted for, since evidentials allow for both uses, i.e. they pattern with drohen and versprechen: (18)

a. b.

Zuerst wollte er nichts davon gewusst haben, . . . ‘First he claimed not to have known anything about it . . . ’ Wo noch will er gestern gewesen sein? ‘Where else does he claim to have been yesterday?’

The first additional component evaluates the proposition as a neutral or welcome (versprechen) or as an undesirable prospect (drohen) according to the attitude of the speaker. The second is the aspectual component. According to evidence, the speaker assumes that a (failure of a) change of state will take place, in other words: Evidence suggests that the situation is in a state which is presumably changing to p. One semi-modal remains to be dealt with. The use of pflegen indicates that the expressed proposition is habitual, customary, or otherwise iterative. Pflegen sounds stilted to most speakers and is used less frequently than the other semimodals. In most instances, some temporal (19a), local (19b) or circumstantial context (19c) provides the frame of reference in which we understand this specific occurrence as typical. Sometimes (as in (1c) above), this context remains implicit. The proposition denotes a factual, repetitive event. Thus, semi-modal pflegen is not an evidential or epistemic modal.

Semi-modal variation

(19)

a. b. c.

305

In den Semesterferien pflegt Caro zu arbeiten. ‘Caro usually works during the semester break.’ Hier pflegt es oft zu regnen. ‘Usually, it tends to rain here.’ Ihm pflegt beim Fahren schlecht zu werden. ‘He tends to feel sick while driving.’

To sum up: The semi-modals scheinen, drohen, and versprechen may best be described as evidential modals signaling that there is some unspecified evidence for the modalized proposition. Drohen and versprechen have an additional temporal-aspectual function of imminence. The semantics of pflegen differs from that of the other verbs. Pflegen expresses some kind of habit or recurring event. These findings cast considerable doubt on the premise that semi-modal verbs form a natural semantic class.

3. Syntactic properties of semi-modals Syntactically, semi-modals are characterized by their selectional properties. They select for an infinitival complement requiring a specific form of this complement (second status/zu-infinitive), and they adopt or raise the suppressed infinitival subject in such a way that it is realized as their own subject. In addition, they form coherent constructions (Bech 1955); they seem to establish a very firm relationship with their complement so that complements and modifiers of the selected head can appear in positions where they would be realized if they were complements or modifiers of the semi-modal matrix verb itself. Furthermore, an embedded negative element may – and often must – have scope over the semi-modal verb (coherent negation). In this section, I will shortly illustrate these properties. A short discussion of optional dative arguments completes the section.

3.1 Status government The label “semi-modal” is mainly due to the fact that the respective verbs resemble modal verbs in that they select for an infinitival complement as modals do, but that their complements differ formally from those of modals. Governed infinitival verbs can be realized in three different statuses. The term “status” was coined by Bech (1955). While modal verbs govern a bare infinitive (first status, cf. (20a)), semi-modals govern an infinitive marked

306

Katarina Klein

by zu (second status, cf. (20b)). A participle (third status) is selected by passive and perfect auxiliaries (cf. (20c)). (20)

a. b. c.

3.2

Wir m¨ussen/k¨onnen/d¨urfen/. . . uns oft treffen. ‘We must/can/may/. . . meet frequently.’ Wir scheinen uns oft zu treffen. ‘We seem to meet frequently.’ Wir haben uns oft getroffen. ‘We have met frequently.’

Raising vs. control

Semi-modals are raising-to-subject verbs. This is a crucial deviation from the behavior of the speech act verbs drohen and versprechen (see examples (8b) and (10b) in Section 2): If used in the speech act reading, drohen and versprechen are control verbs. The distinction between raising and control verbs was not introduced by Bech (1955), but originated in the framework of generative grammar. It can be found in Chomsky (1965) and was elaborated on by Rosenbaum (1967). Raising verbs do not possess a semantic subject but require a subject according to the restrictions that their infinitival complement poses on its subject. In transformational grammar, the subject is thus said to be moved or raised from the subject position of the embedded verb to the matrix subject position. The term “raising” has also been adopted by non-transformational frameworks as in Head-Driven Phrase Structure Grammar (HPSG) (e.g. Pollard and Sag 1994). In HPSG, the subject requirement of the raising verb is said to be identical with that of its complement. The contrast between raising and control verbs is revealed by the following tests. Since raising verbs are semantically not related to their subjects, they allow expletive subjects while control verbs do not. Thus, the sentences in (21) and (22) only make a semi-modal reading available as in (a). A speech act reading as in (b) cannot be obtained. (21)

Es droht zu hageln. a. ‘It is probably going to hail.’ (negative expectation) b. *‘It(expl) is threatening that it will hail.’

(22)

Es verspricht ein Gewitter zu geben. a. ‘There is probably going to be a thunderstorm.’ b. *‘It(expl) promises that there will be a thunderstorm.’

Semi-modal variation

307

A second way of testing raising and control is by passivizing the infinitive. If the matrix verb is a raising verb, each sentence logically implies the other one, as in (23). This test also works with other passive forms as, for instance, the recipient passive (cf. (24)). (23)

a. b.

(24)

a.

b.

Der Hund droht den Eimer umzuwerfen. ‘The dog will probably knock over the bucket.’ Der Eimer droht vom Hund umgeworfen zu werden. ‘The bucket will probably be knocked over by the dog.’ Die K¨onigin verspricht dem Schauspieler eine Medaille zu u¨ berreichen. ‘The queen is about to give a medal to the actor.’ Der Schauspieler verspricht von der K¨onigin eine Medaille u¨ berreicht zu bekommen. ‘The actor is about to be given a medal by the queen.’

Since a control verb assigns a semantic role to its subject, control sentences with active and passive infinitive do not logically imply each other. The sentences in (25a) may be true while those in (25b) are false, and vice versa. (25)

a. b.

Bettina droht/verspricht, Tim zu k¨ussen. ‘Bettina threatens/promises to kiss Tim.’ Tim droht/verspricht, von Bettina gek¨usst zu werden. ‘Tim threatens/promises to be kissed by Bettina.’

One last test should be mentioned.7 Usually, German raising verbs embed impersonal constructions. Examples for German impersonal constructions are psychverbs as in (26a) and (26b), or the passive of an intransitive verb (26c). (26)

a.

b.

c.

Mich frier-t. me.acc be.cold-3sg ‘I’m freezing.’ Mir grau-t vor Hunden. me.dat be.terrified-3sg of dogs ‘I’m scared of dogs.’ Hier wird gearbeitet. here aux.3sg work.pass ‘Work is being done here.’

308

Katarina Klein

As expected, scheinen and pflegen are able to embed impersonal constructions. However, drohen and versprechen differ in this respect. Most speakers judge examples as in (27c) and (27d) as ungrammatical: (27)

a.

Hier scheint gearbeitet zu werden. ‘It seems that work is being done here.’ b. Hier pflegt dienstags getanzt zu werden. ‘Usually, people dance here on Tuesdays.’ c. *? Hier verspricht am Dienstag getanzt zu werden. ‘Probably there will be dancing here at Tuesday.’ d. *? Hier droht ab morgen gearbeitet zu werden. ‘Probably someone will work here from tomorrow on.’

Kiss claims that there are three different patterns of verbal orientation (1995: 81–88). For raising-to-subject and subject control verbs, these variants are as follows: First, raising verbs as scheinen and pflegen share their complete subject subcategorization value with that of their complements, which may thus be empty. Next, raising verbs as drohen and versprechen share their subject with that of their complements. In consequence, a complement that does not provide for a subject leads to an ungrammatical sentence. Subject control verbs, finally, share only the index value of their subject with the index of the infinitival subject. Thus, the differences in grammaticality in (27) are due to the fact that drohen and versprechen belong to another class than scheinen and pflegen. However, impersonal constructions that are not passives lead to much more acceptable sentences. The following examples (taken from grammis8 ) are fully grammatical: (28)

a. b. c.

(29)

a. b. c.

Ihm scheint schlecht zu werden. ‘It seems (as if) he is getting sick.’ Ihm pflegt schlecht zu werden. ‘He tends to feel sick.’ Ihm droht schlecht zu werden. ‘It is likely that he will feel sick.’ Ihn scheint zu schaudern. ‘It seems that he shudders.’ Ihn pflegt an dieser Stelle immer zu schaudern. ‘Usually, he shudders at this point.’ Ihn droht zu schaudern. ‘It is probable that he is going to shudder.’

Semi-modal variation

309

The semantics of versprechen requires that the embedded proposition denotes an event that is not unwelcome. Since impersonal constructions other than impersonal passives hardly denote positive events, sentences analogous to (28) and (29) are highly marked. That this is not a syntactic restriction of versprechen can be demonstrated by adjusting the context to obtain a positive evaluation. For instance, sentence (30) is perfectly grammatical if uttered in the medical context of a thermotherapy. (30)

Ihr verspricht gleich warm zu werden! ‘She probably will feel warm in a moment!’

To sum up, while drohen and versprechen (but not pflegen and scheinen) are restricted with respect to impersonal passives (cf. (27)), they are not restricted with respect to other impersonal constructions (cf. (28) to (30)). Gunkel (2000) shows that the difference in (27) is due to a selectional restriction of the semi-modals based on this observation. He argues that semi-modal drohen and versprechen select for non-intentional events while the control readings select for intentional events. Since impersonal passivization is restricted to intentional events, this predicts correctly the ungrammaticality of (27c) and (27d). Other impersonal constructions (as in (28) and (29)) are not excluded by this explanation. Some verbs are underspecified for intentionality, e.g. zerschlagen ‘to break’ or u¨ berfahren ‘to run over’. If semi-modal drohen or versprechen embed these, contextual disambiguation forces a non-intentional reading of the denoted event. Thus, Gunkel predicts that (31) allows for the readings (a) and (b), but not for the readings (c) or (d).9 (31)

Karl drohte das Geschirr zu zerschlagen. (Gunkel 2000: 116) a. ‘It was probable that Karl would (accidentally) break the dishes.’ b. ‘Karl threatened to smash the dishes (on purpose).’ c. # ‘It was probable that Karl would smash the dishes on purpose.’ d. # ‘Karl threatened to (accidentally) break the dishes.’

However, interpretation (31c) cannot be ruled out completely. Furthermore, there are other counterexamples to the hypothesis, cf. (32). All of the following events need an intentional, controlling agent. (32)

a.

b.

Die feindlichen Truppen drohten die Stadt zu erobern. (Gunkel 2000: 117) ‘The hostile troops were about to conquer the town.’ Oh nein! Er droht sie zu k¨ussen! ‘Oh dear! He might kiss her!’

310

Katarina Klein

c.

Bitte tu schnell etwas, der Vorsitzende droht zur¨uckzutreten! ‘Please do something; the president may resign any moment!’

Taking this data into account, Gunkel suggests that the complements in (32) undergo a semantic shift. He argues that the intentional events in question are conceptualized as non-intentional events, i.e. that they are perceived as being ‘out of control’ (Gunkel 2000: 118).

3.3

Coherence

Bech (1955) calls a construction consisting of a status-governing verb and a governed infinitive coherent if both verbal phrases are entangled in such a way that there seems to be no boundary between them. If such a construction is not coherent, it is called incoherent. The most prominent examples for incoherent constructions are extraposed infinitives, as appear with the speech act readings of drohen and versprechen. (33)

a. b.

. . . dass ich dir drohe, dich jeden Tag danach zu fragen. ‘. . . that I threaten you to ask you about it every day.’ . . . dass ich dir verspreche, dich morgen anzurufen. ‘. . . that I promise you to give you a call tomorrow.’

The infinitival complement of a semi-modal verb must not be extraposed as demonstrated in the following examples (b).10 Thus, semi-modals are called “obligatorily coherent”. (34)

a. . . . dass es morgen zu regnen droht. b. *. . . dass es droht, morgen zu regnen. ‘. . . that it threatens to rain tomorrow.’

(35)

a. . . . dass es ziemlich oft zu regnen pflegt. b. *. . . dass es pflegt, ziemlich oft zu regnen. ‘. . . that it tends to rain quite often.’

(36)

a. . . . dass es zu regnen scheint. b. *. . . dass es scheint, zu regnen. ‘. . . that it seems to rain.’

(37)

a. . . . dass es morgen zu regnen verspricht. b. *. . . dass es verspricht, morgen zu regnen. ‘. . . that it promises to rain tomorrow.’

Semi-modal variation

311

Another property of coherent constructions is that negative elements embedded in the infinitival complement can have scope over the matrix verb. The preferred reading of (38) is (38a), though reading (38b) is also possible (the preference is inverted for drohen because of felicity conditions). (38)

... ... a. b.

weil es keinen Nachtisch zu geben verspricht. because expl no dessert to be promises ‘It is not likely that there will be a dessert.’ ‘It is likely that there will be no dessert.’

Bech (1955) mentions further properties that distinguish coherent and incoherent constructions (see von Stechow and Sternefeld 1988: 406–410 or Kiss 1995: 27–31 for a brief summary). For the purposes of this article, there is no need to list all of them. To sum up: Semantically, semi-modals have only one argument, which is the modalized proposition. Syntactically, they may adopt all or part of the valency of the embedded infinitive. In a sentence with a semi-modal predicate, all nominal arguments depend semantically on the infinitival verb (cf. also Zifonun, Hoffmann and Strecker 1997: 1282–1285). There is only one exception to this observation, and this exception will be the subject of the next subsection: scheinen allows for optional dative arguments.

3.4 Dative arguments Semi-modals differ from each other with respect to another property: While scheinen allows for an optional dative argument, the other verbs do not. (39)

¨ Der Gaspreis scheint mir mit dem Olpreis zu steigen. ‘It seems to me that gas prices are increasing with the oil prices.’ b. *Der Gaspreis verspricht mir bald zu fallen. ‘The gas prices promise me to decrease soon.’ c. *Der Gaspreis droht mir wieder zu steigen. ‘The gas prices threaten me to increase again.’ ¨ d. *Der Gaspreis pflegt mir mit dem Olpreis zu steigen. ‘Usually to me, the gas prices increase with the oil prices.’

a.

The dative argument in (39a) bears the semantic experiencer role. In the particular case of scheinen, this corresponds to the modal evaluator. Since scheinen allows for explicit coding of the modal evaluator, scheinen may be used descriptively, i.e. to attribute the modal attitude to somebody else but the speaker

312

Katarina Klein

(cf. Nuyts 2005: 15). Semi-modal drohen and versprechen do not allow for an optional dative argument. (40)

a.

b.

. . . nach langwierigen Verhandlungen, die sich (*ihm) schon zu zerschlagen drohten. (Askedal 1997: 13) ‘. . . after tedious negotiations that tended (*to him) to fail.’ . . . die großen Erfindungen, die (*ihm) dieser Streit hervorzurufen versprach. (Askedal 1997: 13) ‘. . . the great inventions that this dispute promised (*him) to produce.’

Consequently, neither drohen nor versprechen may be used descriptively. This means that the modal evaluator is always identical to the speaker. If embedded in the subordinate clause of a speech act verb, the evaluator corresponds to the contextually given speaker, i.e. the agent of the matrix verb (41a). This also holds for scheinen (41b). (41)

a. b.

Jan sagt, dass das Fest ein Erfolg zu werden verspricht. ‘Jan says that the party is probably becoming a success.’ Jan sagt, dass (*dir) das Fest ein Erfolg zu werden scheint. ‘Jan says that the party seems (*to you) to become a success.’

4. Problems In the previous sections, semi-modals have been described as a natural class of second-status-governing raising-to-subject verbs that form coherent constructions and express modality in a rather broad sense. This definition faces at least four problems. As Gunkel (2000) and Reis (2005) note, not only semi-modals, but also phase verbs (such as beginnen ‘to start’, anfangen ‘to start’, and aufh¨oren ‘to stop’) exhibit the syntactic properties mentioned. In her investigation of semi-modal drohen and versprechen, Reis (2005: 143) concludes that these two so-called semi-modals are more likely to form a natural class with phase verbs than with epistemic scheinen and other modal verbs. As delineated in Section 2, there are fundamental differences in the semantics of the four verbs. According to Diewald (2001), scheinen may best be described as an evidential marker signaling that there is some strong (but unspecified) evidence for the modalized proposition. The semi-modals drohen and versprechen can be analyzed as evidential verbs as well, but pflegen indicates that the modi-

Semi-modal variation

313

fied proposition is a habitual, repetitive event. Thus, there is considerable doubt that semi-modal verbs form a natural semantic class. A third problem for the given definition of semi-modal verbs lies in the observation that semi-modals differ syntactically, too. While scheinen allows for an additional dative argument, none of the other verbs do. Neither scheinen nor pflegen necessarily require a syntactic subject, i.e. they embed impersonal constructions. It is controversial whether corresponding constructions with drohen and versprechen are grammatical, ungrammatical or slightly awkward (cf. Section 3.2). Finally, semi-modals license constructional variants without infinitival complements. While scheinen can be realized with an expletive subject and a finite complement clause (cf. (2)), drohen and versprechen may construe with nominal complements. To a certain extent, the same holds for pflegen. These variants are explicitly excluded by the current syntactic definition given above – though they intuitively have the same meaning as their raising variants. In the next section, I will focus on two of these constructions, namely drohen and versprechen without infinitival complements.

5. Semi-modals without infinitival complements 5.1 Semi-modal valence alternation This section is concerned with some constructional variants of the semi-modals that are not realized with infinitival complements, in particular those of drohen and versprechen. These constructions are syntactic variants of the semi-modals, i.e. they share the semi-modal meaning of their raising counterparts, but differ with respect to their syntactic realization. Interestingly, they resemble the raising verbs with respect to further properties: In some cases, they are realized with an additional dative argument that is semantically related only to the modified argument. We saw above that scheinen may be realized with an expletive subject and a finite complement clause as propositional complement (cf. (2)). In transformational syntax, this variation is explained by raising. The infinitival construction derives from the construction with a finite complement clause: Though scheinen has no semantic subject, it opens a syntactic argument slot for a syntactic subject. Thus, it is possible to raise the embedded subject into this position. Raising is optional; if it does not take place, the subject position is filled by an expletive.11 As the variations of scheinen have been studied in various works (e.g. Askedal 1998, Diewald 2000, 2001, etc.), I will not further address this issue here.

314

Katarina Klein

At first glance, the same construction seems to be possible with drohen, cf. (42a). But, as (42b) and (42c) demonstrate, the pronoun in (42a) is merely a placeholder (examples from grammis12 ). It is only realized if no other constituent occupies the first position of a verb-second clause. In contrast, the expletive in (2) is an obligatory syntactic argument of scheinen (cf. (42d)). (42)

a.

Es expl b. Seit since c. Dass that d. *Seit since

droht (seit langem), dass Hans gewinnt. threatens (since long) that Hans wins langem droht, dass Hans gewinnt. long threatens that Hans wins Hans gewinnt, droht seit langem. Hans wins threatens since long langem scheint, dass Hans gewinnt. long seems that Hans wins

This variant with a sentential subject is very rare (e.g. only one example in the NZZ 1993 corpus). Since nominal and sentential subjects are interchangeable in many contexts (e.g. with psych verbs), this construction resembles the nominal subject variant of drohen to a large extent as in (3) or the following example (NZZ 1993). (43)

Wegen des in Richtung der Bucht von Bengalen treibenden brennenden ¨ a¨ usserten die indonesischen Schiffahrtsbeh¨orden die Besorgnis, dass Ols ¨ eine Olpest drohe. ‘Because of the burning oil slick that was drifting towards the Bay of Bengal the maritime authorities of Indonesia expressed their concerns about environmental and ecological damage that might be about to occur.’

Semi-modal versprechen can be realized with two nominal complements, where the propositional argument is realized as an accusative-marked complement and cannot be realized as a finite complement clause (cf. also (4)). (44)

Die zu erwartende Zinskonvergenz bei Staaten, die jetzt den Wechsel zum Euro noch nicht vollzogen haben, verspricht gerade in diesen L¨andern in naher Zukunft attraktive Resultate. (NZZ 1993) ‘The likely interest rate convergence of those countries that have not yet introduced the Euro promises attractive returns for assets of such countries in the future.’

The ‘propositional argument’ may refer to a situation or event. Since derived nominalizations in German never denote facts, but events or results, this distinc-

Semi-modal variation

315

tion is not reflected in the grammar (cf. Ehrich 1991: 448).Apart from combining with deverbal nominals, drohen and versprechen are realized with event nouns ¨ as Olpest (roughly ‘the process and result of oil contaminating entire coasts’) or descriptions for situations as Gefahr ‘risk, danger’. Somewhat surprisingly, perhaps, infinitive-less versprechen has an additional referential argument, its nominal (or sentential) subject. Its existence contradicts the assumption that semi-modal verbs have only one semantic argument, namely the modified proposition, while all other arguments depend on the embedded verb. The subject in (44) cannot be a raised argument, since there is no subcategorizing head for it. The semantic function of the subject will be the topic of the next section. Pflegen, finally, may be used with a nominal subject and a nominal complement that denotes some tradition, habit or activity. The readings of these sentences are very close to the so-called “semi-modal” readings, but they are restricted to activities. In this reading, the agentive subject regularly participates in the denoted activity. This variant of pflegen does not express modality in the strict sense, just like the raising variant. (45)

a.

b.

Pflegen Sie deshalb regelm¨assig das ungezwungene Gespr¨ach zwischen den Generationen. (NZZ 1993) ‘Therefore, you should maintain the informal chat between different generations on a regular base.’ Die ,,Neue Z¨urcher Zeitung“ pflegt die Verwendung des inkriminierten Begriffs in liebevoller Weise. (NZZ 1993) ‘The NZZ frequently and carefully uses this incriminated expression.’

5.2 The subject of versprechen Unlike the raising verb versprechen, semi-modal versprechen has two arguments: a complement that corresponds to the modified argument, and a subject. This section is concerned with this subject’s semantic role. (46) presents two hypotheses about the semantic structure of versprechen: (46)

a. b.

subject causes that the speaker thinks that probably p ‘subject makes the speaker believe that probably p.’ the speaker thinks that probably subject causes p ‘The speaker thinks that this is probable: subject will cause p.’

316

Katarina Klein

According to the first hypothesis (46a), the subject serves to specify some evidence for the modal evaluation that probably p. According to the second hypothesis (46b), the speaker believes that probably the subject referent will cause p. The modal operator has scope over the subject only according to the second hypothesis. In both readings the subject is some sort of a cause: in (a) it is the evidence that causes the modal evaluation, in (b) it is a proper causer that causes p in some way or another. Hypothesis (46b) is easily falsified: Clearly, (47) (NZZ 1999) has reading (a), not (b), i.e. the cause-relation is not in the scope of the modal operator. The subject itself is never modalized. (47)

Eifriges Bew¨assern verspricht aber eine recht ansehnliche eager irrigating promises but a quite respectable Ernte. harvest a. ‘But if the fields are irrigated eagerly, there will probably be a good harvest.’ b. # ‘Probably irrigating eagerly will bring about a good harvest.’

The following sentence provides further support for hypothesis (a). In (48), there is no direct causation between the subject (that she did well in the last race) and the proposition (roughly, that she will do well in the next race). (48)

Den kurvigen Abschnitt [. . . ] fuhr sie aber sehr stark, was f¨ur den Super-G vom Freitag doch wieder einiges verspricht. (NZZ 1999) ‘That she did really well on those turns is promising with respect to the Super-G on coming Friday.’

This reading is not restricted to sentential subjects, of course. As before, every sentential subject can easily be transformed into a nominal subject. (49)

Ihr Abschneiden in der Qualifikation verspricht einen spannenden Super-G. ‘Her final scores in the qualification promise an interesting Super-G.’

The semantic analysis of versprechen as evidential (as given in Section 2) allows an integrated treatment of all semi-modal variants, since the evidential argument has a natural place in the argument structure of the evidential verb.

Semi-modal variation

317

5.3 Dative and other arguments Interestingly, infinitive-less drohen is frequently realized with an optional dative argument as in (50) and (52a) below. Unlike optional dative-arguments of scheinen, these datives do not bear an experiencer role, i.e. they are not semantically related to the matrix verb. Instead, they are related to the modalized argument. (50)

Dem Meister droht gar das Ausscheiden in der Hallenmeisterschaft. ‘The elimination in the indoor championship is threatening the champion.’

In (50), the understood subject of the deverbal nominal Ausscheiden ‘being eliminated’is coindexed with the dative argument dem Meister, i.e. it is the champion who will be eliminated. The dative argument is not a semantic argument of the matrix verb. Thus, this brings to mind raising constructions comparable to socalled “raised possessors”. Datives (and other arguments) that are semantically related only to another argument and are not direct arguments of the verb belong to the field of “extra argumentality” (Hole 2006). The semantic relations in (50) are revealed by sentence (51), which paraphrases the infinitive-less sentence (50) by using the raising variant of drohen. (51)

Der Meister droht in der Hallenmeisterschaft auszuscheiden. ‘The champion may even be facing elimination from the indoor championship.’

The semantic role of the dative in the next sentence (52) is not that obvious mainly because Gefahr ‘risk, danger’ is not a deverbal nominal. Furthermore, possessor constructions such as seine Gefahr ‘his risk’ are only used in juridical contexts. However, the paraphrases in (52b) and (52c) provide support for an extra argument hypothesis. (52)

a.

b. c.

Die wirkliche Gefahr, die ihm droht, besteht [. . . ] in der Abstumpfung [. . . ] (NZZ 1993) ‘The actual risk that is threatening him is the dulling.’ Er droht Gefahr zu laufen, . . . he threatens risk to run Er droht abzustumpfen. he threatens to.become.dull

318

Katarina Klein

Such datives are also possible (but rare) with versprechen. As above, the dative fulfills a semantic role with respect to the proposition (‘the bandits make a good haul’ in (53a), ‘he finds protection and a dry place’ in (53b)). (53)

a.

b.

Allerdings versprechen die Touristen, auf die man trifft, eventuellen R¨aubern h¨aufig keine grosse Beute. (NZZ 1993) ‘However, the tourists encountered often do not promise a good haul to potential bandits.’ ¨ Uber vereiste B¨ache gelangte er tiefer und fand schliesslich diesen Felsen, der ihm Schutz und Trockenheit versprach. (NZZ 1993) ‘. . . eventually, he found this rock that promised him protection and a dry spot.’

The paraphrases challenge the assumption that the datives in question are beneficiaries (that have an advantage or a disadvantage from the event) or recipients (that receive something) in the narrow sense, though the argument in (52) might be a maleficiary and that in (53) might be a recipient. The dative in (51) bears neither role. At any rate they are not agents, either. Hole (2006) claims that all extra arguments denote affectees – a role that combines proto-agentive properties (as consciousness) and proto-patientive properties (as causal affectedness) in the sense of Dowty (1991). The datives in the examples fulfill these requirements. However, further research on this issue needs to be conducted.

6. Summary In this paper I have investigated some variants of semi-modal verbs. Semimodal verbs are said to be characterized by the following properties: They select for infinitival complements, they govern the second status, they form obligatorily coherent constructions and they are raising verbs. In German, this class comprises four verbs: drohen ‘threaten’, pflegen ‘usually do/happen’, scheinen ‘seem’, and versprechen ‘promise’. The semantics of semi-modal drohen and versprechen has been outlined. Based on Diewald’s (2000, 2001) analysis of scheinen, I have classified the semi-modal verbs drohen and versprechen as evidential modals. The modal component of these verbs is supplemented by two additional meaning components. The first is a positive or negative attitude towards the proposition, the second is an aspectual component (cf. Reis 2005). Pflegen does not express modality in the strict sense, it has a habitual or iterative reading.

Semi-modal variation

319

The view of semi-modals as a natural class has been called into question before (Gunkel 2000, Reis 2005), and this paper has listed further problems that the definition given faces. I have focused on two semi-modal variants of drohen and versprechen that challenge this definition. These frequently used variants do not govern an infinitival complement, but select for nominal complements. The modalized proposition is realized as subject in the case of drohen, but as object in the case of versprechen. In contrast to drohen, infinitive-less versprechen is realized with a subject specifying the evidence that the modal evaluation is based on. Furthermore, extra dative arguments occur with either variant. These arguments have a semantic relation only with respect to the propositional argument and are not direct arguments of the semi-modal verb. The current definition needs to be amended since it excludes these variants. Future research should be extended to the class of the so-called Modalit¨atsverben (‘modality verbs’), which also convey notions of aspect, modality and modification. The issue of extra argumentality of infinitive-less semi-modals deserves special attention, since a comparison to coherence phenomena might yield promising results.

Notes ∗

1.

2.

3. 4.

This paper has benefited from discussions with many people. In particular, I would like to thank Stefan Engelberg, Peter Gallmann, Daniel Hole, Tibor Kiss and Jacob Mach´e. Special thanks are due to Neil Hickey, Martin Hoelter, Philipp Paulus and Jan Strunk for helpful comments and editorial help. This paper also benefited greatly from comments by an anonymous reviewer. Zifonun, Hoffmann and Strecker (1997) exclude versprechen from the set of semimodal verbs because of selectional restrictions with passivized infinitives, see Section 3.2. Zifonun, Hoffmann and Strecker (1997) slightly deviate from this view. Ignoring differences in terminology, they also mention coherence and raising but add that semi-modals do not have a valency of their own (a view that is challenged with respect to scheinen in Section 3.4) but adopt the valency of the embedded infinitive. In addition, another commonality with modal verbs is mentioned, namely the restriction that semi-modals lack a perfect participle. Askedal (1998) and Diewald (2000, 2001) discuss constructional variants of scheinen. This corresponds to Dietrich’s definition of modality (1992: 24): According to Dietrich, a sentence is modalized if the proposition expressed is not marked as factual, i.e. if the speaker does neither assert nor negate the factuality of the proposition by uttering the sentence.

320

Katarina Klein

5. Traugott (1997) describes the diachronic development of English epistemic threaten and promise. To my knowledge, there is no comparable diachronic investigation on the use of German drohen and versprechen. 6. The intended epistemic readings of (13a) and (14b) are available in free indirect style, though. The following example should provide a context in which the epistemic reading is even more acceptable: Ich wusste, dass er da sein musste (sein Auto stand doch vor der T¨ur). ‘I knew he had to be there (his car was standing in front of the house).’ I would like to thank an anonymous reviewer for pointing out the relevance of this stylistic device. 7. For an overview of the phenomenon and further tests cf. Davies and Dubinsky (2004). ¨ 8. grammis, subsection Ubereinstimmungen zwischen Halbmodalen und Modalverben (last revised: 2002/10/24). grammis is an online information system for German grammar, edited by the Institut f¨ur deutsche Sprache (IdS) in Mannheim. It has been available to the public since 2000 and is constantly updated. See references for the URL. 9. In written language, coherent and incoherent constructions are distinguished by punctuation, i.e. the control reading of (31) would be Karl drohte, das Geschirr zu zerschlagen. Gunkel perceives (31a) as rather awkward. 10. To many speakers, drohen and versprechen are more acceptable with an extraposed complement than scheinen or pflegen. This might be due to the speech act homonyms of drohen and versprechen: Since the infinitival complement of a speech act verb is mostly realized in extraposed position, the extraposed construction sounds familiar also with the semi-modals. 11. This account does not explain why no expletive subject is inserted in the case of infinitival impersonal constructions as in (27a). 12. grammis, subsection Zum generativen Erkl¨arungsmodell f¨ur Konstruktionen mit Halbmodalen (last revised: 2002/03/01).

References Anderson, Lloyd B. 1986 Evidentials, paths of change, and mental maps: typologically regular asymmetries. In Evidentiality. The Linguistic Coding of Epistemology, Wallace Chafe and Johanna Nichols (eds.), 273–312. Norwood:Ablex. Askedal, John Ole 1997 drohen und versprechen als sogenannte ,,Modalit¨atsverben“ in der deutschen Gegenwartssprache. Deutsch als Fremdsprache 34(1): 12– 19. 1998 Satzmustervariation und Hilfsverbproblematik beim deutschen Verb scheinen. In Deutsche Grammatik – Thema in Variationen. Festschrift f¨ur Hans-Werner Eroms zum 60. Geburtstag, Karin Donhauser and Ludwig M. Eichinger (eds.), 49–74. Heidelberg: Winter.

Semi-modal variation Bech, Gunnar 1955

321

Studien uber ¨ das deutsche verbum infinitum. Part 1. Købnhavn: Munksgaard. Bybee, Joan, Revere Perkins, and William Pagliuca 1994 The Evolution of Grammar. Tense, Aspect, and Modality in the Languages of the World. Chicago/London: University of Chicago Press. Chomsky, Noam 1965 Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press. Davies, William D., and Stanley Dubinsky 2004 The Grammar of Raising and Control. A Course in Syntactic Argumentation. Oxford: Blackwell. de Haan, Ferdinand 2001 The relation between modality and evidentiality. In Modalit¨at und Modalverben im Deutschen, Reimar M¨uller and Marga Reis (eds.), 201–216. Hamburg: Buske. Dietrich, Rainer 1992 Modalit¨at im Deutschen. Opladen: Westdeutscher Verlag. Diewald, Gabriele 2000 scheinen als Faktizit¨atsmarker. In Wortschatz und Orthographie in Geschichte und Gegenwart. Festschrift f¨ur Horst Haider Munske zum 65. Geburtstag, Mechthild Habermann, Peter O. M¨uller, and Bernd Naumann (eds.), 333–355. T¨ubingen: Niemeyer. 2001 Scheinen-Probleme: Analogie, Konstruktionsmischung und die Sogwirkung aktiver Grammatikalisierungskan¨ale. In Modalit¨at und Modalverben im Deutschen, Reimar M¨uller and Marga Reis (eds.), 87–110. Hamburg: Buske. 2004 Faktizit¨at und Evidentialit¨at: Semantische Differenzierungen bei den Modal- und Modalit¨atsverben im Deutschen. In Tempus / Temporalit¨at und Modus / Modalit¨at im Sprachenvergleich, Oddleif Leirbukt (ed.), 231–258. T¨ubingen: Stauffenburg. Dowty, David R. 1991 Thematic proto-roles and argument selection. Language 67(3): 547–619. Ehrich, Veronika 1991 Nominalisierungen. In Semantik / Semantics. Ein internationales Handbuch der zeitgen¨ossischen Forschung. An International Handbook of Contemporary Research, Arnim von Stechow and Dieter Wunderlich (eds.), 441–458. Berlin/New York: Mouton de Gruyter. Eisenberg, Peter 1999 Grundriss der deutschen Grammatik. Vol. 2: Der Satz. Stuttgart: Metzler. Engel, Ulrich 2004 Deutsche Grammatik. Neubearbeitung. M¨unchen: Iudicium.

322

Katarina Klein

von Fintel, Kai 2006

Modality and language. In Encyclopedia of Philosophy. 2nd ed., vol. 6, Donald M. Borchert (ed.), 20–27. Detroit: MacMillan.

grammis grammis. Das grammatische Informationssystem des Instituts f¨ur deutsche Sprache (IdS), Mannheim. Mannheim: Institut f¨ur deutsche Sprache. (URL: http://www.ids-mannheim.de/grammis, online since 2000, continuously updated) Grice, H. Paul 1967 Gunkel, Lutz 2000

Logic and conversation. Ms. Published 1975 in Speech Acts, Peter Cole and Jerry L. Morgan (eds.), 41–58. New York: Academic Press.

Selektion verbaler Komplemente. Zur Syntax der Halbmodal- und Phasenverben. In Deutsche Grammatik in Theorie und Praxis, Rolf Thieroff, Matthias Tamrat, Nanna Fuhrhop and Oliver Teuber (eds.), 111–121. T¨ubingen: Niemeyer. Heine, Bernd, and Hiroyuki Miyashita 2004 Drohen und versprechen – zur Genese von funktionalen Kategorien. Neue Beitr¨age zur Germanistik 3(2): 9–33. Hole, Daniel 2006 Extra argumentality – affectees, landmarks and voice. Linguistics 44(2): 383–424. Kiss, Tibor 1995 Infinite Komplementation. Neue Studien zum deutschen Verbum infinitum. T¨ubingen: Niemeyer. 2005 Subjektselektion bei Infinitiven. In Der Infinitiv im Deutschen, JeanFran¸cois Marillier and Claire Rozier (eds.), 115–132. T¨ubingen: Stauffenburg. Kratzer, Angelika 1991 Modality. In Semantik / Semantics. Ein internationales Handbuch der zeitgen¨ossischen Forschung. An International Handbook of Contemporary Research, Arnim von Stechow and Dieter Wunderlich (eds.), 639–650. Berlin/New York: Mouton de Gruyter. Nuyts, Jan 2005 Modality: Overview and linguistic issues. InThe Expression of Modality, William Frawley (ed.), 1–26. Berlin/NewYork: Mouton de Gruyter. Palmer, Frank R. 2001 Mood and Modality (2nd edition). Cambridge: Cambridge University Press. Pollard, Carl, and Ivan A. Sag 1994 Head-Driven Phrase Structure Grammar. Chicago: University of Chicago Press.

Semi-modal variation Reis, Marga 2005

323

Zur Grammatik der sog. ,Halbmodale‘ drohen/versprechen + Infinitiv. In Deutsche Syntax: Empirie und Theorie. Symposium in G¨oteborg 13.–15. Mai 2004, Franz Josef d’Avis (ed.), 125–145. G¨oteborg: Acta Universitatis Gothoburgensis. Rosenbaum, Peter S. 1967 The Grammar of English Predicate Complement Constructions. Cambridge, Mass.: MIT Press. von Stechow, Arnim, and Wolfgang Sternefeld 1988 Bausteine syntaktischen Wissens. Opladen: Westdeutscher Verlag. Traugott, Elizabeth Closs 1997 Subjectification and the development of epistemic meaning: The case of promise and threaten. In Modality in Germanic Languages. Historical and Comparative Perspectives, Toril Swan and Olaf Jansen Westvik (eds.), 185–210. Berlin/New York: Mouton de Gruyter. 2005 Historical aspects of modality. In The Expression of Modality, William Frawley (ed.), 107–139. Berlin/New York: Mouton de Gruyter. Zifonun, Gisela, Ludger Hoffmann, and Bruno Strecker 1997 Grammatik der deutschen Sprache. 3 volumes. Berlin/New York: de Gruyter.

Variation in Komi object marking∗ Gerson Klumpp Abstract. Komi, a Uralic language in Northern Russia, shows a typologically widespread differential object marking pattern in which direct objects ranging high on the scales of definiteness and/or animacy are accusative-marked, whereas objects ranging low on these scales remain unmarked. In several Komi dialects dative case is moreover an additional option for direct object marking, a fact which follows neither from animacy nor from the definiteness status of the object referent. This paper proposes a hypothesis according to which secondary topichood of the object referent triggers the dative marking. The purpose of the paper is to contribute to the understanding of Komi differential object marking as an example of complex variation in grammar.

1. Introduction It is appropriate to make a distinction between standard and non-standard object marking in Komi.1 “Standard” refers to those patterns which are present in all Komi dialects and which are also part of the grammar of the standardized literary languages Komi Zyryan and Komi Permyak. In a nutshell, standard differential object marking assigns nominative (zero-)case to indefinite objects and accusative case to definite objects as illustrated by (1a, b). “Non-standard” refers to object marking patterns which are only found as an additional option in certain dialects. Non-standard differential object marking includes the dative case among the possible markings of direct objects, cf. (1c). (1)

a.

b.

Men jen s’et-is m¨os da o¨ sh. I.dat God give-prt3sg cow and ox ‘God gave me a cow and an ox.’ [Vym’; ZhS71.29: 200] Vaj let’t’s’-am da o¨ sh-s¨o nat’s’k-am, bring.imp2sg go.down-fp1pl and ox-acc3sg butcher-fp1pl a m¨os-s¨o vid’z’-am. but cow-acc3sg keep-fp1pl ‘Come on, let’s go down and butcher the ox, but let’s keep the cow.’ [Vym’; ZhS71.29: 200]

326

Gerson Klumpp

c.

¨ Osh-ly nat’s’k-isny. ox-dat butcher-prt3pl ‘They butchered the ox.’

[Vym’; ZhS71.29: 200]

Komi standard differential object marking will be the topic of Section 2. Nonstandard dative marking of direct objects is presented in Sections 3 and 4. The latter focuses on the Vym’ dialect, for which a hypothesis is put forward according to which the function of dative-marking a direct object is to mark its status as a secondary topic. Finally, Section 5 summarizes and discusses the results. The generalizations in this paper are based on Komi grammars and dialect monographs as well as previous studies on the topic and, most importantly, on Komi dialect texts from collections mainly of the first half of the twentieth century.

2. Komi standard differential object marking 2.1

General remarks

The system of Komi differential object marking is organized according to the discourse pragmatic parameter of identifiability. An immediately identifiable DO referent may be represented by zero-reference (pro-drop), and a pro-identifiable object referent may be represented by an object-marked pronoun, see Section 2.2. The largest array of choice between different object markers is attested for lexically expressed objects. As Table 1 below shows, a basic distinction is made between “possessive”-marked and “non-possessive”-marked objects. Among other functions Komi second and third person singular possessive suffixes are markers of definiteness. Since marking of definiteness is crucial for Komi object marking, Section 2.3 on the “possessive” accusative starts with Table 1. Komi DO expressions and their marking DO Pro-drop expression:

Pronominal expression:

Lexical expression: Possessive

∅

Pronominal accusative

Possessive accusative

Non-possessive non-poss. accusative

nominative

Variation in Komi object marking

327

a short outline of the function of possessive suffixes. Non-possessive-marked objects are discussed in Sections 2.4 and 2.5, the former with an outline of the use of the non-possessive accusative, the latter on the use of the nominative (zero-case). Section 2.6 summarizes the parameters of Komi standard differential object marking.

2.2 Pro-drop objects and pronominal objects Komi allows not only for subject pro-drop, but also for object pro-drop as illustrated in (2). (2)

T’s’el’ad’-t¨o taj abu vaj-¨omyd. Me vetl-a da children-acc2sg mpt isn’t bring-pf2sg I go-fp1sg and vaj¨od-a. bring-fp1sg ‘You didn’t bring the children along. I’ll go and fetch [them].’ [Vym’; Zh98.2: 423]

Personal, demonstrative, reflexive and interrogative/relative pronouns are always marked for accusative case if they function as direct objects, cf. (3) for a third person plural pronoun; the corresponding nominative form, naja ‘they’, would be ungrammatical here. Example (4) illustrates the relative pronoun kod ‘who, which’ with an inanimate referent. (3)

Kork¨o . . . naj¨o kor-isny sud vyl-¨o once they.acc call-prt3pl court on-ill ‘One day . . . they were summoned to court.’ [Vym’; ZhS71.29: 200]

(4)

rumka vyna, kod-es2 myt’t’s’al-eny tsar-len nyy-jas glass vodka which-acc offer-prs3pl tsar-gen girl-pl ‘a glass of vodka, which is offered by the tsar’s daughters’ [Izhma; UK86.29: 56]

Pronouns display different accusative formations. A specific pronominal accusative in -¨o (e.g. me ‘I’: men¨o ‘me’) is restricted to personal pronouns of the first and second singular. Dialectally it appears also with third person personal pronouns (singular and plural, as in e.g. naj¨o ‘them’ in example (3) from the Vym’ dialect) and demonstrative pronouns. In literary Komi Zyryan, the relevant third person accusative forms have the ending of the absolutive accusative, -¨os, (e.g. naj¨os ‘them’ instead of naj¨o), cf. Table 2. The reflexive/intensifying pronominal stem as- inflects for person by way of possessive suffixes, accusative is marked by

328

Gerson Klumpp

Table 2. Object pronouns

NOM. ACC.

1SG

2SG

me menö

te tenö

3SG Vym’ sija sijö

3SG lit. lang. sijö sijös

interrogative/relative pronoun kod kod-ös; kod-tö, kod-sö …

PRONOMINAL ACCUSATIVE

accusative possessive suffixes here, e.g. a’ts’-yd ‘you (yourself)’, accusative: as’-t¨o (for possessive suffixes see 2.3.1 below). Other pronouns take the same possessive or non-possessive accusative endings as nouns, as e.g. kod ‘who’ in (4), cf. also Table 2. Non-referential non-specific interrogative and negative pronouns myj ‘what’ and n’in¨om ‘nothing’ usually remain unmarked in object function, cf. (5a) for a typical instance of myj ‘what’. On the other hand, the same pronoun may have a specific reading and occur in an accusative form, as with myj-s¨o ‘what’ in (5b). The sentence with this pronoun is uttered at a point in the story in which the audience already knows that the dogs saw an approaching avenger. But the storyteller utters this sentence from the perspective of the dog owner who can only guess from the dogs’ barking that there must be something outside the tent. (5)

a.

b.

A myj mi jenm-yd-ly pukt-am? but what we God-2sg-dat put-fp1pl ‘But what do we offer to God?’ [Vym’; ZhS71.29: 199] Pon-jas uut-enys . . . Myj-se, myj-se i dog-pl bark-prs3pl what-acc3sg what-acc3sg and ad’d’z’-isnys? see-prt3pl ‘The dogs are barking . . . What, what did they see?’ [Izhma; Mikushev 1987: 301]

Pronominal adjectives like e.g. o¨ t’i ‘one’ or m¨od ‘(an)other (one)’ have the accusative in -¨os as their default object form, as in (6a). When specific, pronominal adjectives have possessive suffixes (cf. 2.3.1 below), as e.g. third singular -ys in o¨ t’i-ys ‘the one’, m¨od-ys ‘the other one’. Accordingly, they are marked as objects by the accusative form of the possessive suffix as in (6b). (6)

a.

¨ Ot’i pi o¨ t’i-¨os v¨otas’-¨o, m¨od m¨od-¨os. one son one-acc dream-prs3sg other other-acc ‘One son dreams one thing, another one another thing.’ [Zh98.3: 425]

Variation in Komi object marking

b.

329

Sen naja jona t¨odmas’-asny, jona m¨oda m¨od-s¨o there they well get.to.know-fp3pl well other other-acc3sg l’ubit-isny. love-prt3pl [Once two men had worked together.] ‘They got to know each other very well there, and liked one another very much.’ [Lower Vychegda; UK89.137: 212]

2.3 Possessive-marked lexical objects 2.3.1 Komi possessive marking In Komi grammar, possessive suffixes are referred to as indan-asalan suffiksjas ¨ ‘demonstrative-possessive suffixes’ (OKK: 54). There are several motivations for marking a noun with a possessive suffix, expression of its possessor being only one of them. Komi grammars even state that the expression of a possessor via possessive suffixes occurs predominantly with kinship terms, body-part ¨ terms or abstract nouns expressing ideas and feelings (OKK: 55–56). Alternatively a possessor is expressed lexically or pronominally in the genitive, e.g. sy-l¨on mam ‘his mother’ (he-gen mother) or in the ablative case, when the possessum is a direct object, e.g. sy-lys’ mam-¨os ‘his mother’ (he-abl motheracc). The possessive suffix may not only be absent as in this example, it may even differ in person features from the person of the possessor expressed in the genitive form, e.g. mijan plug-jas-yd ‘our ploughs’ (we.gen plough-pl-2sg; Serebrennikov 1963: 137). This is due to the non-possessive function of the second and third singular suffixes to which we will turn below. Besides possession in its basic sense (e.g. my daughter) the possessive relation may be an “associative” one as in e.g. your weather (i.e. ‘the weather you have talked about’) (Leinonen 2005: 130, 2006: 101–102). The first singular possessive suffix also has a vocative function, e.g. nyl-¨oj, lok tat’t’s’¨o! ‘girl-1sg, come here!’ (R´edei 1978b: 62). Another function of possessive suffixes is agreement marking, e.g. in the cross-referencing of a second person agent of perfect participle (lokt-¨om-yd come-pt-2sg ‘your coming/you have come’) or the referent of a reflexive pronoun (as’-nys self-3pl ‘they themself’). The discourse-pragmatic function of second and third singular possessive suffixes, which express definiteness of the noun’s referent, is an important issue in Komi grammar. These suffixes can be used to mark generic definiteness ¨ (shondi-ys ‘the sun’; OKK: 57), and anaphoric or situational definiteness (see

330

Gerson Klumpp

examples below). In contrast to the definite article of e.g. Hungarian, German or English, their use is (often, at least) not obligatory – cf. (7) below, in which a textual definite (given) referent, ‘the priest’, appears once with a possessive suffix, and once without. (7)

Pop-yd ord-yn tulysja vyy-nad vaj-¨oma. A priest-2sg at-ine springtime on-ins2sg bring-pf3sg but pop-l¨on i v¨oli kod’z’ o¨ sh priest-gen and was castrated ox [The cow was pregnant.] ‘When it was nearly spring she calved at the priest’s. But the priest had also an ox.’ [ZhS71.29: 199]

Concerning the difference in meaning between the two markers of definiteness, the notion of “intimizatsiya” (‘intimization’) in connection with the second singular possessive suffix is employed frequently, e.g.: In their determinative-identifying functions, the forms of 2nd and 3rd person singular can replace each other, e.g. Suk parmaad/parmaas mortyd i vosh¨oma ‘In the dense woods the person got lost’ [. . . ]. But their meanings are not the same. The suffix of the 2nd person includes a meaning of higher familiarity and closeness (intimacy), signaling that the subject of the utterance is well known and ¨ 57) close to the speakers.3 (OKK: The s2 suffix -¨ıd tightens the sphere of reference closer to the collocutors, while s3 -¨ıs is neutral in this respect. (Hausenberg 1998: 313) The selection of the second person form signals the speaker’s assumption that his/her interlocutor has a personal, indeed intimate, knowledge of the referent. (Baker 1986: 53)

It should furthermore be mentioned that possessive suffixes in the second and third singular accusative have been grammaticalized as markers of emphasis independently of the possible object status of a referential expression, as in e.g. menym-s¨o?! ‘to me?!’ (I.dat-acc3sg; Leinonen 2005: 131). In this function they appear not only with nouns and pronouns, but also with verbs. The borderline between possessive marking and definiteness marking of possessive suffixes, as studied extensively by Schlachter (1960), for instance, may be neglected for the purpose of the present paper. Suffice it to say that, whatever motivation leads to the marking of a noun with a specific possessive suffix, if this noun is a direct object, then the accusative variant of the possessive suffix is obligatorily used. Table 3 shows the nominative and accusative paradigm of all possessive suffixes. Note, however, that the possessive declension encompasses

Variation in Komi object marking

331

Table 3. Nominative and accusative forms of possessive suffixes sg 1 2 3

nom -¨o(j) -yd -ys

acc -¨os -t¨o -s¨o

pl 1 2 3

nom -nym -nyd -nys

acc -nym¨os -nydt¨o -nys¨o

the whole paradigm of 17 case suffixes. In the following I will present examples for non-possessive uses of accusative second and third singular only.

2.3.2 Accusative second singular -t¨o The possessive second singular suffix expresses shared knowledge of speaker and hearer with several connotations, such as relevance of a referent for the hearer (and the speaker) in a given situation, intimacy or even pejorativity. A typical use is illustrated in (8) in which m¨os-t¨o ‘the cow’ refers to a situationally definite object referent known both to the speaker and the hearer (it is their only cow). (8)

A vot in¨o m¨os-t¨o pukt-am, oz-¨o jenm-yd but look! ep cow-acc2sg put-fp1pl neg.fp3sg-qp God-2sg kyk m¨os s’et. two cow give.cn [But what do we offer to God?] ‘Look, we will offer the cow, let’s see whether this God returns two cows or not.’ [ZhS71.29: 199]

2.3.3 Accusative third singular -s¨o The possessive third singular suffix is more frequent than the second singular suffix; it is, in fact, a default marker for definite nouns. Accordingly, the accusative third singular is the default marker for definite direct objects. A typical anaphoric use is illustrated in (9), in which the object, t’s’un’kytsh ‘ring’, is unmarked upon its first mention, but marked for accusative third singular upon its second mention. Example (10) shows the accusative third singular with an accessible (i.e. situationally given) object. Example (11) shows that the accusative third singular is also used with human referents – a fact to which we shall return in the following Section 2.4. The fact that the possessive accusative also appears with pronouns was already mentioned in Section 2.2 above in connection with example (5b).

332

Gerson Klumpp

(9)

Veres saj-as s’et-as sy saj-e, kod husband behind-ill3sg give-fp3sg (s)he behind-ill who tshapk-as . . . t’s’un’kytsh, kod-es o¨ shed-ema stolb vyl-e. snatch-fp3sg finger.ring which-acc hang.up.pf3sg pole on-ill Kyk vok m¨odet’t’s’-isny mun-ny kare, vid’z’ed-ny, kod two brother get.ready-prt3pl go-inf town.ill see-inf who ored-as t’s’un’kyt-se. tear.off-fp3sg finger.ring-acc3sg ‘[The tsar] will give [his daughter] to the one who . . . snatches a ring which has been hung upon a pole. The two brothers got ready to go to town to see who will tear off the ring.’ [Izhma; UK86.28: 40]

(10)

Pop lokt-is . . . Vos’t-is o¨ d’z’¨os-s¨o priest come-prt3sg open-prt3sg door-acc3sg ‘The priest arrived [at the young woman’s house] . . . He opened the door.’ [Vym’; R78.6: 40]

(11)

N’ekud’z’ zakon-yd starik-s¨o oz myzhoo not.how law-2sg old.man-acc3sg neg.fp3sg convict.cn ‘In no way could the law convict the old man.’ [Vym’; ZhS71.29: 200]

2.4

Lexical objects marked for non-possessive accusative in -¨os4

As was stated above, for a possessive-marked noun assuming the function of an object means to be obligatorily marked for the accusative variant of the possessive suffix at hand. For non-possessive marked nouns, on the other hand, assuming the function of an object has no such automatic consequence. There are two options instead: object marking with the non-possessive accusative in -¨os, or no object marking at all, i.e. nominative. The question is, what are the parameters allowing one to decide whether non-possessive lexical objects are case-marked or not? In an examination of folk tales from the Sysola dialect, Rounds (1990) found significantly frequent accusative marking on pronominal objects, human and animate objects but also on plural objects, whereas inanimate and indefinite objects were very likely to be unmarked. She schematized her results as shown in Figure 1. For pronouns, ranging at the left end of the scale, obligatory object marking was already stated above (see 2.3). Indefinites, at the other end of the scale, according to Rounds, are unmarked, especially when inanimate. On the other hand, for human, animate or plural-marked objects, definiteness is not a necessary condition to receive object marking. This is illustrated by (12) and (13). Example (12) contains an object with a human referent that is marked

Variation in Komi object marking

333

pronoun human animate plural inanimate indefinite

likelihood of likelihood of -ös accusative ∅-accusative Figure 1. Accusative in -ös and ∅-accusative (nominative) with objects in folk tales from the Sysola dialect (Rounds 1990: 229)

for accusative upon its first mention (the object referent soldier has not been previously mentioned in the text). Example (13) contains an accusative-marked plural object with an indefinite inanimate referent.5 (12)

Kaga vaj-is da ses’s’a sar gozja-ys i saldat-¨os child bring-prt3sg and then tsar couple-3sg and soldier-acc yst-isny pi-ys din-¨o. send-prt3pl son-3sg to-ill ‘She gave birth to a child, and then the tsar and the tsarina sent a soldier to their son.’ [Vym’; R78.4: 34]

(13)

Pyr-is pervyj t’erka-as. Set-ys’ ad’d’z’-is mis’t¨om enter-prt3sg first room-ill3sg there-ela see-prt3sg ugly k¨oluj-jas-¨os. rubbish-pl-acc ‘She entered the first room. She only found ugly rubbish there.’ [Middle Sysola; UK89.23: 36]

Komi grammarians often state that the accusative in -¨os appears mostly with ¨ human objects (e.g. OKK 2000: 69). Prokusheva (1988) has examined the frequency of the accusative in -¨os and the nominative with different lexico-semantic groups, cf. Table 4. According to her conclusions, (human) proper names or kinship terms never occur as unmarked objects, other [+human] object expressions rarely do. For animal object expressions being unmarked is expectable, and for inanimate objects it is the only possible form (leaving possessive accusatives out of consideration). Thus, in accordance with Prokusheva (1988), the accusative in -¨os is the expected object marker for a proper noun as in (14), but it is not expected with an inanimate object expression as in (15). – Cf. (18a, b) in the following section for unmarked human objects that are likewise unexpected. (14)

Ivan-¨os myrd-en kysk-¨ony pyzan saj-¨o Ivan-acc force-ins drag-prs3pl table behind-ill ‘They drag Ivan forcibly to the table.’ [Upper Vychegda; R78.84: 212]

334

Gerson Klumpp

Table 4. Accusative in -¨os and ∅-accusative (nominative) with different lexico-semantic groups (Prokusheva 1988: 70) object marking acc. in -¨os unmarked

(15)

Lexico-semantic groups proper kinship terms names Ivan-¨os vok-¨os ‘Ivan’ ‘brother’ – –

other + human expressions zonm-¨os ‘boy’ zon ‘boy’ “rare”

animals

in-animates

oshk-¨os ‘bear’ osh ‘bear’ “unsteady”

– va ‘water, river’

kodi lovz’¨od-is da kyp¨od-is byd yli pel’¨os-¨os which quicken-prt3sg and exalt-prt3sg each distant corner-acc R¨od’ina pas’tala-ys’ motherland width-ela ‘which quickened and exalted each distant corner of the Motherland’ [Zyry. liter. language; Baker 1986: 56]

In fact, cases as in (15) are extremely rare. Baker (1986) checked a corpus of texts from the Komi Zyryan literary language. His findings were that in nearly 95 percent of all instances of the accusative in -¨os the object referent was animate (see Table 5 below). The remaining 5 percent encompass adjectives (see below in connection with example (16)), “groupings of animate entities” (e.g. delegatsijao¨ s ‘delegation-acc’; Baker 1986: 55) – the inanimacy of these objects is certainly questionable (cf. Dahl and Fraurud 1996: 62) – and finally examples like (15), for which he considers a “highly emotive and poetic context” (Baker 1986: 56). Other than Rounds (1990) and Prokusheva (1988), Baker (1986) examined the accusative in -¨os not in contrast with the nominative object, but in contrast with the possessive accusatives in -s¨o and -t¨o. His results in fact strongly corroborate the animacy parameter of Round and Prokusheva. He found that in nearly 90 percent of all instances of the possessive accusative in -s¨o, the object referent was inanimate (see Table 5). Interestingly, for the possessive accusative second singular in -t¨o no such clear tendency could be found. That means that formal marking of definiteness by a third singular possessive suffix is extremely rare Table 5. Usage of -¨os, -s¨o and -t¨o with Komi Zyrian (in)animate objects (Baker 1986: 52) Acc in -¨os: Acc3sg in -s¨o: Acc2sg in -t¨o:

total 397 total 364 total 70

with animate DOs 375 = 94.5% 38 = 10.4% 26 = 37.1%

with inanimate DOs 22 = 5.5% 326 = 89.6% 44 = 62.9%

Variation in Komi object marking

335

with animate nouns, especially with human nouns. Thus, an instance of a human object expression with accusative third singular such as starik-s¨o ‘the old man’ in (11) above is highly marked – at least in the Zyryan literary language. It may be more common in dialects as e.g. in Vym’, from which (11) has been taken. Up to this point the use of the ending of the non-possessive accusative has been discussed only for referential objects. It has been shown that the marking of objects with -¨os is rather due to the animacy and/or humanness of the object referent than to definiteness. Now, this is valid also for non-referential objects as in (16). Here a referential object expression in accusative third singular in -s¨o contrasts with a non-referential object expression in the non-possessive accusative in -¨os. Both object referents are animals. It was also shown that the accusative in -¨os occurs as a default object marker with pronominal adjectives, cf. (6a) above, and with pronouns which have no pronominal accusative, such as e.g. the interrogative/relative pronoun kod ‘who, which’, cf. (4). A clear exception to the animacy generalization was the accusative with an inanimate plural noun as in (13) above. Cases like these are not exceptional in another respect, though, as can be seen by the fact that Rounds (1990) included plural as a parameter favoring accusative marking. Another domain of the accusative, independent from animacy, includes adjectives in those cases in which they do not occur in the attributive position before the head noun. While the adjective kuz’ ‘long’ would not receive object marking in the position before its head noun gez ‘rope’, in the position after the head noun in (17), it is marked for accusative in -¨os. Baker (1986: 54–55) observes similar cases. Komi grammar states that the accusative in ¨ -¨os generally marks object expressions that are derived substantives (OKK: 70). (16)

(17)

Osh-s¨o sij¨o dod’d’al-is, kyd’z’i v¨ol-¨os. bear-acc3sg this.acc harness.up-prt3sg like horse-acc ‘He harnessed up this bear like a horse.’ [Middle Sysola; UK89.24: 42] ¨ Osh kut’s’ik-jas-ys’ v¨ot’s’-am gez kuz’-¨os. ox hide-pl-ela make.fp1pl rope long-acc ‘We’ll make a long rope from ox-hides.’ [Upper Vychegda; UK95.5: 12]

2.5 Unmarked objects The unmarked object is regularly found with quasi-incorporated6 non-referential objects like e.g. t’s’eri ‘fish’ in t’s’eri kyjny ‘catch fish’, which is the standard Komi expression for ‘to fish’. Among referential object expressions a new inanimate object is unmarked, but animate objects may also be unmarked, cf. (1a)

336

Gerson Klumpp

above. For mass, abstract or event nouns, even if given, it is typical, too, to have no object marking at all. But an object with a given human referent may also be unmarked, as e.g. kaga ‘child’ in (18a), which has been introduced before (cf. (12) above), or zh¨on’ik ‘bridegroom’ in (18b). They belong to Prokusheva’s (1988) class of human object expressions, for which it is “rare” to be unmarked, but obviously not excluded – as it would be for a human proper name or kinship term (cf. Table 4 above). (18)

a.

b.

Sar gozja lyd’d’-isny i kaga s’et-isny sy-ly . . . tsar couple read-prt3pl and child gave-prt3pl (s)he-dat ‘The tsar and tsarina read [the letter], and gave her the child.’[Vym’; R78.4: 34] Zh¨on’ik ad’dz’-em b¨or-yn tsar mime kar-is bridegroom see-pf3sg after-ine tsar at.once make-prt3sg pyr. party ‘After he had found the bridegroom the tsar gave a party at once.’ [Izhma; UK86.28: 42]

It was stated above that formal expression of definiteness is not obligatory. Therefore, it is not surprising that we find text passages in which a recurring object expression with a given (inanimate) referent remains unmarked throughout. (Cf. t’s’un’kytsh ‘ring’ in (19) for such a case of consistent zero-marking of anaphoricity.) For the same reason it is also not surprising that a situational definite object expression within a demonstrative noun phrase, as in (20), remains unmarked – in contrast with the demonstrative pronoun, which must be marked for accusative. (19)

Z¨ol¨otn¨oj t’s’un’kytsh med bos’t-asny o¨ k¨on’t’s’a vyl-ys’. . . . golden finger.ring opt take-fp3pl window.frame upon-ela Iz vermy bos’-ny t’s’un’kytsh. . . . T’s’un’kytsh neg.prt3sg be.able.cn take-inf finger-ring finger.ring bos’t-is n’ol’ soda vyl-ys’ take-prt3sg four floor upon-ela ‘They must take a ring from the window frame. . . . He could not manage to take the ring. . . . He took the ring from the fourth floor.’ [Vym’; FF51.75: 353–355]

(20)

T’s’el’ad’-¨o, taj¨o nos¨ovik s’et-a children-1sg this.acc handkerchief give-fp1sg ‘Children, I give [you] this handkerchief.’ [Vym’; Zh98.3: 428]

Variation in Komi object marking

337

2.6 Summary of Komi standard differential object marking A first parameter which co-determines the form of a direct object in Komi object marking is definiteness. Definiteness of an object referent may be contextual (givenness), situational (anchoredness or accessibility), or universal. No matter what source definiteness has, when the definiteness of an object referent is encoded formally with a possessive suffix, the accusative variant of the possessive suffix is obligatory. A similarly strong generalization can be stated for pronominal objects. A referent which can be referred to by a referential pronoun must be definite, consequently a pronoun may be understood as a formal expression of definiteness marking which obligatorily triggers object marking. (Definiteness is of course also crucial for the identifiability of pro-drop objects. But these are not subject to object marking.) Definiteness may also be relevant with proper names, which are always definite because of their monoreferentiality. For lack of data I cannot decide here whether object marking would be obligatory for proper names referring to inanimate or abstract referents. If it is only animate proper names which trigger obligatory object marking, animacy would be the relevant parameter here. Animacy has been shown to be a second parameter in Komi which triggers object marking, independent of definiteness. Other than definiteness, animacy does not trigger marking for possessive accusative, but, rather, marking for non-possessive accusative. With human (i.e. highly animate) objects, marking of definiteness by the possessive suffix third singular is actually disfavored, and, consequently, the non-possessive accusative is preferred over the accusative third singular. Since this is not the case with the possessive second singular suffix, we cannot say that animacy outranks definiteness; rather, only that a certain formal encoding of definiteness – namely possessive suffix third singular – is disfavored when a referent is human. Regarding animacy the question is whether an object marking rule could be stated which contains a “must” instead of a “may”. If we trust analyses such as Prokusheva’s (1988) we might say that object expressions high in animacy such as human proper names and kinship terms are never unmarked. Perhaps humanness should be a parameter of its own then, but since even human object expression may remain unmarked, cf. examples (18a, b), the answer to this question is beyond the scope of this paper and must be left to further research. Neither the definiteness nor the animacy parameter cover object marking with indefinite inanimate plural nouns as in (13), with postponed adjectives with an indefinite inanimate head noun as in (17), or with non-referential pronominal adjectives as in (6a). In the two latter cases, one might think of specificity as a further parameter, since a postponed adjective yields a specific reading (‘a long one’) and pronominal adjectives like m¨od ‘other’ allow for the specific

338

Gerson Klumpp

reading ‘another one’. As En¸c (1991: 24) has shown, specificity involves a “link [. . . ] of being a subset of or standing in some recoverable relation to a familiar object” and is weaker than the definiteness link of identity of reference, but still specificity triggers the accusative marking of Turkish direct objects in En¸c’s study. For Komi this would mean that accusative marking is triggered not only by definite but also by specific direct objects. It will also be left to future research whether specificity acts as a sufficient condition for certain choices of differential object marking in Komi, or whether it is only a factor with less predictive potential. In the case of plural objects, it is important to state that plural marking is a strategy to individualize referents (cf. the sentence from the Zyryan literary language in note 5, where a marked plural object, k¨or-jas-¨os ‘reindeer-pl-acc’ contrasts with two unmarked plural objects, m¨os ‘cow(s)’ and v¨ov ‘horse(s)’). One may, thus, assume that plural marking triggers accusative marking. The results, so far, are summarized in Table 6 (“>” = ‘preferred over’). Table 6. Object marking parameters in Komi standard differential object marking DO context properties: DO expression properties:

DO referent properties:

+ definite + pronoun + possessive-marked noun + plural-marked noun + derived noun + human proper name + human + animate + specific + count noun

accusative > -∅ accusative! accusative! accusative! accusative > -∅ accusative! accusative > -∅ accusative > -∅ accusative > -∅ accusative > -∅

Up to this point, Komi differential object marking parameters have been in complete accordance with crosslinguistic generalization as treated, e.g., in Aissen’s (2003) optimality theoretic approach (cf. especially her summarizing table on page 459). As will be shown in the following paragraphs, some Komi dialects go beyond that.

Variation in Komi object marking

339

3. Non-standard dative-marking of direct objects in Komi dialects 3.1 Dative direct objects The phenomenon of dative marked direct objects in Komi has been widely ignored in overview studies on European or Uralic differential object marking as e.g. Wickman (1955), Comrie (1975), Bossong (1998). (However Alvre 1987: 10–11 does mention it.) In Komi dialectology the dative-marking of direct objects is perceived as a matter of variation, and the treatment of the phenomenon is typically stated as a desideratum, cf. Baker (1985: 202–221), Bartens (2000: 333–334), Batalova (1975: 141–146) for Northern Permyak, Saharova and Sel’kov (1976: 45–47) for Izhma; also Zhilina (1998: 57–58) for Vym’ and Zhilina (1985: 43–44) for Luza-Letka; cf. also Klumpp in print. The most detailed cross-dialectal description is due to Baker (1985: 202–221). A dativemarked direct object was already presented in (1c) above. Another example is (21) from a Northern Permyak folk tale. It includes three participants, a subject causer (eta sar’‘this tsar’), an indirect object causee (t’s’as¨ov¨oj-l¨o ‘guard-dat’) whose dative marking is governed by the verb tsh¨oktyny ‘to order’, and a direct object patient (et¨o vrat’s’-l¨o ‘this.acc physician-dat’). While dative marking with the indirect object is standard, dative marking with the direct object of vijny ‘to kill’ is not (the accusative marking of the demonstrative et¨o follows the regular pattern). Furthermore, it is somehow surprising that a dialect of Komi, despite the wide range of its distinct accusative markers (cf. Section 2), should favour the non-distinction of a direct and an indirect object over distinct markings, as in (21). (21)

Eta sar’ et¨o vrat’s’-l¨o t’s’as¨ov¨oj-l¨o tsh¨okt-as this tsar this.acc physician-dat guard-dat order-fp3sg vij-ny kill-inf ‘This tsar orders the guard to kill this doctor.’ [Kosa-Kama; UK85.29: 88]

The Komi dative fulfills “classical” dative functions like the expression of recipient and beneficient – rarely also maleficient, where the dative is in competition with the ablative in -lys’, and goal, where the dative is in competition with the illative and with postpositions. Moreover, it is used in modal constructions, for external possessors and with certain postpositions.

340

Gerson Klumpp

The dative marking of a direct object is not verb-governed. This can easily be demonstrated by examples in which one and the same verb may appear with an accusative or a dative object, cf. (22a, b). In addition, my material at present contains about 900 instances of dative direct objects with approximately 300 different verbal stems, among them all kinds of aktionsart derivations like frequentative or momentaneous verbs. These verbs hardly constitute a special class which governs dative marking of a direct object. (22)

a.

b.

¨ Osh-ly nat’s’k-isny. ox-dat butcher-prt3pl ‘They butchered the ox.’ [Vym’; ZhS71.29: 200] ¨ Osh-s¨ o nat’s’k-am. ox-acc3sg butcher-fp1pl ‘We’ll butcher the ox.’ [Vym’; ZhS71.29: 200]

An objection might be that dative marking of a direct object expresses partial affectedness of the object while accusative marking would express total affectedness (as e.g. with genitive objects in Russian or partitive objects in Finnish). Baker (1985: 205–208) has already demonstrated that cases like these are rare and only occur with a small number of verbs, as e.g. with lyjny + acc ‘shoot (dead)’ vs. lyjny + dat ‘shoot at’. Baker (1985: 207) cites (23) from Kosa-Kama in which a dative-marked and an accusative-marked direct object are coordinated. He takes this example as evidence for the identical syntactic functions of the differently marked objects. Still, it could be claimed that the two object referents differ in specificity – one is a collective noun, the other one a possessive-marked noun with a singular referent. I therefore add (24) from the Izhma dialect in which the consultant offered dative and accusative third singular as variants. These variants cannot be explained by differences in object affectedness and they clearly show that dative marking fulfills a function which cannot be explained by the classical semantics of the dative case (cf. above). One may say that dative marking of direct objects can occur with all transitive verbs, and the triggering factors must be identified beyond the realms of classical “deep case” accounts. (23)

Eta starik . . . kor-¨o samej med bur ot’yr-l¨o i this old.man call-prs3sg sup sup good people-dat and von-s¨o g¨os’t’it-ny brother-acc3sg be.guest-inf ‘This old man . . . invites the best people and his brother as guests.’ [Kosa-Kama; UK85.23: 54]

Variation in Komi object marking

(24)

341

Shybit-is gen’eral d¨oremgat’-se (∼ d¨oremgat’s’-ly) throw.off-prt3sg general shirt.trouser-acc3sg shirt.trouser-dat i pyr-is va-as. and enter-prt3sg water-ill3sg ‘And the general threw off his clothes and stepped into the water.’[Izhma; UK86.29: 62]

3.2 Dialectal distribution of dative-DOs Dative marking of direct objects occurs in three dialects of Komi-Zyryan, viz. Vym’, Izhma and Luza-Letka, and in one Komi-Permyak dialect, viz. Kosa-Kama (or Northern-Permyak), cf. Figure 7. The dative ending in Vym’ and Izhma is -ly, in Luza-Letka and in Kosa-Kama it is -l¨o. In Luza-Letka it seems that the dative ending in -l¨o has replaced the ending -¨os in its function as the possessive accusative first person singular (or an older accusative first singular in *-me, cf. Zhilina 1985: 40, 43). In this dialect the use of the dative case for direct objects seems thus to be restricted to a specific semantic function. Vym’ and Izhma, which are neither adjacent to Luza-Letka nor to KosaKama, are historically related because the Izhma dialect is a comparatively young variety which arose after a northwards movement of Vym’ speakers and others after the fifteenth century (see Saharov and Sel’kov 1976: 4–5). The Upper Vym’ area can therefore be considered the centre of the development for the subgroup of Vym’ and Izhma, but not as a centre for the whole Komi dialect area. Table 7. Komi dative DO varieties Komi-Zyryan

Komi-Permyak

Vym’ (-ly) Izhma (-ly) [Luza-Letka (-l¨o)]

Kosa-Kama (-l¨o)

3.3 “Accusative-dative”-marked pronominal objects The cooccurence of dative case and accusative or nominative is found only with non-possessive marked lexical direct objects. In addition, in two of the dative varieties mentioned in 3.1, namely in Vym’ and Luza (a subdialect of Luza-Letka), dative marking occurs also with pronominal direct objects. Other

342

Gerson Klumpp

than with nouns, dative proper and direct object dative are not homonymous. With a direct object pronoun the dative ending appears with the accusative form of the pronoun. This affects personal and demonstrative pronouns which have a special pronominal accusative form in -¨o (see Section 2.2 above). For this reason the resulting combined form is called “accusative-dative”, cf. Table 8 and example (25).7

Table 8. “Accusative-dative” pronouns nom

acc

1sg me

1sg men¨o

3sg sija

3sg sij¨o

Vym’ Luza

(25)

3.4

dat 1sg men(ym) 3sg syly

“acc-dat” 1sg men¨oly 3sg sij¨oly

1sg me(ny)m 3sg si(ja)l¨o

1sg men¨ol¨o 3sg sij¨ol¨o

Ivan, men¨o vaj-an, on? . . . Vot jes’li-k¨o te Ivan I.acc take-fp2sg neg.prs2sg look! if-cp you vaj-an men¨o-ly, te asy lok . . . ! take-fp2sg I.acc-dat you tomorrow come.imp2sg ‘Ivan, will you marry me, or won’t you? . . . So if you then marry me, you come . . . tomorrow!’ [Vym’; R78.1: 14]

Frequency of dative direct objects

In those dialects that have them, dative direct objects occur in narrative texts (fiction and non-fiction), songs and epic songs, but also in descriptive non-fiction by speakers who were born roughly between 1880 and 1940. For Izhma, data of older speakers from the nineteenth century are available (Castr´en 1878). Data from younger speakers as well as elicited data are not available (yet). Table 9 illustrates differences in the frequency of dative marked direct objects in three varieties, two of them known for their dative direct objects (Izhma and Kosa-Kama), the third (Upper Vychegda) without dative direct objects. The table summarizes statistical data on direct object expressions from three dialect versions of a folk tale called “Stupid Ivan”. As can be seen from the version from Kosa-Kama, dative-marked direct objects amount to up to a quarter of all object expressions whereas in Ihzma they only amount to up to 4 percent. The data from Upper Vychegda, a dialect in which dative marking of direct objects does not occur at all, are provided for comparison.

Variation in Komi object marking

343

Table 9. Interdialectal differences in the frequencies of dative DOs DO expressions in “Stupid Ivan”

pro-drop acc pronoun lex. nom lex. acc lex. acc3sg-DO lex. acc2sg-DO lex. acc.possX-DO lex. dat

UK86.28 Izhma 1221 words 123 DOs 21 = 17.1% 5 = 4.1% 20 = 16.3% 10 = 8.1% 62 = 50.4% – – 5 = 4.1%

UK85.24 Kosa-Kama 1603 words 239 DOs 89 = 37.2% 18 = 7.5% 51 = 21.3% – 16 = 6.7% 1 = 0.4% 2 = 0.8% 62 = 25.9%

R78.84 UpperVychegda 1748 words 122 DOs 47 = 38.5% 7 = 5.7% 27 = 22.1% 13 = 10.7% 14 = 11.5% 14 = 11.5% – –

Differences in frequency can not only be observed between different dialects, but also between speakers of a single dialect. For instance, the speaker V. S. Isaev (born 1909; UK.12–32) from Gajny, a district within the dialect area of Kosa-Kama, utters an average of 3.5 dative direct objects per transcript page. For another speaker from the same area, N. Isaev (born 1929; A05.53, 54), the figure is only 0.7. Another question is whether the frequency of dative marked direct objects may depend on the text type. There is some interesting pertinent data from the Izhma dialect (V´aszolyi 1967–1968). U. A. Koskova, an informant from Kolva, sings epic narratives and also gives prose versions of their contents. Interestingly, we find only two dative direct objects among the 2200 words of epic songs, but 13 dative direct objects among the 2000 words of prose, cf. (26a, b). From this it may cautiously be concluded that the use of dative-marked direct objects has something to do with structuring a narrative rather than being the result of idiomatization. (26)

a.

b.

Jaran d’etina torj-yd taj / pes-se Nenets children piece-2sg and firewood-acc3sg keralis . . . make-prt3sg ‘And the Nenets child / made the firewood . . . ’ [Izhma; V67/68.7: 417] Sya pes-ly keral-is ... (s)he firewood-dat make-prt3sg ‘He made the firewood . . . ’ [Izhma; V67/68.8: 418]

344

Gerson Klumpp

3.5 Views on function and origin of dative direct objects So far no explanation of the function and meaning of dative marked direct objects in Komi dialects has been provided. The idea that -ly/-l¨o is an “emphatic particle”, which is homophonous with the dative (Lyashev 1977; Serebrennikov 1963: 43–46) has not much explanatory value. Another idea brought forward is that the dative marking of direct objects is an archaism from earlier times, in which the Proto-Permian predecessor of the dative, *-le, allegedly marked indirect and direct objects (Lytkin 1955: 141). This would be an interesting case of variation, but since Komi and Udmurt accusative marking can be traced back to Proto-Uralic, I doubt whether there could have been a stage of Proto-Permian in which the indirect and direct object were systematically undistinguished. An areal parallel was noted by Baker (1985: 219), namely the homophonous Chuvash marker of the dative and definite accusative, e.g. tu ‘hill’: tˇava ‘hill.acc/dat’. But this observation does not fit together with the Komi situation. In Komi different markers are used to express allegedly identical functions, in Chuvash one marker fulfills two different functions. In the past, scholars have observed that dative marked direct objects are always definite and that dative marking occurs with animate and inanimate object referents. However, it does not occur with possessive-marked nouns. The intuition of earlier scholars has therefore been that a dative direct object is “somehow more definite than an unmarked DO, but not as definite as an accusative marked DO” (Baker 1985: 211–212). This statement raises at least two questions: (i) Since definiteness can be formally expressed by possessive suffixes, what could be the additional function of the dative regarding definiteness? (ii) And furthermore, what would be the difference between an accusative pronoun men¨o ‘me’ and an accusative-dative pronoun men¨oly ‘me’? Certainly not an increase or decrease in definiteness! The following section presents a hypothesis to account for dative marking of direct objects in the Vym’ dialect. It seems reasonable to start research on the function of dative marked direct objects for each dialect separately. Vym’ is a good starting point, because dative marking of direct objects is well developed (it affects nouns and pronouns), but dative marked direct objects are not as frequent as in the bulk of the available texts from Kosa-Kama (cf. Table 9), for which an extension of functions could be assumed.

Variation in Komi object marking

345

4. A hypothesis for Vym’ 4.1 Topical/thematic vs. focal/rhematic direct object The aforementioned intuition of “more” or “less” definite may be due to the idea that the dative direct object ranges between the indefinite nominative and the definite accusative direct object. It seems more plausible to me that the dative direct object as a definite lexical object expression ranges between the two other definite object expressions: namely the pronominal (or pro-drop) direct object and the accusative direct object. Thus, the question is not which of these object expressions is more definite; but rather, what category it is that requires a further distinction between different definite object expressions. Baker (1985: 212) already thought that the answer was topicality. Topicality is used here as defined by Lambrecht (1994: 127): “A referent is interpreted as the topic of a proposition if in a given discourse the proposition is construed as being about this referent, i.e. as expressing information which is relevant to and which increases the addressee’s knowledge of this referent.” Definiteness is a good precondition of topicality (cf. e.g. the topic acceptability scale in Lambrecht 1994: 165). On the other hand, not every definite expression is a topic. For different definite object expressions topicality may be a property that some have and that others do not have. Thus we assume that it is the information structural category of topicality which differentiates between the markings of definite object expressions; cf. Figure 2. Definiteness

indefinite

focal/rhematic

definite

focal/rhematic

topical/thematic

Figure 2. Definiteness and topicality

The function of the dative when used with a direct object would then be to mark its topicality, i.e. make the direct object a topical expression and not a focal one. The difference between o¨ sh-ly ‘ox-dat’ and o¨ sh-s¨o ‘ox-acc3sg’ in (22a, b), repeated here as (27a, b), is thus one in topicality: while the sentence with the dative-marked direct object can answer implicit questions “what did they do to the ox?” or “did they (really) butcher the ox?”, where the ox is part

346

Gerson Klumpp

of the presupposition, the one with the accusative-marked direct object answers questions like “what do we do?” or “what do we butcher?”, where the ox is part of the focus. The presupposed direct object in (27a) is topical, the direct object in (27b), which is not presupposed, is focal. (27)

a.

¨ Osh-ly nat’s’k-isny. ox-dat butcher-prt3pl ‘They butchered the ox.’ [Vym’; ZhS71.29: 200] S O [V]F8 (verb-focal sentence) when answering the question “What did they do the ox?” ∼ S OV [. . . ]F (fully topical core clause; verum focus) when answering the question “Did they butcher the ox?”

b.

¨ Osh-s¨ o nat’s’k-am. ox-acc3sg butcher-fp1pl ‘We’ll butcher the ox.’ [Vym’; ZhS71.29: 200] S [O V]F (VP-focal sentence) when answering the question “What do we do?” ∼ S [O]F V (object-focal sentence) when answering the question “What do we butcher?”

The idea as developed so far is that dative marking of a direct object in Vym’ expresses topicality of its referent. As will be shown, this idea has to be refined, but for the time being we can assume that direct object expressions in Vym’ are organized as presented in Table 10. Table 10. Direct object expressions in Vym’ DO referent DO expression

4.2

topical pro-drop pronoun: acc-dat lexical: dat

focal definite – pronoun: acc lexical: acc

focal indefinite – – lexical: nom

Secondary topic

The problem with the above idea is that not every topical object is marked with dative. In the case of the anaphoric pronoun in (28) which has a topical referent currently under discussion (the hops) we would expect the accusativedative form sij¨oly ‘it’ here. Instead, we find the accusative form sij¨o, which was claimed to express a focal object according to Table 10.

Variation in Komi object marking

347

(28) “Mam-¨o, tagj-yd vol-is iz, me ystyl-i mom-1sg hops-2sg arrive-prt3sg neg.prt3sg I send-prt1sg da.” yes “J¨oj mort-¨o, sij¨o ed t¨oo¨ r¨oz’n’it-is.” stupid man-1sg (s)he.acc mpt wind scatter.about-prt3sg ‘“Mom, did the hops arrive or not; I have sent them.” – “You’re fool, they must have been scattered about by the wind.”’ [Vym’; Zh98.2: 423] There is a difference between (28) and (27a) concerning the subject referent. (27a) has a topical pro-drop subject while in (28) the subject referent, the wind, is a new, lexically expressed one. Thus in (28) the object is a topical expression while the subject is not, whereas in (27a) both constituents, subject and object, are topical. Within a subcorpus of Vym’ texts I found dative marking of lexical direct objects comparetively often in sentences with a topical third person prodrop subject. In numbers: 22 dative marked direct objects vs. only 7 accusative marked direct objects. This amounts to a ratio of 3:1, which contrasts with the fact that, overall, accusative-marked direct objects (in Vym’) are far more frequent than dative marked direct objects. The observation thus is that a topical direct object is dative marked in the presence of a topical subject, in other words, dative marked objects are secondary topics. The idea of secondary topics was briefly considered by Lambrecht (1994: 147–150) and further developed by Nikolaeva (1999, 2001, 2002, with references to predecessors; see also Skribnik 2000). Nikolaeva (2001: 17) defines secondary topic as “an entity such that the utterance is construed to be about the relationship between it and the primary topic” (her emphasis) and, slightly differently, as “a referent which stands in a certain pragmatic relation to the primary topic referent, and the uttering communicates new information on their relation”9 (Nikolaeva 2002: 277). Skribnik (2000: 230) speaks of “relative topicality” and distinguishes a “‘continuous discourse topic’ (‘main hero’ of the discourse)” and a “‘topic of a thematic paragraph”’ in Mansi narratives. Both authors investigated Ob-Ugric languages for which it has been shown by several authors (Nikolaeva 1999, 2001, Nikolaeva, Kovgan, and Koˇskar¨eva 1993, Skribnik 2000) that the topic role is always assigned to the subject of a sentence. Put differently, a topical referent must always be encoded as the subject. The ObUgric passive serves to promote whichever syntactic function into the subject function when needed, cf. (29) from Northern Khanty [Northern Ostyak].

348

Gerson Klumpp

(29)

What about Peter? (i)

(Luw) Juwan (s)he J. ‘He hit John.’

r¯esk-´s. hit-prt3sg.sc

(ii) (Luw) Juwan-na r¯esk-´s-a. (s)he J.-loc hit-prt3sg-pass ‘John hit him.’/ ‘He was hit by John.’ (Nikolaeva 2001: 16) Nikolaeva, whose definition of secondary topic was quoted above, states that while “in English, the secondary topic can be expressed by unaccented nonsubject pronouns [. . . ] Ostyak has a syntactic device that unambiguously marks the secondary topic, object agreement.” (Nikolaeva 2001: 9). Ugric languages have two conjugations, a subjective and an objective conjugation. It has often been claimed that the latter is triggered by the definiteness of a direct object. While this seems to be true for Hungarian, in Ob-Ugric definiteness of an object does not necessarily trigger object conjugation (or object agreement). Thus there is a similar situation as with dative marked objects in Komi dialects: Dative marking/object agreement occurs with definite objects, but not with all of them. Nikolaeva (2001) has shown that it is topicality of the object which triggers object agreement. Since in Ob-Ugric the (primary) topic is restricted to subject function, a topical object is a secondary topic. For Ostyak, Nikolaeva (2001: 17) states that “the secondary topic is systematically encoded either as an overt object NP or as referential null in the object position, but in both cases the verb must be marked for object agreement.” Example (30a) illustrates a non-topical, focal definite object, which does not trigger object agreement. In (30b) the object is topical and triggers object agreement on the verb. (30)

a.

What did you do? Man [tam kalaN w¯el-s-´m.]F I this reindeer kill-prt-1sg.sc ‘I [killed this reindeer.]F’ (focal definite direct object)

b.

What did you to this reindeer? Man tam kalaN [w¯el-s-¯em.]F I this reindeer kill-prt-1sg.oc ‘I [killed]F this reindeer.’ (topical object [secondary topic]) (after Nikolaeva 2001: 10)

Variation in Komi object marking

349

Marking a direct object as a (secondary) topic expression takes it out of the focus domain, thereby yielding a more precise identification of the focus, i.e. of that part of the assertion which constitutes new information. In terms of the distinction between head-marking and dependent-marking strategies (Nichols 1992: 48–56), Ob-Ugric, by employing object agreement on the verbal head, follows a head-marking strategy. Vym’, by changing the case affix from accusative to dative, follows a dependent-marking strategy.10 The hypothesis for dative marking of direct objects in Vym’ is stated in (31): (31)

Hypothesis: A dative marked DO in Vym’ is a secondary topic.

The following text passages are chosen to corroborate the hypothesis in (31). For reasons of simplicity, I do not always make a distiction between syntactical objects and their referents.

4.3 Text examples In (32) the protagonist, Ivan, is sent by his mother to his sister’s house to ask for hops. In sentences (i) and (ii) tag ‘hops’ is not object marked, because in both contexts it is an inanimate indefinite object. Beginning from sentence (iii) the plot is about the sister and the hops. In (iii) the hops are referred to only indirectly by the quantifier expression tuu kyk ‘about two pounds’. Sentence (iv) adds more information about the sister, who is a topical pro-drop subject, and the hops, which constitute a secondary-topic object. They are expressed lexically, perhaps because they have been referred to only indirectly in the preceding sentence. The object marker is the dative as predicted. In the following sentence (v) the subject role switches from the sister to her equally topical brother, the topical object is still the hops, which, by now, are identifiable enough to undergo pro-drop. (32)

[Vym’; Zh98.2: 423] (i)

S [V O]:

(ii) [S V O]:

“Mun tag kor.” go.imp2sg hops ask.imp2sg “‘Go and ask for hops!”’ “T’s’oj-¨o, mam-¨o tag kor-is.” sister-1sg mother-1sg hops ask-prt3sg “‘Sister, mother asked for hops.”’

350

Gerson Klumpp

(iii) S [V] O:11

T’s’oj s’et-is tuu kyk. sister give-prt3sg pound two ‘The sister gave [him] about two pounds.’

(iv) S [V] O:

T’s’yshjan-¨o tag-ly turbal-is. kerchief -ill hops-dat wrap-prt3sg ‘She wrapped the hops up in a kerchief.’

(v) S [V] O:

“In kis’ty!” neg.imp2sg scatter.cn “‘Don’t scatter [it] about!”’

Example (33) is about a cow which is offered to God. In sentence (ii) it is mentioned for the first time, and since it is a definite object well known to the speakers, it is object marked with accusative second singular. Sentence (iii) provides new information on the topical subject, the old couple, and the secondary-topic object, the cow, which is now dative-marked. (33)

[Vym’; ZhS71.29: 199] (i)

“A myj mi jenm-yd-ly pukt-am?” but what we God-2sg-dat put-fp1pl “‘But what do we offer to God?”’

(ii) S V [O]:

“A vot in¨o m¨os-t¨o pukt-am . . . but look! ep cow-acc2sg put-fp1pl “‘Look, we will offer the cow . . . ”’

(iii) S [V] O:

M¨os-ly domal-asny i bos’t-asny m¨os-ly cow-dat rope-fp3pl and take-fp3pl cow-dat nu¨od-ny jen-ly. lead-inf God-dat ‘They roped the cow and set out to bring God the cow.’

Example (34) provides evidence from a chain verse. In this passage from a children’s song a new referent is introduced in every other line. The new referent is directly addressed for introduction (with first singular possessive suffix in vocative function). Each newly introduced referent is supposed to punish the preceding referent, who has not acted as requested by the speaker. Once the referent has been addressed and thereby established, the referent switches into

Variation in Komi object marking

351

the role of topical subject, the previous referent into the role of topical object. This topical object is again marked as dative. (34)

Chain verse [Vym’; MCh95.41: 104] ...

Len’ snail “Gezj-¨o, rope-1sg Gez rope “Shyr-¨o, mouse-1sg Shyr mouse “Kan’-¨o, cat-1sg

sija this gezj-¨o, rope-1sg sija this shyr-¨o, mouse-1sg sija this kan’-¨o, cat-1sg

oz neg.fp3sg len’-ly snail-dat oz neg.fp3sg gez-ly rope-dat oz neg.fp3sg shyr-ly mouse-dat

pyshjy. run.cn dzhag¨od!” strangle.imp2sg dzhag¨od. strangle.cn keroo!” cut.imp2sg keroo. cut.cn kyj!”. . . catch.imp2sg

...

‘This snail won’t run. “Rope, rope, strangle the snail!” This rope won’t strangle. “Mouse, mouse, bite the rope in two pieces!” This mouse won’t bite. “Cat, cat, catch the mouse!” This cat won’t catch.’ . . .

In (35), finally, a father goes on a hunting trip in the forest with his sons. After arriving at the intended place they set about to build a hut, which in sentence (i) is introduced as an unmarked object. In sentence (ii) it occurs again, now as a topical object. The new information in the sentence is how the building of the hut is performed, both subject and object are continual topics, and the object is dative marked. (35)

[Vym’; Zh98.3: 425] (i)

S [V O]: . . . zav¨od’it-isny v¨or kerka kar-ny. go.about-prt3pl forest house make-inf ‘. . . they went about to build a forest hut.’

352

Gerson Klumpp

(ii) S V O:

Naja setsh¨om n’eshtshas’l’iveja kut’t’s’-isny they so unluckily begin-prt3pl kerka-ly kar-ny: Kodi ki-s¨o house-dat make-inf who hand-acc3sg ran’it-¨o, kodi kok-s¨o. hurt-prs3sg who foot-acc3sg ‘They began to build the hut in such an unlucky way that one of them hurt his hand, another one his foot.’

The text passages in (32)–(35) display topical subjects and dative-marked secondary-topic objects. The constellation of topical object and topical subject seems to trigger this additional object marking strategy at least in Vym’. Of course, further research has to show that the hypothesis in (31) really covers all cases of dative-marked direct objects. The quoted examples all display OV word order, but other examples with dative marked objects, namely (23), (24) and (25), have the opposite VO order. The same variation is found with accusative marked objects, only unmarked objects show a strong tendency for OV word order (cf. Vilkuna 1998: 187). A strong claim concerning correlations between object marking and word order must be postponed to further research.

5. Summary and discussion The purpose of the present paper is to find explanations for object patterns in Komi – standard as well as non-standard ones – which were often perceived as a matter of variation. The standard patterns which involve possessive and non-possessive accusative as well as nominative could have been shown to be in accordance with widespread cross-linguistic patterns of differential object marking.The main parameters involved are definiteness, animacy and specificity (see the summary in Section 2.6 for more details). An even more challenging problem is the non-standard dative marking of direct objects which occurs in dialects of Komi. The hypothesis proposed in Section 4 explains this phenomenon as a strategy to encode topical objects in the presence of topical subjects, i.e. secondary topic objects. As shown in 4.2 this pattern has a parallel in neighbouring Ob-Ugric languages. Thus the Komi dialectal pattern is also in accordance with a pattern outside of Komi, furthermore, it may be described in terms of a linguistic area: Ob-Ugric languages are spoken to the east of Komi and were formerly also spoken in areas to the west. Further research has to outline the details, not only for the Vym’ dialect, but also

Variation in Komi object marking

353

for the other dialects which dative-mark direct objects (see Section 3). In addition, it must be explained, why of all of the cases, it is the dative which took over this function. This paper can only be a starting point for more detailed research. Still, it has become clear that it is possible to explain patterns of Komi differential object marking which were perceived as variation. This is an important first result for the task of finding more explanations and rules which help to sharpen the borderline between variation and choice. The concept of choice here relates to the freedom of a speaker in encoding an object referent as a pro-drop viz. a pronominal object expression or a lexical object expression. Such a choice is hardly fully determined by the rules of grammar. On the other hand, there is reason to engage in the search for the object marking rules of Komi (dialectal) grammar which operate as soon as the choice for a certain expression type has been made.

Notes ∗

I would like to express my gratitude to the editors of the present volume, to an anonymous reviewer, to Daniel Hole, Elena Skribnik and Christianna Stavroudis. 1. Komi, together with Udmurt, belongs to the Permian branch of the Uralic language family. It is spoken by approximately 350,000 people in Northern Russia. Two varieties of the Komi dialect continuum have been standardized as literary languages, viz. the Sykytvkar dialect in the Komi Republic (Komi-Zyryan) and the Kudymkar dialect of the Komi-Permyak okrug (Komi-Permyak). Komi has 17 cases with a non-possessive and a possessive declension in two numbers. It is basically agglutinative. Singular possessive suffixes fulfill non-possessive functions (see 2.3.1). Komi allows for zero expression of participants (pro-drop). Verbal TMA-morphology is developed moderately; there is no rigid aspect system as in Russian, but there are more tenses, with evidentiality as a crosscutting dimension (cf. Leinonen 2000); negation is expressed by a negative auxiliary, and there is only one conjugation, no “determinative” or “object” conjugation as in Mordvin or in the neighboring ObUgric and Samoyed languages. Word order is rather flexible with SVO as the basic order (Vilkuna 1998: 178, 186). For general treatments see e.g. Bartens (2000), Batalova (1993), Hausenberg (1998), R´edei (1978b) (the latter includes a map of Komi dialects). 2. Izhma has e pro o¨ in non-first syllables. The ending of the accusative therefore is -es, not -¨os. 3. “Indan-urit’s’an funktsijayn 2 da 3 morta o¨ tka lyda formajasys verm¨ony vezh-ny ota-m¨ ¨ odnys¨o, shuam, Suk parmaad/parmaas mortyd i vosh¨oma . . . No vezh¨ortasys nal¨on abudzhyk otkod’. ¨ 2 morta -yd suffiks kut¨o t¨odandzhyk, mortysly matyndzhyk (intimn¨oj) vezh¨ortas, bytt’¨ok¨o s’ornitantorjys’ s’ornitys’jasly bura t¨odsa da matyn.”

354

Gerson Klumpp

4. The ending of the non-possessive accusative, -¨os, is homonymous with the possessive accusative first singular (cf. the paradigm in Table 3). Historically this is the result of a replacement of an older accusative first singular in *-me by the non-possessive accusative in -¨os (cf. the Udmurt accusative first singular -me vs. non-possessive accusative -ez). An explanation of this process is outside the scope of this paper. 5. Another example, not from Sysola but from the Zyryan literary language, comes from Baker’s study (1985: 117–118). He assumes that accusative marking with k¨orjas ‘reindeer-pl’ is cataphoric here because the following paragraphs are about reindeer breeding, whereas the cattle and the horses play no further role. All three referents are mentioned for the first time. He also assumes that plural marking occurs here because of the prominent role of the reindeer. (i)

V¨od’it¨ony m¨os, v¨ov, vid’z’¨ony k¨or-jas-¨os keep.3pl cow horse tend.3pl reindeer-pl-acc ‘They keep cattle, horses and they tend reindeer.’

6. Cf. Dahl’s (2004: 216–219) discussion of noun incorporation. 7. In Kosa-Kama no pronominal dative objects can be found, but for Izhma it seems that the regular dative form of a personal pronoun can – albeit very rarely – also be used as a direct object, cf. (i) from Izhma (Saharova and Sel’kov 1976: 67). (i)

Ti me pomlas’ buraa kejme, ye I for good.adv pray.imp2pl a me b¨or loktygam ti-ly zahvat’ita and I back come.cv1sg ye-dat take.with.fp1sg ‘You pray well for me and I’ll take you with me on my way back.’

8. S = subject, V = verb, O = direct object; F = focal/rhematic part of the sentence. 9. My translation from Russian, G.K.; the original reads as follows: “Vtorichnyj topik: referent, kotoryj stoit v opredelennom pragmaticheskom otnoshenii k referentupervichnomu topiku, i vyskazivanie soobshchaet novuyu informatsiyu ob e˙ tom otnoshenii”. 10. I am grateful to an anonymous reviewer for this hint. 11. Instead of “S [V] O” a more adequate representation might be “S [VF [OT ]]F ”, i.e. a topical object is part of the focus.

Variation in Komi object marking

355

Text sources FF51.75 Folk tale no. 75. Fokos-Fuchs 1951: 352–366. MCh95.41 Len’¨o, len’¨o, pyshjy [‘Snail, snail, run’]. Mikushev, Chistalev and Rochev 1995: 104–105. R78.1 & 4 & 6 & 84 Gozja i ot’ik ¨ pi [‘A married couple and a son’] & The Tsarevich and the girl who was driven out & The soldier and the priest & Lazy Ivan. R´edei 1978a: 14–18, 32–36, 40–45, 208–217. UK85.23 & 29 [‘The knapsack’] & [‘Ivanov and the wife of the Russian’]. Uotila 1985: 52–56, 78–90. UK86.28 & 29 Ivanushko durat’s’ok [‘Stupid Ivan’] & Ivan tsarevit’s’, Ivan generalovit’s’ [‘Ivan Tsarevich, Ivan Generalovich’]. Uotila 1986: 38–64. UK89.23 & 24 & 137 [‘The girl and the nine exiles’] & V¨oralys’ [‘The hunter’] & Mojd-kyv [‘A fairy tale’]. Uotila 1989: 34–43, 212–219. UK95.5 Ivan tsarevit’s’ [‘Ivan Tsarevich’]. Uotila 1995: 10–19. V67/68.7,8 [‘Song of the frozen servant’]. V´aszolyi 1968: 408–420. Zh98.2 & 3 Duren’ [‘The fool’] & Shom Vukva [‘Shom Vuk river’]. Zhilina 1998: 423–432. ZhS71.29 Starik bogomolet’s’ [‘The believer’]. Zhilina and Sorvacheva 1971: 199–203.

Abbreviations acc: accusative case; cn: connegative stem; dat: dative case; ela: elative case; ep: emphatic particle; fp: future present tense; gen: genitive case; ill: illative case; imp: imperative mood; ine: inessive case; inf: infinitive; ins: instrumental case; loc: locative case; mpt: modal particle; oc: object conjugation; opt: optative particle; pl: plural number; pf: perfect tense; prs: present tense; prt: preterite tense; pt: participle; qp: question particle; sc: subject conjugation; sg: singular number; sup: superlative particle.

356

Gerson Klumpp

References Aissen, Judith 2003 Alvre, Paul 1987

Baker, Robin 1985 1986 Bartens, Raija 2000

Differential object marking: iconicity vs. economy. Natural Language & Linguistic Theory 21: 435–483. O padezhah ob”ekta (pryamogo dopolneniya) v finno-ugorskih yazykah. In XVII vsesoyuznaya finno-ugorskaya konferentsiya 1: Yazykoznanie (tezisy dokladov), Mihail G. Atamanov (ed.), 3–14. Ustinov. The Development of the Komi Case System: A Dialectological Investigation. (M´emoires de la Soci´et´e Finno-Ougrienne 189.) Helsinki. The role of animacy in Komi direct object marker selection. UralAltaische Jahrb¨ucher N. F. 6: 47–60.

Permil¨aisten kielten rakenne ja kehitys [The structure and development of Permian languages]. (M´emoires de la Soci´et´e FinnoOugrienne 238.) Helsinki. Batalova, Raisa M. 1975 Komi-permyatskaya dialektologiya. Moskva. 1993 Komi(-zyryanskij yazyk). Komi-permyatskij yazyk. In Yazyki mira: Ural’skie jazyki, Yu. S. Eliseev, K. E. Majtinskaya (eds.), 214–239. Moskva: Nauka. Bossong, Georg 1998 Le marquage diff´erentiel de l’objet dans les langues d’Europe. In Actance et valence dans les langues de l’Europe, Jack Feuillet (ed.), 193–258. (Empirical approaches to language typology 20. EUROTYPE 2.) Berlin/New York: Mouton de Gruyter. Comrie, Bernard 1975 Subjects and direct objects in Uralic languages: a functional explana´ tion of case-marking systems. Etudes Finno-Ougriennes 12: 5–17. ¨ Dahl, Osten 2004 The Growth and Maintenance of Linguistic Complexity. Amsterdam/Philadelphia: John Benjamins. ¨ Dahl, Osten, and Kari Fraurud 1996 Animacy in grammar and discourse. In Reference and Accessibility, Thorstein Fretheim and Jeanette K. Gundel (eds.), 47–64. Amsterdam/ Philadelphia: John Benjamins. En¸c, M¨urvet 1991 The semantics of specificity. Linguistic Inquiry 22: 1–25.

Variation in Komi object marking

357

Fedyun¨eva, Galina V. (ed.) ¨ 2000 Oniya komi kyv. Morfologiya [Modern Komi. Morphology]. Syktyvkar: Komi neb¨og l˙edzanin. Fokos-Fuchs, David R. 1951 Volksdichtung der Komi (Syrj¨anen). Budapest: Akad. Kiad´o. Hausenberg, Anu-Reet 1998 Komi. In Uralic languages, Daniel Abondolo (ed.), 305–326. London/New York: Routledge. Klumpp, Gerson In print On dative marked direct objects in Komi dialects – The T’s’hun’-kytsh (‘ring’) episode. To appear in Congressus 10. Internationalis Fennougristarum, Yoshkar-Ola 2005. Leinonen, Marja 2000 Evidentiality in Komi-Zyrian. In Evidentials. Turkic, Iranian and Neighbouring Languages, Lars Johanson and Bo Utas (eds.), 419– 440. (Empirical approaches to language typology 24.) Berlin/New York: Mouton de Gruyter. 2005 Upotreblenie prityazhatel’nyh suffiksov v nesobstvenno prityazhatel’nyh funktsiyah v komi yazyke. In Istoriya, sovremennoe sostoyanie, perspektivy razvitiya yazykov i kul’tur finno-ugorskih narodov, A. A. Popov (ed.), 129–132. Syktyvkar: IJaLI. 2006 Omistussuhteen ulokkeita: komin possessiivisuffiksin ei-possessiivisista funktiosta [Extension of possessive relations: On the nonpossessive function of Komi possessive suffixes]. Journal de la Soci´et´e Finno-Ougrienne 91: 93–114. Lyashev, Vladimir A. 1977 Nekotorye distinktivnye priznaki severnogo nul’-˙e lovogo areala komi yazyka. Sovetskoe Finnougrovedenie 13: 119–123. Lytkin, Vasilij I. 1955 Dialektologicheskaya hrestomatiya po permskim yazykam. Moskva: Izdat. Akademii Nauk SSSR. Mikushev, Anatolij K. 1987 Komi narodnyj epos. ˙ Moskva: Nauka. Mikushev, A. K., P. I. Chistalev and Yu. G. Rochev 1995 Komi narodnye pesni 3: Vym’ i Udora. Syktyvkar: Komi knizhnoe izdat. Nichols, Johanna 1992 Linguistic Diversity in Space and Time. Chicago/London: University of Chicago Press. Nikolaeva, Irina 1999 Object agreement, grammatical relations, and information structure. Studies in Language 23: 331–376.

358

Gerson Klumpp 2001

Secondary topic as a relation in information structure. Linguistics 39: 1–49. Nikolaeva, Irina, Elena Kovgan and Natalia Koˇskar¨eva 1993 Communicative roles in Ostyak syntax. Finnisch-Ugrische Forschungen 51: 125–167. Prokusheva, T. I. 1988 Leksiko-semanticheskie razryadi sushchestvitel’nyh s pokazatelem akkuzativa -¨os. In Leksikologiya i leksikografiya komi yazyka, 59– 71. (Trudy IYaLI Komi NTs Uro AN SSSR 41.) Syktyvkar: Komi nauchnyi tsentr Ur O AN SSSR. R´edei, K´aroly 1978a Zyrian Folklore Texts. Budapest: Akad. Kiad´o. 1978b Syrj¨anische Chrestomathie. (Studia Uralica 1.) Wien: Verband der ¨ Wiss. Ges. Osterreichs. Rounds, Carol 1990 The distribution of the accusative case in Komi folktales. In Congressus Septimus Internationalis Fenno-Ugristarum. Vol. 3C, 227–232. Debrecen. Saharova, M. A., and N. N. Sel’kov 1976 Izhemskij dialekt komi yazyka. Syktyvkar: Komi knizhnoe izdat. Schlachter, Wolfgang 1960 Studien zum Possessivsuffix des Syrj¨anischen. (Finnisch-ugrische Studien 3.) Berlin: Akademie-Verlag. Serebrennikov, Boris A. 1963 Istoricheskaya morfologiya permskih yazykov. Moskva: Izdat. Akademii Nauk SSSR. Skribnik, Elena 2000 Pragmatic structuring in Northern Mansi. In Congressus Nonus Internationalis Fenno-ugristarum Pars VI: Dissertationes sectionum: Linguistica III, T˜onu Seilenthal (ed.), 222–239. Tartu. Uotila, Toivo E. ¨ 1985 Syrj¨anische Texte I. Komi-Permjakisch. Ubersetzt und herausgegeben von Paula Kokkonen. (M´emoires de la Soci´et´e Finno-Ougrienne 186.) Helsinki: Suomalais-ugrilainen seura. 1986 Syrj¨anische Texte II. Komi-Syrj¨anisch: Iˇzma-, Peˇcora- und Vym¨ Dialekte. Ubersetzt und hg. von Paula Kokkonen. (M´emoires de la Soci´et´e Finno-Ougrienne 193.) Helsinki. 1989 Syrj¨anische Texte III. Komi-Syrj¨anisch: Luza-Letka, Ober-Sysola, Mittel-Sysola, Prisyktyvkar-, Unter-Vyˇcegda und Udora-Dialekte. ¨ Ubersetzt und hg. von Paula Kokkonen. (M´emoires de la Soci´et´e Finno-Ougrienne 202.) Helsinki.

Variation in Komi object marking 1995

V´aszolyi, Erik 1967–68 Vilkuna, Maria 1998

Wickman, Bo 1955

359

Syrj¨anische Texte IV. Komi-Syrj¨anisch: Ober-Vyˇcegda-Dialekt. ¨ Ubersetzt und hg. von Paula Kokkonen. (M´emoires de la Soci´et´e Finno-Ougrienne 221.) Helsinki. ´ Eszaki z¨urj´en epikus e´ nekek [Northern Zyrian Epical Songs] I–III. Ethnographia 78(3): 438–451, 79(1): 76–91, 79(3): 408–420. Word order in European Uralic. In Constituent Order in the Languages of Europe, Anna Siewerska (ed.), 173–233. (Empirical approaches to language typology 20. EUROTYPE 1.) Berlin/New York: Mouton de Gruyter.

The Form of the Object in the Uralic Languages. (Uppsala Universitets Årsskrift 6.) Uppsala/Wiesbaden: Lundequist/Harrassowitz. Zhilina, Tat’yana I. 1985 Luzsko-letskij dialekt komi yazyka. Moskva: Nauka. 1998 Vymskij dialekt komi yazyka. Syktyvkar: Prolog. Zhilina, Tat’yana. I., and Valentina A. Sorvach¨eva 1971 Obraztsy komi-zyryanskoj rechi. Syktyvkar: Akad. Nauk.

How lexicalization reflected in hyphenation affects variation and word-formation Britta Mondorf Abstract. Comparative alternation, i.e. the choice between synthetic forms of comparison (prouder) and their analytic variant (more proud), is a showcase of grammatical variation that poses an intriguing challenge to linguistic theorybuilding. The present article highlights the lexicalisation and choice of comparative forms in adjectival compounds/derivatives (e.g. harder-nosed vs. more hard-nosed). Findings reported in Mondorf (2000) are revisited and complemented on the basis of novel data – including a cross-variety perspective – in order to arrive at new insights concerning the systematic correlation between the lexicalization of compounds and the choice of comparative form. By gauging lexicalization in terms of spelling (i.e. highrisk, high-risk, high risk), we can predict that the more closely integrated the elements of compounds, the less likely they are to take the synthetic comparative form, and vice versa. While this tendency might at first glance appear to reflect a reluctance to permit intervening -er suffixes within compounds (e.g. shorter-lived), it can now for the first time be shown to hold irrespective of the prospective position of the -er suffix (e.g. street-wiser).

1. Introduction The area of comparative formation raises intriguing issues for any theory of language. On the one hand, we have a situation in which categorical rules apply, such as ‘adjectives that are rendered longer than three syllables in the synthetic comparative form take the analytic comparative’.1 On the other hand, there are a vast number of adjectives that allow a choice between both variants. The following cline illustrates the situation for morphologically simple adjectives. At the left end of the cline, we find adjectives such as big, which would not (or only extremely rarely) take the analytic comparative form. It lies at the 100% synthetic end. At the other extreme we find adjectives such as important, whose comparative form would be tetrasyllabic if the -er suffix were attached. These two adjectives illustrate the categorical choices. In the middle ground, however, we observe all shades of intervening stages, exemplified here by sure

362 bigger

Britta Mondorf surer (76%) more sure (24%)

readier (42%) righter (21%) more ready (58%) more right (79%)

100% synthetic

more important

0% synthetic

Figure 1. Variation cline of morphologically simple adjectives (data: British pilot corpus described in Section 4)

with a synthetic percentage of 76%, via ready for which the likelihood of occurrence of the two variants is roughly balanced to right, which takes the synthetic comparative in 21% of all cases. The present article demonstrates that this degree of variability also applies to adjectival compounds. While long-term strongly favours synthetic comparison, right-wing almost exclusively takes the analytic form. The chances for highprofile to occur with any of the two comparative variants are roughly equal. longer-term (97%) more long-term (3%)

100% synthetic

higher-profile (53%) more high-profile (47%)

righter-wing (2%) more right-wing (98%)

0% synthetic

Figure 2. Variation cline of morphologically complex adjectives (data: British pilot corpus described in Section 4)

This distribution calls for an explanation, unless we choose to believe that such choices are entirely arbitrary. The present paper investigates in how far the selection of synthetic vs. analytic comparative forms is conditioned by the degree of lexicalization, integration or fusion as reflected in the spelling.2 The notion lexicalization is meant to refer to the process whereby frequently used morphologically complex lexemes tend to become single lexical entities with the concomitant effect of gaining a more specific meaning and losing part of their structural transparency (cf. Lipka 1977: 155). It corresponds to Brinton and Traugott’s (2005: 47) historical dimension of lexicalization as fusion, optionally accompanied by univerbation. The following four types of adjectival compounds which are formed by currently productive word-formation strategies are analyzed in order to trace the relation between lexical complexity and choice of comparative form. In comparison to the data provided in Mondorf (2000), especially those compounds taking

How lexicalization reflected in hyphenation affects variation and word-formation

363

Table 1. Types of compounds investigated (based on Mondorf 2000: 35)3 -ing -ed ADJ+N N+ADJ

Positive Form long-lasting broad-minded high-risk environment-friendly

Internal Structure [ADJ + [V + -ing]V ]ADJ [ADJ + [N/V + -ed]ADJ ]ADJ [ADJ + N]ADJ [N + ADJ]ADJ

the form of noun+adjective combinations will be shown to merit reconsideration in the light of the extended database available for the present study. The English language permits two comparative variants for these adjectives (e.g. higher-risk vs. more high-risk) and three potential spellings (i.e. as one word, with a hyphen or as two words). This leaves us with six combinations for the comparative form of the morphologically complex adjectives investigated here. Table 2. Variant spellings of the positive and the comparative4 (based on Mondorf 2000: 36) Spelling 1 word hyphen 2 words

Positive highrisk high-risk high risk

Synthetic Comparative higherrisk higher-risk higher risk

Analytic Comparative more highrisk more high-risk more high risk

The present article revisits and extends earlier analyses (presented in Mondorf 2000) which show that the relation between the choice of comparative form systematically corresponds to the degree of integration in morphologically complex adjectives as reflected in spelling conventions concerning hyphenation. The more closely integrated the elements of compounds, the less likely they are to take the synthetic comparative form, and vice versa. Since 2000, the corpus collection available in the research project ‘Determinants of Grammatical Variation in English’ has quadrupled. This enables us to place the analysis on more solid foundations as well as to widen the scope of research to new issues concerning an explanation in terms of processing considerations and to cross-variety comparisons of British and American English. The present article now for the first time provides data showing that the tendency to avoid the -er comparative in compounds does not reflect a reluctance to permit intervening -er suffixes (e.g. shorter-lived), but rather holds irrespective of the prospective position of the -er suffix (e.g. street-wiser). In Mondorf (2000) I have challenged two claims concerning comparative formation of adjectival compounds and derivatives.

364

Britta Mondorf

The first claim concerns the comparative formation strategy chosen by morphologically complex adjectives. While Jespersen (1977: 223) was quite aware of the fact that adjectival compounds take both the synthetic and the analytic comparative, e.g. harder-working, more old-fashioned, Bauer (1988: 134) stipulates that “(. . . ) compound adjectives never take synthetic comparison in -er, and so are never inflected; (. . . )”. In Mondorf (2000), I have rejected this claim on the basis of a wide range of counterexamples from British English corpora comprising ca. 600 million words.As the corpus studies introduced in the present article show, the rejection of this claim can now be extended to American English data (cf. Section 5.1). The second claim concerns the allegedly random use of hyphenation in compounds. On the basis of the four types of compound adjectives shown in Table 1, pilot analyses in Mondorf (2000) indicate that at least for three groups there is a correlation between the choice of comparative form and the degree of integration between the parts of compounds. Further support for this claim concerning the fourth group of adjectives as well as new insights in terms of explaining the distribution of the variants is derived from a detailed analysis of noun+adjective compounds presented in Section 5.2. The article is structured as follows: Section 2 outlines theoretical considerations relevant to the topic under investigation. The state of the art is presented in Section 3, while Section 4 introduces methodological prerequisites and the database. Two claims concerning comparative formation and integration in adjectival compounds (gauged in terms of spelling as one word, with a hyphen or as two words) are revisited and supplemented by additional empirical analyses in Section 5. The next section then deals with the explanatory potential of more-support as a compensatory strategy reducing the processing effort exerted by adjectival compounds. Finally, Section 7 relates the results to ongoing discussions of language processing and word-formation theory.

2. Theoretical considerations The competence-performance distinction has been under vivid attack ever since it has been formulated. Especially for those working in the field of language variation, relegating non-categorical choices to performance has always been an unsatisfying and unilluminative makeshift solution. While some reacted to the fact that most aspects of language variation were thus declared out of bounds of linguistic theory-building by setting out to develop processing-based performance theories (e.g. Hawkins 1994, 1999), others rejected the division of competence and performance as being void from the outset (c.f. e.g. Giv´on 1979).

How lexicalization reflected in hyphenation affects variation and word-formation

365

Recently, scholars are increasingly trying to bring together insights from formerly irreconcilable camps by bridging the gap between formal and functional linguistics. The present volume is a case in point. What we lose by restricting linguistic theory-building to categorical choices is conspicuously illustrated by comparative alternation, i.e. the choice between synthetic forms of comparison (e.g. prouder) and analytic ones (e.g. more proud). If we arbitrarily confine our attention to categorical choices at the expense of the most telling middle-ground comprising non-categorical constraints, we miss out on the opportunity to observe and monitor important generalizations and principles that govern language use and affect, shape and ultimately constitute grammar. As regards comparative alternation, further complications arise from the fact that the variants themselves stem from different levels of linguistic analysis, i.e. syntax and morphology, that have traditionally been claimed to form largely independent, autonomous modules. We come to realize that one and the same linguistic function, i.e. that of marking comparative degree, can be fulfilled by either a syntactic (more) or a morphological (-er) variant, which calls for an explanation with reference to the morphology-syntax interface. Neither can traditional word-formation models account for the comparative formation of adjectival compounds. Thus, cases of synthetic comparatives such as environment-friendlier pose serious problems to lexical phonology models assuming level ordering as advocated in Siegel (1974) and Allen (1978), since their claim that inflections are added after compounding leads to a six-syllable base [environment-friendly]-er taking a synthetic comparative, a situation which is unparalleled by any other adjective in the English language system. Thus, the issue of comparative formation of compounds in English adds another inconsistency to the so-called level ordering paradoxes discussed for instance in Booij and Rubach (1984), Kiparsky (1982), Selkirk (1982: 101), Booij (1993: 36) and Plag (1999).5 More recent approaches investigating the cognitive basis of lexical storage and processing observe interesting determinants of processing complexity, such as family size effects (cf. Schreuder and Baayen 1997) and relative frequency (Hay 2001), incorporating both whole-word access and decomposition.6 The issue of comparative alternation is further complicated by the fact that previous research (cf. Mondorf 2003, 2009b) indicates that the mere decision of whether to choose the synthetic or analytic comparative form is constrained by at least 24 determinants from all levels of linguistic analysis (phonology, morphology, syntax, semantics and pragmatics), whose effects can largely be explained in terms of processing considerations. These findings clearly call for

366

Britta Mondorf

linguistic theories that are able to model synergetic and competing forces of highly diverse factors. What is more, the competition between both comparative variants can by no means be described as a temporary situation in a transitional phase in the diachronic development from one stable linguistic system to the next, since comparative alternation has been shown to be around for more than 500 years and appears to be presently evolving towards a division of labour between synthetic and analytic comparatives (cf. Mondorf 2004). The aspects of comparative formation that are given centre-stage in the present article concern morphologically complex adjectival compounds or derivatives. The article focuses on four different groups of morphologically complex adjectives in order to illustrate which factors cause language users to prefer one variant over the other. The variation encountered is systematically constrained and functionally motivated. This means that functional theories of processing complexity will be given precedence, particularly those which do not tend to isolate language matters from other cognitive processes. The present paper thus strongly advocates incorporating principles and insights from our general cognitive design into the analysis of language processing. More concretely, this paper assumes that language users can be considered to weigh the pros and cons between [...] less form processing [...] but more dependent processing on the one hand, and more form processing (explicit marking) with less dependent processing on the other. (Hawkins 2003: 200)

As regards comparative alternation language users correspondingly trade-off the more explicit analytic variant for the more dependent synthetic -er variant. The concept of more-support which is derived from Rohdenburg’s (1996) complexity principle states that: more-support: In cognitively more demanding environments, which require an increased processing load, language users tend to make up for the additional effort by resorting to the analytic rather than the synthetic comparative. (Mondorf 2003: 252)

This principle applies to instances of cognitive complexity in the form of morphological complexity exerted by bimorphemic rather than monomorphemic adjectives, of syntactic complexity produced by the presence of argument complexity, of semantic complexity effected by abstract rather than concrete senses of formally identical adjectives, of phonological complexity triggered by identity effects, etc. (cf. Mondorf 2002, 2003, 2009b and Boyd 2007).

How lexicalization reflected in hyphenation affects variation and word-formation

367

One advantage of the analytic comparative variant in cognitively complex environments is that it can signal early in the degree phrase that a comparative follows, thus rendering phrase structure easily identifiable. Functional processing theories a` la Hawkins (2000, 2003) predict that early recognition of phrase structure demands less processing from working memory than late recognition. Hawkins’ Principle of Mother Node Construction (1994: 60) states that a word that can uniquely determine a phrase will immediately be used to construct a representation of that phrase. Relating this principle to comparative alternation, we can reasonably expect that an early occurrence of more is a relatively – though not completely – safe signal that a degree phrase follows.7

3. The state of the art According to reference grammars, it is the number of syllables and the nature of the final segment that condition the choice between synthetic comparatives in -er (e.g. fuller) and the analytic comparatives with more (e.g. more important). There is general consent in the literature that trisyllabic words take the historically more recent analytic comparative, while monosyllables take synthetic variants, with disyllabic words being subject to variation. The situation is, however, considerably more complex (cf. Figure 3). In Mondorf (2009b), 24 factors determining the choice of comparative form are discussed and investigated. Szmrecsanyi (2006) additionally provides empirical support for the effect of lexical persistence. The status of compounds is, however, still unclear.

4. Methodology In order to retrieve all adjectives of the syntactic templates listed in Table 1, the British pilot corpus (cf. Table 3 below) has been searched by means of concordance software and all hits have been manually post-edited. Only those adjectives which take both the synthetic and the analytic variant in the data and which additionally occur at least ten times in both the synthetic and the analytic forms have been selected for further analysis. The database for the present study predominantly comprises newspaper data, the only exception being the British National Corpus (BNC).A list of the corpora together with information on their approximate size is provided in the following tables.8 Totting up the figures for the British and the American English data, the database for the present study amounts to 2.7 billion words.

368

Britta Mondorf

Phonology Avoidance of Identity Effects I: Stress Clash Avoidance of Identity Effects II: Morpho-phonologically Identical Segments Avoidance of Identity Effects III: Consonant Clusters Final Segment Morphology Morphological Complexity Syntax Prepositional Complements Infinitival Complements Position Semantics Concrete vs. Abstract Literal vs. Figurative Weak vs. Strong Gradability Diachrony Morphological Complexity Argument Complexity

Lexicon Length Frequency Compounds Parallel Structures Lexical Persistence (cf. Szmrecsanyi 2006)

moreSupport

Pragmatics End-Weight Proximity Cumulative Comparatives Emphasis Gradual Increase Establishment in Discourse Variety British vs. American English

Figure 3. The domains of more-support (based on Mondorf 2009b) Table 3. British pilot corpus9 British pilot corpus British National Corpus (BNC) Guardian 1990–94 (incl. Observer 1994) Daily Mail and Mail on Sunday 1993–94 Daily Telegraph and Sunday Telegraph 1991–94 Times and Sunday Times 1990–94 Total

Mio words 100 141 38 128 192 599

Table 4. Extended British corpus Extended British corpus Guardian 1990–2005 (incl. Observer 1994–2005) Daily Mail and Mail on Sunday 1993–2000 Daily Telegraph and Sunday Telegraph 1991–2000, 2002, 2004 Times and Sunday Times 1990–2004 Total

Mio words 640 207 430 725 2002

How lexicalization reflected in hyphenation affects variation and word-formation

369

Table 5. American corpus American corpus Washington Times 1990–92 Los Angeles Times 1992–99 Detroit Free Press 1992–95 Total

Mio words 89 569 103 761

5. Investigating two claims on comparative formation of compounds 5.1 Why do adjectival compounds take fewer synthetic forms of comparison than simplexes? Putting Bauer’s (1988: 134) claim concerning the non-existence of compounds taking the synthetic comparative to the test, we find extensive amounts of counterevidence. Table 6 shows that all four groups of morphologically complex adjectives investigated allow the synthetic comparison. What is more, for those adjectives in -ed and adjective+noun combinations, the synthetic variant is even the preferred form rather than the exception. These results led to the rejection of the claim that “(...) compound adjectives never take synthetic comparison in -er, and so are never inflected; (...)” (Bauer 1988: 134). The only way to uphold this claim would be to declare a large number of words out of bounds by classing them as syntactic groups instead, a makeshift solution that has been rejected as arbitrary and theoretically unfounded in Mondorf (2000: 41–42). What is striking about the distribution shown in Table 6 is the extreme scarcity of synthetically formed noun+adjective combinations. Only 20 instances of adjectives such as street-wiser, voter-friendlier, etc. are found in the extended 2 billion word corpus of British English. This scarcity also prevails in the American English data (cf. Table 7). We will return to the issue of noun+adjective combinations and its relevance for the explanation of variation patterns in adjectival compounds in Section 5.1.3. However, before we proceed to a possible explanation of the variation patterns, a look at the distribution of synthetic and analytic comparatives for the individual adjectives is in order.

370

Britta Mondorf

Table 6. Comparative alternation in four groups of adjectival compounds/derivatives in British English10 Type -ing -ed ADJ+N N+ADJ Total

Synthetic 219 498 3464 20 4201

Synthetic % 43% 59% 67% 1% 50%

Analytic 292 351 1689 1572 4263

Analytic % 57% 41% 33% 99% 50%

N 511 849 5153 1592 8464

The adjectives investigated are: -ed: bloody-minded, broad-based, broad-minded, deep-rooted, far-sighted, finegrained, full-blooded, full-bodied, full-flavoured, hard-nosed, heavy-handed, high-minded, high-pitched, high-priced, longer-lived, long-winded, lowpriced, old-fashioned, short-lived, short-sighted, sure-footed, tough-minded -ing: easy-going, far-reaching, fast-moving, hard-working, long-lasting, longrunning, long-standing, slow-moving, wide-ranging ADJ+N: close-run, free-market, hard-line, heavy-duty, high-brow, high-class, highpressure, high-profile, high-quality, high-risk, high-tech, high-technology, large-scale, left-wing, low-grade, low-key, low-profile, low-tech, old-style, right-wing, short-term, small-scale N+ADJ: accident-prone, audience-friendly, batsman-friendly, child-friendly, customer-friendly, driver-friendly, environment-friendly, family-friendly, listener-friendly, media-friendly, parent-friendly, passenger-friendly, patientfriendly, people-friendly, punter-friendly, radio-friendly, reader-friendly, street-wise, stress-free, trigger-happy, trouble-free, user-friendly, viewerfriendly, visitor-friendly, voter-friendly, woman-friendly, women-friendly, world-weary

Table 7. Comparative alternation in two groups of adjectival compounds/derivatives in British (BrE) and American English (AmE) (based on the extended British corpus and the American English corpus described in Section 4 Type ADJ+N BrE ADJ+N AmE N+ADJ BrE N+ADJ AmE

Synthetic 3464 1878 20 9

Synthetic % 67% 81% 1% 2%

Analytic 1689 428 2343 569

Analytic % 33% 19% 99% 98%

N 5153 2306 2363 578

How lexicalization reflected in hyphenation affects variation and word-formation

371

5.1.1 Earlier results on -ing and -ed formations The first group of adjectives investigated in Mondorf (2000: 39) were those ending in -ing. All nine -ing compounds investigated allowed the synthetic comparative. Moreover, for five of the nine -ing compounds the synthetic comparative turned out to be the rule rather than the exception. Furthermore, even highly frequent dimension adjectives such as long occurred with the analytic comparative form if they formed part of compounds. This is indicative of a lexical complexity effect which triggers an increased use of more-support. The second group of adjectives investigated in Mondorf (2000: 41) consisted of 21 adjectival compounds/derivatives in -ed. While in general, highly frequent dimension adjectives almost exclusively select the -er variant, the data revealed that even the dimension adjectives broad, high, long and short require the analytic comparative when they occurred in combination with another lexeme. What was particularly striking was that so-called morphological families11 , i.e. formally similar sets of compounds such as full-blooded (9%), full-bodied (59%) and full-flavoured (94%) displayed highly divergent ratios for the synthetic variant, which suggests that the choice of comparative form is not solely determined by the first element of the compound. We will return to this issue in Section 6.

5.1.2 Revisiting adjective+noun combinations The analysis of the third group of adjectival compounds offers new insights into the mechanisms of word-formation. It comprises ADJ+N combinations, such as large-scale or high-risk. The comparative formation of this type is particularly interesting in terms of a statement by Reis (1983: 123–124) who comments on the productivity of affixes in German: “N verbindet sich nicht mit suffigiertem A als Erstglied.” [N does not combine with a suffixed ADJ as a first element.]. If this restriction also holds for English, we should expect that forms such as larger-scale, higher-risk, etc. are not permitted. Contrary to this, the British and American English ADJ+N compounds analysed in the present study display the distribution shown in Table 8. All adjectival compounds allow both comparative variants. The percentages for the synthetic variant in the British English data range from 1% for left-wing to 99% for low-grade. This distribution is hardly surprising in view of the correlation between gradability and choice of comparative form (cf. Graziano-King and Smith Cairns 2005: 348 and Mondorf 2009b). Left-wing is only very weakly gradable and should hence require a high ratio of more-support if compared at all. Conversely,

372

Britta Mondorf

Table 8. Synthetic comparative forms of ADJ+N compounds in British and American English (NSynthetic+Analytic = 7459) Hyphen Synthetic BrE Synthetic AmE Synthetic (%) BrE Synthetic (%) AmE close-run 30 0 86% – free-market 10 0 27% 0% hard-line 28 19 25% 29% heavy-duty 9 10 43% 83% high-brow 17 10 50% 83% high-class 43 17 86% 89% high-pressure 6 0 75% – high-profile 305 197 46% 72% high-quality 265 285 94% 98% high-risk 718 170 97% 99% high-tech 31 34 20% 29% high-technology 13 6 76% 75% large-scale 273 69 96% 96% left-wing 2 0 1% 0% low-grade 210 45 99% 100% low-key 143 45 29% 24% low-profile 191 115 92% 93% low-tech 34 18 59% 62% old-style 106 19 91% 90% right-wing 1 0 0% 0% short-term 569 643 85% 99% small-scale 460 176 99% 99% TOTAL 3464 1878 67% 81%

N Br N Am 35 0 37 5 112 65 21 12 34 12 50 19 8 0 660 272 283 292 741 172 152 119 17 8 284 72 199 8 212 45 495 185 208 124 58 29 117 21 295 16 669 652 466 178 5153 2306

low-grade is easily gradable and hence does not require more-support to the same extent. Again, the use of the analytic variant with dimension adjectives, such as long, short, large, low, high, etc. appears to be triggered by the morphological complexity and their lexical status as compounds. After all, the analytic comparative of these monosyllabic adjectives is not even found once with the corresponding simplexes in the British pilot corpus. This distribution is indicative of a lexical complexity effect associated with compounds, which triggers the use of more-support. The American English data also merits our attention. Firstly, we observe that just as their British English counterparts, the American English newspapers make use of both synthetic and analytic comparative formation strategies. What is surprising, however, is that the sum total reveals that American English makes more use of the synthetic comparative than British English. This observation stands in contrast to findings from highly different text genres by Kyt¨o and Romaine (2000) and Mondorf (2009a). Theoretically, one possible explanation

How lexicalization reflected in hyphenation affects variation and word-formation

373

could be forwarded in terms of different conventions concerning hyphenation in both national varieties. Assuming that spelling conventions (as one word, with a hyphen or as two words) reflect the degree of integration or lexicalisation, this correlation deserves further attention. If, for instance, American English used more loosely-linked compounds than British English, we would expect a larger ratio of synthetic comparatives in American English. We will return to this issue in Section 5.2. If more-support as a compensatory strategy for the lexical complexity of compounds is responsible for the selection of the analytic form with adjectives that do not take the analytic form as simplexes, then this support strategy should apply to the fourth and final group of adjectival compounds investigated.

5.1.3 Revisiting noun+adjective combinations Noun+adjective combinations such as environment-friendly, street-wise, etc. were too scarce in Mondorf’s (2000) data to warrant the formulation of meaningful claims. With only four instances of the synthetic comparative (three for environment-friendlier and one for street-wiser), even a 600 million word corpus proved too small to render meaningful results. In the meantime the database has been quadrupled, which allows us to revisit an important issue raised by the comparative formation of this group of adjectives. The question that had to remain an open issue was what motivates the choice of comparative variants with morphologically complex adjectives. In Mondorf (2000: 43) it was pointed out that theoretically, whenever we choose the synthetic variant for those adjectives in -ed, -ing or ADJ+N compounds, we form a complex lexeme in which the -er affix separates the two elements of the compound, e.g. shorter-lived, wider-ranging or lower-key. If the compound is perceived as a highly lexicalized lexeme, the -er affix might be regarded as something having an almost infix-like status.12 This infix-like status of -er could easily be avoided by means of the analytic comparative. Thus, both lexical complexity and a constraint disallowing affixation within compounds could reasonably be held responsible for the comparatively high ratio of more-support with adjectival compounds. Noun+adjective compounds allow us to show that the avoidance of affixation within compounds cannot be the whole story: Synthetic comparisons of noun+adjective combinations such as street-wise do not involve insertion of an affix, because the adjective is the second rather than the first element in the compound. If the main reason for choosing the more-variant in the remaining three groups of compounds were the avoidance of affix insertion, we would expect that noun+adjective combinations use the more-variant less often than

374

Britta Mondorf

Table 9. Synthetic comparative forms of noun+adjective compounds in British and American English (NSynthetic+Analytic = 1993) Adjective audience-friendly child-friendly customer-friendly environment-friendly family-friendly listener-friendly media-friendly street-wise trigger-happy user-friendly voter-friendly world-weary TOTAL

Synthetic BrE AmE 0 1 2 0 1 0 4 0 3 2 0 6 1 0 2 0 1 0 4 0 1 0 1 0 20 9

Synthetic (%) % BrE %AmE 0% 7% 2% 0% 1% 0% 6% 0% 1% 5% 0% 67% 2% – 4% 0% 6% 0% 0% 0% 1% 0% 3% 0% 1% 2%

N N Br 19 84 89 63 235 20 50 47 17 851 66 31 1572

N Am 13 14 28 5 40 3 0 2 2 279 3 3 392

the remaining groups and consequently are more prone to take the -er variant. Table 9 indicates that this is not the case. On the contrary, we find an exceptionally high rate of analytic comparatives for these adjectives. Thus, avoidance of affixation within the compound does not account for the variation patterns of adjectival compounds. This leaves us with an explanation in terms of the lexical complexity of compounds which is likely to call for more-support arising from a tendency to counterbalance an added processing effort associated with lexically complex compounds.

5.2

Is the use of hyphenation random?

It has widely been believed that morphologically complex words are accessed in the mental lexicon in basically two ways. Either the whole-word route is used, which means that the lexeme is directly accessed as one word. Alternatively, the decomposed route is chosen, i.e. the lexeme is stored in form of its component morphemes and assembled on the spot. More recent research findings based on eye tracking experiments suggest dual or even multiple route models which make use of all possible cues available in processing compounds (Kuperman et al. 2008: 3–6). Neural network models of the spreading activation type13 further assume that the entrenchment of concepts is linked to their frequency of use, i.e. the more often a neuron is activated, the more likely it is to fire. Its resting activation level decreases with use, so that it can be more easily activated in the future.

How lexicalization reflected in hyphenation affects variation and word-formation

375

Given that frequency plays a role in determining the resting activation of lexical items, it is clear that every access via the whole-word route strengthens the wholeword representation, whereas access on the decomposed route reinforces the representation of the decomposed morphemes and the decomposability of the complex word. (Plag 2003: 176)

The present article assumes that integration vs. decomposition of morphologically complex words is a matter of degree and that the differential degrees of integration are eventually systematically reflected in their spelling. Thus, the comparative form of one and the same complex adjective can be spelt as one word, as two words and – reflecting an intermediate degree of integration – with a hyphen. higherrisk more highrisk Integration

higher- risk more high-risk

higher risk more high risk Decomposition

Figure 4. Cline of integration

Conceiving of hyphenation as a measure of integration in compounds challenges widely accepted claims pertaining to the allegedly random use of hyphenation in compounds. Thus, Bauer (1988: 801) states that (...) hyphenation in English is totally random, and does not necessarily prove anything at all about the linguistic status of strings of elements. (Bauer 1988: 101)

He exemplifies this observation by quoting three dictionaries, each using a different spelling convention for the compound girl-friend. girlfriend One word Hamlyn’s Encyclopaedic World Dictionary girl-friend Hyphen The Concise Oxford Dictionary [also OED convention] girl friend Two words Webster’s Third New International Dictionary In the same vein, Jespersen (1977: 136) remarked on hyphenation that “(. . . ) prevailing usage is little short of chaotic”. However, research findings by Mondorf (2000: 43) clearly indicate that the degree of lexicalization is mirrored in compound spelling. Additionally, Sepp (2006: 96) and Plag et al. (2007: 215) discern systematic correspondences

376

Britta Mondorf

between compound stress and spelling: The more closely integrated in terms of spelling, the more likely is the compound to be stressed on the first element. We can now turn to ascertaining the effect of lexicalisation or integration on the choice of comparative form. It is argued that the perceived degree of integration of compound adjectives finds its expression in the spelling, i.e. as one word (higherrisk), with a hyphen (higher-risk) or as two words (higher risk). We will find that spelling conventions concerning hyphenation are not as random as has been suggested. Crucially, the choice of spelling variant also systematically corresponds to the choice of comparative variant. Figure 5 presents the correlation between integration (gauged in terms of spelling) and the choice of comparative form in the four groups of adjectival compounds/derivatives.14 Synthetic (%) 100 % 99

90 %

2141

80 %

66

3464

70 % 60 %

498

50 %

219

40 %

-ed -ing Adj + N N + Adj

30 % 20 % 10 % 0% Most integrated

1 5

0 0 1 Word

20

Hyphen

6

2 Words Least Integrated

Figure 5. Degree of Integration and Comparative Form in Four Groups of Adjectival Compounds/Derivatives (NSynthetic+Analytic = 11728) (data: Extended British corpus and American corpus described in Section 4)

Each curve is additionally labelled with the number of occurrences of the synthetic comparative in each of the three spellings in order to indicate whether the sample is large enough to allow the deduction of meaningful claims. The closer the elements of adjectival compounds are in spelling, the less likely are they to permit the -er affix and hence the more likely are they to resort to the more-variant. Conversely, the less closely integrated the elements of a

How lexicalization reflected in hyphenation affects variation and word-formation

377

compound are in spelling, the more likely are they to attach the -er affix and the more reluctant will they be to use the analytic form of comparison. This correlation can also be shown to hold individually for each of the two national varieties investigated. Figure 6 exhibits the correlation between the degree of integration and the choice of comparative form for adjective+noun combinations for British and American English. The curves are labelled by the number of synthetic comparatives in relation to the total number of both comparative variants (synthetic + analytic). The framed numbers are those showing the AmE values. For each national variety, we observe that the less closely integrated the compound, the larger the share of the -er variant and vice versa. Synthetic (%) 100% 1878/2306

336/387

80% 2141/2451 3464/5153

60% 40%

BrE AmE

20% 0%

5/685

0/18

1 Word

Hyphen

2 Words

Figure 6. Degree of Integration and Comparative Form in ADJ+N Compounds (NSynthetic+Analytic = 11000) (data: Extended British corpus and American corpus described in Section 4)

The corresponding figures for noun+adjective compounds are depicted in Figure 7. Though the occurrences of synthetic comparatives are rare, the variation pattern can be shown to systematically correlate with the degree of integration of the compound. For both national varieties, the use of the -er variant increases as the elements of the compound become more loosely bound. The present study is, for the first time, able to show that it is not the avoidance of something resembling an infix-like affix (see however note 12) that triggers the reluctance to use -er with highly integrated compounds. In fact, noun+adjective combinations (e.g. street-wise, trigger-happy, media-friendly) are even less permissive towards -er, even though with this type of adjectival compound the suffix would be attached at the end of the compound rather than

378

Britta Mondorf

Synthetic (%) 10% BrE AmE

8% 6% 2/58

4% 9/401

6/276

2% 0/0 0%

0/3 1 word

20/1592 Hyphen

2 words

Figure 7. Degree of Integration and Comparative Form in noun+adjective Compounds (NSynthetic+Analytic = 2330) (data: Extended British corpus and American corpus described in Section 4)

in the middle. This indicates that avoidance of affixation within the compound does not cause speakers to opt for the more-variant. While both Bauer (1988: 101) and Erdmann (2009) observe a tendency in American English as compared to British English to disprefer hyphenated words, this tendency is not observable for the two groups of adjectival compounds contrastively analysed in British and American English. As Table 10 illustrates American English (85%) has a larger share of hyphenated adjective+noun compounds than British English (62%). Both national varieties prefer hyphenation (e.g. higher-risk or more high-risk) over the remaining

Table 10. Hyphenation of ADJ+N Compounds and N+ADJ Compounds in British and American English (Database: British extended corpus and American English corpus described in Section 4) Compound ADJ+N

Spelling 1 Word Hyphen 2 Words

Total N+ADJ Total

1 Word Hyphen 2 Words

BrE 685 5153 2451 8289 3 1592 276 1871

AmE 8% 62% 30% 0% 85% 15%

18 2306 387 2711 0 401 58 459

1% 85% 14% 0% 87% 13%

How lexicalization reflected in hyphenation affects variation and word-formation

379

two spellings, but this preference is considerably more marked in American than in British English. Both one word spellings (higherrisk or more highrisk) and spellings without hyphen (higher risk and more high risk) are less often used in the American English data. For noun+adjective compounds, the percentages are almost identical in British and American English. An explanation for the American preference for synthetic comparatives on the basis of less closely integrated components can thus be ruled out.

6. Support for more-support from adjectival compounds/derivatives One argument in favour of an explanation of comparative alternation in terms of more-support with closely integrated compounds emanates from the observation that it is not only the increase in length that triggers the analytic form. Though word length measured by syllables is certainly related to complexity, length cannot be the only factor involved. Otherwise it would be hard to explain why disyllabic adjectives, such as street-wise should so strongly trigger moresupport, while in general the inflectional comparative is quite common with disyllabic adjectives. What is more, the individual adjectives behave very heterogeneously even if they are identical in length gauged in terms of the number of syllables and the first element is identical, as in full-blooded (93% formed analytically) and full-flavoured (6% formed analytically). Here the decisive factor that conditions the choice between the synthetic -er variant and the analytic more variant can obviously not be the number of syllables. Another factor that cannot solely account for the distribution of comparative variants in the compounds investigated is so-called left constituent frequency. Eye-tracking experiments by Kuperman et al. (2008) indicate that more frequent compounds are easier to process. This frequency effect proved stronger if the first element in a compound was relatively rare, while it was weaker if the first element in a compound was frequent. In the above-mentioned examples this effect cannot come to bear, since identical left elements (e.g. full) yield different percentages for each comparative variant. The members of so-called morphological families of constituents, which share a constituent (such as full-blooded, full-flavoured or family-friendly, listener-friendly) have been reported to elicit shorter lexical decision latencies by De Jong et al. (2000, 2002), Dijkstra et al. (2005) and Kuperman et al. (2008). Though this factor cannot explain the divergent patterns of comparative forma-

380

Britta Mondorf

tion observed in this study within the sets of compounds sharing a member, it might well be one of the reasons why we find so many instances of compound adjectives sharing e.g. friendly as second element. Stein (1976: 31) has observed positional restrictions on literal as opposed to figurative -ed formations, an aspect which is also likely to have a bearing on the choice of the comparative form but which would require additional semantic analyses that cannot be accommodated within the scope of the present article. In any case, both the syntactic factor of adjective position and the semantic factor of concreteness have been found to affect the choice of comparative form. Several studies on positional effects concur in that non-attributive uses trigger an increased ratio of the more-variant than attributive uses (cf. Braun 1982: 89, Leech and Culpeper 1997: 366, Lindquist 2000: 125, Mondorf 2003: 287 and Szmrecsanyi 2006). Mondorf (2007) additionally shows that abstract senses of adjectives are more likely to be used with analytic forms of comparison than concrete senses, a finding interpreted as a semantic complexity effect triggering more-support. A factor producing higher percentages for the more-variant is the existence of parallel structures. Thus, while a full body is a fairly well established phrase, a full blood would hardly ever occur.15 Analyses of the influence of parallel structures reveal that the less entrenched the parallel structure, the higher is the use of the more-variant (cf. Mondorf 2007: 224).

7. Conclusion In Mondorf (2000), the degree of lexicalisation or integration between the elements of a compound has been shown to correlate with the adjective’s likelihood to take the synthetic comparative. By quadrupling the database as well as by including data from different varieties of English, the present article confirms the initial observations. In addition to that, the present study is, for the first time, able to show that it is not the avoidance of an infix-like -er affix (e.g. shorterterm) that triggers the incremental use of the more-variant in highly lexicalized compounds. Otherwise, noun+adjective combinations (e.g. street-wise, triggerhappy, media-friendly) should be less prone to resort to the analytic comparative, since the -er suffix is attached at the end of the compound rather than in the middle. Exactly the opposite is the case: those adjectives which do not run the risk of affix insertion in the synthetic comparative form are the ones to exhibit the lowest proclivity towards synthetic comparatives. This indicates that avoidance of affixation within the compound does not cause language users to opt for the more-variant.

How lexicalization reflected in hyphenation affects variation and word-formation

381

Different degrees of lexicalisation or integration gauged in terms of hyphenation are shown to systematically affect comparative alternation in both British and American English. The more tightly integrated the elements of adjectival compounds are in spelling, the less prone are they to permit the -er affix and the more likely are they to resort to the analytic variant. Conversely, the less closely integrated the elements of a compound are in spelling, the more prone are they to attach the -er affix and the more reluctant will they be to use the morevariant. Graphemic conventions in compounds thus correlate with the choice of comparative variant. The empirical analysis of the theoretical issues raised in the present study shows that hyphenation is not as random as is often suggested and language users have strong intuitions about the degree of lexicalisation of a compound. Moreover, we can clearly falsify statements claiming that compound adjectives do not take inflections and the rejection of this claim can now even be extended to American English data. The correlation between lexicalization and choice of comparative form is well in line with the second maxim of Haiman’s (1983: 782–783) iconically motivated Distance Principle which states, “the linguistic separateness of an expression corresponds to the conceptual independence of the object or event which it represents.” Highly integrated lexemes do not combine expressions of conceptually independent objects but each of their elements requires an interpretation that is highly dependent on the other elements within the compound. The choice of the analytic comparative for highly integrated compounds might be indicative of a use of more-support that aims at compensating for an emergent lexical complexity. In addition, the analysis of compound adjectives – again – demonstrates that the number of syllables is not the only or even most important factor constraining the choice between both comparative variants. And finally, a clear delimitation of compound vs. syntactic group appears arbitrary and theoretically unfounded. Problems not only arise from the often discussed borderline cases, such as those in -ed, but also from the need to account for more and less prototypical compounds that fulfil phonological, morphological, semantic and syntactic conditions of ‘compoundness’ to different degrees. The transition in lexicalisation appears to be observable for all four types of morphologically complex adjectives investigated in the present study. And the extent to which a morphologically complex adjective is considered a single entity is reflected in the comparative formation strategy it prefers.

382

Britta Mondorf

Notes 1. A notable exception are prefixed adjectives such as unfriendlier (cf. e.g. Mondorf 2009b). 2. Gauging the degree of lexicalization in terms of spelling allows us to employ a parameter that has the advantage of being well operationable since it is less prone to interpretation than, for instance, semantic criteria and since it is easily amenable to corpus-retrieval. This is not to belittle the relevance of other parameters of lexicalization, such as phonological, morphological, syntactic, semantic, pragmatic and frequency-related dimensions discussed, for instance, in Brinton and Traugott (2005). 3. The compound status of adjectives in -ed, such as broad-based, has been subject to some debate (cf. Hirtle 1969, Hudson 1975, Beard 1976, Lljung 1976, Stein 1976, Welte 1982 to cite only a few). The alternative terms ‘morphologically complex’ adjectives or ‘derivatives’ have the disadvantage that they additionally denote adjectives, such as those in (e.g. lucky), which are irrelevant for the issue under investigation. As regards the status of -ed adjectives as compounds, the point made here is, however, sufficiently buttressed by the remaining three types of adjectives and the -ed adjectives do not behave markedly different. 4. No adjective in the data occurs in all nine spellings. The form more highrisk is actually not attested in the data. However, corresponding forms do occur for other adjectives falling into the [ADJ+N]ADJ template, e.g. more highbrow, more hardline, etc. 5. Cf. Mondorf (2000: 37–39) for a more detailed discussion. 6. Their impact on the integration of adjectival compounds can, however, not be ascertained within the confines of the present article, but it provides a promising avenue for further research. 7. A more detailed discussion of the problem of defining cognitive complexity and why the more-variant might facilitate processing is found in Mondorf (2009b). 8. Several corpora contain multiple entries of identical occurrences, which have been eliminated from the tally.As these doublets had to be discarded, the resultant databases are slightly smaller than is suggested by their word quantities. 9. Though the BNC comprises spoken samples, the effect of the medium is considered negligible, since the proportion of spoken to written data amounts to approximately 10:900 million words, i.e. merely 1% spoken English. 10. The figures for those two adjective groups ending in -ing and -ed are taken from Mondorf’s (2000: 35) investigation of the British pilot corpus described in Section 4. The figures for the two adjective groups ADJ+N and N+ADJ compounds are based on the extended British corpus (cf. Section 4). The reason why larger corpus sizes became necessary lies in the scarcity of comparatives for N+ADJ compounds, such as street-wiser or more street-wise, which will be given centre stage in the successive analyses. In Mondorf (2000: 40) only 30 instances were found altogether. ADJ+N combinations (which serve as a control group) have been additionally analysed in the extended British corpus in order to assess whether the modification of the database might distort results. The table provides the result for hyphenated spellings.

How lexicalization reflected in hyphenation affects variation and word-formation

383

11. Cf. Kuperman et al. (2008: 6). 12. The term ‘infix-like’ is used here for want of a more appropriate term. By definition, infixation requires the insertion of an affix within a morpheme. In our case, the affix is clearly inserted at morpheme boundaries and hence not a case of infixation. Similarly, the term ‘linking element’ (cf. N¨ubling and Szczepaniak 2009) seems not completely suitable here. Linking elements need not be morphemic, while -er clearly is. 13. Cf. e.g. McClelland and Rumelhart (1981), Dell (1986), MacKay (1987) or Berg (2005). 14. Note that the present study investigates the correlation between the spelling of the comparative form and the choice of comparative variant. This is not to deny potential effects of the spelling of the positive form (e.g. a high-risk approach) on comparative alternation or that of the spelling of parallel structures (e.g. she realized that she was taking a high risk). The impact of parallel structures is summarized on page 23 and investigated in detail in Mondorf (2007: 224). 15. Rare instances of full blood can be found in the data. They belong to medical terminology denoting blood testing procedures.

References Primary sources British National Corpus (BNC) 1995 Version 1.0. BNC Consortium/Oxford University Computing Services. Detroit Free Press on CD-ROM 1992–1995 Copyright: Knight-Ridder Information Inc. Software copyright: Dialog Information Services Inc. Los Angeles Times on CD-ROM 1992–1999 Copyright:Times Mirror Company. Software copyright: Dialog Information Services Inc./Knight-Ridder Information, Inc. The Daily Mail and The Mail on Sunday on CD-ROM 1993–2000 Produced by Financial Times Information. Distributed by Chadwyck-Healey. The Daily Telegraph and Sunday Telegraph on CD-ROM 1991–2004 Produced by FT Profile/FT Electronic Publishing. Distributed by Chadwyck-Healey. Copyright: The Telegraph PLC/The Telegraph Group Ltd. Software copyright: Personal Library Software Inc. The Guardian (including The Observer 1994–2005) on CD-ROM 1990–2005 Copyright: Chadwyck-Healey/Guardian Newspapers Ltd. Software copyright: The Open University. The Times and Sunday Times Compact Disk Edition 1990–2004 Copyright: Times Newspapers Ltd. Software copyright: The Open University/Chadwyck-Healey. Washington Times 1990–92 Wayzata Technology.

384

Britta Mondorf

Secondary sources Allen, Margaret 1978 Bauer, Laurie 1988 Beard, Robert 1976 Berg, Thomas 2005 Booij, Geert 1993

Morphological Investigations. Unpublished Ph.D. Thesis, University of Connecticut. Introducing Linguistic Morphology. Edinburgh: Edinburgh University Press. Once More on the Analysis of -ed Adjectives. Journal of Linguistics 12: 155–157. A StructuralAccount of Phonological Paraphasis. Brain and Language 94 (1): 104–129.

Against Split Morphology. In Yearbook of Morphology, Geert Booij and Jaap van Marle (eds.), 27–49. Dordrecht: Kluwer. Booij, Geert and Jerzy Rubach 1984 Morphological and Prosodic Domains in Lexical Phonology. Phonology Yearbook 1: 1–27. Boyd, Jeremy 2007 Comparatively speaking: A Psycholinguistic Study of Optionality in Grammar. Ph.D. Thesis, University of California, San Diego. Braun, Albert 1982 Studien zur Syntax und Morphologie der Steigerungsformen im Englischen. (Schweizer Anglistische Arbeiten 110.) Bern: Francke. Brinton, Laurel J. and Elizabeth Closs Traugott 2005 Lexicalization and Language Change. (Research Surveys in Linguistics.) Cambridge: Cambridge University Press. De Jong, Nivja H., Robert Schreuder, and R. Harald Baayen 2000 The morphological family size effect and morphology. Language and Cognitive Processes 15: 329–365. De Jong, Nivja H., Laurie B. Feldman, Robert Schreuder, Matthew Pastizzo, and Harald R. Baayen 2002 The Processing and Representation of Dutch and English Compounds: Peripheral Morphological and Central Orthographic Effects. Brain and Language 81(1–3): 555–567. Dell, Gary S. 1986 A Spreading Activation Theory of Retrieval in Sentence Production. Psychological Review 93: 283–321.

How lexicalization reflected in hyphenation affects variation and word-formation

385

Dijkstra, Ton, Ferm´ın Moscoso del Prado Mart´ın, B´eryl Schulpen, Robert Schreuder, and Harald R. Baayen 2005 A Roommate in Cream: Morphological Family Size Effects on Interlingual Homograph Recognition. Language and Cognitive Processes 20: 7–41. Erdmann, Peter 2009 Compound Verbs in American and British English. In One Language, Two Grammars? Differences between British and American English, G¨unter Rohdenburg and Julia Schl¨uter (eds.), 38–59. (Studies in English Language) Cambridge: Cambridge University Press. Giv´on, Tom 1979 On Understanding Grammar. New York: Academic Press. Graziano-King, Janine and Helen Smith Cairns 2005 Acquisition of English Comparative Adjectives. Journal of Child Language 32: 345–373. Haiman, John 1983 Iconic and Economic Motivation. Language 59: 781–819. Hawkins, John A. 1994 A Performance Theory of Order and Constituency. Cambridge: Cambridge University Press. 1999 Processing Complexity and Filler-gap Dependencies across Grammars. Language 75(2): 244–285. 2000 The Relative Order of Prepositional Phrases in English: Going beyond Manner-Place-Time. Language Variation and Change 11: 231–266. 2003 Why Are Zero-marked Phrases Close to their Heads? In Determinants of Grammatical Variation in English, G¨unter Rohdenburg and Britta Mondorf (eds.), 175–204. (Topics in English Linguistics 43.) Berlin: Mouton de Gruyter. Hay, Jennifer 2001 Lexical Frequency in Morphology: Is Everything Relative? Linguistics 39(6): 1041–1070. Hirtle, Walter H. 1969 -Ed Adjectives like ‘Verandahed’ and ‘Blue-eyed’. Journal of Linguistics 6: 19–36. Hudson, Richard A. 1975 Problems in the Analysis of Ed-Adjectives. Journal of Linguistics 11: 69–72. Jespersen, Otto 1977 Essentials of English Grammar. Second Edition. London: George Allen and Unwin. Kiparsky, Paul 1982 Word Formation and the Lexicon. In Proceedings of the 1982 MidAmerica Linguistics Conference, Lawrence, University of Kansas,

386

Britta Mondorf

Frances Ingemann (ed.), 3–29. Lawrence: Department of Linguistics, University of Kansas. Kuperman, Victor, Robert Schreuder, Raymond Bertram, and Harald R. Baayen 2008 Reading Polymorphemic Dutch Compounds: Towards a Multiple Route Model of Lexical Processing. Unpublished Manuscript, Max Planck Institut, Nijmegen. Kyt¨o, Merja and Suzanne Romaine 2000 Adjective Comparison and Standardisation Processes in American and British English from 1620 to the Present. In The Development of Standard English, 1300–1800, Laura Wright (ed.), 171–194. Cambridge: Cambridge University Press. Leech, Geoffrey and Jonathan Culpeper 1997 The Comparison ofAdjectives in Recent British English. In To Explain the Present: Studies in the Changing English Language in Honour of Matti Rissanen, Terttu Nevalainen and Tarrka Leena Kahlas (eds.), 353–373. (Memoires de la Soci´et´e N´eophilologique de Helsinki 52.) Helsinki: Soci´et´e N´eophilologique. Lindquist, Hans 2000 Livelier or More Lively? Syntactic and Contextual Factors Influencing the Comparison of Disyllabic Adjectives. In Analyses and Techniques in Describing English: Papers from the Nineteenth International Conference on English Language Research on Computerized Corpora, ICAME 1998, John M. Kirk (ed.), 125–132. Amsterdam: Rodopi. Lipka, Leonhard 1977 Lexikalisierung, Idiomatisierung und Hypostasierung als Probleme einer synchronischen Wortbildungslehre. In Perspektiven der Wortbildungsforschung, Herbert Ernst Brekle and Dieter Kastovsky (eds.), 155–164. Bonn: Bouvier. Lljung, Magnus 1976 -Ed Adjectives Revisited. Journal of Linguistics 12: 159–168. MacKay, Donald G. 1987 The Organization of Perception and Action. A Theory for Language and Other Cognitive Skills. New York: Springer. McClelland, James L. and David E. Rumelhart 1981 An Interactive Activation Model of Context Effects in Letter Perception, Part I, An Account of Basic Findings. Psychological Review 88: 375–407. Mondorf, Britta 2000 Wider-Ranging vs. More Old-Fashioned: Views on Comparative Formation in Adjectival Compounds/Derivatives. In Proceedings of the Anglistentag 1999, Mainz, Bernhard Reitz and Sigrid Rieuwerts (eds.), 35–44. (Proceedings of the Conference of the German Association of

How lexicalization reflected in hyphenation affects variation and word-formation

387

University Teachers of English 21.) Trier: Wissenschaftlicher Verlag Trier. 2002 The Effect of Prepositional Complements on the Choice of Synthetic or Analytic Comparatives. In Perspectives on Prepositions, Hubert Cuyckens and G¨unter Radden (eds.), 65–78. T¨ubingen: Niemeyer. 2003 Support for More-Support. In Determinants of GrammaticalVariation in English, G¨unter Rohdenburg and Britta Mondorf (eds.), 251–304. Berlin: Mouton de Gruyter. 2004 Comparative Alternation from Early Modern English to Present Day English. Paper Presented at the Thirteenth International Conference on English Historical Linguistics (ICEHL 13), Vienna, 23–28 August 2004. 2007 Recalcitrant Problems of Comparative Alternation and New Insights Emerging from Internet Data. In Corpus Linguistics and the Web, Marianne Hundt, Nadja Nesselhauf, and Carolin Biewer (eds.), 211– 232. (Language and Computers 59.) Amsterdam: Rodopi. 2009a Synthetic and Analytic Comparatives. In One Language – Two Grammars? Grammatical Differences between British and American English, G¨unter Rohdenburg and Julia Schl¨uter (eds.), 86–107. (Studies in Language). Cambridge: Cambridge University Press. 2009b More Support for More-Support: The Role of Processing Constraints on the Choice between Synthetic and Analytic Comparative Forms (Studies in Linguistic Variation 4). Amsterdam: Benjamins. N¨ubling, Damaris and Renata Szczepaniak 2009 On the Way from Morphology to Phonology: German Linking Elements and the Role of the Phonological Word. Morphology 18: 1–25. Plag, Ingo 1999 Morphological Productivity: Structural Constraints in English Derivation. Berlin: Mouton de Gruyter. 2003 Word-Formation in English. (Cambridge Textbooks in Linguistics) Cambridge: Cambridge University Press. Plag, Ingo, Gero Kunter, and Sabine Lappe 2007 Testing Hypotheses about Compound Stress Assignment in English: A Corpus-Based Investigation. Corpus Linguistics and Linguistic Theory 3(2): 199–233. Reis, Marga 1983 Gegen die Kompositionstheorie der Affigierung. Zeitschrift f¨ur Sprachwissenschaft 2(1): 110–131. Rohdenburg, G¨unter 1996 Cognitive Complexity and Increased Grammatical Explicitness in English. Cognitive Linguistics 7(2): 149–182.

388

Britta Mondorf

Schreuder, Robert and Harald R. Baayen 1997 How Complex Simplex Words Can Be. Journal of Memory and Language 37(1): 118–139. Selkirk, Elisabeth O. 1982 The Syntax of Words. Cambridge, MA: MIT Press. Sepp, Mary 2006 Phonological Constraints and Free Variation in Compounding: A Corpus Study of English and Estonian Noun Compounds. Ph.D. Thesis, City University of New York. Siegel, Dorothy 1974 Topics in English Morphology. Ph.D.Thesis, MIT. NewYork: Garland. Stein, Gabriele 1976 Semi-Productive Lexical Rules: A Note on -Ed Adjectives. Journal of English Linguistics 10: 30–33. Szmrecsanyi, Benedikt 2006 Morphosyntactic Persistence in Spoken English: A Corpus-Study at the Intersection of Variationist Sociolinguistics, Psycholinguistics and Discourse Analysis. (Trends in Linguistics. Studies and Monographs 177.) Berlin: Mouton de Gruyter. Welte, Werner 1982 Synchrone und diachrone Aspekte kontrastiver Wortbildung: Adjektivkomposita im Englischen und Deutschen. In Sprachtheorie und angewandte Linguistik. Festschrift f¨ur Alfred Wollmann zum 60. Geburtstag, Werner Welte (ed.), 161–183. (T¨ubinger Beitr¨age zur Linguistik 195.) T¨ubingen: Narr.

Variation in German adjective inflection: A corpus study Said Sahel Abstract. The present paper reports the results of a quantitative corpus study on the variation of German adjective inflection after so-called pronominal adjectives. Focusing on the four pronominal adjectives all- ‘all’, beid- ‘both’, manch‘some’, and s¨amtlich- ‘all’, issues concerning the extent of variation in this field of German adjective inflection as well as the factors which determine it are addressed. It is found that the distribution of strong and weak adjective inflection differs after the same pronominal adjective. However, this variation is neither free nor arbitrary, but rather determined by morphosyntactic factors like case, number and government. The implications of the findings for the description and systematization of the variation in this field of adjective inflection are discussed.

1. Introduction German adjective inflection is characterized in some areas by variation. However, up to now, research in German syntax has paid little attention to this variation. This lack of interest could be related to the view that the use of German adjective inflection is largely consistent and that the variation which does occur is a peripheral phenomenon. In fact, there are some constellations in which variation in the use of German adjective inflection is observed. One of these constellations is where an adjective, being part of a noun phrase, is preceded by a so-called pronominal adjective. In such a case, strong as well as weak adjective inflection is used. Most German reference grammars (e.g., Dudenredaktion 2005: 969; Jung [1966] 1990: 296) contain references to the variation of adjective inflection after pronominal adjectives and consider this variation as a particularity of German adjective inflection. However, further statements concerning the extent of the variation are notable by their absence. In studying linguistic variation, the quantification of the variation is one of the most interesting issues to be addressed. At least two types of linguistic evidence obtained using two different methods are relevant when dealing with this issue. The first type is grammaticality judgments elicited by questioning native speakers. Potential preferential asymmetries in the

390

Said Sahel

grammaticality judgments can provide information about the extent of the variation. The second type of linguistic evidence is corpus data. In this case, the frequency distributions of strong and weak adjective inflection in the studied corpus can reveal the extent of the variation. As far as methodological aspects are concerned, corpus data have certain advantages over grammaticality judgments, the most important one being their natural emergence context. A corpus is a collection of texts that originate from natural language use and is therefore not specifically constructed for studying a particular linguistic phenomenon. In contrast, grammaticality judgments do not stem from a natural situation of language use but are the result of metalinguistic tasks in which competent speakers have to decide on the grammaticality of a given utterance. Grammaticality judgments represent an atypical way to make use of our linguistic competence because we normally produce and understand language but we do not give grammaticality judgments about utterances (Klein 1998: 25). A further advantage of corpus data is the availability of large electronic corpora in which the retrieval of a large amount of data samples can be achieved automatically within a reasonable time span. For the reasons mentioned above, corpus data are used in the present paper in order to study the variation of adjective inflection in actual usage of German. The focus of interest will be on the following two issues: 1) the extent of variation after so-called pronominal adjectives, and 2) the factors that determine variation in this field of German adjective inflection.

2. German adjective inflection Like other Germanic languages, German exhibits two types of adjective inflection (for an overview, see contributions in K¨onig and Auwera (eds.) 1994) with a distinction being made between so-called strong and weak inflection. Adjectives in German can therefore take a strong or a weak inflectional ending. As mentioned above, dual inflection of adjectives is characteristic not only of German, but also of other Germanic languages (e.g., Scandinavian languages, Dutch). However, the particularity of German adjective inflection lies in the principles of its distribution in the noun phrase. Unlike all other Germanic languages which use strong and weak inflection, the distribution of strong and weak adjective inflection in German is not governed by the definiteness of the noun phrase in which the adjective occurs (for the distribution of adjective inflection in Germanic languages, see contributions in K¨onig and Auwera (eds.) 1994). In these languages, weak adjective inflection indicates the definiteness of a noun phrase, whereas strong adjective inflection indicates its indefiniteness.

Variation in German adjective inflection: A corpus study

391

Table 1. German adjective inflection Singular

Nom Acc Dat Gen

Strong inflection Masculine Feminine Neuter klein-er klein-e klein-es klein-en klein-e klein-es klein-em klein-er klein-em klein-en klein-er klein-en

Weak inflection Masculine Feminine Neuter klein-e klein-e klein-e klein-en klein-e klein-e klein-en klein-en klein-en klein-en klein-en klein-en

Plural Nom Acc Dat Gen

Strong inflection klein-e klein-e klein-en klein-er

Weak inflection klein-en klein-en klein-en klein-en

In German, however, the distribution of adjective inflection is governed by a formal principle according to which the adjective requires a weak inflection if preceded by a determiner with an inflectional ending and in other cases the adjective takes a strong inflection (Dudenredaktion 2005: 368). (1)

dieses klein-e Haus this small-Weak house ‘this small house’

(2)

mein klein-es Haus my small-Strong house ‘my small house’

(3)

klein-es Haus small-Strong house ‘small house’

(4)

Peters klein-es Haus Peter’s small-Strong house ‘Peter’s small house’

In example (1), the adjective kleine takes the weak inflectional ending -e because it is preceded by an inflected determiner, namely dieses. The demonstrative pronoun dieses consists of the stem dies and the inflectional ending -es. In the examples (2) to (4), the adjective kleines is preceded by the possessive pronoun mein without an inflectional ending (2), by a so-called zero-determiner (3), or

392

Said Sahel

by a prenominal genitive Peters (4). In these three cases, the adjective takes a strong inflection because none of the elements preceding the adjective has an inflectional ending. Obviously, there is a complementary distribution of strong and weak inflectional endings in the noun phrase: either the determiner or the adjective takes the strong inflectional ending. This complementary distribution of strong and weak inflectional endings in the noun phrase is known in German linguistics as the principle of monoflexion.

3. Adjective inflection after pronominal adjectives The variation of adjective inflection after a pronominal adjective is traditionally described as being lexically motivated. There is broad agreement in the descriptive literature that the variation of adjective inflection after pronominal adjectives is mainly due to the ambiguous word category of these lexemes (e.g., Behagel 1923: 200, Paul [1919] 1954: 102, W¨alterlin 1941: 25, Dudenredaktion 2005: 969). Pronominal adjectives range between determiners and adjectives. After the same pronominal adjective, an attributive adjective can take a weak or a strong inflectional ending. If the pronominal adjective preceding the attributive adjective is considered by the speaker/writer as a determiner, the attributive adjective takes a weak inflectional ending (see example (5)). If, on the other hand, the pronominal adjective preceding the attributive adjective is considered an adjective, the attributive adjective takes a strong inflectional ending (see example (6)). (5)

s¨amtliche politisch-en Beobachter all political-Weak observers ‘all political observers’

(6)

s¨amtliche elektronisch-e Medien all electronic-Strong media ‘all electronic media’

Pronominal adjectives are not a homogenous word class, but rather a fuzzy category. The term pronominal adjective reflects the morphosyntactically inconsistent influence of these lexemes on the inflection of the subsequent attributive adjective. At times, they influence the inflection of a subsequent adjective as a pronoun or a determiner and at other times as an adjective. Pronominal adjectives are mainly made up of quantifiers such as all- ‘all’, s¨amtlich- ‘all’, einig- ‘some’, manch- ‘some’, viel- ‘many’, and beid- ‘both’ or demonstrative adjectives such as solch- ‘such’. In the literature, the delimitation of pronominal adjectives is

Variation in German adjective inflection: A corpus study

393

neither clear nor consistent as there are no established criteria which help to distinguish pronominal adjectives from common adjectives or determiners. Since the delimitation of pronominal adjectives is not the focus of interest in the present study, we dispense here with further discussion of this issue and consider those lexemes which inconsistently influence the inflection of subsequent adjectives as pronominal adjectives.

4. The corpus study This section presents the corpus study. Section 4.1 describes the database used for studying the variation of adjective inflection after pronominal adjectives and the criteria we applied to collect the data samples. In Section 4.2, we introduce the factors whose impact on the variation will be examined and the statistical method applied for these purposes. In Section 4.3, the results of the corpus analysis are reported.

4.1 Material and method The corpus study is based on data from theTagesspiegel corpus, which is part of a large electronic corpus collection in the DWDS project. The Tagesspiegel corpus contains the online editions of the German daily newspaper Der Tagesspiegel published between January 1996 and June 2005 and consists of 170 million running words which have been morphologically annotated. The annotation contains information about the lemma and the word class of each word form and makes an automated extraction of samples possible. In the present study, relevant data samples are those noun phrases which consist of an inflected pronominal adjective followed by an attributive adjective and a noun, such as in (7): (7)

s¨amtliche politischen Beobachter pronominal adjective attributive adjective noun

Relevant data samples were extracted automatically by entering a query into the search engine of the corpus. The query contained the respective pronominal adjective followed by the tags of attributive adjectives and nouns. For each query, the first 500 hits were considered and all were then manually checked in order to filter out irrelevant samples. Most irrelevant samples were due to the syncretism characteristic of German inflectional paradigms. All data samples containing a pronominal adjective with inflectional features other than those intended were filtered out.

394

Said Sahel

The corpus was searched for the four pronominal adjectives all- ‘all’, beid‘both’, manch- ‘some’, and s¨amtlich- ‘all’. For each of them, two conditions were defined. The choice of these four lexemes to tackle the variation of adjective inflection after pronominal adjectives and its determinants is motivated by several reasons, one of them being that these four lexemes are among the most frequent pronominal adjectives in contemporary language use. It can thereby be ensured that the lexemes under investigation occur in an appropriate quantity. Furthermore, the choice of the lexemes is based on the Duden reference grammar classification (Dudenredaktion 2005: 970–974) according to which the pronominal adjectives all- ‘all’, beid- ‘both’, manch- ‘some’, and s¨amtlich‘all’ respectively belong to the inflectional behavior of the subsequent adjective in two different groups. According to this classification, no variation of adjective inflection is expected after all- ‘all’ or beid- ‘both’. These two lexemes behave like definite articles, triggering weak adjective inflection. After manch‘some’ and s¨amtlich- ‘all’, however, strong as well as weak adjective inflection can occur. A further argument concerns the lexeme manch- ‘some’ in particular. The choice of manch- is primarily ascribed to the fact that this lexeme is one of few pronominal adjectives that frequently occurs in singular. By means of manch-, the influence of the category of number on the extent and direction of adjective inflection should be assessed. For the purpose of this study, it is not of particular relevance which pronominal adjective is chosen to assess the influence of the factors mentioned below (Section 4.2). More fundamental is the issue as to what extent the adjective inflection varies after the same lexeme to different degrees depending on factors other than lexical ones. For all- ‘all’, the corpus was scanned for the genitive plural form aller in two conditions of case government: 1) case is assigned to aller by a noun (8) and 2) case is assigned to aller by a preposition (9): (8)

die Untersuchung all-er m¨oglichen Faktoren the analysis all-Gen.Pl possible factors ‘the analysis of all possible factors’

(9)

trotz all-er guten Vors¨atze despite all-Gen.Pl good intentions ‘despite all good intentions’

For the pronominal adjective beid- ‘both’, the corpus was searched for two plural forms differing only in case: 1) the nominative plural form beide (10) and 2) the genitive plural form beider (11).

Variation in German adjective inflection: A corpus study

(10)

beid-e großen Parteien both-Nom.Pl big parties ‘both big parties’

(11)

die Ideen beid-er großen Parteien the ideas both-Gen.Pl big parties ‘the ideas of both big parties’

395

For the pronominal adjective manch- ‘some’, the corpus was scanned for two forms differing only in number: 1) the nominative masculine singular form mancher (12) and 2) the nominative plural form manche (13). (12)

manch-er alte Fan some-Nom.Sg.Masc old fan ‘many an old fan’

(13)

manch-e arabische Staaten some-Nom.Pl Arab countries ‘some Arab countries’

For the pronominal adjective s¨amtlich- ‘all’, the corpus was searched for the nominative plural form s¨amtliche in the two conditions of adjacency: 1) the pronominal adjective s¨amtliche and the attributive adjective are adjacent to each other (14) or 2) the pronominal adjective s¨amtliche and the attributive adjective are not adjacent to each other (15). (14)

s¨amtlich-e touristischen Aktivit¨aten all-Nom.Pl touristic activities ‘all touristic activities’

(15)

s¨amtlich-e f¨ur Eritrea wichtigen Vertr¨age all-Nom.Pl for Eritrea important agreements ‘all agreements important for Eritrea’

4.2 The factors The purpose of collecting data for four pronominal adjectives in two different conditions is to assess the influence of the morphosyntactic factors of – – – –

case, number, adjacency and government

396

Said Sahel

on the variation of adjective inflection after pronominal adjectives. In many German reference grammars, as well as in the literature on this specific field of adjective inflection, we find the assumption that these four factors might influence the variation of the inflection. In the present study, the effect of these factors will be investigated by means of a statistical analysis.

4.2.1 Case In most reference grammars of German, the grammatical category of case is implicitly considered as a factor influencing the variation of adjective inflection after pronominal adjectives, especially with plural forms. For many pronominal adjectives, it is stated that the extent of the variation of adjective inflection differs after nominative plural forms from that after genitive plural forms. The factor case has, in the present study, the two values: 1) nominative and 2) genitive. The impact of this factor on the variation of adjective inflection after pronominal adjectives is assessed by studying the distribution of the strong and weak adjective inflection after two plural forms of the pronominal adjective beid- ‘both’: beide (Nom.Pl) and beider (Gen.Pl).

4.2.2 Number Many reference grammars of German assume that the grammatical category of number has an impact on the variation of adjective inflection after pronominal adjectives. According to these reference grammars, the extent of variation depends on number, i.e. on whether the pronominal adjective is plural or singular. The factor number has two values: 1) singular and 2) plural. In order to assess the influence of this factor, we analyzed the distribution of the strong and weak adjective inflection after two forms of the pronominal adjective manch- ‘some’: mancher (Nom.Sg.Masc) and manche (Nom.Pl).

4.2.3 Adjacency The idea that adjacency might influence the variation of adjective inflection after pronominal adjectives is based on Hotzenk¨ocherle’s (1968) assumption that adjacency influences adjective inflection in general. The author cites corpus samples in which noun phrases consisting of an attributive adjective and a noun are preceded by prenominal genitives and where the pronominal genitive and the attributive adjective are not adjacent. (16)

Troxlers vom m¨achtigen Strome der abendl¨andischen Mystik gen¨ahrten psychologischen Einsichten [. . . ] passten nicht in eine Psychologie

Variation in German adjective inflection: A corpus study

397

ohne Seele. (H. Barth, Neue Z¨urcher Zeitung 20 Aug 1949, cited in Hotzenk¨ocherle 1968: 3) ‘Troxler’s psychological insights fed by the powerful currents of occidental mysticism [. . . ] did not fit into soulless psychology.’ Although the attributive adjectives gen¨ahrten and psychologischen in (16) are preceded by the prenominal genitive Troxlers and not by an inflected determiner, they take the weak inflectional ending -en rather than the strong one -e. It is assumed that this variation is caused by the fact that the prenominal genitive and the attributive adjectives are not adjacent. The factor adjacency has two values: 1) adjacent and 2) non-adjacent. The influence of this factor on the variation of adjective inflection after pronominal adjectives is assessed on the basis of the distribution of the strong and weak adjective inflection after s¨amtliche (Nom.Pl) ‘all’ in the adjacent and in the non-adjacent condition.

4.2.4 Government The factor government refers to the syntactic constellation in which case is assigned to a noun phrase containing a pronominal adjective. Based on the distinction made by generative linguistics between lexical and structural case assignment (e.g., Gallmann and Lindauer 1994: 14; Fanselow 2000), the impact of this factor on the variation of adjective inflection after pronominal adjectives is assessed. The factor government has two values: 1) case is assigned by a preposition (lexical case assignment) and 2) case is assigned by a noun (structural case assignment).The influence of this factor is examined by means of the distribution of the strong and weak adjective inflection after aller (Gen.Pl) ‘of all’ in the two conditions of government mentioned above.

4.3 Results In this section, the results of the quantitative corpus analysis are presented. Our interest will focus on the following two questions: 1) How much variation is there? and 2) Do the factors discussed above influence this variation? The extent of variation is revealed by the frequency distribution of the strong and weak inflectional endings of the attributive adjectives after the four pronominal adjectives used in the present corpus study. A statistical exploration was performed in order to find out whether the variation is influenced by the four factors discussed. The quantitative difference in the distribution of the strong and weak inflection in

398

Said Sahel

the two conditions defined for each factor was tested for statistical significance. If the difference is significant, the influence of the factor on the variation of adjective inflection after pronominal adjectives can be considered established.

4.3.1 The effect of case The assessment of the influence of case on the variation of adjective inflection after pronominal adjectives is based on two plural forms of beid- ‘both’: beide (Nom.Pl) and beider (Gen.Pl). The corpus contains 74 relevant samples of beide (Nom.Pl) and 78 relevant samples of beider (Gen.Pl). The distribution of the strong and weak adjective inflection after the two pronominal adjective forms is shown in Table 2: Table 2. The distribution of the strong and weak adjective inflection after beide (Nom.Pl) and beider (Gen.Pl) strong inflection 34% (n = 25)

Singular weak inflection 66% (n = 49)

strong inflection 67% (n = 52)

Plural weak inflection 33% (n = 26)

Table 2 shows completely different distributions of the strong and weak adjective inflection after the two case forms of the pronominal adjective beid-. After beide (Nom.Pl), the weak inflection type is more frequent (66%) whereas after beider (Gen.Pl), the strong inflection occurrence rate (67%) is higher than that of the weak inflection. Interestingly, the distribution of the two inflection types after beide (Nom.Pl) and beider (Gen.Pl) is not only different, but the reverse. After beider (Gen.Pl), strong inflection is dominant almost to the same extent that weak inflection dominates after beide (Nom.Pl). The statistical analysis revealed a highly significant difference in the distribution of the strong and weak adjective inflection after beide (Nom.Pl) and beider (Gen.Pl) and, as a result, a main effect of case on the variation of adjective inflection after pronominal adjectives: chi-square (1, N = 152) = 16.42, p < .0001.

4.3.2 The effect of number The influence of number on the variation of adjective inflection after pronominal adjectives is assessed by using mancher (Nom.Sg.Masc) and manche (Nom.Pl). The corpus contains 157 relevant samples of mancher (Nom.Sg. Masc) and 126 samples of manche (Nom.Pl). The distribution of the strong

Variation in German adjective inflection: A corpus study

399

and weak inflection after these two forms of the pronominal adjective manchcan be seen in Table 3: Table 3. The distribution of the strong and weak adjective inflection after mancher (Nom.Sg.Masc) and manche (Nom.Pl) strong inflection 10% (n = 15)

Singular weak inflection 90% (n = 142)

strong inflection 41% (n = 52)

Plural weak inflection 59% (n = 74)

Table 3 shows a different distribution of the strong and weak adjective inflection after mancher (Nom.Sg.Masc) compared to manche (Nom.Pl). While the weak inflection is more frequent after mancher (Nom.Sg.Masc), the opposite is found for manche (Nom.Pl); here, the strong inflection dominates. The statistical analysis revealed a highly significant difference in the distribution of the strong and weak adjective inflection after mancher (Nom.Sg.Masc) and manche (Nom.Pl) and, accordingly, a main effect of number on the variation of adjective inflection after pronominal adjectives: chi square (1, N = 283) = 38.91, p < .0001.

4.3.3 The effect of adjacency In order to assess the influence of adjacency on the variation of adjective inflection after pronominal adjectives, samples of the pronominal adjective form s¨amtliche (Nom/Acc.Pl), in the two conditions of adjacency defined above, were collected (see Section 4.2.3). We found 341 adjacent condition examples and 75 non-adjacent condition examples. The distribution of the strong and weak adjective inflection in these two conditions is shown in Table 4: Table 4. The distribution of the strong and weak adjective inflection after s¨amtliche in the conditions adjacent and non-adjacent strong inflection 21% (n = 72)

Adjacent weak inflection 79% (n = 269)

Non-adjacent strong inflection weak inflection 12% 88% (n = 9) (n = 66)

At first sight, the weak adjective inflection seems to be the more frequent inflection type in both adjacency conditions. Furthermore, it is apparent from Table 4 that the rate of weak inflection is higher in the non-adjacent condition than in

400

Said Sahel

the adjacent condition (88% vs. 79%). However, the statistical analysis did not reveal an effect of adjacency as the difference in the distribution of the strong and weak adjective inflection in the two conditions of adjacency failed to reach significance: chi square (1, N = 416) = 3.25, p = 0.5. A possible explanation for this failure to find an effect of adjacency is the small number of samples found for the non-adjacent condition.

4.3.4 The effect of government The influence of government on the variation of adjective inflection after pronominal adjectives is assessed by using the pronominal adjective form aller (Gen.Pl). Examples are collected in the two conditions of government defined above (see Section 4.2.4). In one condition, the genitive case is assigned to aller (Gen.Pl) by a noun, whereas in the other, the genitive case is assigned to aller (Gen.Pl) by the preposition trotz ‘despite’. We refer to the former condition of government as government by a noun and the latter condition as government by a preposition. We found 325 government by a noun condition examples and 200 government by a preposition condition examples. The distribution of the strong and weak inflection in these two conditions is presented in Table 5: Table 5. The distribution of the strong and weak adjective inflection after aller in the conditions government by noun and government by preposition Government by noun strong inflection weak inflection 3% 97% (n = 9) (n = 316)

Government by preposition strong inflection weak inflection 13,5% 86,5% (n = 27) (n = 173)

The weak adjective inflection is the most frequent inflection type after aller in both conditions. The table also shows a higher frequency of the strong adjective inflection in the government by a preposition condition compared to the government by a noun condition. The statistical analysis revealed a highly significant difference in the distribution of the strong and weak adjective inflection in these two conditions of government and accordingly a main effect of government: chi square (1, N = 525) = 22.32, p < .0001.

Variation in German adjective inflection: A corpus study

401

5. Summary and discussion This paper presents a case of variation in German grammar. This variation concerns the inflection of attributive adjectives after so-called pronominal adjectives. After the same pronominal adjective, an attributive adjective can take a strong as well as a weak inflectional ending. Applying methods of corpus linguistics, we addressed two core issues in respect of the variation of adjective inflection after pronominal adjectives. The first issue concerns the extent of the variation. The second issue looks at whether the observed variation is completely free or whether there are systematic restrictions on this variation. These issues were assessed by means of the four pronominal adjectives all- ‘all’, beid- ‘both’, manch- ‘some’ and s¨amtlich- ‘all’. First of all, it must be emphasized that our findings and the implications drawn from them are valid only for the newspaper text genre and the period between 1996 and 2005. The results cannot be generalized and transferred to other text genres nor to the linguistic modality of spoken language. The data analysis reveals variation of adjective inflection after all four pronominal adjectives including all- ‘all’, beid- ‘both’ that, following the Duden reference grammar (Dudenredaktion 2005), exclusively trigger weak inflection of the subsequent adjective. One further finding is that the adjective inflection after the same pronominal adjective varies to a different scope; the extent of variation seems to be dependent on the morphosyntactic constellation in which the pronominal adjective occurs. For the pronominal adjective beid-, the quantitative analysis revealed roughly equal variation after beide (Nom.Pl) und beider (Gen.Pl). However, the distribution of the strong and weak adjective inflection after these two inflection forms of beid- is almost exactly inverse. After beide (Nom.Pl), the weak inflection is the more frequent inflection type (66%); after beider (Gen.Pl), the strong inflection is more frequent (67%). How can this completely different distribution of strong adjective inflection be accounted for? Is this an effect of the category case itself, or does the asymmetry result from the different phonetic-phonological features of the inflection endings -e in beide (Nom.Pl) versus -er in beider (Gen.Pl)? Wiese (2004: 19) claims an influence of phonetic-phonological properties of inflection ending on the inflection of the subsequent adjective. He argues that the tendency toward weak adjective inflection after the pronominal adjective strengthens with increasing ending weight, but that the tendency toward strong adjective inflection also increases with decreasing ending weight. However, this claim cannot account for the different distribution of strong and weak adjective inflection after beide (Nom.Pl) and beider (Gen.Pl) because -e [´] and -er [å] are central vowels with almost the same weight. The rhythm also can be considered as a potential phonetic-

402

Said Sahel

phonological determinant of the observed asymmetry. It can be assumed that due to rhythmical reasons, the inflection ending -er preferably triggers a further (strong) ending -er, whereas the ending -e prefers the (weak) ending -en. Although a similar asymmetry concerning the endings -er and -e also has been shown for other pronominal adjectives than beid- (see note 14), the rhythm cannot explain all distributions of weak and strong inflection revealed by the data analysis. For the pronominal adjective manch-, it is found that after mancher (Nom.Sg.Masc) as well as manche (Nom.Pl), the weak inflection dominates (90% and 59%, respectively). In contrast to the forms of the genitive plural, the inflection ending -er in mancher (Nom.Sg.Masc) alternates with the (weak) inflection ending -e and triggers in only 10% of the cases the (strong) inflection ending -er. It can be assumed that this discrepancy either arises from the difference in the number or probably has the function of marking the difference between the two categories of number. In plural, the sequence -er – -er seems to be preferred over the alternation -er – -en, whereas in singular the alternation -er – -e is the more preferable variant compared to -er – -er. A further corpus analysis regarding mancher (Gen.Pl) provides further evidence for this assumption. After mancher (Gen.Pl) the strong adjective inflection is more frequent (86%) than the weak adjective inflection (14%). Like the other pronominal adjectives mentioned above, it is also the case in manch-, that in the genitive plural the sequence -er – -er is preferred over the alternation -er – -en. After aller (Gen.Pl), the adjective inflection shows the fewest variations. In both conditions of government (government by noun versus government by preposition), weak adjective inflection is clearly the more frequent inflection type. The pronominal adjective all- generally behaves like a definite article triggering, in most cases, the weak inflection of the subsequent adjective. This can be attributed to the fact that this pronominal adjective includes all members of the class denoted by the subsequent substantive. In previous studies, it has been often assumed that the adjective inflection after a pronominal adjective is predictable through its semantic features (e.g., Brinkmann [1962] 1971; Schatte 1994); weak adjective inflection follows pronominal adjectives with denotation properties like that of all-. On the other hand, pronominal adjectives which only include an indefinite amount of the members of the class denoted by the subsequent substantive (e.g., manch- ‘some’) trigger strong adjective inflection. Nevertheless, the rate of the strong adjective inflection after aller (Gen.Pl) is higher when the genitive is assigned by a preposition than by a noun. This difference can be ascribed to the two different syntactic constellations for the assignment of the genitive. The assignment of the genitive by the noun occurs under structural conditions, whereas the prepositional case is assigned under lexical conditions.

Variation in German adjective inflection: A corpus study

403

The results of the present study suggest that the variation of adjective inflection after pronominal adjectives is neither free nor arbitrary, but rather the subject of linguistic restrictions. Furthermore, the finding that morphosyntactic factors like case, number and government have an effect on the variation in this area of adjective inflection has implications on the description and systematization of this variation. The findings and the discussion make clear that the lexical features of the pronominal adjectives cannot exclusively determine the extent and the direction of the variation of adjective inflection after pronominal adjectives and that other linguistic factors play a role in determining this variation. Accordingly, it is obvious that monocausal attempts to describe this variation are untenable because, at best, they provide an incomplete description of the variation. The more satisfactory way of describing this variation is to take into account more linguistic factors whose influence on the variation is supported through the study of data from real language usage. In this study, it has been shown that corpus linguistic methods are appropriate when studying linguistic variation. The availability of large, annotated electronic corpora allows us to shed light on relevant issues concerning linguistics such as the extent of variation or the factors determining it.

Notes 1. The term pronominal adjective refers to those lexemes influencing the inflection of a subsequent attributive adjective sometimes as a pronoun or a determiner and at other times as an adjective (Dudenredaktion [1983] 2003: 1247). This term is not meant to presuppose any decision about the word class of these lexemes. 2. In contrast to grammatical judgments traditionally used in generative linguistics as linguistic evidence and based mainly on introspection, grammatical judgments here are meant in a quantitative sense. They are the result of data gathered by questioning a large number of native speakers. In the case of linguistic variation, the questioning could be designed as a preference decision task where participants have been asked to decide between two or more variations of an utterance. 3. See Stefanowitsch (2005) for further details about the suitability of corpus data for linguistic research. 4. In German, only adjectives used in an attributive function take an inflectional ending. Adjectives with predicative or adverbial functions always occur without inflectional endings. 5. For this reason, the dichotomy definite and indefinite adjective forms is used synonymously with the dichotomy strong and weak inflected adjectives in the grammars of these languages.

404

Said Sahel

6. In an earlier stage of German, the distribution of the strong and weak adjective inflection was also governed by the semantic principle of definiteness. 7. In other Germanic languages, the adjective in constellations such as in examples (2) and (4) does not – in contrast to German – take a strong, but a weak inflectional ending, because the respective adjective occurs in a definite noun phrase. In example (2), definiteness is indicated by the possessive pronoun, in example (4), by the prenominal genitive. 8. All the samples containing a pronominal adjective in this paper are taken from the newsletter corpus used in this study. 9. DWDS project: Das (D)igitale (W)¨orterbuch der (d)eutschen (S)prache des 20. Jahrhunderts [the digital lexicon of the German language of the twentieth century]. For more information, see http://www.dwds.de/. 10. In addition to genuine adjectives, present and perfect participles used in the function of an attributive adjective are also considered as attributive adjectives here. 11. The hyphen marks the end of the stem to which an ending can/must be affixed. 12. The restriction to nominative and genitive is due to the fact that 1) the pronominal forms of nominative and genitive plural have morphologically distinct inflection endings (-e versus -er) and 2) that, in contrast to dative (-en versus -en), the strong and weak inflection endings are not homonymous (-e versus -en, for nominative and -er versus -en, for genitive). Accusative plural is morphologically identical with nominative plural. 13. The test of significance we used to assess the influence of the factors defined above is the Chi-square-test. This statistical method is commonly used for frequency comparisons and tests whether two empirical distributions are similar by comparing the observed and the expected frequencies. In our case, the test compares whether the distribution of the strong and weak adjective inflection is similar for the two conditions of a factor. We chose a confidence interval of 95%, i.e. the likelihood that the different distribution of the strong and weak adjective inflection occurs by chance is less than 5%. Accordingly, the influence of a tested factor is considered as being established if the P-value of the Chi-square-test is not higher than 0.05. 14. This asymmetry in the distribution of strong and weak inflection is not restricted to the lexeme beid-. Sahel (2005) found that after s¨amtliche (Nom.Pl) and solche ‘such’ (Nom.Pl) the weak inflection type is more frequent (84% and 95%, respectively), whereas after s¨amtlicher (Gen.Pl) and solcher ‘such’ (Gen.Pl) the strong inflection occurrence rate (79% and 58%, respectively), is higher than that of the weak inflection.

Variation in German adjective inflection: A corpus study

405

References Behagel, Otto 1923

Deutsche Syntax. Eine geschichtliche Darstellung. Band I. Die Wortklassen und Wortformen. A. Nomen, Pronomen. Heidelberg: Winter. Brinkmann, Henning 1971 Die Deutsche Sprache: Gestalt und Leistung. Reprint 2nd ed. D¨usseldorf: Schwann. Original edition, D¨usseldorf: Schwann, 1962. Dudenredaktion (ed.) 2003 Deutsches Universalw¨orterbuch. 5th ed. Mannheim/Leipzig/Wien/ Z¨urich: Dudenverlag. Reprint. Original edition, Mannheim/Leipzig/ Wien/Z¨urich: Dudenverlag, 1983. 2005 Die Grammatik. Duden Band 4. 7th ed. Mannheim/Leipzig/Wien/ Z¨urich: Dudenverlag. Original edition, Mannheim/Leipzig/Wien/ Z¨urich: Dudenverlag. Fanselow, Gisbert 2000 Optimal exceptions. In The Lexicon in Focus, Barbara Stiebels and Dieter Wunderlich (eds.), 173–209. Berlin: Akademie Verlag. Gallmann, Peter, and Thomas Lindauer 1994 Funktionale Kategorien in Nominalphrasen. Beitr¨age zur Geschichte der deutschen Sprache und Literatur 116(1): 1–27. Hotzenk¨ocherle, Rudolf 1968 Gegenwartsprobleme im deutschen Adjektivsystem. Neuphilologische Mitteilungen 69: 1–28. Jung, Walter 1990 Grammatik der deutschen Sprache. Reprint 10th ed. Mannheim/ Leipzig: Bibliographisches Institut. Original edition Leipzig: Bibliographisches Institut, 1966. Klein, Wolfgang 1998 Ein Blick zur¨uck auf die Variet¨atengrammatik. Sociolinguistica 12: 22–38. K¨onig, Ekkehard, and Johan van der Auwera (eds.) 1994 The Germanic Languages. London/New York: Routledge. Paul, Hermann 1954 Deutsche Grammatik. Band III. Teil IV: Syntax (erste H¨alfte). Reprint 2nd ed. Halle (Saale): Niemeyer. Original edition, Halle (Saale): Niemeyer, 1919. Sahel, Said 2005 Die Variation der Adjektivflexion nach Pronominaladjektiven und einige ihrer Determinanten. Eine empirische Untersuchung. Deutsche Sprache 33(4): 355–381.

406

Said Sahel

Schatte, Christoph 1994 Eine Flexionsregel f¨ur Adjektiv- und Partizipialattribute nach pluralischen Indefinitpronomen und quantifizierenden Adjektiven. Deutsch als Fremdsprache 31: 95–99. Stefanowitsch, Anatol 1995 Quantitative Korpuslinguistik und sprachliche Wirklichkeit. In Von der Wirklichkeit zur Wissenschaft. Aktuelle Forschungsmethoden in den Sprach-, Literatur- und Kulturwissenschaften, Christiane SollteGresser, Karin Struve, and Natascha Ueckmann (eds.), 141–155. Hamburg: LIT-Verlag. W¨alterlin, Kurt 1941 Die Flexion des Adjektivs hinter Formw¨ortern in der neueren deutschsprachigen Presse. Z¨urich: Lang. Wiese, Bernd 2004 Zur Systematisierung der Schwankungen zwischen starker und schwacher Adjektivflexion nach Pronominaladjektiven. (www.idsmannheim.de/gra/texte/wi5.pdf; 22 Sep 2008)

Index acceptability judgment, 204, 217 accusative, 325–335, 337 adjacency, 396, 399 adjective inflection (weak vs. strong), 390–403 age, 164, 167, 170, 172 agent, 83 agreement, see agreement (in prepositional phrases), agreement (object/verb), agreement (pronoun/be), agreement (subject/verb), Northern Subject Rule agreement (in prepositional phrases), 173 agreement (object/verb), 87–89, 348 agreement (pronoun/be), 278–291 agreement (subject/verb), 271, 272. See also agreement (pronoun/be), Northern Subject Rule allostructional model, 186–189 alternation of schwa (German), 119, 120, 125 ambiguity (categorial) , 99, 104, 106, 107 ambiguity (structural), 110, 111 American English, 363, 371–379 attitude, 304, 311, 312 auxiliary, 26, 28–34, 37, 40, 49, 60–62 bare noun (as a predicative), 100–108 base-generation, 141–145 Burzio’s Generalization, 86 case, 396, 398. See also accusative, case assignment, case matching, dative, dative substitution, nominative object, oblique case case assignment, 164–170 case matching, 150, 153, 154 categorial underspecification, 112 category change, 103, 107

cause, 315, 316 causer, see cause change of state, 302–304 Cimbrian, 22, 31–33 coherent construction, 310, 311 competence/performance dichotomy, 6–8 competing grammars, 10, 60 Complexity Principle, 364 compositionality, 191 compound (adjectival), 362–381 conditioned variation, 150 Construction Grammar, 12, 272, 273, 289, 290 constructionist approach, 272, 274, 279, 282, 292. See also construction grammar contraction preposition+article, 166, 169, 172 conversion, 102–107 corpus, 60, 234, 367 (English); 393 (German); 205 (Heerlen Dutch); 256 (Qu´ebec French). See also FRED, plays dative, 135–154, 311, 313, 317, 318, 339–344. See also dative substitution dative substitution, 90–92 declarative analysis, 122, 125–129 dependent marking, 349 descriptive linguistics, 164, 169 dialect attrition, 174 dialect/standard continuum, 164, 174, 175 dialectal variation, 8, 146–148 diatopic variation, see dialectal variation differential object marking, 326, 337 Distance Principle, 381 Dutch, 37, 38–46, 203–221, 225–250. See also Heerlen Dutch

408

Index

English, 59–75, 183–196, 271–291, 361–381. See also American English entrenchment, 190, 274, 275, 278–280, 282–285 epistemic modality, 300–303 establishedness, 190 evidentiality, 300–305, 312 expletive subject, 306, 313, 314 explicit articulation, 119, 120, 121, 130 extra argument, 317, 318 extralinguistic determinants, 3, 4 extraposition, 231, 238 focus, 39, 192–194, 345 four verb cluster, 38 FRED (Freiburg English Dialect Corpus), 278, 286 free variation, 5, 64–67, 183, 194, 198 French, 23. See also Qu´ebec French frequency, 171, 172, 273, 287, 288, 291, 292. See also entrenchment functional head, 82, 83 generalization in paradigms, 278–290 German, 23–26, 35, 38–46, 91, 99–112, 117–132, 139, 164–166, 301–323, 389–403. See also Cimbrian, Pladen German, Ruhrdeutsch, Swiss German got, see possessive got government, 397, 400 gradience, 5 grammatical complexity, 119, 120, 131 grammatical variation, 4, 168–170, 174, 175 have, see possessive have head marking, 349 Heerlen Dutch, 205–212 hyphenation, 363, 374–379 Icelandic, 81–92. See also Old Icelandic idealization, 6, 59 identifiability, 326, 337

idiolectal variability, 204, 205, 213–215, 219 impersonal construction, 307–309 implicational tendency, 281, 285 infinitival complement, 298, 299, 305, 306, 317 information structure, 345. See also focus, secondary topic, topic information value, 240, 245, 246 inherence, 232, 241, 245 interrogative marker -tu (Qu´ebec French), 255, 259–261, 264–266 inter-speaker variation, 8, 148, 204, 208 intonation (interrogative), 255, 258, 261 intralinguistic determinants, 4 intra-speaker variation, 8, 149, 204, 208 inversion (interrogative), 255, 256, 259, 261–266 Italian, 23 Komi, 325–353 language change (in real time), 257, 261–264 language documentation, 163, 167 language use, 206 lexeme, 118, 122 lexical choice, 60, 67 lexical integrity, 196 lexicalization, 184, 189–191, 362, 375, 386 literalness, 184, 191–194 logistic regression, 227 microvariation, 1, 23–26, 203, 205 middle field, 231, 240 Minimalist Program, 7, 11, 145, 151, 155 modality, 298–302, 311, 312, 316 more-support, 366, 368, 371–374, 379 morphological category, 126, 131 morphological families, 371 morphologically conditioned variation, 125–129 movement, 142–145 multivariate, 234, 248

Index nominative object, 87–89 non-compositionality, 191 Northern Subject Rule, 274, 275, 277 noun incorporation, 108–110 number, 396, 398 Ob-Ugric, 347–349 oblique case, 139–142, 145 Old Icelandic, 89, 91 opacity, 191 operationalization, 235, 237–243 Optimality Theory, 11, 34–37, 45, 152–154 OV language, 21, 22, 26–34, 40, 46 participle in verb clusters (Dutch): categorial status, 232, 233, 236, 241 participle in verb clusters (Dutch): frequency, 235, 242, 243, 249 participle in verb clusters (Dutch): morphology, 231, 238 particle placement alternation, 183–198 passive, 307–309 persistence, 233, 243 phonological word, 128, 131 phonologically conditioned variation, 125, 126 phrasal verbs, 184–198 Pladen German, 22, 27–31 plays, 256, 262 possessive got, 63, 67–73 possessive have, 60, 62, 73 possessive suffix, 329–332 predicative (as a complement), 103, 110 predicative position, 99–104, 106, 107 prepositional phrase, 164–176 Principles & Parameters Theory, 59, 82–84 processing difficulties, 249 pronominal adjective, 389, 392 pronoun, see agreement (pronoun/be), expletive subject, Northern Subject Rule, resumptive pronoun

409

Qu´ebec French, 255–261, 264–267 question formation, 255–267 raising verb, 298, 306–310, 313, 317 real-time: see language change (in real time) reanalysis, 110 regional standard variety (Dutch), 205 relative clauses, 135–154 resumption, see resumptive pronoun resumptive pronoun, 137–145 rhematic, see focus rhyme, see syllable rhyme right periphery, 21–24, 27, 39, 47 Ruhrdeutsch, 163, 164, 166–176 secondary topic, 346–349 semi-modal verb, 297–323 social variation, 205 sociolinguistics, 7, 59, 170, 171, 176 sonority, 130 spelling, 363, 373, 375, 376 standard articulation, 119, 121, 122 standardization, 204, 221 stress, 229, 230, 237 strong inflection of adjectives: see adjective inflection (weak vs. strong) stylistic variation, 119, 126, 127, 260, 264 subject, see agreement (subject/verb), expletive subject, Northern Subject Rule Swiss German, 135–154 syllable (full), 129,131 syllable (reduced), 120–122, 127, 129–131 syllable rhyme, 129–131 teenage slang, 105–107 thematic, see topic three verb cluster, 37, 40, 213–215, 217–220 (Dutch); 38, 40 (German) topic, 345. See also secondary topic transparency, 191, 193

410

Index

two verb cluster, 40, 205–212, 216, 225–250 (Dutch); 40 (German) underspecification, see categorial underspecification usage-based model, 186, 273, 289 variable rule analysis, 6, 259–266 variation, see conditioned variation, dialectal variation, free variation, grammatical variation, idiolectal variability, inter-speaker variation, intra-speaker variation, lexical choice, microvariation, morphologically conditioned variation, phonologically conditioned variation, stylistic variation

verb cluster, see four verb cluster, three verb cluster, two verb cluster verb projection raising, 29 verb second, 27–29 verbal particle, 183–197 (English); 108, 109, 111 (German). See also verb-particle construction verb-particle construction, 188, 189 VO language, 21, 22, 28–32, 47 weak inflection of adjectives: see adjective inflection (weak vs. strong) word order, see four verb cluster, OV language, right periphery, three verb cluster, two verb cluster, verb projection raising, verb second, VO language