Constructions and Language Change 9783110211757, 9783110198669

Citation preview

Constructions and Language Change

≥

Trends in Linguistics Studies and Monographs 194

Editors

Walter Bisang (main editor for this volume)

Hans Henrich Hock Werner Winter

Mouton de Gruyter Berlin · New York

Constructions and Language Change

edited by

Alexander Bergs Gabriele Diewald

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 Constructions and language change / edited by Alexander Bergs and Gabriele Diewald. p. cm. ⫺ (Trends in linguistics. Studies and monographs ; 194) Includes bibliographical references and index. ISBN 978-3-11-019866-9 (hardcover : alk. paper) 1. Linguistic change. 2. Grammar, Comparative and general. I. Bergs, Alexander. II. Diewald, Gabriele. P142.C67 2008 4171.7⫺dc22 2008032045

ISBN 978-3-11-019866-9 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 2008 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, Berlin. Printed in Germany.

Table of contents Introduction: Constructions and Language Change Alexander Bergs and Gabriele Diewald

1

The grammaticalization of NP of NP patterns Elizabeth Closs Traugot

23

Constructions and constructs: mapping a shift between predication and attribution Mirjam Fried

47

Constructional idioms as products of linguistic change: the aan het + infinitive construction in Dutch Geert Booij

81

Where did this future construction come from? A case study of Swedish komma att V Martin Hilpert

107

Bedusted, yet not beheaded: The role of be-’s constructional properties in its conservation Peter Petré and Hubert Cuyckes

133

Negative verbal clause constructions in Puyuma: exploring constructional disharmony Malcom Ross

171

Borrowed rhetorical constructions as starting points for grammaticalization Marianne Mithun

195

(De)grammaticalisation as a source for new constructions: the case of subject doubling in Dutch Gunther De Vogelaer

231

Syntax as a repository of historical relics Wallace Chafe

261

Subject index

269

Introduction: Constructions and Language Change Alexander Bergs, Gabriele Diewald 1. “Constructions” In their seminal 1968 article Weinreich, Labov, and Herzog show with regard to the linguistic theories developed by Paul, Saussure, Bloomfield, and Chomsky that “each refinement in the theory of language structure [. . . ] had the following potential effects: (a) a reclassification of observed changes according to new principles; (b) proposal of fresh constraints on change; and (c) proposal of new causes of change.” (1968: 126; emphasis original). This volume focuses on one particular refinement in the theory of language structure: the (re-)introduction of the theoretical notion of “constructions”, and how constructional approaches to language deal with problems in linguistic variation and change. The notion of “constructions” is of course not new in linguistics. In the sense of “formulaic, fixed sequences” the idea can be traced back at least to the mid-nineteenth century and can be found in writings of Saussure, Jespersen, Bloomfield, Firth, and many more (see Wray 2002: 7–8 for an overview). The term “construction” in linguistic analyses can be traced back even to Antiquity and the Modistae of the 13th century (for an overview, see Sch¨onefeld 2006 and Goldberg and Casenhiser 2006). In recent times interest in constructions has culminated in the development of Construction Grammar (also abbreviated as “CxG” in the following), a family of linguistic approaches which focuses ¨ exclusively on the structure and function of constructions in grammar (Ostman and Fried 2004, Croft and Cruse 2004, Fischer and Stefanowitsch 2006 provide informative overviews over the different strands and their development). Despite or perhaps even because of this long-standing history, there is (still) no agreement as to what constructions actually are and what their place in language and linguistic theory actually is (see e.g. Wray 2002: 9 for an informative list of diverging terminological decisions). Nevertheless, there is a common core of theoretical positions shared by virtually all scholars working on Construction Grammar, which is specific to this approach and distinguishes it from generative grammar as well as from more traditional grammatical schools. The following issues are among the most important ones: (1) Constructions are defined as pairings of form and meaning, ranging from the morphemic to the utterance level of linguistic structure; (2) constructions are organized in complex hierarchical networks with inheritance, polysemy and synonymy relations;1 (3) the scope of the

2

Alexander Bergs, Gabriele Diewald

notion of construction ranges from ‘lexicalized’ or ‘idiomatic’ items to abstract, productive patterns; (4) constructions are highly sensitive to frequency as well as to their respective co- and con-texts.2 These positions – among other, more specific ones – mark Construction Grammar as an innovative and integrative linguistic approach, which, by combining traditional concepts with results and positions of modern linguistic theories of cognitive, functional as well as formal origin, increases the range as well as the depths of linguistic investigation. Most important for this volume is the fact that the central tenets of Construction Grammar lead on to theoretical and methodological consequences that make CxG a particularly suitable tool for investigating and describing language change (see below). While the present volume aims at improving our understanding of “constructions”, it is also in line with the present state-of-affairs as it is not committed to any single definition of the notion of construction. Nor is it explicitly couched in any specific Construction Grammar framework. Most of the contributions in this volume can be conveniently arranged into three different groups which naturally derive from the kaleidoscope of current research in constructions. The first group subsumes contributions which clearly can be classified as state-ofthe-art construction grammar analyses (e.g. Fried, Hilpert, Booij). The second group includes contributions which use the term “construction” in a rather nontechnical, but still very illuminating sense, and whose theoretical framework also allows for more traditional grammatical notions (e.g. Mithun, Ross). The contributions in group three take a more distanced point of view and focus on more theoretical issues in construction grammar and constructional approaches in contrast to other frameworks and traditions (e.g. Traugott, Chafe). It will become clear that these different points of view of course also lead to different research questions. Studies in the framework of construction grammar need to address theoretical questions pertaining to that very framework while studies using constructions as an additional notion in a more traditional framework rather need to think about what kind of advantages and disadvantages that notion can have for them. Moreover, the multitude of different approaches also has some consequences for the different presentations in this volume. Readers will immediately notice that there is no single uniform notation for constructions. This of course stems from the different approaches themselves, but it is also one of the credos of con¨ structional approaches and construction grammar in particular (cf. Ostman and Fried 2004; Boas and Fried 2005). While often being highly formal, these do not require a strict, uniform notation, on the contrary. Constructional approaches begin with the data and the phenomena and only then develop the necessary formalism on the basis of what they find and deem necessary. It should be emphasized here that this seeming inattentiveness towards an a priori given formal

Introduction: Constructions and Language Change

3

apparatus is in fact a pragmatic necessity for diachronic studies transcending merely theoretical reflections on change and in addition working with empirical data. As diachronic data-based studies are notoriously confronted with the trivial but far-reaching problems of lack of native speaker competence for past language stages and insufficient language data (cf. Labov 1994: 11), descriptive models requiring a complete analysis of the whole linguistic structure in terms of a preset formalism inevitably run the danger of anachronistic distortions. In contrast to that, the descriptive practice favored by constructional approaches, which allows structural descriptions with varying granularity, provides for analytical solutions that avoid over-specified, non-provable descriptions and analyses (cf. Diewald 2006). ¨ Thus, the present volume follows the policy outlined in Fried and Ostman (2004) and Boas and Fried (2005) in that every paper can use its own notation as long as the procedure and formalism is explained to the reader. This introduction is organized as follows. First, we will discuss some more general issues pertaining to linguistic change from a constructional point of view. These are, for us, the units of change (since constructional approaches often do not work with traditional linguistic units such as phoneme, morpheme, and phrase, including the traditional division between variable and variant) and context sensitivity, i.e. contextual factors in linguistic change. In a second step, we will concentrate on some more special issues, especially as they emerge from the contributions to this volume. In particular, the question of how new structures can emerge in linguistic change will be our focus here. The topics to be treated include analogy, reanalysis, frequency, explorative expressions, and language contact. The third section of this introduction then gives detailed summaries of the individual papers in their specific theoretical context.

2. Constructions in language variation and change 2.1.

Units of change and context sensitivity

The notion of “constructions” has already figured prominently in many approaches to linguistic change (Traugott, this volume), but very few of these systematically investigate the potential of this notion. When we look at it closely, two basic ideas seem to underlie most constructional approaches to linguistic change: (a) linguistic change often does not affect only single linguistic items, like words, morphemes, or phonemes, but also syntagmatic structures up to the sentential and utterance levels (i.e. the relevant co-text comprises all levels of

4

Alexander Bergs, Gabriele Diewald

explicitly expressed linguistic material) and (b) linguistic change can be very con-text-sensitive, i.e. motivated, triggered or influenced by pragmatic extralinguistic factors. For example, as Diewald and Habermann (2005) show, the development of the German construction werden & infinitive as a future marking device results from a highly complex interplay of co-textual factors, like the acquisition of different types of complements (including the bare infinitive) of werden (which enlarged the syntagmatic combinability), and con-textual factors, like the influence of rhetoric traditions, text-type choices and theological positions in the 16th centuries, which enriched the new syntagmatic formation with a distinct meaning (in specific text types), thus leading to the rise of a specific future-marking construction not found in related languages. Let us look a bit more closely into the question of the units of change. Recent studies in grammaticalization have shown that strict distinctions between the individual elements in linguistic structure and their independence can often not be upheld in linguistic change: “grammaticalization does not merely seize a word or morpheme [. . . ] but the whole construction formed by the syntagmatic relations of the element in question” (Lehmann 1982: 406; cf. Traugott 2003: 625; Himmelmann 2004: 31; Wiemer 2004: 271ff). The development of present-day Dutch progressive construction aan het + infinitive, for example, involves the preposition aan and the determiner het, which function as specific markers of the progressive when combined with infinitival forms of verbs. Any explanation which solely focuses on the linguistic elements in isolation misses some important factors in this development (Booij, this volume). Wiemer (2004) analyzes the evolution of passives in Baltic and Slavic languages and comes to the conclusion that this can also be regarded as one form of grammaticalization, despite the fact that not one single element, but a full construction undergoes change. He thus suggests that new criteria for grammaticalization are badly needed which take account of constructional factors. Himmelmann (2004: 33) argues in a similar vein and claims that the grammaticalization of single elements (the traditional point of view), despite the fact that it still forms some part of the definition of grammaticalization, is actually epiphenomenal to the construction-based approach. It should have become clear that constructional approaches to linguistic change, and construction grammar in particular, not only raise some interesting new questions here, but that they are also well suited for dealing with these problems and for treating multiple elements as single units, either in the form of constructions or what could be termed constructionalization, i.e. the formation of new units (constructions) out of hitherto independent material. However, what has rarely, if ever, been discussed in this context is the issue of language variation as the basis of linguistic change, and thus also of gram-

Introduction: Constructions and Language Change

5

maticalization. To begin with a rather uncontroversial idea: language change is inextricably entwined with language variation. As early as 1968, Weinreich, Labov, and Herzog claimed: “Not all variability and heterogeneity in language structure involves change; but all change involves variability and heterogeneity” (1968: 187). This sounds much more innocent and straightforward than it actually is. Linguistic change usually does not simply arise from some kind of wild and random variability. It essentially requires structured heterogeneity. Change can then be described as the generalization of one particular alternation, i.e. it is characterized by some sort of directionality. For example (cutting a million corners), word order in present-day English (PdE) probably evolved out of general syntactic variability in Old and early Middle English combined with information structuring principles. In the transition period between Old and Middle English the topic domain might have been reanalyzed as the (fixed) subject position. Information structuring thus gave directionality to the basic variability in syntax, which in turn was also dependent on and influenced by the changes in the morphosyntactic marking of constituents3 . So the question arises how linguistic variability, let alone directionality, can be modeled in constructional approaches. Martin Hilpert (this volume), fruitfully combines corpus linguistics with grammaticalization theory and CxG approaches. This work as well as many others (cf. Gries and Stefanowitsch 2004a) impressively document the quantitative dimension of variability in language and language change. What still needs to be discussed is the qualitative nature of this variability. Fortunately, previous research has led to several converging insights, which can be used as the starting point for tackling this problem. It is common practice to distinguish between constructions on the one hand and constructs on the other, or as suggested by Cappelle (2006), between constructions and so-called allostructions. Constructions are seen as the more abstract blueprints which license well-formed linguistic expressions, while constructs or allostructions are the more concrete realizations of constructions, i.e. actually occurring expressions or types of expressions (see for example Goldberg 1995 as well as most CxG scholars). These notions, of course, are set up in analogy to central concepts of structuralism, like the notions of phoneme, ¨ allophone and phone, or morpheme, allomorph and morph (Fried and Ostman 2004: 18f, Cappelle 2006). However, in contrast to structuralism, CxG is explicitly performance-based, and there is a continuum between schematic and concrete constructions, i.e. a continuum of embedded, multi-layered type-token relations (see, e.g., Croft 2001; Tomasello 2005). In particular, this performance-oriented nature of CxG in combination with the central tenets listed in Section 1 brings into close contact the two issues which have to be kept theoretically distinct. The first issue

6

Alexander Bergs, Gabriele Diewald

is the relation between virtual linguistic signs (types) and the instances of their realization (tokens), the second issue is the systematic stratification of constructions, i.e. the relation between more abstract constructions and their more concrete “descendants”, i.e. constructions which may show formal as well as functional and semantic variations between each other and thus create polysemy and synonymy relations (for inheritance relations see for example Goldberg 1995, Booij, this volume, Ross, this volume). An attempt to account for the stratification in constructions is the distinction between “macro-constructions”, “meso-constructions”, and “micro-constructions” suggested by Traugott (this volume). “Macro-constructions” are defined as higher-level, more abstract functional constructions, “meso-constructions”, encountered on the next level, are seen as groupings of similar-behaving constructions, and finally, we find single, basic constructions (“micro-constructions”). Traugott (this volume) points out that all three levels are to be interpreted as abstractions, as types. These need to be distinguished from actual utterances, which she calls constructs, in line with ¨ Fried and Ostman’s analysis. Some examples may be helpful in illustrating these issues. While the SubjectPredicate Construction, for example, clearly is a macro-construction in the way defined above, which is actualized in a concrete realization (construct) like “Pete ate the bagel”, an idiomatic entity like when push comes to shove, on the other hand, would be the realization (the construct) of a micro-construction, i.e. of a construction whose formal as well as functional and semantic features are fixed. Still unsolved is the question of the place of allostructions (which at least in one reading can be equated with meso-constructions) in the linguistic system, in particular, the distinction between “constructions” on the one hand, and their individual “allostructions” on the other. What makes “bring in the criminal” and “bring the criminal in” two allostructions of the verb-particle construction? Gries (2003) claims that there are more differences than similarities between the two and that they do not form one single category. Capelle (2006) claims there are enough similarities to warrant one basic construction with two variants. So how do we define the cut-off point in dissimilarity? And, what do we need: formal differences, functional differences, or both? While most researchers seem to agree that similar function in a similar context results in different constructions in competition with each other, this is not entirely clear. A further problem having to do with the interrelations among constructions is the issue of compositionality. The original definition of constructions as conventionalized pairings of form and meaning at all levels of linguistic structure necessarily treats every new form as a new construction. Originally, one caveat was added to this definition: constructional meaning had to be non-compositional, i.e. not predictable from the individual components (see, e.g., Goldberg 1995: 4). More

Introduction: Constructions and Language Change

7

recent treatments (e.g., Goldberg 2006, Goldberg and Jackendoff 2005) also include compositional constructions “as long as they are sufficiently frequent” (Goldberg and Jackendoff 2005: 533). In sum, it seems highly desirable to give the structure and function of variability in construction grammar more thought. We have made great progress in the statistical description of variants and their developments, but we still know very little about their theoretical and cognitive status. The whole dilemma, which will and can not be solved in an introduction like this, can be summarized in one simple question: If something is not a construction, what is it? The second issue that was mentioned at the beginning of this section is context sensitivity. Linguistic change in general is often very context-sensitive, i.e. it involves factors which can be found outside the linguistic system as such. Not without good reason does Himmelmann (2004: 31–34) describe grammaticalization as “context expansion”. One of his examples is the development of demonstratives and articles. The evolution of articles out of demonstratives typically involves three kinds of change, i.e. context expansion on three different levels: construction internally, in a larger syntactic context, and in semantic and pragmatic contexts. Articles can typically co-occur with proper names or nouns designating unique entities, demonstratives usually can’t. This can be seen as construction internal context expansion. Similarly, Himmelmann argues (2004: 32), emerging articles often first occur in core argument positions and rarely, if ever, in adpositional expressions. When they grammaticalize further, they may also become possible, even obligatory in these expressions and other syntactic contexts. This exemplifies the second type of context expansions, i.e. in larger syntactic contexts. Finally, adnominal demonstratives only occur in expressions involving deictic, anaphoric or ‘recognitional’ reference. Articles, in contrast, are used in much wider and more general contexts in which demonstratives do not occur (e.g. associative anaphoric uses). This is an illustration of semanticpragmatic context expansion (see Himmelmann 2004 for a full account of this model). Diewald (2002, 2006) is another important study which deals with the role of context in linguistic change and grammaticalization in particular. On the basis of the development of modal verbs in German, she also suggests three different stages in grammaticalization, which are specifically tied to particular types of context. The first stage involves the preconditions for grammaticalization: the grammaticalized element is embedded and used unspecifically in a number of new (syntagmatic, semantic, pragmatic) contexts. These contexts are called ‘untypical contexts’. The new meaning, which is to be further grammaticalized, arises in these contexts via conversational implicature. Interestingly, untypical contexts show either morphological or semantic incompatibilities, but never both. The second type of context, the so-called critical context, is the

8

Alexander Bergs, Gabriele Diewald

actual starting point for the grammaticalization process proper. The essential point is that in this second context type, morphological and semantic incompatibilities occur. The third type of context is the so-called isolating context, in which new incompatible meanings and functions independent of the original interpretation are conventionalized (see Diewald 2002, 2006 for a more detailed account). Constructional approaches, and construction grammar in particular, generally include contextual (i.e. co-textual and con-textual factors as introduced in Section 1) in their analyses (cf. the papers in Bergs and Diewald, in prep.; Diewald 2006). Co-textual factors are often already captured through syntagmatic configurations, but in fact, CxG also explicitly calls for an inclusion of contextual factors: “by construction I intend a conventional association of any or all of the following kinds of grammatical information: syntactic, semantic – including ‘pragmatic’, lexical and phonological” (Kay 2002: 1). Goldberg is even more explicit on this point: “Another notion rejected by Construction Grammar is that of a strict division between semantics and pragmatics. Information about focused constituents, topicality, and register is presented in constructions alongside semantic information” (Goldberg 1995: 7). Again, this means that constructional approaches and CxG in particular are perfectly compatible with recent findings in studies of language change, and grammaticalization in particular. If the road is kept open for two-way traffic, this relationship will certainly be rather sym- than antibiotic.

2.2.

Some more specific problems for a theory of change in constructional terms

Apart from these two central issues, units of change and context sensitivity, there are some other interesting problems regarding language variation and change from a constructional point of view. When we talk about variability and linguistic change, we also have to consider the innovation of new forms and structures. In constructional terms this means that we need to ask how new constructions can be added to the supposedly structured inventory of constructions (or maybe they are only activated?). Traditionally, we see performance, contact, reanalysis, and expressivity as some of the central factors in the genesis of new linguistic forms and structures. Some changes on the level of sound and morphology somehow go back to what Croft called ‘imperfect replication’. That is, word final sounds are likely candidates for weakening and deletion (e.g. [and] > [an], [give] > [giv]), some rare irregular verb forms like heft, bereft are likely to be regularized through analogical level-

Introduction: Constructions and Language Change

9

ing into heaved and bereaved, some of the developments in the English analytic comparative are also of this kind (see Mondorf 2006). Very little has been said about these phenomena from a constructional perspective, apart maybe from usage and frequency factors which have moved closer to the center of attention ¨ in many of these studies (cf. Fried and Ostman 2004; Gries and Stefanowitsch 2004b). As far as analogy as one factor is concerned, the present volume contains an important investigation into its role (Traugott, this volume) and sheds light on the question of how constructions actually work. Constructions operate in this respect much like ordinary lexical items since they, like ordinary lexical items, are defined as conventionalized pairings of meaning and form. CxG, being performance based, operates with frequency as a factor. This means that frequent exposure to certain constructions – or perhaps rather constructs – can tip the scales in a certain direction. Which means that analogy also becomes explicable in terms of connectionist approaches psychology and recent theories of learning as advocated by Tomasello (2005, 2006), for instance. Moreover, they are easily extendable to syntactic structures, as they, too, count as constructions. The role and relationship of analogy and reanalysis in linguistic change have been subject to many discussions (see Traugott, this volume). Reanalysis, which for construction grammar means the very concrete dissolution and creation of new constructions in the inventory, allows for some interesting perspectives, e.g. on the grammaticalization of the English “going to” future. This seems to be an area where construction grammar can have a profound effect in terms of reclassification, constraints, and causes. We start with going to, a complex, but regular construction for physical movement, consisting of the verb of movement complemented by a structurally independent adverbial (usually a prepositional phrase) expressing the direction or goal. In order to derive a future marker like gonna out of these sources, a new construction which includes both the verb going and the prepositional marker to is required. This new construction has a number of new formal and semantico-pragmatic properties, like new cooccurrence restrictions (it need not be followed by a noun phrase expressing the direction, but can be followed by a purposive phrase and finally also verbs which are incompatible with the source meaning “physical movement”, like love or go). We might want to call this process uniconstruction, or constructionalization. In this context, it seems important to note that the development apparently does not begin with constructions as such, but rather with constructs (i.e. concrete realization which are reanalyzed as different constructions), as Traugott (this volume) points out. Finally, after having gone through some intermediate stages with ambiguous structures, the two elements are univerbated through phonetic erosion into gonna, which is incompatible with physical movement (*I’m gonna Texas). Traditional grammatical models can describe reanalyses of this kind –

10

Alexander Bergs, Gabriele Diewald

no doubt. But they seem to have a hard time incorporating semantico-pragmatic (contextual) factors, which are part and parcel of this process, as has been pointed out above. CxG, by definition, includes both semantics and pragmatics in constructions. Also, since CxG is explicitly cognitive and combines insights from usage and frequency, it allows for some interesting observations about cognitive aspects of reanalysis and might even help to identify causes of change which have not been fully explored yet (syntactic priming might be such a promising approach). Gradualness in linguistic change is another issue that more traditional approaches have to struggle with. Constructional approaches, in contrast, can indeed model gradualness as a step-by-step change in the individual features of a particular construction, or rather through its concrete constructs. In CxG, there is no theory internal reason to assume that certain changes must leap from one stage to another. Finally, some forms are created as explorative expressions, as Harris and Campbell (1995) call them, i.e. innovative (non-conventional) ways of expressing the same propositional content (cf. also Haspelmath 1999). To some extent, reanalysis mostly focuses on language change initiated by the hearer, while explorative expressions rather focus on the speaker. Examples from morphology are plentiful if you watch just one episode of Seinfeld: here we learn new words like sidler, re-gifter, sponge-worthy, happy festivus, susher, shushee, and unshushables. Traditional examples can be found in the development of negation in French and English, where emphatic markers are added to single negative markers for expressivity (ne V > ne V noght > Ø V noght > V not > do not V). How about this kind of expressivity in CxG? Although constructions in the traditional sense have slot-filler patterns, this does not correspond to what we know about expressivity as a factor. Constructions are conventionalized formmeaning pairings. Explorative expressions are, by definition, non-conventional. So what we need is combinability of constructions in innovative ways, and the use of constructions (as new, non-conventional constructs) in new co- and contexts. One promising approach here comes from recent work (Michaelis 2003, 2004) on coercion. Coercion, in a nutshell, is “syntactically and morphologically invisible: it is governed by implicit contextual reinterpretation mechanisms triggered by the need to resolve [semantic] conflicts” (de Swart 1998: 360; cf. Michaelis 2003, but also Traugott 2007; Ziegeler 2007). In other words, coercion is what we see when we order two beers, since beer should be an uncountable mass noun, but the combination with the numeral coerces a new interpretation, namely two units (e.g. pints) of beer. This appears to be a very elegant way of capturing the introduction of new structures as a result of expressivity. Some new forms and structures arise in language contact situations (cf., for example, Heine and Kuteva 2006). Much of the English lexicon is not English,

Introduction: Constructions and Language Change

11

the th-pronouns are Proto-Norse, derivational morphemes like –able are French, and the PdE relativizer system perhaps was modeled on the basis of Latin. How can we deal with this in CxG? It is perhaps fair to say that research in this domain from a constructional point of view practically did not exist, but that we are now witnessing rapid and promising progress (exemplified, for example, by Heine and Kuteva (2006: 44) where replicating use patterns in language contact are compared to constructions). Mithun (this volume) is one of the most detailed and systematic studies to investigate the borrowing of syntactic elements from a constructional perspective. Suffice it to say that the notion of construction (because of its proximity with ‘lexical element’) is very compatible and helpful for investigations in linguistic typology and the effects of language contact, and that in the future some more excellent research results in this domain can be expected.

3. This volume The volume assembles papers working with construction grammar informed concepts of constructions as well as studies which may be subsumed under the label “constructional approaches”. The languages investigated in the case studies include several Germanic languages (English, Dutch, and Swedish), Czech, North-West Coast Indian languages, African languages, Puyuma, and Japanese. The linguistic phenomena treated range from the development of word formation (participles, prefixed verbs) to the rise of periphrastic verbal categories (progressive in Dutch, future in Swedish), from case marking changes in connection with word order to multi-phrasal rhetorical processes, from the development of quantifiers to that of modals. According to their theoretical and thematic orientation, the papers are grouped into three sections. The first section, “Construction grammar and language change”, contains papers making extensive use of a particular models of Construction grammar, representing the data in construction grammar formats, and discussing the suitability of construction grammar approaches for modeling change. Elizabeth Closs Traugott’s paper “The grammaticalization of NP of NP patterns” unites a theoretical evaluation of the merits of construction grammar for grammaticalization studies with a detailed case study of the development of degree modifiers from partitive constructions in English. The first part takes up reflections on constructions and grammaticalization Traugott had proposed in an earlier paper and develops them further in the direction of the radical construction grammar approach suggested by Croft. It is argued that a con-

12

Alexander Bergs, Gabriele Diewald

struction grammar approach is better equipped to tackle the notorious problems of correlation between form, meaning and function that are most important in grammaticalization processes, and also improves the ways of modeling different degrees of co-evolution (or its absence) between form and meaning. Traugott assumes several layers of constructional organization as well as different degrees of entrenchment of a construction, which is judged to be helpful for solving the problem of identifying the “locus of change”. It is suggested that innovation starts in constructs, and, if conventionalized, becomes an inherent part of the higher constructional levels, “starting from micro-constructions to mesoconstruction-types and ultimately reaching the highest level of constructions” The case study presented in the paper explores the rise of quantifying noun modification in English, i.e. the development of items like “bits of NP”, as an instance of grammaticalization. While the overall picture of the usefulness of a construction grammar approach to grammaticalization is judged to be positive, Traugott also points to some problematic aspects. For one thing, she warns that the holistic approach of construction grammar might prove a hindrance for the description of semantic change in grammaticalization, as the holistic model of meaning might lead to a neglect of the investigation of the complexities of semantic and pragmatic interrelations that are at work. Another word of caution is expressed with reference to the notion of “coercion”, i.e. the notion of a construction exerting force on the meaning of an item inserted in that construction: This concept might stand in the way of recognizing important and typical phenomena connected with semantic change like the retention of older meanings and contingent restrictions on the use of the construction containing that item. The paper by Mirjam Fried with the title “Constructions and constructs: mapping the shifts between predication and attribution” is inspired by the same general objective as the first paper. It explores the usefulness of construction grammar for the investigation of grammaticalization with special attendance to the gradience of grammatical change and to the need of defining the exact locus of change. The theoretical positions are tested in a case study on morphological change in the development of transpositional morphology in the case of the Old Czech participles, more precisely, the development of the so-called long form of the present participle. The paper starts with a succinct explication of the positions of construction grammar and the particular model Fried adheres to, which ¨ is based on Fillmore 1988, taken up by Fried and Ostmann 2004 and Croft 2001. In accordance with the preceding paper, it is emphasized that the distinction between constructions as abstract elements of “grammar” and constructs as actual realizations of constructions is an important one, and that change starts out in particular constructs. The tension between holistic meaning and the possibility for small steps of change, which Traugott takes to be a problem for uniting

Introduction: Constructions and Language Change

13

construction grammar and grammaticalization theory, is resolved here via the definition of construction. Fried states that “non-compositionality in the narrow semantic sense is not a necessary condition for constructional status” Further important aspects are the tension between the holistic meaning of constructions on one side, and particular features that are affected by change on the other, as well as Croft’s assertion that the syntactic function of lexical categories is dependent of particular constructional patterns (and thus not describable in universal terms). The merits of construction grammar for representing diachronic processes are seen in the following points: First, construction grammar provides the means for focusing on contextual features that motivate and trigger change, and to identify individual properties that together lead to a change in a holistic construction. Second, Fried points out that constructions are “blueprints” (generalizations over constructs) and as such presuppose variation and change as an inherent part of grammar. Third, multi-layered representation with equal weight for each layer accommodates non-compositionality, and, finally, as construction grammar does not require a full specification, this way of representation is most suited for diachronic stages. Section 2, “Verbal constructions and grammaticalization” contains papers on verbal word formation and on the development of periphrastic analytic verbal categories in Germanic languages, and thus tackles a classical topic of grammaticalization studies in constructional terms. The paper by Geert Booij “Constructional idioms and grammaticalization: the aan het + INFINITIVE construction in Dutch” takes up the development of verbal periphrasis for the encoding of progressive meaning, which is a well-known area of grammaticalization. The focus of this paper is not on the diachronic development of the progressive from aan het + INFINITIVE but on the synchronic distribution of this form and its integration into the verbal paradigm. It takes its starting point from the wellknown observation that grammaticalization processes do not affect linguistic items in isolation but are bound to specific contexts or constructions. Applying the notion of construction in an informal way, which focuses on the association of a structural pattern with partially non-compositional meaning, Booij argues that aan het + INFINITIVE may be defined as a “constructional idiom” in the sense of Jackendoff (1977; 2002) and Goldberg (1995). He further suggests that the notion of construction is suitable in accounting for the reflexes of grammaticalization on a synchronic grammatical system, in particular on the productivity of certain multi-word combinations, i.e. constructional idioms. After discussing two cross-linguistically frequent sources for progressives, namely postural verbs (like items meaning ‘sit’), and locative constructions like the Dutch aan hetconstruction, the latter construction is shown to have progressed far on the grammaticalization cline (semantics, generalization to other infinitives, typical

14

Alexander Bergs, Gabriele Diewald

restriction for progressives concerning the infinitive etc.) and that it is integrated as a periphrastic verbal construction into the verbal paradigms of Dutch as it interacts with inflectional means to express progressive aspect. In particular it functions as an alternative realization to the present participle and may have blocked its use in predicative position. Furthermore, it is argued that the regular aspects of the constructional idiom, which coexist together with the idiomatic ones, can be accounted for by the notion of “inheritance tree” in the sense of Goldberg 1995 and Jackendoff 2002, which allows a depiction of the transfer of valency features (predicate argument structure) from the verb to the whole construction within the inheritance tree (although this is not inheritance as defined in the classic way). The second paper in this section, Martin Hilpert’s investigation on the rise of the Swedish COME-future “Where did this future construction come from? The case of Swedish komma att V”, too, takes up one of the favorite topics of grammaticalization studies, and explores the constructional status of the different stages of the development, which are documented by a diachronic corpus (starting from Old Swedish texts from between 1300 to 1450 up to Modern Swedish). The aim is to test which of two alternative suggestions for grammaticalization paths for future grams holds for the Swedish data: either the suggestion by Bybee, Perkins and Pagliuca (1994), who describe the development of devenitive future via a stage encoding “intention”, or the alternative hypothesis by Dahl (2000), who suggests that Germanic de-venitive futures develop from inchoative notions and do not require the stage of intentional meaning. For a definition of the Swedish periphrastic forms as constructions, Hilpert follows the classical definitions in Goldberg 1995, Fillmore, Kay and O’Connor 1988, ¨ Fried and Ostman 2004. Accordingly, he postulates that “constructions, as conventionalized sequences of morphemes have direct semantic representation”, thus focusing on the aspect of “non-compositionality” (contra Fried). Furthermore, again with reference to the authors quoted, he understands constructions as polysemous “having a number of conceptually interrelated functions”. The corpus analysis reveals the existence of three construction types using COME for each period and their development through time. Hilpert points out that the data clearly are in favor of Dahl’s suggestions and that the grammaticalization path MOTION > INTENTION > FUTURE has to be replaced by MOTION > INCHOATIVE > FUTURE in the case of the Swedish de-venitive future. The word formation of English prefix verbs is the topic of the paper by Peter Petr´e and Hubert Cuyckens “Bedusted, yet not beheaded: The role of be-’s constructional semantics in its conservation”. It treats the retention and grammaticalization of the construction with the inseparable prefix be- in English, and motivates it by the “salience” of the be-construction in contrast to other pre-

Introduction: Constructions and Language Change

15

fixed verbs, which have become extinct. Petr´e and Cuyckens make explicit use of Goldbergian constructional schemata, whereby the resultative constructions with unselected objects like He sneezed the napkin off the table are directly relevant for their data. In contrast to other prefixes, a subtype of the be-construction had special semantic and valency features – Petr´e and Cuyckens call that “constructional salience” – which helped this construction to survive, while other prefix constructions got lost because of the change from OV to VO (astigen > come down: object or other specification of verb must follow verb), and even gained ground and specialized in the formation of deverbal nouns and adjectives. The third section “Word order and argument structure (including clitics and case marking)” units papers which, though dealing with languages distributed across the globe (American North-West Coast, Taiwan, Europe), converge in their aim to tackle word order changes and their concomitant morphosyntactic changes in a constructional model. They demonstrate that constructional approaches are well-suited for this purpose. In his paper “Negative verbal clause construction in Puyuma: exploring constructional disharmony”, Malcolm Ross deals with the rise and accommodation of constructional disharmony, i.e. with the relation among constructions in a paradigmatic field. The definition of constructional (dis)harmony rests on the fact that, while harmonic constructions are systematic and expected, disharmonic constructions are not, as they display an unexpected combination of parent constructions. The case discussed in the paper is the negative transitive construction in Puyuma, an Austronesian language of Taiwan, which is described as a disharmonic construction as it does not pattern regularly with its neighboring constructions, i.e. all combinations of the values transitive/intransitive and positive/negative. Although the paper suggests a possible scenario for the complex diachronic development of the present-day situation, which rests on frequency arguments (following Bybee 2005, and Bybee and Hopper 2001), its main interest is in the relations among the paradigmatically associated constructions. The paper raises the question how disharmonic relations among coexisting constructions within a grammatical paradigm can be adequately represented by an inheritance network. The argumentation and solution proposed is based on the constructional works of Fillmore, Kay and O’Connor 1988, Kay and Fillmore 1999, and in particular Croft 2001, whose suggestions for the representation of constructional inheritance are taken up and applied to the Puyuma case. Modifying Croft’s model of representation of constructional relations, Ross suggests a network of inheritance relations instead of the hierarchically ordered taxonomy proposed by Croft 2001. He argues that a network can accommodate disharmonic constructions better, because in a network, as opposed to a taxonomy, constructions do not have to be derived from the same parents as their neighboring harmonic constructions.

16

Alexander Bergs, Gabriele Diewald

The paper by Marianne Mithun “Borrowed rhetorical constructions as starting points for grammaticalization” combines diachronic linguistics with areal linguistics and language contact. It is concerned with contact-induced changes of constructional patters that give rise to a “misplaced marking of argument structure” in two neighboring, though genetically unrelated, language families of the Northwest Coast of North America, Wakashan and Tsimshianic. Mithun works with the definitions of constructions a` la Goldberg (1995), and shows that the rhetorical pattern she is dealing with are constructions in the technical sense, showing highly complex and specific features on several levels of linguistic organization. They involve multi-phrasal demonstrative structures with a cataphoric demonstrative referring to a NP in a separate prosodic unit, with which it agrees, but cliticising to the element preceding it, thus laying the ground for the genesis of misplaced argument marking. The hypothesis put forward is that those highly idiosyncratic constructions are due to the fact that the relevant entities that were transferred from one language to the other are not concrete linguistic items, but constructions seen as abstract templates, as patterns of usages. The origin of the constructions responsible for this case marking pattern lies in a common discourse pattern favored in ceremonial speech, which finally evolved into a conventionalized rhetorical strategy. This strategy (and its constructional make-up) but not the concrete linguistic material was borrowed between the languages in contact, and in the further history developed according to language specific features. Gunther De Vogelaer takes up the issue of “(De)grammaticalisation as the source for new constructions: the case of subject doubling in Dutch”. He provides an empirical study combining dialect geography and grammaticalization issues to treat the question of the rise and spread of constructions with double subject marking in Dutch dialects. Working with a loose and pre-theoretical concept of construction, he argues that subject doubling originates in inverted word order constructions with second person subjects, a construction which is restricted to spoken language. The hypothesis is that grammaticalization as well as degrammaticalization may lead to the rise of new constructions. It is exemplified by the case of pronominal subject doubling in some Dutch dialects; this construction is treated by De Vogelaer as the result of a degrammaticalization process. The volume concludes with a paper by Wallace Chafe dealing with “Syntax as a Repository of Historical Relics”. In this programmatic paper, Chafe addresses the very fundamental question of what syntax (and language) actually is, how constructions as syntactic units may come into being, and what their function is. He assumes a primacy of meaning point of view and suggests that thoughts give rise to semantic structures which are then modeled in (histori-

Introduction: Constructions and Language Change

17

cally developed) syntactic structures. The historical developments which lead to certain syntactic structures include the creation of metaphors, collocations, and idioms. Similarly, grammaticalization and phonological changes may play a role here. Thus, syntax in general is interpreted as “a museum” of a number of changes, each of which removes the linguistic expressions from its origins in linguistic thought.

Notes 1. Inheritance relations deal with similarity and compositionality among different constructions; polysemy refers to semantic and/or functional differentiation in one given form (isomorphism), synonymy refers to similar meaning and/or function of different constructions and the related question of competition between constructions. 2. In this use of the terms ‘co-text’ and ‘con-text’, which was introduced in translation studies in the nineteen-sixties, ‘co-text’ refers to the strictly linguistic environment of a given item, while ‘con-text’ encompasses extra-linguistic, communicative, and pragmatic factors (cf. Catford 1965). 3. This fairly straightforward analysis already poses some problems for more recent constructional treatments, like Construction Grammar (CxG). Traditionally, language variation and change cuts across the boundaries of linguistic levels: changes in phonology affect morphology affect syntax affect discourse, and vice versa. How can we capture these concomitant and interdependent movements in a theoretical framework that does not explicitly distinguish between the different levels?

References Bergs, Alexander and Gabriele Diewald (eds.) In prep. Context and Constructions Grammar. Amsterdam: John Benjamins. Boas, Hans C. and Mirjam Fried (eds.) 2005 Construction Grammar: Back to the Roots. Amsterdam: John Benjamins. Bybee, Joan 2005 The impact of use on representation: grammar is usage and usage is grammar. Unpublished ms. Bybee, Joan and Paul Hopper (eds.) 2001 Frequency and the emergence of linguistic structure. Amsterdam: John Benhamins. Bybee, Joan L., William Pagliuca and Revere D. Perkins 1994 The Evolution of Grammar: Tense, aspect and mood in the languages of the world. Chicago: University of Chicago Press.

18

Alexander Bergs, Gabriele Diewald

Cappelle, Bert 2006

Particle Placement and the Case for “Allostructions”. In: Doris Sch¨onefeld (ed.), Constructions all over: Case Studies and Theoretical Implications (Constructions Special Volume 1– 7/2006). (http://www.constructions-online.de/articles/specvol1/683 [21.04.2008])

Catford, John C. 1965 A Linguistic Theory of Translation. London: Oxford University Press. Croft, William 2001 Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford: Oxford University Press. Croft, William and D. Alan Cruse 2004 Cognitive Linguistics. Cambridge: Cambridge University Press. ¨ Dahl, Osten (ed.) 2000 Tense and Aspect in the Languages of Europe. Berlin/New York: Mouton den Gruyter. De Swart, Henri¨ette 1998 Aspect Shift and Coercion. Natural Language and Linguistic Theory 16: 347–385. Diewald, Gabriele 2002 A Model for Relevant Types of Contexts in Grammaticalization. In: Gabriele Diewald and Ilse Wischer (eds.), New Reflections on Grammaticalization. International Symposium, Potsdam, 17–19 June, 1999, 3–120. Amsterdam: John Benjamins. 2006 Context Types in Grammaticalization as Constructions. In: Doris Sch¨onefeld (ed.), Constructions all over: Case Studies and Theoretical Implications (Constructions Special Volume 1–7/2006.). (http://www.constructions-online.de/articles/specvol1/ [25.01.2008]) Diewald, Gabriele and Mechthild Habermann 2005 Die Entwicklung von werden & Infinitiv als Futurgrammem: Ein Beispiel f¨ur das Zusammenwirken von Grammatikalisierung, Sprachkontakt und soziokulturellen Faktoren. In: Torsten Leuschner, Tanja Mortelmans and Sarah De Groodt (eds.), Grammatikalisierung im Deutschen, 229–250. Berlin: de Gruyter. Fillmore, Charles J. 1988 The mechanisms of ‘Construction Grammar’. BLS 14: 35–55. Fillmore, Charles J., Paul Kay and Mary O’Connor 1988 Regularity and idiomaticity in grammatical constructions: The case of let alone. Language 64(3): 501–538. Fischer, Kerstin and Anatol Stefanowitsch (eds.) 2006 Konstruktionsgrammatik. Von der Anwendung zur Theorie. T¨ubingen: Stauffenburg.

Introduction: Constructions and Language Change

19

¨ Fried, Mirjam and Jan-Ola Ostman (eds.) 2004 Construction Grammar in a Cross-Language Perspective. Amsterdam: John Benjamins. ¨ Fried, Mirjam and Jan-Ola Ostman 2004 Construction Grammar: A thumbnail sketch. In: Fried, Mirjam and ¨ Jan-Ola Ostman (eds.), Construction Grammar in a Cross-Language Perspective, 11–86. Amsterdam: John Benjamins. Goldberg, Adele E. 1995 Constructions: A Construction Grammar Approach to Argument Structure. Chicago: Chicago University Press. 2006 Constructions at Work: The Nature of Generalization in Language. Oxford: Oxford University Press. Goldberg, Adele E. and Ray Jackendoff 2004 The English Resultative as a Family of Constructions. Language 80(3): 532–568. Goldberg, Adele E. and Devin Casenhiser 2006 English Constructions. In: Bas Aarts and April McMahon (eds.), The handbook of English linguistics, 343–335. Malden, Mass.: Blackwell. Gries, Stefan Th. 2003 Multifactorial Analysis in Corpus Linguistics: A Study of Particle Placement. New York: Continuum International Publishing Group Ltd. Gries, Stefan Th. and Anatol Stefanowitsch 2004a Extending Collostructional Analysis: A Corpus-based Perspective on ’Alternations’. International Journal of Corpus Linguistics 9(1): 97– 129. 2004b Co-varying Collexemes in the IntoCausative. In: Michel Achard and Suzanne Kemmer (eds.), Language, Culture, and Mind 225–336. Stanford, CA: CSLI. Harris, Alice C. and Lyle Campbell 1995 Historical Syntax in Cross-Linguistic Perspective. Cambridge: Cambridge University Press. Haspelmath, Martin 1999 Why is grammaticalization irreversible? Linguistics 37(6): 1043– 1068. Heine, Bernd and Tania Kuteva 2006 Language Contact and Grammatical Change. Oxford: Oxford University Press. Himmelmann, Nikolaus 2004 Lexicalization and Grammaticization: Opposite or Orthogonal? In: Walter Bisang, Nikolaus P. Himmelmann and Bj¨orn Wiemer (eds.), What makes Grammaticalization? A Look from its Fringes and its Components, 21–24. Berlin: Mouton de Gruyter.

20

Alexander Bergs, Gabriele Diewald

Jackendoff, Ray 1977 The architecture of the language faculty. Cambridge: MIT Press. 2002 Foundations of Language. Oxford: Oxford University Press. Kay, Paul 2002 An Informal Sketch for a Formal Architecture for Construction Grammar. Grammars 5: 1–19. Kay, Paul and Charles Fillmore 1999 Grammatical constructions and linguistic generalizations: The What’s X doing Y? construction. Language 75(1): 1–33. Labov, William 1994 Principles of Linguistic Change. Volume 1: Internal Factors. Malden, Mass.: Blackwell. Lehmann, Christian 1982 Thoughts on Grammaticalization: A Programmatic Sketch. K¨oln: Institut f¨ur Sprachwissenschaft. Michaelis, Laura A. 2003 Headless Constructions and Coercion by Construction. In: Elaine J. Francis and Laura A. Michaelis. (eds.), Mismatch: Form-Function Incongruity and the Architecture of Grammar, 259–310. Stanford: CSLI Publications. 2004 Type Shifting in Construction Grammar: An Integrated Approach to Aspectual Coercion. Cognitive Linguistics 15: 1–67. Mondorf, Britta 2006 More Support for More-Support: The Role of Processing Constraints on the Choice between Synthetic and Analytic Comparative Forms. Unpublished postdoctoral thesis. ¨ Ostman, Jan-Ola and Mirjam Fried 2004 Construction Grammar: A thumbnail sketch. In: Mirjam Fried and ¨ Jan-Ola Ostman (eds.), Construction Grammar in a Cross-Language Perspective, 1–10. Amsterdamm: John Benjamins. Sch¨onefeld, Doris 2006 Constructions. In: Doris Sch¨onefeld (ed.), Constructions all over: Case Studies and Theoretical Implications (Constructions Special Volume 1–7/2006). (http://www.constructionsonline.de/articles/specvol1/667 [21.04.2008]) Tomasello, Michael 2005 Constructing a Language: A Usage-Based Theory of Language Acquisition. Cambridge, Mass.: Harvard University Press. 2006 Construction Grammar for Kids. In: Doris Sch¨onefeld (ed.), Constructions all over: case Studies and Theoretical Implications (Constructions special Volume 1–7/2006). [http://www. constructionsonline.de/articles/specvol1/689 [21.04.2008])

Introduction: Constructions and Language Change

21

Traugott, Elizabeth Closs and Paul Hopper 2003 Grammaticalization. Cambridge: Cambridge University Press. Traugott, Elizabeth Closs 2003 Constructions in Grammaticalization. In: Brian D. Joseph and Richard D. Janda (eds.), The Handbook of Historical Linguistics, 624–647. Oxford: Blackwell. 2007 The concepts of constructional mismatch and typeshifting from the perspective of grammaticalization. Cognitive Linguistics 18(4): 523– 557. Weinreich, Uriel, William Labov and Marvin I. Herzog 1968 Empirical Foundations for a Theory of Language Change. In: Winfried P. Lehmann and Yakov Malkiel (eds.), Directions for Historical Linguistics, 95–95. Austin: University of Texas Press. Wiemer, Bj¨orn 2004 The Evolution of Passives as Grammatical Constructions in Northern Slavic and Baltic Languages. In: Walter Bisang, Nikolaus P. Himmelmann and Bj¨orn Wiemer (eds.), What makes Grammaticalization? A Look from its Fringes and its Components, 271–331. Berlin: Mouton de Gruyter. Wray, Alison 2002 Formulaic Language and the Lexicon. Cambridge: Cambridge University Press. Ziegeler, Debra 2007 A word of caution on coercion. Journal of Pragmatics 39: 990–1028.

The grammaticalization of NP of NP patterns Elizabeth Closs Traugott 1. Introduction ‘Constructions’ have been mentioned at least since Giv´on (1979) and Lehmann (1995 [1982]) as input, along with lexical items, to grammaticalization.1 However, it is often not clear what is meant by ‘construction’. At best the term appears to have been used as weakly synonymous with ‘collocation’, ‘string’, ‘phrase’, ‘constituent’, or even ‘syntagmatic context’. The recent development of varieties of construction grammar2 has provided the opportunity to begin to refine what is meant by the statement that constructions grammaticalize. An investigation of the histories of three English Partitive constructions in terms of construction grammar, especially as represented by Radical Construction Grammar (Croft 2001), suggests some of the potential advantages and difficulties of this approach. In this paper I provide overviews of earlier characterizations of the role of constructions in grammaticalization (section 2), and of construction grammar, especially Radical Construction Grammar (section 3). After a sketch of the development of the Complex Determiner/Quantifiers a kind/bit/shred of from [NP1 [of NP2]] > [[NP1 of] NP2], etc. (section 4), I assess some aspects of a construction grammar approach to grammaticalization (section 5). Section 6 provides a brief conclusion.

2. Some claims about “constructions” and grammaticalization in earlier work Meillet considered ‘lexical items’ to be the source of most instances of grammaticalization, but he also included word order and phrases, or lexical items in context, as for example in his discussion of French suis ‘I am’ in the context of phrasal chez moi ‘at home’ as opposed to lexical parti ‘left’ (1958 [1912]: 131). Such phrases have been called ‘constructions’ in the literature on grammaticalization and have been seen as source as well as outcome of grammaticalization by e.g., Giv´on (1979), Bybee, Perkins, and Pagliuca (1994), Heine (2003), Hopper and Traugott (2003 [1993]), Traugott (2003), and Himmelmann (2004). Some representative statements are:

24

Elizabeth Closs Traugott [G]rammaticalization does not merely seize a word or morpheme . . . but the whole construction formed by the syntagmatic relations of the elements in question (Lehmann 1992: 406). It is the entire construction, and not simply the lexical meaning of the stem, which is the precursor, and hence the source, of the grammatical meaning (Bybee, Perkins, and Pagliuca 1994: 11).

One of the key hypotheses in construction grammar, as in Cognitive Linguistics (e.g., Langacker 1987, 1991), has been that form and meaning are paired as equals; meaning is not interpreted from syntax, as in generative grammar. Although the pairing of form and meaning has not always been explicitly envisioned in approaches to grammaticalization, nevertheless, correlations among levels of grammar have been considered as of the highest importance in most work on the subject.3 For example, Lehmann (1995 [1982]) proposes six parameters which form a correlated set of paradigmatic and syntagmatic constraints (integrity, paradigmaticity, paradigmatic variability, structural scope, bondedness, syntagmatic variability), and says: We may say that grammaticalization as a process consists in a correlative increase or decrease of all the six parameters taken together (Lehmann 1995 [1982]: 124).

While semantics is backgrounded in Lehmann’s work, it is foregrounded in most functionalist work on grammaticalization (see e.g., Bybee, Perkins, and Pagliuca 1994, Hopper and Traugott 2003 [1993], Himmelmann 2004, and especially Heine, Claudi, and H¨unnemeyer 1991). Furthermore, Bybee, Perkins, and Pagliuca hypothesize that there is coevolution of semantics, morphosyntax, and morphophonology: Our hypothesis is that the development of grammatical material is characterized by the dynamic coevolution of meaning and form (Bybee, Perkins, and Pagliuca 1994: 20).

Most recently, Himmelmann (2004) has considered the correlation of three types of expansion (host-class, syntactic, semantic-pragmatic) essential to grammaticalization. His examples are from the development of articles out of demonstratives. He illustrates host-class expansion by the co-occurrence of articles with “proper names or nouns designating unique entities (such as sun, sky, queen, etc.), i.e. nouns they typically did not co-occur with before”; syntactic expansion by the emergence of articles with non-core (e.g., adpositional) argument positions later than core (subject, object) positions; and semantic/pragmatic expansion by increase of usage contexts from “deictic . . . anaphoric or recognitional reference” to “larger situation uses (the queen, the pub) and associated anaphoric uses (a wedding–the bride)” (Himmelmann 2004: 32–33). According

The grammaticalization of NP of NP patterns

25

to Himmelmann, in grammaticalization all three contexts expand. However, in lexicalization the first does not expand, while syntactic or semantic-pragmatic contexts may stay the same, expand, or narrow (Himmelmann 2004: 36). Such correlational approaches to grammaticalization are consistent with much work on construction grammar, and at least in the case of Croft (2001)’s Radical Construction Grammar, have in fact contributed to the development of the theory. It is therefore appropriate to consider to what extent construction grammar can inform and be informed by work on grammaticalization.

3. Construction grammar approaches Key to most versions of construction grammar is the concept of grammar as a holistic and usage-based framework, i.e. no one level of grammar is autonomous, ¨ or “core” (see Fried and Ostman 2004: 24). Rather, semantics, morphosyntax, phonology, and in some models also pragmatics, work together in a construction. This is clearly very much in keeping with many approaches to grammaticalization. However, while the general position is that all of language is constructional, and that “Construction Grammar can (or could) be a universal theory of gram¨ mar” (Ostman and Fried 2004: 6), in practice earlier work, such as Goldberg (1995) and Kay and Fillmore (1999), focused primarily on patterns not strictly predictable from their component parts. Indeed, the construction grammar program, as initiated by Goldberg, explicitly focused on idiosyncrasies: [A] distinct construction is defined to exist if one or more of its properties are not strictly predictable from knowledge of other constructions existing in the grammar (Goldberg 1995: 4).

To the extent that the focus of data analysis is on idiosyncrasies, the objectives of practitioners of grammaticalization and construction grammar diverge, since most proponents of grammaticalization have sought to account for general, indeed universal, patterns in grammaticalization. Furthermore, the origins of grammaticalization have been sought in universal processes, whether generalized invited inferences arising in particular contexts of use (see e.g., Traugott 2002, Hopper and Traugott 2003 [1993]: 81–84), metaphorical extensions (e.g., Heine, Claudi, and H¨unnemeyer 1991), or the development of lexical into functional heads (e.g., Roberts and Roussou 2003). Nevertheless, some attempts to approach grammaticalization, largely synchronically, from the perspective of Goldberg’s and Kay and Fillmore’s approaches to construction grammar have been made, see e.g., Wiemer and Bisang (2004), No¨el (2005).

26

Elizabeth Closs Traugott

Recently, however, there has been a shift away from idiosyncrasy to constructions of all types (see e.g., Croft 2001, Goldberg 2003, 2006). This is encapsulated in the aphorism that it’s “constructions all the way down” (Goldberg 2006: xx). Because Radical Construction Grammar (Croft 2001) was developed in part to account for grammaticalization, this model seems particularly appropriate for conceptualizing grammaticalization. In Radical Construction Grammar constructions are symbolic units conceived as in Figure 1:

syntactic properties morphological properties phonological properties

| semantic properties pragmatic properties discourse-functional properties

CONSTRUCTION

FORM

symbolic correspondence link (CONVENTIONAL) MEANING

Figure 1. Model of the symbolic structure of a construction in Radical Construction Grammar (Croft 2001: 18, Croft and Cruse 2004: 258; reprinted by permission)

The link between form and conventional meaning is construed as internal to the construction. Conceptual structures may be universal, but in other respects constructions are language-specific. With respect to grammaticalization, Radical Construction Grammar assumes correlations such as have been outlined in section 2, e.g.: a. In the grammaticalization process, the construction as a whole changes meaning (Croft 2001: 261). b. The [newly conventionalized] construction is polysemous with respect to its original meaning . . . the new construction undergoes shifts in grammatical structure and behavior in keeping with its new function (Croft 2001: 127). c. Extension of constructions to new uses is a change in the distribution of that construction, and such changes are theorized to follow connected paths in conceptual space (Croft 2001: 130). In the next two sections I briefly present some examples and raise questions about some of the assumptions that have been introduced in this section.4

The grammaticalization of NP of NP patterns

27

4. Some examples: the development of three degree modifier constructions English has a large number of expressions with the form NP1 of NP2.5 This syntactic string has a wide array of functions, including locative the back of the house, partitive a piece of a plate, measure a cup of tea, approximative (a) sort of a frog, and emotionally charged epithets such as an idiot of a teacher, as well as subjective genitives like a painting of (=by) Lee Krasner, and objective genitives like a portrait of a hunter. NP1 of NP2 strings therefore participate in a number of different constructions, where syntax, meaning, and pragmatic function can be construed at various degrees of granularity, ranging from general classes to individual idiosyncratic combinations which differ with respect to the kind of noun or determiner that can realize either NP, or the pragmatics associated with the string. In this paper I consider a set of English binominal patterns, all “constituency constructions” (Zwicky 1994: 611): (1)

a. A side of the barn b. A kind of hawk c. A bit of an apple d. A shred of a robe

b’. (a) kind of a problem c’. a bit of a liar d’. (not) a shred of honor

Syntactically patterns exemplified in (1 b.–d.) undergo grammaticalization to patterns exemplified by (1 b’.–d’.) via the changes in (2) (“Cxn” is short for “construction”):6 (2)

[NP1 [of NP2]] Head = NP1 Partitive Cxn

> > >

[[NP1 of] NP2] Head = NP2 Complex Determiner/Quantifier Cxn

There are at least three robust criteria for determining when the Complex Determiner/Quantifier (CompDet/Quant) construction, as opposed to the Partitive7 construction is involved (see Denison 2002, 2005): a) agreement patterns: in the Partitive construction the initial determiner agrees in number with N1 (these kinds of skill), but in the CompDet/Quant construction it agrees with N2 (these kind of skills), at least in colloquial use b) in the Partitive construction NP2 may be preposed (of an apple a bit) but not in the CompDet/Quant construction (*of a liar a bit) c) in the CompDet/Quant, but not in the Partitive construction, a N1 of can be replaced by one word: a bit of/rather/quite a talker.

28

Elizabeth Closs Traugott

Some CompDet/Quant expressions develop further into Degree Modifiers which serve as approximators (kind of ), or scalars (a bit) that can modify not only NPs but also adjectives (kind of red, a bit red) and verbs (I kind of ran), and ultimately into Free Adverbs that can be used without heads as responses to questions, as in Q. Did you like it? A. Kind of/(Not) a bit. Semantically (1 a.–d.) all involve part-whole relationships. Despite these similarities, they have rather different histories. (1 a.) involves a part that is a spatial adjunct. The absence of an obviously quantificational inference seems to have prevented side of from being reinterpreted as CompDet/Quant (in many languages such body-part expressions become case markers, see Svorou 1993, Heine 1997, Heine and Kuteva 2002). By contrast, in certain contexts (1.b.d.) could be inferred to have quantificational properties and were reinterpreted as CompDet/Quant. Subsequently (1.b’. and 1c’.) were further reinterpreted as adverbial Degree Modifiers that scale their heads up or down on a scale of closeness to a prototype, but with different constraints, as summarized below (for fuller details, see Denison 2002 on a kind/sort of, Traugott Forthc a on a lot of, Forthc b on a lot/deal/piece of; also Brems 2003, Brems and Davidse 2005 on the development of “size nouns” in, general). As Denison (2002) points out, all the constructions in question, and others of this type in English, depend crucially on the reanalysis in Middle English after Step I of Old English of ‘out of’ as the analytic equivalent of the genitive inflection, and eventually as the default preposition in English. The following outlines are highly schematic and ignore many details, including distinctions among determiners within NP1 of NP2 patterns.

4.1.

(a) kind of (see also a sort of )

(3)

Step I, OE: gecynd ‘kind, nature, (superordinate) class in a taxonomy’: – Crist . . . . wearð acenned of menniscum gecynde of þam Iudeiscum cynne ‘Christ . . . was born of human kind of that Judaic kin’(c.1000 ÆLS, Maccabees 514 [DOE]) > Step II , early 16thC: kind ‘individual included in a class/member of class’, i.e. the superordinate term has been reinterpreted as a member of the set: – A newe kynde of sicknes came sodenly ... into this Isle (a1548 Hall Chron., Hen. VII 3b [OED kind II.14a]) – a Cotton or hempy kind of moss (c.1654 Howell, Lett. II. 54 [OED hempy]) > Step III , CompDet/Quant, early 18thC: ‘member of class not possessing full characteristics of the class’:

The grammaticalization of NP of NP patterns

29

– kind of what d’ye call ‘em ... a sort a Queen or Wife (1752 Foote, Taste II [OED what d’ye call ‘em]) > Step IV , DegreeAdverb, late 18thC: – I kind of love you [1804 Fessenden Poems 100 [OED kind 14d]) > Step V , Free Adverb, mid 19thC: – Yes, kind of , sort of (1852 Aiken, Uncle Tom’s Cabin (dramatic version) [UVa]); here kind of, sort of downtone the strength of the assent.

4.2.

a bit (of) (see also a lot/bunch (of))

(4)

>

>

>

>

Step I , OE: bita ‘bite (out) of’: – Hu he wrec in adam þe bite of an eappel ‘How he avenged in Adam the bite of an apple’ (c.1230 Ancr. 91a [MED]) – þis appyl a bete þerof þou take ‘this apple a bite thereof thou take’ (= take a bite of this apple) (c.1475 Ludus C. 23/220 [MED]); note the preposed NP2 Step II , Partitive, 16thC: bite ‘mouthful/morsel/small part of’; by metonymy from action of biting to result: – Gif God was made of bits of breid (= bread) (c.1550 Scot. Poems C. II. 197 [OED]) Step III , CompDet/Quant, mid 17thC: ‘inadequate small part of’: – if you think to scape with sending mee such bitts of letters you are mistaken (1653 DOsborne: Lett. 36.771 [PCEEC]) Step IV , DegreeAdverb, mid 18thC: – I would not be a bit wiser, a bit richer, a bit taller, a bit shorter, than I am at this Instant (1723 Steele, The Conscious Lovers III.i [UVa]) – You need not be a bit afraid of going on with me (1863 Trollope, Rachel Ray I [UVa]) Step V , Free Adverb, 18thC; usually with not: – A. Hear me. B. Not a bit (1739 Baker, The Cit Turn’d Gentleman [LION; English Prose Drama])

Note that the last two ‘steps’ appear at about the same time, at the beginning of the 18thC. The sequence postulated is, however, not only logical, but also supported by evidence that a lot developed its Free Adverb function a hundred years after its Degree Modifier one.

30

Elizabeth Closs Traugott

4.3.

(not) a shred of (see also (not) an iota/drop/jot (of))

(5)

Step I , OE schrede ‘bit cut or broken off from fruit, vegetable, textile, coin, vessel’: – gif heo mei sparien eni poure schreaden ‘if she may spare any poor shreds’ (a1225 Ancrene Riwle 416 [OED]) > Step II , Partitive, 14thC: ‘small part of’: – The white brent than rede, / That of him nas founden a schrede / Bot dust ‘The white burned the red, so-that of him not-was found a shred but dust’ (= The white [dragon] burned the red so that not an iota was found of him except dust) (c1300 Arthur and Merlin 1540 [MED and OED]); the preposing and the contrast with ‘dust’ confirm the Partitive reading In the 16thC a shred of was generalized to language, mankind, nature; it was favored with but not restricted to mass nouns, and used to imply an insufficient exemplar compared to the expected norm: – Suche shredis of sentence strowed in the shop of ancient Aritippus ‘Such scraps of wisdom strewn in ancient Aritippus’ shop’ (1529 Skelton, Sp. Parrot 94 [OED]) > Step III , CompDet/Quant; usually used in negative polarity constructions (negatives, interrogatives, etc.), 19thC: ‘some’: – Loto has not a shred of beauty (1867 Ouida, Under Two Flags [LION; 19thC fiction]) but it is not limited to negative polarity contexts: – The multicultural cast gives a shred of substance to what’s otherwise a standard adolescent gross-out flick [2005 Google, Christ. Sci. Monitor])

5. Assessment of advantages and disadvantages of the construction grammar and radical construction grammar approaches to grammaticalization 5.1. A sketch of the main changes in terms of Croft’s model A first step in assessing the advantages and disadvantages of the Radical Construction Grammar approach is to consider how Figure 1 could be used to show correlations among the changes, from the perspective of both internal structure and relationships among the individual constructions. Figure 2 is an attempt to capture some major aspects of the summaries in 4.1.–4.3. Except for the lefthand item (‘Source’, which is an eclectic group of [NP [of NP]] expressions that

The grammaticalization of NP of NP patterns PrePart Step I SY MO PH

[NP1 [of NP2]] of proclitic a kind of

SM PG

class of container

Part Step II

>

= = = = = = = = = = = =

>

DegMod/Q Step III

>

member of contained +> inad mem

[[NP1 of] NP2] of enclitic = = = =

>

DegAdv Step IIIa

>

downtoner approximator

>

FreeAdv Step IV Free Adv = = = = = = = =

>

>>

= = = = = = = =

= = = = = = = =

hedge

DF 1000

1550

SY MO PH

[NP1 [of NP2]] of proclitic a bite of

SM PG

ablative act 3D shape

>

1750

= = = = = = = = a >bit of

>

member of small part of +> inad mem

[[NP1 of] NP2] = = = = = = >= = downtoner approx

DF

1200

1550

1650

SY MO PH

[NP1 [of NP2]] of proclitic a shred of

= = = = = = = = = = = =

[[NP1 of] NP2] = = = = = = = =

SM PG

elative

>

DF 1300 Key:

Adv__A/V kind of kinda

SY MO PH SM PG DF

syntax morphology phonology semantics pragmatics discourse function

>

member of small part of +> inad mem

quantifier pos evl of NP2

1500

1850 ‘= = = =’ +> ‘inad mem’ ‘approx’ ‘neg pol’ ‘pos evl’

tag, response

1800

1850

Adv__A

>

a bit

>

>

>

Adv = = = = = = >= =

==== ====

= = = = = = = =

hedge

response; favors neg pol

1750

31

1750

favors neg pol

‘carries over from prior cxn’ ‘implies’ ‘inadequate member’ ‘approximator’ ‘negative polarity’ ‘positive evaluation’

Figure 2. Model of the development of a kind of, a bit of, a shred of

are historically prior stages in English), the top line identifies what I consider to be functional superordinate ‘macro-constructions’: meaning-form pairings that are defined by function. The individual constructions, a kind of, a bit of, a shred of, as presented here, are at the lowest level in the hierarchy of relevant constructions. However, they are representatives of sets of similar individual constructions at an intermediary level that I will call ‘meso-constructions’. ( A) kind of has as its congener (a) sort of, while a bit (of) has as its congeners a lot (of), a bunch (of), a tad (of ), and a shred of has a drop of, an iota of, a jot of.

32

Elizabeth Closs Traugott

In other words, there is a hierarchy of idealized category levels of which only the first and third are represented in Figure 2: i) macro-constructions (Partitive, CompDet/Quant, etc.), ii) meso-constructions (groupings of similarly-behaving constructions, e.g., a kind of, a sort of, as a set distinct from the set a bit of, a lot of, etc.), iii) constructions (individual constructions such as a kind of, a bit of ). For consistency I will call them ‘micro-constructions’. These three levels are abstractions: types, as distinct from actual utterances. The latter are data-points, or ‘constructs’. In making these hierarchical distinc¨ tions, I follow the spirit of Fried and Ostman (2004) and add hierarchic levels to Radical Construction Grammar as discussed in Croft (2001). Note also that the representations in Figure 2 give a discreteness to the constructions under review that is antithetical to the very slight distinctions among (micro-)constructions in “near-continuous multidimensional space” that Croft envisages both synchronically and diachronically (2001: 314). It also obscures the highly local structural shifts normally subsumed under ‘gradualness’ (in the sense of local step-by-step development, see Lichtenberk 1991). However, schematic presentation is useful to highlight analytic points. In Figure 2 the phonology is represented by modern spelling. Discourse function (DF) is only minimally specified. The DF is presumably meant to specify the larger contexts in which constructions themselves appear (see Bergs 2005 on the importance of investigating the contexts of constructions as ‘factors and products’of grammaticalization). For example, it is presumably relevant to the development of the constructions in question that at Step I the constructs that came to be used in the new ways that led to Step II, appear to have been used primarily in oblique position, in other words, in contexts where they were new information (focus) (this coincides with their indefiniteness). DF in Croft’s model presumably also refers to external situational factors, but how to constrain such external contexts is difficult to determine for particular constructions such as these.

5.2.

Some potential advantages of a Radical Construction Grammar approach

While the correlations displayed in Figure 2 can be stated seriatim in any analysis, constructional or not, a constructional approach has the obvious advantage of allowing us to visualize the correlations simultaneously. They therefore allow us to evaluate instantaneously various hypotheses that have been put forward about grammaticalization. For example, they support the hypothesis cited at the beginning that “[i]t is the entire construction, and not simply the lexical meaning

The grammaticalization of NP of NP patterns

33

of the stem, which is the precursor, and hence the source, of the grammatical meaning” (Bybee, Perkins, and Pagliuca 1994: 11). However, other hypotheses are called into question, most notably the hypothesis, also cited at the beginning, that there is “coevolution” of form and meaning (Bybee, Perkins, and Pagliuca 1994: 20), at least in a strict interpretation of simultaneity.8 Note that phonological strings remain constant through Steps II – III, and IV – V. Another hypothesis that is falsified is that there is a ‘cycle’ from syntax > morphology > 0 that is a ‘cause’ for later renewal (among many works citing loss as a cause for renewal, see Heine and Reh 1984; also van Gelderen 2005 on loss of French ne via ne pas > pas [“Jespersen’s Cycle”]). The Partitive and adverbial Degree Modifier constructions are shown to have been repeatedly renewed over a short period of time, without prior or even subsequent loss of competing constructions. None of the construction-types illustrated disappears, although some of the particular meanings attested at Stage I do.9 Therefore, it is still possible to refer to a class of things, even if the word kind is not used with this meaning, and to a bite out of an apple, even though bite and bit have diverged phonologically. Constructions serve as attractor sets and give substance to the observation that gradualness is not a “slippery slope” but rather has “cluster-points” (Hopper and Traugott 2003 [1993]: 6). No¨el (2005) suggests constructions are ‘responsible’ for individual grammaticalizations, and Croft (2001: 127) that “the new construction at least partially imposes the conceptualization of its original structure and function”. In other words, constructions highlight the force of analogy rather than reanalysis. Each entering item undergoes local reanalysis, but the attracting force is analogy, alignment with an already existing pattern (see Traugott Forthc b).10 From this perspective, we can see ‘family resemblances’ emerging at various levels of abstraction: Partitive versus CompDet/Quant versus Degree Modifier, or within each of these construction-types, sets that behave in similar ways but yet are different. The congeners of (a) kind of and a bit (of) differ in whether the article or of has a phonological reflex and in the type of Degree Modifier they become (approximator vs. downtoner). By contrast, a shred of and its congeners cannot be used as Degree Modifiers (a bit bare, but not *a shred bare), or as Free Adverbs in answer to a question. But they share with a bit (of) the propensity to be used in negative polarity contexts, a factor which indicates overlap with another set of constructions.11 It appears then that the ‘imposition’ by a construction is only partial. A question to be researched is at what hierarchic level this imposition (to the extent that it occurs) is most likely to occur. The concept of ‘attractor sets’ has the potential to allow for detailed study of the degree to which any particular lower-level construction is firmly established in all six dimensions (see Figure 1) of the higher construction, in other

34

Elizabeth Closs Traugott

words, how well it fits the characteristics of the next higher-level construction. Once attraction/analogy has occurred, it should be possible to track the degree to which syntactic, morphological, phonological, semantic, pragmatic, and discourse-functional properties have been attracted. We can call this paradigmatic (‘vertical’) entrenchment. While measures of the strength of association on this paradigmatic dimension appear not to have been studied yet, suggestive work has been done on synchronic syntactic (‘horizontal’) entrenchment: the strength of association between a schematic slot in a (meso-)construction and the lexical fillers of that slot (see Geeraerts, Tummer, and Speelman 2005 on “collostructions”, citing concepts in Gries and Stefanowitsch 2004; also Hoffmann 2004, Rice and Newman 2004). In diachronic terms this could be thought of as assessment of the strength of individual host-class expansions (see Hilpert 2007). One could, for example, measure the strength of association over time between a bit of and apple, bread, liar, or evidence, compared with that of a shred of with the same NP2s. One could also perhaps extend the technique to measuring changing strengths of association with grammatical contexts, such as negative polarity contexts in the case of a bit of and a shred of, etc., compared with a kind of. In other words, collostructional analysis could give insights into ways in which prototypical membership within a meso- or macro-construction changes over time with respect to both lexical/substantive constituency and grammatical/schematic structure.

5.3.

Some issues for further consideration

While the potential advantages of looking at grammaticalization from the perspective discussed here are promising, some potential problems need to be recognized as well. They suggest ways in which construction grammar and Radical Construction Grammar as formulated in Croft (2001) might be refined. One considerable concern is that, if taken seriously, the holistic focus on constructions adopted by most proponents of construction grammar may detract from attention to the complexities of the semantic and pragmatic processes that have been identified as precursors to grammaticalization and as constraints on change. More specifically, as we have seen, the construction grammar literature assumes that constructions impose (or ‘coerce’) meaning. While this is certainly true in part, it should not overshadow the fact that grammaticalized items tend to be constrained in many cases by the constructions from which they derive, in other words, the attractive force of the meso- or micro-construction may be counter-balanced long-term12 by the ‘backward pull’ of earlier uses of a micro-construction (which are of course still available as polysemies, at least in

The grammaticalization of NP of NP patterns

35

the early stages of development). Bybee and Pagliuca (1987: 117) suggest, for example, that the differences between will and be going to as future markers can in part “be understood as continuations of their original lexical meanings”. In the data discussed here, there are differences between what is asserted by different NP1 of NP2 constructions. A speaker using the X is a bit of a Y construction asserts that X is a Y (but not much of one); thus Your friend is a bit of a liar, means ‘your friend (X) is a liar (Y), but not to a high degree’; in other words, this expressions asserts that the friend is a liar. However, a speaker using the X is kind of a Y construction asserts that X approximates Y; if I say Your friend is a kind of a liar I am not necessarily asserting that your friend is a liar, only that he or she is/behaves somewhat like one. These differences would seem to have more to do with the origins of a bit of and a kind of than with the macroconstruction they are in. As we have seen, a bit of originates in a bite (out) of, which presupposes a concrete whole and implies a concrete part resulting from the bite. By contrast, there is no such presupposition with kind of, which originates in the abstraction ‘class, nature of’ and implies a set, the members of which may be null.13 To regard a construction as a conventionalized formmeaning pairing on all six levels specified in Figure 1, with a holistic, noncompositional meaning, would be problematic in accounting for the ‘backwardpull’effect. It would also be problematic in accounting for invited inferences that develop and become conventionalized over time before a polysemous new use can come into being; for example, it was necessary for the invited inference that a bit of something is less than the expected norm to become conventionalized before it could become a downtoning Degree Modifier.14 Among the continuing problems in historical linguistics is that of identifying the locus of change. A solution proposed within the generative tradition has been that grammars change, as a result of reinterpretation of the output of grammars in the process of language acquisition (Kiparsky 1968). In the functionalist literature, there has been growing consensus that it is not grammars but use that changes (a position developed in Croft 2000, but anticipated in e.g., Weinreich, Labov and Herzog 1968). Where is the locus of change from the perspective of Radical Construction Grammar? It would seem to follow from the hypothesis that language change is change in use that the locus of change should be in constructs, the tokens that are produced by the speaker on analogy with a different but conceptually similar construction, or interpreted by the addressee as having some different structure than that assigned by the speaker. This is the point at which innovation or actuation of change occurs. To count as a change, innovation needs to be followed by spread or actualization (Milroy 1992, 2003, Andersen 2001). However, not having the notion of ‘construct’ at his disposal, Croft conceptualizes a construction as changing and says: “[t]he first step in the

36

Elizabeth Closs Traugott

process [of grammaticalization, ECT] is that the construction is extended to a new function” (2001: 127). If the construction is a holistic unit, then it would be expected to change as a whole, but as we have seen, it does not always do so, a fact that Croft recognizes. When he says that a construction is “extended to a new function” this seems to imply that function is independent of the rest of the construction. Likewise when he says “[i]n the grammaticalization process, the construction as a whole changes meaning” (Croft 2001: 261), this seems to imply that meaning is also independent. Indeed, later we read “[T]he real locus of innovation in language change is in symbolic relations . . . The remapping of symbolic relations between form and function is the primary mechanism for innovation of ALTERED REPLICATION of grammatical structures” (Croft 2001: 366, caps original). This seems to be correct, with the added proviso that the remapping occurs in the production or perception of constructs. One of the theoretical contributions of research in grammaticalization has been the hypothesis of unidirectionality. The change-schema in Figure 2 is unidirectional in that a free response adverb is not likely to become a complex determiner, and a complex determiner is not likely to become a partitive. As Kiparsky (Forthc) has pointed out, an exclusively exemplar-based approach to unidirectionality will always have counterexamples, even if they are only rare, since not obviously related constructions can function as counter-attractors. Kiparsky’s solution is to reconstrue analogy as optimization based in Universal Grammar (UG). Such optimization “projects UG constraints that are not positively instantiated in the language” (Kiparsky Forthc).15 This is not the place to elaborate on this complex theoretical debate, only to point out that the question of how to understand reanalysis and analogy will be crucial to further work on the role of constructions in grammaticalization. It may be that “Humans are simply analogical animals” (Anttila 2003: 438), and that much language change involves the emergence of family resemblances, but where the families come from is an important issue, as is the way in which new membership in a family comes about. Work on analogy from the perspective of construction grammar and on reanalysis from the perspective of grammaticalization has the potential for accounting for ‘local reanalysis’ (for further discussion, see Traugott Forthc b). Speakers and hearers presumably match parts of constructions. By hypothesis, an innovated construct-token may be matched to any one or more layers of a construction-type. If the innovation is replicated and conventionalized by other speakers, stronger integration with the multiple layers of a constructiontype may occur. This integration may lead to change: the development of a micro-construction-type that is partially aligned with meso-construction-types and ultimately with hierarchically high-level functions (similarly Fried, this volume, using a different model of construction grammar).

The grammaticalization of NP of NP patterns

37

6. Conclusion I have shown that Radical Construction Grammar can help model the multiple correlations between aspects of form, meaning and function that have long been considered of fundamental importance in the grammaticalization literature. It can also model the degree to which there is or is not coevolution of form and meaning. It highlights analogical processes, but does not (and should not) exclude reanalysis as a major mechanism of change. On the other hand, there is an area where the model might prove a hindrance: the holistic approach taken to meaning in the construction. An area of considerable interest for recent approaches to grammaticalization is the potential for accounting in construction grammar terms for the degree of entrenchment within a construction, either syntagmatically in collostructions or paradigmatically with respect to degrees of alignment to the various levels within a construction and from one level of constructions to another. This would ideally require a quantitative approach. Such an approach has not been taken here, but see Bybee (2003) and papers in Lindquist and Mair (2004) on grammaticalization and frequency effects.16

Notes 1. Many thanks to participants at the ICHL workshop, especially Mirjam Fried, and also to Henning Andersen, Dirk No¨el, Craige Roberts, Graeme Trousdale, and two anonymous reviewers for comments and discussion. For fuller, related versions of this paper, see Traugott (Forthc a, b). 2. Different conventions regarding the use of upper and lower case have arisen; Croft and Cruse (2004: 257) distinguish Construction Grammar as developed in e.g., Kay and Fillmore (1999) from construction grammar as developed in e.g., Goldberg (1995). I will use lower case as the default. 3. This is not to deny that some researchers in grammaticalization have focused on one aspect of grammaticalization over others in an attempt to push understanding of that particular aspect further, see e.g., Heine, Claudi, and H¨unnemeyer (1991) on semantic aspects; Roberts and Roussou (2003), van Gelderen (2004) and Batllori et al. (2005) on syntactic aspects. Such focus does not, however, undermine the basic observation that grammaticalization theorists typically view grammaticalization as a multi-layered process. 4. For other diachronic studies of grammaticalization in construction grammar terms, see Traugott (2007, Forthc a, b), and Trousdale (Forthc a, b). 5. So do most European languages (Geert Booij, p.c.).

38

Elizabeth Closs Traugott

6. In Present Day English a kind of in the Partitive sense of ‘member of a set’ typically requires a bare NP2, while the Complex Determiner and especially the Degree Modifier senses require an indefinite article. This does not change the schema at the level of generalization in (2). 7. Not all the constructions are strictly speaking ‘partitive’ in the semantic sense (a kind/sort of are taxonomic, a bit/shred of are meronymic, see Croft and Cruse 2004) although the cover term Partitive is often used for them, and I follow suit. Denison (2005) uses “qualifying construction” for a kind/sort of in the Partitive use. 8. See also Bisang (2004) for arguments against necessary coevolution of form and meaning. The more restrictive claim in Bybee and Dahl (1989: 68) that coevolution is involved in the development of affixes does, however, seem better substantiated. 9. In some other cases, however, various uses are discontinued, e.g. a piece of was briefly used as a Degree Modifier, but this has not survived, and a deal of, which was originally Partitive, is not used in this meaning, and its Degree Modifier use has survived only in relatively fixed expressions such as a great deal of (see Traugott Forthc a, b). 10. Croft appears to identify innovation primarily with reanalysis rather than analogy (Croft 2001: 366); see also Petr´e and Cuyckens (This volume). In the grammaticalization literature reanalysis has traditionally been regarded as the prime change mechanism from Meillet 1958 [1912], to Roberts and Roussou (2003). In recent years, however, this view has been changing, see Lehmann (2004), Hoffmann (2004), Kiparsky (Forthc), Fischer (2007). 11. Negative polarity items form a large set, of which any, hardly etc. are well-known. What they have in common is a scalar meaning; in this sense, negative polarity may be regarded as a functional macro-construction, the members of which are scalar constructions, including some CompDets/Quants and Degree Modifiers (see Israel 2004). 12. However, Croft assumes that such effects occur “(if at all) at the initial stage of the process, when the new construction is not yet conventionalized” (2001: 127). 13. It might be argued that speakers have no conscious knowledge of or access to such differences; while this is no doubt true of speakers in the twenty-first century who no longer have access to a meaning of kind that was lost long ago, at the time that the older meaning still coexisted with and was polysemous with the newer meanings (a period of well over a hundred years) at least some speakers would, however, have been influenced by it. The OED cites examples of the original meaning of kind through the end of the nineteenth century. Older meanings may in other words come to be entrenched, however unconsciously, as part of newer meanings. 14. For models of semantic change construed as invited inferences becoming conventionalized and then semanticized, see Traugott and Dasher (2002), Enfield (2003: 29–30). 15. Note that the specifics of Kiparsky’s UG hypothesis are not compatible with a construction-based approach of the type discussed here since they are syntacticallybased, and Radical Construction Grammar rejects the universality of syntactic categories.

The grammaticalization of NP of NP patterns

39

16. Bybee proposes that significant increase in tokens triggers grammaticalization. This has, however, been shown not always to be the case: grammaticalization can occur prior to significant token or even type increase, see Hundt (2001) and Mair (2004).

Sources of data DOE. LION. MED.

Dictionary of Old English http://ets.umdl.umich.edu/o/oec/ Chadwyck Healey website http://lion.chadwyck.com The Middle English Dictionary. 1956–2001. Ann Arbor: University of Michigan Press. (see also http://www.hti.umich.edu/dict/med/) OED. Oxford English Dictionary. 3rd ed. (in progress) http://dictionary.oed.com/ PCEEC. Parsed Corpus of Early English Correspondence, Ann Taylor, Anthony Warner, Susan Pintzuk, Terttu Nevalainen, and Arja Nurmi (eds.). Oxford Text Archive UVa. University of Virginia, Electronic Text Center, Modern English Collection http://etext.lib.virginia.edu/modeng/modeng0.browse.html

References Andersen, Henning (ed.) 2001 Actualization: Linguistic Change in Progress. Amsterdam/Philadelphia: Benjamins. Anttila, Raimo 2003 Analogy: The warp and woof of cognition. In: Brian D. Joseph and Richard D. Janda (eds.), The Handbook of Historical Linguistics, 425– 440. Malden, MA/Oxford: Blackwell. Batllori, Montse, Maria-Llu¨ısa Hernanz, Carme Picallo and Francesca Roca (eds.) 2005 Grammaticalization and Parametric Variation. Oxford: Oxford University Press. Bergs, Alexander 2005 Grammaticalizing constructions – constructing grammaticalization. Paper presented at New Reflections on Grammaticalization 3, Santiago de Compostela July 17th -20th . Bisang, Walter 2004 Grammaticalization without coevolution of form and meaning: The case of tense-aspect-modality in East and mainland Southeast Asia. In: Walter Bisang, Nikolaus Himmelmann, and Bj¨orn Wiemer (eds.),

40

Elizabeth Closs Traugott What Makes Grammaticalization - A Look from its Fringes and its Components, 109–138. Berlin/New York: Mouton de Gruyter.

Brems, Lieselotte 2003 Measure noun constructions: An instance of semantically-driven grammaticalization. International Journal of Corpus Linguistics 8(2): 283–312. Brems, Lieselotte and Kristin Davidse 2005 Type noun constructions: paths of grammaticalization. Paper presented at FITIGRA. Bybee, Joan L. 2003 Mechanisms of change in grammaticization: The role of frequency. In: Brian D. Joseph and Richard D. Janda (eds.), The Handbook of Historical Linguistics, 602–623. Malden, MA/Oxford: Blackwell. Bybee, Joan L. and William Pagliuca 1987 The evolution of future meaning. In: Anna Giacalone Ramat, Onofrio Carruba and Giuliano Bernini (eds.), Papers from the 7 th International Conference on Historical Linguistics, 108–122. Amsterdam/Philadelphia: Benjamins. ¨ Bybee, Joan L. and Osten Dahl 1989 The creation of tense and aspect systems in the languages of the world. Studies in Language 13: 51–103. Bybee, Joan L., Revere Perkins and William Pagliuca 1994 The Evolution of Grammar: Tense, Aspect, and Modality in the Languages of the World. Chicago: University of Chicago Press. Croft, William 2000 Explaining Language Change: An Evolutionary Approach. Harlow/Essex: Longman. 2001 Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford: Oxford University Press. Croft, William and D. Alan Cruse 2004 Cognitive Linguistics. Cambridge: Cambridge University Press. Denison, David 2002 History of the sort of construction family. Second International Conference on Construction Grammar (ICCG2), University of Helsinki, Sept. 6th -8th . 2005 The grammaticalisations of sort of, kind of and type of in English. Paper presented at New Reflections on Grammaticalization (NRG) 3, University of Santiago de Compostela, July 17th -20th . Enfield, Nick J. 2003 Linguistic Epidemiology: Semantics and Grammar of Language Contact in Mainland Southeast Asia. London/New York: RoutledgeCurzon.

The grammaticalization of NP of NP patterns Fischer, Olga 2007 FITIGRA 2005

41

Morphosyntactic Change: Functional and Formal Perspectives. Oxford: Oxford University Press. From Ideational to Interpersonal: Perspectives from Grammaticalization. Conference at University of Leuven, February 10th –12th .

Fried, Mirjam This volume Constructions and constructs: mapping a shift between predication and attribution. ¨ Fried, Mirjam and Jan-Ola Ostman 2004 Construction Grammar: A thumbnail sketch. In: Mirjam Fried and ¨ Jan-Ola Ostman (eds.), Construction Grammar in a Cross-Linguistic Perspective, 11–86. Amsterdam/Philadelphia: Benjamins. Geeraerts, Dirk, Jos´e Tummers and Dirk Speelman 2005 Measuring the interplay between items and constructions. Paper presented at FITIGRA. Gelderen, Elly van 2004 Grammaticalization as Economy. Amsterdam/Philadelphia: Benjamins. 2005 Linguistic cycles: Grammaticalization as economy. Paper presented at Blankensee Colloquium on Language Evolution: Cultural and Cognitive Factors, July 14th -16th . Giv´on, Talmy 1979 On Understanding Grammar. New York: Academic Press. Goldberg, Adele E. 1995 Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. 2003 Constructions:A new theoretical approach to language.Trends in Cognitive Sciences 7: 219–224. 2006 Constructions at Work: The Nature of Generalization in Language. Oxford: Oxford University Press. Gries, Stefan Th. and Anatol Stefanowitsch 2004 Co-varying collexemes in the into-causative. In: Michel Achard and Suzanne Kemmer (eds.), Language, Culture, and Mind, 225–236. Stanford, CA: CSLI Publications. Heine, Bernd 1997 Cognitive Foundations of Grammar. Oxford/New York: Oxford University Press. 2003 Grammaticalization. In: Brian D. Joseph and Richard D. Janda (eds.), The Handbook of Historical Linguistics, 575–601. Malden, MA/Oxford: Blackwell.

42

Elizabeth Closs Traugott

Heine, Bernd, Ulrike Claudi and Friederike H¨unnemeyer 1991 Grammaticalization: A Conceptual Framework. Chicago: University of Chicago Press. Heine, Bernd and Tania Kuteva 2002 World Lexicon of Grammaticalization. Cambridge: Cambridge University Press. Heine, Bernd and Mechthild Reh 1984 Grammaticalization and Reanalysis in African Languages. Hamburg: Helmut Buske. Hilpert, Martin 2007 Germanic Future Constructions – A Usage-based Study of Grammaticalization. Ph.D. Dissertation, Rice University. Himmelmann, Nikolaus P. 2004 Lexicalization and grammaticalization: Opposite or orthogonal? In: Walter Bisang, Nikolaus Himmelmann and Bj¨orn Wiemer (eds.), What Makes Grammaticalization – A Look from its Fringes and its Components, 19–40. Berlin/New York: Mouton de Gruyter. Hoffmann, Sebastian 2004 Are low-frequency complex prepositions grammaticalized? In: Hans Lindquist and Christian Mair (eds.), CorpusApproaches to Grammaticalization in English, 171–210. Amsterdam/Philadelphia: Benjamins. Hopper, Paul J. and Elizabeth Closs Traugott 2003 [1993] Grammaticalization. Cambridge: Cambridge University Press. Hundt, Marianne 2001 What corpora tell us about the grammaticalisation of voice in getconstructions. Studies in Language 25: 49–87. Israel, Michael 2004 A partial catalogue of polarity sensitive constructions in English. TS, University of Maryland. Kay, Paul and Charles J. Fillmore 1999 Grammatical constructions and linguistic generalizations: The What’s X doing Y construction. Language 75(1): 1–33. Kiparsky, Paul 1968 Linguistic universals and linguistic change. In: Emmon Bach and Robert T. Harms (eds.), Universals in Linguistic Theory, 171–202. New York: Holt, Rinehart and Winston. Forthc Grammaticalization as optimization. In: Diana Jones (ed.), The Emergence of Grammaticalization. Oxford: Oxford University Press. (http://www.stanford.edu/∼kiparsky/ [18.01.2008]) Langacker, Ronald W. 1987 Foundations of Cognitive Grammar, Vol. I. Theoretical Prerequisites. Stanford: Stanford University Press.

The grammaticalization of NP of NP patterns 1991

43

Foundations of Cognitive Grammar, Vol. II. Descriptive Application. Stanford: Stanford University Press. Lehmann, Christian 1992 Word order change by grammaticalization. In: Marinel Gerritsen and Dieter Stein (eds.), Internal and External Factors in Syntactic Change, 395–416. Berlin/New York: Mouton de Gruyter. Lehmann, Christian 1995 [1982] Thoughts on Grammaticalization. Munich/Newcastle: LINCOM EUROPA. Lehmann, Christian 2004 Motivation in language: Attempt at systematization. http://www.unierfurt.de/sprachwissenschaft/personal/lehmann/d lehmann.html [18.01.2008] Lichtenberk, Frantisek 1991 On the gradualness of grammaticalization. In: Elizabeth Closs Traugott and Bernd Heine (eds.), Approaches to Grammaticalization, Vol. I: 37–80. Amsterdam/ Philadelphia: Benjamins. Lindquist, Hans, and Christian Mair (eds.) 2004 Corpus Approaches to Grammaticalization in English. Amsterdam/Philadelphia: Benjamins. Mair, Christian 2004 Corpus linguistics and grammaticalisation theory: Statistics, frequencies, and beyond. In: Hans Lindquist and Christian Mair (eds.), Corpus Approaches to Grammaticalization in English, 121–150. Amsterdam/Philadelphia: Benjamins. Meillet, Antoine 1958 [1912] L’´evolution des formes grammaticales. In: Antoine Meillet (ed.), Linguistique historique et linguistique g´en´erale, 130–148. Paris: Champion. Milroy, James 1992 Linguistic Variation and Change: On the Historical Sociolinguistics of English. Oxford: Blackwell. 2003 On the role of the speaker in language change. In: Raymond Hickey (ed.), Motives for Language Change, 143–157. Cambridge: Cambridge University Press. No¨el, Dirk 2005 The productivity of a ‘source of information’ construction: Or, where grammaticalization theory and Construction Grammar meet. Paper presented at FITIGRA. ¨ Ostman, Jan-Ola and Mirjam Fried 2004 Historical and intellectual background of Construction Grammar. In: ¨ Mirjam Fried and Jan-Ola Ostman (eds.), Construction Grammar in

44

Elizabeth Closs Traugott

a Cross-Linguistic Perspective, 1–10. Amsterdam/Philadelphia: Benjamins. Petr´e, Peter and Hubert Cuyckens This volume Bedusted, yet not beheaded: The role of be-’s constructional properties in its conservation. Rice, Sally and John Newman 2004 Aspect in the making: A corpus analysis of English aspect-marking prepositions. In: Michel Achard and Suzanne Kemmer (eds.), Language, Culture, and Mind, 313–327. Stanford, CA: CSLI Publications. Roberts, Ian and Anna Roussou 2003 Syntactic Change: A Minimalist Approach to Grammaticalization. Cambridge/New York: Cambridge University Press. Svorou, Soteria 1994 The Grammar of Space. Amsterdam/ Philadelphia: Benjamins. Traugott, Elizabeth Closs 2002 From etymology to historical pragmatics. In: Donka Minkova and Robert Stockwell (eds.), Studies in the History of the English Language: A Millennial Perspective, 19–49. Berlin/New York: Mouton de Gruyter. 2003 Constructions in grammaticalization. In: Brian D. Joseph and Richard D. Janda (eds.), The Handbook of Historical Linguistics, 624–647. Malden, MA/Oxford: Blackwell. Forthc a Grammaticalization, constructions and the incremental development of language: Suggestions from the development of degree modifiers in English. In: Regine Eckardt and Gerhard Jaeger (eds.), Language Evolution: Cognitive and Cultural Factors, 219–250. Berlin/New York: Mouton de Gruyter. Forthc b Constructions, emergent constructions, and the notion of “newness”. Paper presented at HDLS, Albuquerque, NM, November 9th – 11th 2006. German translation to appear as “Grammatikalisierung und emergente Konstruktionen” in: Anatol Stefanowitsch and Kerstin Fischer (eds.), Konstruktionsgrammatik und grammatische Konstruktionen. T¨ubingen: Stauffenburg. Forthc c The concepts of constructional mismatch and type-shifting from the perspective of grammaticalization. Cognitive Linguistics, 18: 523– 557. Traugott, Elizabeth Closs and Richard B. Dasher 2002 Regularity in Semantic Change. Cambridge: Cambridge University Press. Trousdale, Graeme Forthc a Words and constructions in grammaticalization:The end of the English impersonal construction. In: Donka Minkova and Susan Fitzmaurice

The grammaticalization of NP of NP patterns

45

(eds.), Empirical and Analytical Advances in the Study of English Language Change. Berlin: Mouton de Gruyter. Forthc b Constructions and grammaticalization and lexicalization: Evidence from the history of a composite predicate construction in English. In: Graeme Trousdale and Nikolas Gisborne (eds.), Constructional Approaches to English Grammar. Berlin/New York: Mouton de Gruyter. Weinreich, Uriel, William Labov and Marvin I. Herzog 1968 Empirical foundations for a theory of language change. In: W. P. Lehmann and Yakov Malkiel (eds.), Directions for Historical Linguistics, 95–189. Austin: University of Texas Press. Wiemer, Bj¨orn and Walter Bisang 2004 What makes grammaticalization? An appraisal of its components and its fringes. In: Walter Bisang, Nikolaus Himmelmann and Bj¨orn Wiemer (eds.), What Makes Grammaticalization - A Look from its Fringes and its Components, 3–20. Berlin/New York: Mouton de Gruyter. Zwicky, Arnold 1994 Dealing out meaning: Fundamentals of syntactic constructions. Berkeley Linguistic Society 20: 611–625.

Constructions and constructs: mapping a shift between predication and attribution Mirjam Fried 1. Introduction∗ The basic premise of this study is the hypothesis that change in grammatical organization can be adequately articulated only as a gradual conventionalization of patterns of understanding, in which morphosemantic structure, syntactic function, communicative function, and lexical meaning form an integrated whole. At the same time, it has been shown for various grammatical phenomena that the gradualness of change consists in discrete partial changes that involve specific features or aspects of a larger pattern before they affect the full pattern completely (Timberlake 1977; Andersen 1987, 2001; Traugott 2003; Harris 2003). Reconciling these two characterizations of linguistic change – the holistic hypothesis with the internal mechanics of production and uptake that eventually result in the new conventionalization – presupposes a model of language in which individual linguistic patterns can be treated as complex signs that are formed by clusters of various properties (formal, functional, semantic) which either individually or collectively participate in a given diachronic process. A constructional approach to language seems like a natural candidate for providing such a model. Indeed, the relevance of a construction-based analysis has been argued for and in recent years increasingly accepted as crucial in explaining various diachronic processes in syntax. It has been found particularly useful in that strand of grammaticalization studies which link the shifts in grammatical structure to the communicative and interactional principles that govern language use (e.g. Bybee, Perkins and Pagliuca 1994, Bisang 1998, Hopper 1998, Traugott 2003 and this volume, Harris 2003, Wiemer and Bisang 2004). However, in diachronic studies, the notion ‘construction’ is usually invoked as the generally understood grammatical environment that delimits the domain of a specific morphosyntactic change; constructions in this sense thus mean nothing more than the traditional notion of ‘syntagmatic strings’ and as such do not carry much explanatory potential. In Construction Grammar, on the other hand, constructions are accorded theoretical status of basic analytic objects, which endows them with the ability to capture systematic associations between form and meaning and to express

48

Mirjam Fried

generalizations about grammatical patterning, while also leaving room for the necessary detail in capturing the internal structure of linguistic signs. However, it remains an open question to what extent and in what way Construction Grammar, which has been designed for synchronic purposes and, to my knowledge, has not been used on diachronic data, can help us be more precise about articulating the emergence of grammatical structure, instead of just comparing discrete synchronic stages. A close examination of the gradience of grammatical change is the main theoretical concern of the present work. It is intended as a test of the conceptual and representational potential of Construction Grammar in articulating diachronic relationships across constructions, thereby capturing the essence of grammatical change with its gradient nature, layering, and the richness of detail: syntactic, semantic, morphological, pragmatic.

1.1.

From fluid categoriality to a conventionalized syntactic function

The Slavic and Baltic languages (with some parallels in German) are known for the emergence of the so-called ‘long’ participles and I will focus on one of them: the long present active participle as it was used in Old Czech (OCz). For reasons that will become clear in the analysis, I will call this form a ‘participial adjective’ (PA). The PAs blur, by their very nature, the grammar/lexicon distinction, as they straddle the boundary between inflection and derivation. They also raise the question of categorial and functional status, which will be the analytic focus of this study. The examples in (1) are a preliminary illustration of the OCz PA’s functional range: as modifiers (1a), as predicates heading non-finite adverbial clauses (1b), and as actor nouns (1c). The PAs are boldfaced and in (1), their English equivalents are underlined; when minimal surrounding context is needed, it will be enclosed in curly brackets{}.1 Because of space limitations, I will discuss only the adnominal uses shown in (1a-b); some of the issues concerning OCz PAs as actor nouns have been addressed elsewhere (Fried 2005, Forthc). (1)

a. a za smrtelneho muˇze neumierajicieho krale zyˇscˇ eˇs and for mortal.adj.acc man.acc neg.die.pa.acc king.acc find.pres.2sg2 ‘and instead of a mortal husband, you will find an immortal king’ [spiritual poetry; mid 1300s; LegKat 59a] b. kdyˇz opˇet s kerchova jdieˇse, uzˇrel opˇet when again from graveyeard.gen go.impf.3sg see.pst.sg.m again d’´abla s seb´u chodiecieho devil.acc.sg with self.ins walk.pa.acc.sg ‘as he was again leaving the graveyard, he saw the devil again walk along with him’ [moralist narrative; late 1300s/early 1400s; PovOl 250a]

Constructions and constructs: mapping a shift between predication and attribution

49

c. na vuoli vˇerˇ´ıcieho jest {diel od jednoho spolurukojm´ı vzieti} on will.loc.sg.f believe.pa.gen.sg aux.3sg ‘it is [left] to the discretion of the/a creditor {to take a down payment from one of the guarantors} [manual for legal writing; late 1400s; ProkArs 168]

While the attributive, adjective-like usage (1a) in a range of meanings is the most frequent in my corpus, we will see that the PA’s function can often be determined only from the specific context in which the form occurs, and sometimes even the context leaves us without an unambiguous analysis. This functional indeterminacy raises the question of the relationship between the PA’s morphosemantic structure and its contextual distribution, both of which are crucial to our understanding of how and why the PA developed the functional range it did in OCz. My goal is thus two-fold: (i) to examine how the known functional shifts between predication and attribution were actualized in OCz (i.e., what partial transitions can be identified and what factors played a crucial role in them) and (ii) to motivate the fact that those transitions never amounted to a full categorial change, leaving the Czech PA as a truly transitional category. I will argue that the resistance has to do with resolving the conflict between the PA’s morphosemantic structure and the syntagmatic context in which it was used. Put in constructional terms, at issue will be the interaction between the internal properties of a morpho logical construction (a complex word-form) and the syntactic constructions it occurred in. This will allow me to isolate “cluster-points” (Hopper and Traugott 20032 : 6) that fix the PA in a particular function, which may go against its unambiguously adjectival inflection. I will show that the PA on its own did not provide enough clues as to its grammatical status. Instead, it was the syntagmatic and pragmatic context that shaped the PA’s functional and categorial status. The analysis is based on the assumption that the relevant functions are best understood as functional prototypes, in the sense of Croft’s (2001: 87) classification in terms of relationality, transitoriness, gradability, and stativity. Thus predication is defined as a relational, transitory, ungradable process (prototypically expressed by verbs), while modification is a relational, permanent, and gradable state (prototypically expressed by adjectives). This is also consistent with Hopper and Thompson’s (1984) prototype-based understanding of transitivity as a gradient notion, which will prove useful to the present analysis as well. The material comes from an extensive corpus of authentic data excerpted manually from OCz texts that offer a representative sample of genres (historical, biblical, administrative, expository, and didactic texts, legal documents, spiritual and secular poetry, popular entertainment, correspondence, drama, instruction manuals, etc.) and provenance (original compositions, translations, or loose adaptations of foreign material). The corpus covers the full OCz period, from the

50

Mirjam Fried

first PA attestations well before 1300 until the early 1500s. Many manuscripts can be dated quite precisely, but many others can only be placed within an estimated time period (a decade or more); I include this information with each example, using the dating practice established by the Old Czech Dictionary (Staroˇcesk´y slovn´ık 1968). I excerpted about 74 different texts of various lengths in their entirety (ranging from poems of several lines to texts consisting of hundreds of folia), which has yielded more than 55% of the PA tokens in the corpus. The remaining 45% come from about 120 additional texts (about one quarter of them ´ Cˇ in Prague. The corpus biblical) and were collected from the OCz archive at UJ contains more than 1200 tokens of PAs, which represent over 240 different roots.

1.2.

Construction Grammar and diachronic processes

The approach tested in this study reflects a particular variant of constructional analysis, the one most closely associated with Fillmore’s original conception of Construction Grammar (e.g. Fillmore 1988, 1989; Fillmore, Kay and O’Connor ¨ 1988; Fried and Ostman 2004; Lambrecht 2004) and further enriched and expanded by the insights of Croft’t (2001) Radical Construction Grammar.3 One of the defining features of Construction Grammar (CxG) is its assumption that grammar consists of networks of partially overlapping patterns organized around shared features (formal, semantic, pragmatic, prosodic, etc.); properties of such networks have been explored in dealing with various synchronic issues of constructional representations, either from a typological perspective or in specific languages. In this study, I explore the possibility of enlisting this network-based view of grammar in accounting for layering effects in grammatical change. I will analyze syntactic, semantic, and communicative factors involved in the development of a specific morphological construction (the categorially undertermined PA) and its relationship to an independently existing syntactic template (Modification construction). The theoretical focus thus will be two-fold: (i) on the inner workings of the diachronic process within a particular construction and (ii) on the status of emerging constructional patterns in grammatical change. With respect to CxG as a way of capturing the details of grammatical change, the investigation will revolve around several specific issues relevant to the theme of this volume: – factors in resolving the conflict between maintaining a transparent internal structure of a linguistic form and developing new functional associations that give rise to unpredictable form-function pairings; – the role of constructions and constructs in grammatical change;

Constructions and constructs: mapping a shift between predication and attribution

51

– the clustering of features that appear to be instrumental in the partial transitions a given change consists of; – finally, touching on the issue of representing the structure of grammar, I will propose a functional/constructional map, rather than a rigid inheritance hierarchy, as a plausible picture of grammatical change. All of this together should lead to a clearer understanding of what it means that constructions are the locus of change, as it has been invoked in the grammaticalization literature.

1.3.

Constructions as multidimensional grammatical objects

Constructions in CxG are cognitive objects that represent generalizations about speakers’ linguistic knowledge. By definition, they allow for both the holistic view of linguistic patterning (unlike formal theories of language) and for keeping track of the internal properties of larger patterns (like any other grammatical theory). CxG thus makes a systematic distinction between what conventionally identifies a construction as a whole vs. what is characteristic of its constituents; the former is referred to as the external properties (a set of constraints on how a given expression fits in larger grammatical patterns), while the latter represents the internal make-up of a construction. This distinction is crucial in that it gives a theoretical status to the observation that a construction is not just the sum of its parts but may have its own idiosyncratic properties, unpredictable from the properties of its constituents.4 The effects of an external/internal mismatch are also at the heart of the analysis developed in this study. The external/internal contrast is directly related to an issue that often causes misunderstanding about what properly constitutes a construction: it is the question of whether constructions have ‘meaning’ and if so, is it by definition noncompositional. The answer can be easily gleaned from one of the first definitions in published sources, which explicates constructions as objects of syntactic representation that “are assigned one or more conventional functions. . . together with whatever is conventionalized about its contribution to the meaning or the use of structures containing it” (Fillmore 1988: 36). A similar understanding is then echoed in Croft’s (2001: 18) formulation that constructions are “pairings of form and meaning that are at least partially arbitrary”. None of this implies that constructions necessarily have a meaning in the sense of specific semantic content. Some do, to be sure, as is also addressed by Petr´e and Cuyckens, this volume. However, describing syntactic patterns such as the Modification construction discussed in this paper clearly does not involve meaning in that same sense. The relevant question thus is this: does a particular string of words,

52

Mirjam Fried

or morphemes, reveal a construction in the technical, theoretical sense if the meaning of the string is actually a sum of the meanings of its parts? The existing definitions do provide an answer that is sufficient for our present purposes: noncompositionality in this narrowly semantic sense is not a necessary condition for constructional status. Constructions also constitute an integral part of Frame Semantics, which forms the semantic component of CxG. Linguistically relevant semantic information is organized and structured in “interpretive” frames (Fillmore 1982), which represent the complete background scene associated with a given linguistic expression: the scene’s participants, settings, and any other unique semantic features that are necessary for speakers’ native understanding of what the lexical item means and how it can be used in context. Frames also contain information about the conventional expression of the event participants as they manifest themselves in the syntactic organization of sentences. Finally, it is important to stress that CxG makes a distinction between constructions and constructs, the former being abstract generalizations over the latter. Constructions are pieces of grammar, while constructs are actual physical realizations of constructions, i.e. utterance-tokens that instantiate constructions in discourse. This distinction is also crucial to tracing diachronic changes: the end result of a series of actualizations may be a new construction or a reorganization of an existing one, but the changes themselves necessarily originate in language use, which is to say, in constructs. The paper is organized as follows. After introducing the PA form in section 2 and a brief review of existing approaches to hybrid morphology, section 3 focuses on the PA’s verbal potential, as it is encoded in its morphosemantic structure. Section 4 analyzes the shift toward attributiveness in particular contexts and identifies the features that were instrumental in supporting the shift toward a generalized modification function. In section 5 I elaborate on the constructional representation of the diachronic processes that are discussed in section 4. Section 6 summarizes the potential of Construction Grammar for modeling the incremental nature of grammatical change.

2. Participial adjectives in old Czech The PA contrasts with, and is derived from, the so-called ‘short’ form, a true participle (here labeled part), illustrated in (2a). The label PA reflects the categorial mismatch between its external and internal morphology: a morphologically adjectival case/number/gender suffix is attached to a verbal (NT -participial) stem. This is illustrated in (2b) with an example based on the root chod- ‘walk’;

Constructions and constructs: mapping a shift between predication and attribution

53

the -NT - label is the traditional way of classifying this participle, based on the morphology of its Latin cognate. (2)

a. [[Vroot – Pres. stem ] – NT ]part ‘[while/when/if. . . ] V-ing’ b. [[[Vroot – Pres. stem ] – NT ]part – c/n/g ]PA ‘(the one) V-ing’ ]PA ‘(the one) walking’ [[[chod – ie ] – c ]part – ´ı

Roughly speaking, the OCz PA resembles present active participles in other conservative I-E languages, but there are also some important differences.5 In particular, we have to keep in mind the contrast between the PA (2b) and the true participle (2a). The short form has always been used only as a non-finite predicate, typically expressing a circumstance of the main event, in (3) illustrated by the concessive meaning. (3)

p´ani ber´uc lichvu nechtie sl´uti master.nom.pl.m take.part usury.acc.sg.f neg.want.pres.3pl be.called.inf lichevn´ıci usurer.nom.pl.m ‘[the] masters, while they practice usury, do not want to be known as usurers’ ˇ ıtMuz 91 (Gebauer 1958)] [expository religious prose; 1450; St´

It follows from the predicative function of the short form that it could be substituted for the PA in the clause-like usage in (1b), but not in (1a) and (1c); indeed, PA/short form alternations are sometimes found in different manuscripts of the same text, always in the predicative function. But this relationship is not necessarily symmetrical, as will become clear in the analysis; for now we note that replacing the short form ber´uc in (3) with a PA would result in losing the concessive meaning. The genesis of the PA is straightforward: it arose in Common Slavic from the fusion of the present active (-NT -)participle and a postposed pronoun, following a general pattern of forming ‘long’ adjectives (mlad-´y ‘the young one’) out of ‘short’ ones (ml´ad ‘young’). What exactly the pronoun marked is a matter of some dispute, but for the purposes of determining the PA’s function in OCz, Kurz’s (1958) well-argued analysis of the adjectives seems to provide the best starting point. On the basis of word order patterns vis-`a-vis information structure of adjective-modified NPs, Kurz concludes that the postposed pronoun was a demonstrative and its original function had to do with expressing contrastiveness: the long form was used to draw attention to the meaning of the adjective in contrast to some other adjective (previously mentioned or presupposed) associated with a given noun. This pragmatic function was gradually lost, as the pronoun grammaticalized into an adjectival cng suffix. The PA is formed by that same suffix, giving rise to the opposition shown in (2), but the PA’s development

54

Mirjam Fried

is considerably less direct and conclusive than with true adjectives, due to its mixed-category nature. Like all participles, the PA involves a tension between verbal and adjectival properties. It is part of the inflectional verbal paradigm by various criteria, such as productivity, generality, and compositional meaning (cf. Bybee 1985, Haspelmath 1996), although it is not a typical inflectional form, since it is defective in expressing certain verbal categories. Existing analyses of participial forms tend to concentrate on the loss of verbal properties and treat the categorial conflict and its resolution in purely morphological terms, as a shift from a verbal stem to an adjective, both synchronically and diachronically; this view is common particularly in the Slavic tradition, e.g. Lamprecht et al. 1986, Gebauer 1958. In a syntax-centered explanation, Haspelmath (1996) correlates the morphologically marked loss of verbal status with the (potential) loss of transparent internal syntax, while accepting the traditional view that the participles are simply adjectives. Following Tesni`ere (1959), Haspelmath suggests for these forms a two-layer representation, each layer having consequences for a different set of the word-form’s syntactic properties. The category of the lexeme is relevant for the form’s “internal syntax” (i.e., the extent to which the form’s complementation structure resembles finite verbs) and the category of the word-form determines the syntactic function of the form as a whole (the “external syntax” of adjectives). Thus the example in (2b) would be represented as in (4) in terms of its two-tiered categorial status. (4)

[ [ chodiec]V (“lexeme part of speech”) – ´ı

]A (“word-form part of speech”)

It has been observed, though, that present active participles “do not so easily become attributes”, especially when the meaning of their verbal root is close to expressing an “instantaneous action” (Hopper and Thompson 1984: 729), or what Bolinger (1967: 9) calls “fleeting”, “temporary states”. Hopper and Thompson thus offer a wider list of criteria for such categorial transitions: in addition to low transitivity, which is compatible with Haspelmath’s loss of internal syntax, they argue that significant contributors to the loss of verbal status are the meaning of the verb root (stative, non-punctual, atelic) and the back grounding function of the form as a whole. The Czech PA is also known to resist a full categorial shift, even more so in the modern language than in OCz. Moreover, we must also keep in mind the striking functional indeterminacy noted in 1.1: the OCz PA’s three-way ambiguity invites three possible interpretations (reference, modification, or predication). This means that in determining the PA’s external function, we cannot go on the assumption that underlies the traditionally accepted morphology-based

Constructions and constructs: mapping a shift between predication and attribution

55

analysis, namely, that an a priori given lexical category determines external syntax. Instead, the criteria for evaluating the functional range of the OCz PAs can be arranged into two poles that characterize the two functional domains, predicative and attributive, in their prototypical form; a preliminary summary is given in Diagram 1 (the significance of the italicized items will become clear in section 5). These properties will prove crucial in tracing the relative loss of the PA’s predicative potential and the strengthening of its attributive function. Prototypical predicative Syntax

Semantics

Prototypical attributive

- non-subject complements - verbal government - active voice - Subj

- voice neutralized -

- Vs of action/process - tense (contemporaneousness) - animate subject

- any verb - atemporal - any NP

Diagram 1. Prototypical predicative and attributive PAs

3. Verbal potential of participial adjectives in an adnominal position The stem marks explicitly several verbal categories: tense (through the presenttense stem, in a paradigmatic contrast to a past-tense stem), aspect (inherent in the root or marked in an aspectual stem), voice (the -NT -suffix), and verbal valence, contributed by the root. All of this represents the verbal potential of the PA, predisposing it, at least in principle, toward uses expressing predication, comparable to the short participle in (1a). Such usage is shown in (1b) and (6); the PA’s subject is always a constituent of the main clause and the PA predicates something about that constituent, agreeing with it in case, number, and gender. For easier orientation, the PA with its non-subject arguments is enclosed in brackets and the PA’s subject is underlined: (6)

a. uslyˇsel zˇ a´ cˇ ka < dˇre´ veˇreˇcen´y verˇs zpievaj´ıcieho > he.heard youth.acc.sg.m aforementioned song.acc.sg.m sing.pa.acc.sg.m ‘{and when he again secretly entered the church on Friday,} he heard a youth sing that aforementioned song’ [popular entertainment; late 1300s/early 1400s; PovOl 255]

56

Mirjam Fried b. strach cˇ lovˇeka < pˇred v´as papeˇze pˇredstupuj´ıcieho > fear.nom man.acc.sg.m before 2pl.acc pope.acc come.up.to.pa.acc.sg.m naplˇnuje fill.pres.3sg ‘fear fills a person when coming face-to-face with you, the Pope’ [diplomatic message; 1462; KorPosA 93a]

These examples display prototypical properties that preserve the PA’s participial origin (the predicative pole in Diagram 1). In terms of syntax, the PA is accompanied by its non-subject arguments, which show verbal, not nominal, government, and the voice is active. Semantically, the PA is typically based on verbs of action, which is presented as contemporaneous with the main event, and the PA’s subject is animate. In this usage, the PA can be described as morphosemantically transparent, with a fully compositional meaning that can be glossed as ‘[who] Vs at the time of the main event’; each of the morphemes that make up the PA contributes exactly the meaning we would expect. This usage is most frequent in the earliest examples, but it shows persistence throughout the OCz period, is by no means marginal, and never disappears completely. The PA is attested even in absolute constructions, such as the genitive absolute in (7), where it expresses a temporal or possibly causal circumstance of the main event. Admittedly, the absolute uses were rare, limited to translations from Latin, and relatively short-lived; nevertheless, they confirm the general observation that the PA could serve the function of a non-finite predicate. (7)

a < jeˇscˇ e jich nevˇerˇ´ıc´ıch >... vece jim and still 3pl.gen neg.believe.pa.gen.pl say.pres.3sg 3pl.dat ‘and as/because they [=disciples] still don’t believe [him=Jesus]. . . , he says to them’ [biblical; late 1300s; EvZimn L 24,41]

The constellation of the properties that result in the predicative function can be constructionally represented as in Diagram 2. Understanding the details of the diagrams in this paper requires a brief digression into the CxG formalism, here substantially simplified. The nested boxes always reflect the hierarchical structure of constituents. The properties of each constituent are expressed by clusters of attribute-value pairs. Most of the abbreviations used here are selfexplanatory, such as cat(egory), sem(antics), prag(matics), val(ence), frame. The ‘values’ can be binary or come from a list of possibilities, or they can be left unspecified, which is indicated by empty brackets []. Diagram 2 can be read as follows. It is a template for forming the PA and it is a (morphological) construction in the CxG sense by virtue of providing a ‘recipe’ for combining a stem of a certain type (NT -PART, INthe left box) with a particular suffix (the right daughter constituent), the result of which is a specific

Constructions and constructs: mapping a shift between predication and attribution

57

inflectional word-form (the outside box). Its external function, however, remains open, as indicated by keeping the exernal cat(egory) unspecified (cat []). This representation says that whatever function the form will serve in a larger pattern, it will have to be motivated by the internal (in this case, verbal) category, since that is the only categorial requirement that is explicitly part of this word-form. The only external property that must be stated directly is the fact that the root’s valence expects its agent argument to be supplied by some larger pattern in which the PA can appear; this property, shared with all non-finite verb forms, is indicated by the PA’s val(ence) requirement, at the top of the outside box. The stem is of the -NT -verbal category, marking tense as contemporaneous with the main event, and voice as active. The root brings along a frame that contains the knowledge structure associated with the verb’s lexical meaning; the -NT -stem only specifies that the verb is prototypically expected to express an action or process and must contain minimally one participant (labeled FE ‘frame element’) that will prototypically have an animate referent. The root’s valence indicates that this particular event participant plays the agent role; the notation #i []* simply says if the root brings along any other arguments (and there may be none), they are unconstrained with respect to their semantic role. However, the inherit statement at the top of the stem box says that if such additional arguments are present, they will receive the same coding as they would in finite clauses (i.e., ‘verbal government’). The representation of the cng suffix (the right box) is explicit only about the agreement features; its category is left open. The lform attribute stands for ‘lexical form’ and indicates that the construction is a word-form of a particular type that of course cannot be spelled out as part of the general template but will always have a specific value (i.e., the actual form, such as zpievaj´ıc´ı, chod´ıc´ı, etc.; the dots are to be read as a shorthand for this fact). There are a few things to note about this construction. (i) The PA’s predicative use is no different from the way the short participle functions; the stem is the short participle. The only difference between the two is that the PA adds the nominal agreement categories, which the OCz short form provided only in minimal and inconsistent vestiges in the nominative case. (ii) The external category cannot be determined without any context, as the suffix itself is categorially underspecified. (iii) The pragmatic contribution of the PA suffix and hence the whole form is also unclear at this stage. Synchronically, it was still showing its original contrastive function, but only marginally so (Kurz 1958). For example, neither of the examples in (6) can be construed as contrastive: (6a) describes an event that repeated itself for several days in a row (the protagonist hearing a mysterious song being sung each day) and in (6b) the noun that instantiates the PA subject is mentioned for the first time. Hence the impossibility of interpreting the PAs

58

Mirjam Fried cat [ ] val {#1 [Agt]}

lform [...]

prag [contrastive +]

cat vNT-part. syn sem

[voice

inherit Verbal Linking active]

[ tense contemp.] [ frame action/process] FE #1 [anim +]

cat [ ] morph.

lform [...] adj

case [] number [] gender []

val {#1 [Agt], #i [ ]* }

Diagram 2. Constructional representation of the categorially undetermined PA.

as expressing any kind of contrast with a previously mentioned property of those subjects. The fading salience of the contrastive function is indicated by the gray color in Diagram 2, as a typographical approximation of the general observation that old functions of grammatical items often linger on, and as an attempt to capture this diachronic relationship between old patterns and newly emerging ones. Finally, (iv), the PA still needs a constituent that satisfies the agent requirement of its root.

4. Erosion of PA’s verbal status Not all attestations are as clear as what we have in (1, 6, 7). The following set shows cases in which the interpretation of the PA and its syntactic function is much less clear: (8)

a. {mˇejte mysl k bohu, v dobrotˇe... hledajte jeho,} nebo v duˇsi < zˇ a´ dajuc´ ´ ı zl´eho > for into soul.acc.sg.f desire.pa.acc.sg.f evil.gen.sg.n {nevende duch milosti} (i) ‘{turn your mind toward God, seek him through good life. . . , for the spirit of mercy will not enter} into the souli if/when iti desires evil things’ (ii) ‘. . . into an evil-minded soul’ (lit. ‘habitually desirous of evil things’) [homily, end of 14th cent.; MatHom 42a] b. nerodil si pˇrestati a k < tepuciemu ´ > neg.want.pst.sg.m aux.2sg stop.inf and toward hit.pa.dat.sg.m s´udci sˇe obr´atiti {ale vetch´ym hˇriechom. . . mnoˇz´ısˇ sˇkody} judge.dat.sg.m refl turn.inf {. . . }

Constructions and constructs: mapping a shift between predication and attribution

59

‘you didn’t want to stop {committing sins} and turn toward a judge for punishment {and instead you keep piling up damage through more sins} (ii) ‘you didn’t want to . . . turn toward a punishment-giving judge . . . [social satire; early 1400s; Budyˇs 9a]

(i)

In (8a), it is entirely plausible to understand the PA as expressing an eventuality that holds ‘here and now’ or presents a condition that applies in a particular instance: this would yield the interpretation in (i), along the lines of ‘in the moment of committing a bad deed you are not in God’s good graces‘. But given the broader context of this utterance (a homily) it is equally possible to understand the PA as a general admonition to be good (ii), not just in an idividual intance but always. A similar uncertainty arises in (8b). It is not readily apparent whether the speaker presents the judge as someone whose customary job it is to mete out punishment, which would be the interpretation in (ii), or as someone who will punish the protagonist on this particular occasion (i). The latter is the one favored by the general context, but the former is not out of the question. There are various reasons for these ambiguities, having to do with clusters of conflicting features associated with the PA in a given context. I will discuss those conflicts in 4.2.2. A clearer departure from the predicative usage toward an atemporal (or at least habitual) interpretation are presented in (1b) or (9a–c). (9)

a. poˇceli obˇetovati kaˇzd´y zlat´y peniez start.pst.pl offer.inf everyone.nom.sg gold.adj.acc.sg.m coin.acc.sg.m < maj´ıc´ı na sobˇe obraz anjelsk´y > have.pa.acc.sg on self.loc picture.acc angelic.acc ‘everyone started offering a gold coin, which had on it a picture of an angel’ [moralist narrative; late 1300s/early 1400s; PovOl 276b] b. pro spletenie zˇilek a svazk´ov < svazuj´ıc´ıch for network.acc veins.gen.pl and ligaments.gen.pl bind.pa.gen.pl to miesto > that spot.acc ‘because of the network of veins and ligaments holding that place together’ [medical text; early 1400s; L´ekSalM 505] c. {voly zaj´ımal u Helfenburka... i hnali na Vitmberg} a prodali Matlovi < tu sed´ıciemu > and sell.pst.3pl Matl.dat.sg.m here sit.pa.dat.sg.m ‘{he caught the oxen at Helfenburg. . . and then they headed for Vitmberg} and sold [them] to Matl, who was residing there’ [executioner’s records; 1429; PoprRoˇzmb 46b]

Such cases often have non-subject complements, just like (6), and they maintain the active orientation of the -NT -stem. Yet, in contrast to (6), these PAs clearly

60

Mirjam Fried

express states of affairs that hold independently of the main event: the presence of the picture of an angel on the coin in (9a) is a permanent feature of the coin, not delimited by the event of offering it, and similarly for the function of the ligaments in (9b) or the presence of one Matl at the Vitmberg homestead in (9c). For now we note that the departure from the pattern represented in Diagram 2 seems to involve two semantic features: an inanimate PA subject or a stative verb. These features seem to correlate with the modifier-like interpretation of the PA, but it also must be pointed out that the PAs in (9) still express a kind of background information related to their subjects, in a participle-like fashion (cf. Fox 1983, Thompson 1983; for Czech specifically Lamprecht et al. 1986: 367, Gebauer 1958: 625, Hrabˇe 1957: 385), rather than serving to restrict the class of subject referents, as truly attributively used modifiers would. The PA is semantically in a coordination relation to the main clause, rather than an adjective-like modifier of the PA subject. For example, (9a) says that everybody was offering a coin and the coin possessed certain properties; the context in which this sentence is used does not allow the interpretation that only coins with an angel on it were offered, in contrast to other kinds of coins. Similarly the description of the veins and ligaments in (9b): the relevant passage describes the difficulties of getting to a spot in a joint because it is surrounded by a thick network of veins and ligaments. Finally, the proper noun in (9c) makes it clear that the PA only adds some background detail about its subject, reminiscent of non-restrictive relative clauses. The shift toward functioning as a modifier of sorts is thus evident; at this stage it is perhaps best understood as a predicative modifier. The question is what features specifically contributed to the shift and how exactly the morphological construction re-organized itself as a result. Let us start with the PA-internal properties: transitivity, internal syntax, verb meaning, tense, and aspect.

4.1.

Internal, verb-related factors

4.1.1. Syntactic criteria: complementation and transitivity Examples such as (9) show that the presence of non-subject arguments does not preclude a modification function. It is true that the PA’s diminished verbal status often correlates with low informativeness of its complements (Fried 2005, 2007), as would be expected, but indefiniteness or pragmatic predictability per se is not a precondition for such a reading to arise. For example, the picture of an angel in (9a) is new information, albeit part of a backgrounded sub-event. There are also cases in which complementation is not an issue to begin with since the PA is formed from an intransitive verb, and by the syntactic argument, such

Constructions and constructs: mapping a shift between predication and attribution

61

examples should be prime candidates for losing their verbal character. Yet, it is not difficult to find cases of intransitive PAs that express a predication dependent on the main event, such as we see in (10). (10) a. skˇrek cˇ lovˇeka < volaj´ıc´ıho > m´a b´yti slyˇseˇ n shriek.nom man.gen.sg call.pa.gen.sg have.pres.3sg be.inf hear.pass.sg ‘the shriek of a man, when he’s calling out, ought to be heard’ [allegorical dispute; early 1400s; TkadlS 2a] b. {kaˇzd´emu sv´u ruku na jich prsy vzkl´adaje...} Kter´ezˇ on vˇsecky < sp´ıc´ı > naˇsel which.acc.pl 3sg.m.nom all.acc.pl sleep.pa.acc.pl find.pst.sg.m ‘{he [=the king] felt their chests with his hand. . . } and he found them all asleep {except for the one who’d just come from his tryst with the queen} [popular entertainment; late 1400s; HynRozpr 140a]

In fact, it is interesting to compare the frequency distribution of these syntactic criteria – complementation and transitivity. Admittedly, it is somewhat tricky to use and interpret quantitative information when working with an incompletely attested language and hence a potentially skewed corpus. Nevertheless, if we apply the necessary caveats about drawing any absolute conclusions from the numbers, they certainly provide discernable patterns. Table 1 summarizes the relative frequencies of transitive vs. intransitive roots among the adnominal PAs in my corpus, distributed over three functional possibilities: predication, modification, and the cases of functional ambiguity, such as we saw in (8). Table 2 provides the token counts of PAs with non-subject complements (left columns) out of the total count (numbers in parentheses) across the different functions, again relative to transitivity. Table 1 brings out the point that intransitivity is not a strong predictor of the PA’s diminished verbal status (contrary to Thompson 1983): intransitive roots are not only more common across the board, but the transitive ones also appear to be less frequent in the predicative function than in modification. Table 2 confirms the expectation that the predicative uses should be more likely than the modification function to preserve the internal syntax of the stem, but given that overall only about half of the predicative attestations contain an internal complement and that the presence of those complements also plays a major role Table 1. Relative frequency of transitive and intransitive roots. Predication

Modification

Ambiguous cases

Intransitive roots Transitive roots

50 20

71% 29%

45 38

54% 46%

27 11

71% 29%

Total

70

100%

83

100%

38

100%

62

Mirjam Fried

Table 2. Relative frequency of PAs expressing their non-subject complements. Tokens with compl.:

Predication

Modification

Ambiguous cases

Intransitive Transitive

39 21

43% (88) 90% (28)

23 23

16% (144) 29% (78)

29 66% (44) 7 54% (13)

Total

60

53% (116)

46

21% (220)

36 63% (57)

in creating functionally ambiguous expressions, complementation per se is not a strong predictor of anything either. The most reliable conclusion we can draw from Table 2 is the fact that transitive predicates with complements are least likely to present the hearer with a functional ambiguity between predication and modification; such examples indeed occur only rarely. 4.1.2. Semantic criteria: temporal grounding and verb meaning We can again follow Hopper and Thompson’s semantic criteria for identifying the potential factors in fixing the PA’s function toward attribution: the non-active verb meaning, aspect, and temporal grounding. For Czech, we can essentially discount the aspectual dimension as relevant. While unbounded and durative event structures generally correlate with diminished verbal status in the sense of not reporting actions with specific conceptual boundaries (cf. also Hopper and Thompson 1983: 57, 61 or Thompson and Hopper 2001: 35), this potential cannot be used as an explanation for the PA’s functional development since the -NT -stem is inherently compatible only with imperfective verbs, in contrast to the passive participles (‘short’ or ‘long’), which are primarily derived from perfective stems. Granted, the imperfective aspect is naturally compatible with developing a habitual reading, going from ‘X is V-ing’ to ‘X has the general habit/property of V-ing’, but this can obviously be only one feature among several that jointly invite the attributive reading, since we have seen in (1b, 6) that the reinterpretation does not obliterate the ‘X is V-ing’ meaning altogether. It is for this same reason that we cannot assign the source of the change to the tense category either. The complete absence of temporal grounding (i.e., the “fleeting”, “event-reporting” meaning in Hopper and Thompson’s 1984 terms), such as in (1a, 9), is the consequence of establishing a habitual interpretation in specific contexts, but could not have originated spontaneously in the form itself because then we would expect the same effect in all instances of PA use. This leaves us with the lexical meaning of the root as a potential motivating factor. While I will show that its effect on the PA development is limited in specific ways, it at least provides an explanation for the futility of using transitivity

Constructions and constructs: mapping a shift between predication and attribution

63

as a criterion. One of the reasons that the PAs display such a high incidence of ‘intransitive’ roots in the corpus and that such a high percentage of these intransitives is accompanied by non-subject complements is due to the fact that lots of the PAs are based on verbs of motion or location, e.g. pˇredstupuj´ıc´ı ‘coming before sb.’ (6b); sed´ıc´ı ‘sitting/residing’ (9c). Those verbs often either require or at least prefer to specify a spatial relation or a manner of motion, which makes them syntactically elaborate, although not transitive in the usual sense, whether semantically or syntactically. There are also non-motion verbs with similar properties (plaˇc´ıc´ım na ‘crying over sb.ACC ’; cˇ ekaj´ıc´ı na ‘waiting for st.ACC ’, boj´ıc´ı se ‘fearingreflexive of st.GEN ’, sluˇsej´ıc´ı k ‘belonging to st.DAT ’, etc.). The distribution in Tables 1 and 2 is based on the usual understanding of transitivity (semantically transitive event, formally encoded as nominative-accusative or, in a few cases, nominative-dative). Somewhat more important than transitivity is the semantic distinction active/stative. We see this in Table 3, which shows the overall distribution of the two semantic classes of roots. Table 3. Distribution of active and non-active roots. Predication

Modification

Ambiguous cases

Active roots Stative (&Psych) roots

46 24

66% 34%

52 31

63% 37%

19 19

50% 50%

Total

70

100%

83

100%

38

100%

The likelihood of finding a stative predicate in the modification function is slightly higher than in the predicative function, as expected. However, the difference is not dramatic and when we consider the proportion of active vs. non-active roots across the functional categories, the difference is even less significant. Moreover, the meaning of the root seems to have no effect on creating a functionally ambiguous usage. Yet, it is also clear from all the raw numbers that the modification usage is the most frequent in the corpus. We thus must look for an explanation of this fact outside the form itself, in the syntagmatic and semantic context in which the PA occurs.

4.2.

Factors external to the PA form

4.2.1. Subject animacy One striking feature is the semantics of the PA’s subject. We noted earlier that the compositional meaning specifies an animate agent, exemplified in (6). This is

64

Mirjam Fried

consistent with the prototypical meaning of the PA’s participial stem: actions are typically carried out by animate entities. Notice, however, that in the examples which functionally depart from the participial template, as in (9), some of the nouns denote inanimate entities (peniez ‘coin’, svazky ‘ligaments’), i.e., not very good agents. In fact, inanimate subjects dominate in the modification function throughout the corpus, as summarized in Table 4. Not only is there a clear difference in animacy between the two functions, but inanimate NPs are also a significant contributing factor in the sentences that are functionally unclear. Table 4. Animacy of PA subjects. Predication Animate subjects Inanimate subjects Total

Modification

Ambiguous cases

72 44

62% 38%

73 147

33% 67%

24 32

43% 57%

116

100%

220

100%

56

100%

Relaxing the animacy restriction on the PA’s subject is semantically and pragmatically consistent with noun modification: properties can be attributed to any referent, regardless of animacy, degree of specificity or individuation, or any other semantic feature. As a result, even PAs of action, such as svazuj´ıc´ı ‘binding’ in (9b), can be interpreted as stative, i.e., as expressing a durative property rather than an action, if the NP referent is inanimate. 4.2.2. Word order The PA occurred predominantly in three different linear patterns that are summarized in (11); in the patterns A (11a) and B (11b) the PA follows its NP, while in C (11c) it precedes. (The brackets enclose the PA with its non-subject complements, labeled ‘xp’.) (11) A. NPPA-subject B. NPPA-subject

C. NPPA-subject

All three orders occur throughout the OCz period, but they are not in free variation. Order A necessarily applies to PAs that are accompanied by their nonsubject complements; this is indicated by the ‘[Kleene] +’ symbol, to be read as ‘one or more complements (xp) must be present’. This order is most commonly associated with the predicative interpretation (1b, 6), although we see in (9c) that an attributive reading is not excluded altogether under favorable semantic conditions (subject animacy and/or stative roots). Order B is the most frequent overall, and occurs both with the PA’s internal complements present, as in (8a, 9a–b) and in a bare form (10); this is indi-

Constructions and constructs: mapping a shift between predication and attribution

65

cated by the ‘[Kleene] *’ in (11), which means ‘zero or more xp’. This linear pattern cannot be easily associated with a particular syntactic function, except that relative to A, it is considerably more common in the attributive interpretations, regardless of presence or absence of internal complements; statistically speaking, the B pattern is about as likely to express an event contemporaneous with the main verb as it is to express an atemporal property of the PA’s subject. Finally, order C, which differs from A and B in the relative position of the PA and the NP, involves almost exclusively bare PAs, such as we have in (1a, 8b, 12), and occurs predominantly in the modification function, as we will see in a moment. The pattern occurs with transitive (12a) and intransitive (12b) roots alike: (12) a. zˇ a´ dajuc´ ´ ımu lidu, {jeˇsto minul´e b´ıdy a strasti pamatuje, demand.pa.dat.sg.m people.dat.sg.m {...} bezpeˇcenstvie a pokoj zdali by optala} ‘{in order for our royal mind to provide security and peace} for the anxious nation {with its memories of past hardships} [legal code, end of 14th cent.; MajCar 72] b. at’ patˇr´ıme na tv´oj kaj´ıc´ı zˇivot so.that look.pres.1pl on your.acc.sg.m repent.pa.acc.sg life.acc.sg.m {a n´asledujeme tebe} ‘so that we look at your life full of repentance {and follow your example}’ ˇ 122a] [expository religious prose, early 1400s; V´yklSal

Of interest is particularly the comparison between B and C, since in both of them the PA and its subject are immediately next to each other and the adjacency is a strong motivation for interpreting the PA as a modifier of the neighboring NP. At a minimum, the proximity creates an opportunity for perceiving the two elements as a conceptual unit that identifies the referent of the NP as being of a particular kind. But it cannot be just the adjacency that is responsible for the shift toward an atemporal reading and hence functional reinterpretation of the PA, since there are also many instances of the B order that unambiguously express a temporally grounded relation (predication). The crucial factor is the relative order, as captured in Table 5, which summarizes the interaction between linear adjacency and the presence/absence of complementation. It is clear from the modification-to-predication ratios that the pattern C correlates very strongly with interpreting the PA as a modifier. This is true regardless of complementation although the absence of complements only strengthens this functional status, unsurprisingly. The latter is also true for the B order, but otherwise B appears to be more sensitive to the presence vs. absence of complementation. (The double checkmark in Table 5 indicates very high incidence, parentheses indicate extremely sporadic incidence.)

66

Mirjam Fried

Table 5. Word order, complementation, and syntactic functions PA & Subj are adjacent with complements – modif : pred bare PA– modif : pred √ √ 8:1 order C: PA - Subj ( ) 3:1 √√ √√ 1:1 3:1 order B: Subj - PA

It thus makes sense to ask at this point if there was an established or at least preferred linear order in the language for expressing a modification relation. OCz word order was quite flexible and the flexibility extended even to the NP level, much more so than is the case in ModCz. However, the order within an NP was not free. This is argued and amply documented in Vondr´ak’s (1908) study, which compares the word order preferences of adjectives in contrast to adnominal genitives. On the basis of comparative evidence from the earliest original Czech texts of any length and late Old Church Slavonic texts, in which no Greek or Latin influence can be expected, Vondr´ak concludes that OCz overwhelmingly preferred the Mod–NP order, inherited from Common Slavic. This means that there was an abstract syntactic pattern (a construction in the CxG sense) that speakers understood as a conventional expression of a modification function, even though the pattern was evidently in some competition with a few other variants (related, but distinct constructions). One such variant was the pattern which reversed the order, NP-Mod. As argued by Kurz (1958), this pattern was originally associated with a specific pragmatic function (contrastiveness), but it is further worth noting that it became particularly common in biblical and religious texts and remained in those genres long after the pragmatic function had been completely lost and the old Mod-NP pattern became grammaticized as the only neutral order for NPs. This genre-based effect is clearly documented also with the PA, but the space of this paper does not permit further discussion of this aspect. The Modification construction is represented in Diagram 3. It consists of two syntactic daughters whose mutual relationship (the information that uniquely identifies this syntactic combination as a conventional pattern) is represented by the attribute role, with the corresponding values in each constituent. The only other information that needs to be specified as an otherwise unpredictable property is the case, number, and gender agreement and the relative order: the modifier precedes the head. Categorially, the construction is a NP, expressed as [cat n], carried over from its head. Notice, however, that the category of the modifier remains unspecified; this notation indicates that this construction licensed

Constructions and constructs: mapping a shift between predication and attribution

67

Modification cat n prag ['restrict reference of the noun (#2) by the property expressed in #1'] #1

cat [ ] role modification morphol. case #i [ ] number #j [ ] gender #k [ ]

#2 cat n role head morphol. case number gender

#i [ ] #j [ ] #k [ ]

Diagram 3. OCz Modification construction

constructs that contained all kinds of modifying words, such as demonstratives, possessive pronouns, adjectives, certain numerals, etc. How does the PA fit in, or what does it have to ‘give up’ in order to be compatible with the slot of the modifier? The construction forces an attributive interpretation, favoring the expression of durable, characteristic, de-individuated properties, independent of any specific temporal frame concerning an individual instance, and manifests itself by certain formal and semantic features: particular word order; adjacency of the constituents; attributive semantics of the left daughter; no semantic restrictions on the head noun. And indeed, these are the characteristics we can identify in different clusters across the three word orders found with the PAs, with the C pattern matching this construction most closely. We can now revisit the distribution of all the PA-related features, internal and external, formal and semantic, in relation not only to the functional status but also across the three word orders. The summary is in Table 6. The columns represent the three functional possibilities and the distribution of the orders A, B, and C within each domain. The rows summarize the relative frequency of the three criteria that seem to show the greatest potential for affecting the verbal character of the PA: the semantics of the root (active/non-active), the animacy of the PA’s subject, and the preservation of internal syntax (complementation). The roots are counted only as distinct roots, the other two criteria are based on the number of all PA tokens (excluding the absolute constructions, as word order is not an issue there). The numbers in gray indicate the actual counts on which the percentages are based; the items in bold will be commented on below. We can draw several conclusions from the patterning in Table 6. (i) The transparent, compositional interpretation of the PA (i.e., as a predicate expressing a temporally grounded situation) is clustered in the A and B orders, both also showing a high concentration of active roots and animate subjects, as expected.

68

Mirjam Fried

Table 6. Distributional patterns in different word orders Predication Modification Ambiguity A B C A B C A B active root 62% 69% 37% 39% 64% 43% 47% 58% 18 (29) 20 (29) 3 (8) 7 (18) 39 (61) 10 (23) 8 (17) 11 (19) animate SUB 62% 56% 50% 48% 31% 33% 45% 37% 20 (32) 27 (48) 6 (12) 10 (21) 36 (116) 27 (83) 9 (20) 10 (27) complements 100% 31% 17% 100% 16% 7% 100% 59% 32 (32) 15 (48) 2 (12) 21 (21) 19 (116) 6 (83) 20 (20) 16 (27)

C 86% 6 (7) 56% 5 (9) 11% 1 (9)

Note, however, that syntactic complementation is not a major criterion in the B order and that the semantic features appear to be sufficient to uphold the verbal potential of the form. In contrast, to the extent that the C order is found in a predicative function at all, it seems that the crucial criterion is the animacy of the PA’s subject, not the PA-internal features, whether semantic or syntactic. (ii) In modification, the B order also shows a very high (in fact, the highest across the board) proportion of inanimate subjects (69%). With respect to the PA-internal features, we still find a high proportion of active roots (64%) but considerably fewer complements (16%). It appears that for the B order to facilitate the modification the PA must be used bare. (iii) Finally, these distributions seem to be confirmed by the cases that are functionally ambiguous. In the B order, the ambiguity appears to arise from the high incidence of complementation (59%) and an equally high proportion of active roots; the semantics of the PA subject does not seem to be enough to resolve the conflict. In the C order, on the other hand, the ambiguity can be attributed squarely to the semantics, both internal and external: predominantly active roots and a noticeable proportion of animate subjects. Both of these features suggest a verblike usage, which puts them in direct conflict with the attributive interpretation suggested by the linear pattern itself (PA-NP). We can now comment further on the ambiguities in (8), repeated below: (13) a. v duˇsi < zˇ a´ dajuc´ ´ ı zl´eho > {nevende duch milosti} into soul.acc.sg.f desire.pa.acc.sg.f evil.gen.sg.n (i) ‘into the souli if/when iti desires evil things { the spirit of mercy will not enter}’ (ii) ‘into an evil-minded soul’ (lit. ‘habitually desirous of evil things’) b. nerodil si pˇrestati a k < tepuciemu ´ > neg.want.pst.sg.m aux.2sg stop.inf and toward hit.pa.dat.sg.m s´udci sˇe obr´atiti judge.dat.sg.m refl turn.inf (i) ‘you didn’t want to stop {committing sins} and turn toward a judge for punishment’ (ii) ‘you didn’t want to . . . turn toward a punishment-giving judge . . . ’

Constructions and constructs: mapping a shift between predication and attribution

69

The PA in (8a) is used in the B pattern, has an animate subject (duˇse ‘soul’), and is accompanied by its object, but the root zˇ a´ d- ‘desire’, though transitive, is not truly a verb of action. This combination – stative verb, subject adjacent to the PA, and the overall context – creates an opportunity for an atemporal analysis of the PA, in spite of the internal structure and subject animacy. In (8b), the conflict is between having an animate subject and an active transitive root (favoring a verblike interpretation) but no direct object and, perhaps most significantly, the PA is used in the C pattern, which strongly favors an attributive conceptualization. It had to be constructs like these that opened up the path toward the loss of the compositional structure of the PA and toward fixing its functional status as an atemporal modifier. We can thus conclude that the structurally and semantically transparent PA construction shown in Diagram 2 gradually shifted to a less compositional configuration, in which a cluster of internal and external changes yields a grammatical entity with a distinct syntactic function, namely, modification. A formal constructional representation of the latter is given in Diagram 4. The verbal features that are being ‘demoted’ are shown in gray, to indicate their diminished salience; the original pragmatic function of the PA-forming CNG suffix is lost completely; and the lexical category of the PA still is best left unspecified as there is no evidence of a complete shift to a full-fledged adjective class (for example with respect to derivational processes that target true adjectives). The adjective-like status cannot be posited as an inherent feature of the PA, despite its external morphology; the adjective-like behavior only follows from the PA’s use in a particular syntagmatic string (licenced by a syntactic modification construction) in a particular pragmatic context, one that favors describing the habits of entities, rather than their actions in specific, individuated instances. The accat [ ] sem ['prone to V-ing' ] val {#1 [Agt]} cat vNT-part. syn sem

inherit Verbal Linking

[voice active] [ tense contemp.] [ frame action/process] FE #1 [anim + ]

Habitual PA

lform [...] cat [ ] morph.

lform [...] adj

val {#1 [Agt], #i [ ]* } Diagram 4. Constructional representation of the PA as a modifier

case [] number [] gender []

70

Mirjam Fried

quired habitual, atemporal semantics is captured by the boldface sem statement at the external level: this is clearly a feature not predictable from the morphosemantic structure and must be, therefore, marked as a special property of this newly emerging PA construction.

4.3.

From modifiers to lexical adjectives

The erosion of the internal features and the external semantic requirements manifests itself to the extreme in examples such as (1a), in which we have neumieraj´ıc´ı ‘immortal’ (< lit. ‘non-dying’), or (14) below, with the PA zˇ a´ daj´uc´ı ‘desired/desireable’ (< lit. ‘desiring’): (14) < zˇ a´ dajuc´ ´ ıho > v´ıtˇezstv´ı trojzvuk sˇcastnˇe pˇrijal desire.pa.gen.sg.n victory.gen triad.acc.sg.m happily accept.pst.sg.m ‘he joyously accepted the sound of the desirable/welcome victory’ ˇ adKor 42b] [administrative, late 14th cent.; R´

Here we see a dramatic shift from the transparent morphosemantic structure. In these cases, the PA is always bare, the subject may or may not be animate but, crucially, it does not fill the role of an agent with transitive roots. Instead, the active orientation of the stem is lost, giving way to a resultative or passive reading (znaj´ıc´ı ‘full of knowledge’ < ‘knowing’; nast´avaj´ıc´ı ‘present’ < ‘up-coming’; zˇ a´ daj´ıc´ı ‘desired’ < ‘desiring’), often also shifting toward purpose meanings, found especially in specialized vocabulary – administrative, medical, legal, etc. (vˇec uzraluj´ıc´ı ‘substance for maturing’, olej posiluj´ıc´ı ‘oil for strengthening’, list napom´ınaj´ıc´ı ‘letter of reprimand’; cf. Mich´alek 1963). In the case of intransitive roots, such as neumieraj´ıc´ı ‘immortal’ (1a), the loss of the diminished active orientation manifests itself especially in the shift in modality (e.g. ‘[who] not dying’ > ‘[who] cannot die’), but the purpose meaning is also very common. These changes reflect a reconfiguration of the semantic participants contributed by the participial stem, whereby the PA becomes semantically fully dissociated from its active orientation signaled by the -NT -morphology. The NP that forms a conceptual unit with the PA is, then, just that: an NP whose referent bears no event-role relationship to the verbal meaning of the PA’s stem. Its inherent semantics is irrelevant, and the PA can only be interpreted as attributing some characteristic to its adjacent NP, as in any other modification relationship. A constructional representation of this extreme shift is in Diagram 5. The internal features contributed by the NT -STEMhave been obliterated, the only piece of information that remains is the meaning of the root (through the frame specification), which, however, is not semantically constrained beyond ensuring

Constructions and constructs: mapping a shift between predication and attribution

cat

71

lform [...] adj sem [ {result of V-ing; V-able; for the purpose of V-ing} ]

cat vNT-part. sem

frame [ ] FE #1 [ ]

cat [ ] morph.

adj

lform [...] case [] number [] gender []

Diagram 5. Representation of an adjective-like PA construction

that zero-valent roots cannot appear (i.e., only frames that contain at least one syntactically expressed participant are allowed, which excludes certain verbs of atmospheric states, physical or mental states, etc.). At the same time, the external properties (in boldface) are completely unpredictable from the structure of the stem and must be specified directly, as idiosyncratic constructional features: categorially, this is as close to an adjective as a PA can get, and the construction also has non-compositional semantics. We have seen that the meaning comes in several flavors; in this diagram they are represented in an abbreviated way as a list of possibilities in the sem statement (each of them should, strictly speaking, be presented as a distinct sub-construction, along the lines of Traugott’s, this volume, classification into meso- and micro-constructions).

5. Constructional representation of grammatical reorganization The PA clearly went through a long period of shifting toward an attributive usage, in co-existence with truly predicate-like uses motivated by the morphosemantic structure (namely, that of an inflectional member of the verb paradigm). The factors that contributed to the shift toward modification have to do both with the internal properties of the PA itself and with the syntagmatic and semantic environment the PA occurred in. With respect to the PA-internal features, the partial changes concerned primarily the loosening of semantic restrictions on the root (from verbs of action to allowing also verbs of states and perception). Less prominent, but still relevant in many cases, were the loss of internal syntax, the erosion of temporal grounding, and, as an extreme shift, the emptying of the -NT -suffix of its voice-marking content. The factors external to the form itself (in Diagram 1 italicized) involve a combination of abandoning semantic

72

Mirjam Fried

restrictions on the subject NP (loss of constraints on the inherent noun semantics and neutralization of any event-role distinctions introduced by the stem) and syntagmatic patterns that put the PA and its subject in adjacent positions, thus suggesting a tighter conceptual unit. The data show, however, that whatever functional shifts took place, they cannot be attributed to any one of these factors individually; instead, we must conclude that various clusters of those factors may be equally capable of inviting a shifted interpretation of a given PA token. Thus, returning to the functional predicative-attributive continuum that was presented in Diagram 1, we can now conclude that the less prototypical cases of either function represent various points in between the two poles, depending on which features prevail in a given instance. As noted in 4.2.2, the source of innovation must have been the ambiguous constructs, which presented language users with various degrees of mismatches between the PA’s morphosemantic structure (licensed by a familiar construction, represented in Diagram 2) and the syntactic and semantic requirements of a particular modification structure (another, independently existing grammatical construction) in which the PA appeared to be used. The presence of constructs that allowed multiple interpretations of such mismatches led to ‘analyses’ that gradually adjusted the internal organization of the PA construction, giving rise to additional (types of) PA constructions. The emergence of these different constructional outcomes can be summarized in a representation that is akin to the functional/semantic maps used in cognitively oriented typological research (Haspelmath 1997, 2003, Croft 2001). A functional map representing the diachronic relationships between different PA constructions is in Diagram 6. The list of (boldfaced) properties in the center of the map corresponds to the constructional features, both internal and external, that must be referred to in representing the PA morphosemantic structure. The symbols ‘’read as ‘less common/favored’and ‘more common/favored’, respectively; the doubling, ‘’ indicates an overwhelming preference in the corresponding direction. The rectangles delimit the sets of features and preferences of individual PA constructions, here identified by their meaning (italicized boldface in single quotes). The dashed line around the rectangle labeled as cat(egory) ADJ indicates that this set of innovations was both more recent than the other two and relatively short-lived, not surviving much beyond the OCz period delimited by the early 1500s. The fact that this adjectival shift consisted of several semantic variants is indicated by the branches leading out of the general representation. The map shows that the transitions for a given feature can only be expressed as tendencies. Nevertheless, the give-and-take of the conflicting pressures is

Constructions and constructs: mapping a shift between predication and attribution

73

'prone to V-ing' bare
> animate

adjacent


thing.baked >

OBJ

OBJ2

Figure 2. Unification of the ditransitive construction with the verb bake

Constructions not only occur at argument-structure level, but also at the morphological level (cf. Croft 2001: 17; Goldberg 2006: 5): not only such complex syntactico-semantic pairings as studied in Goldberg’s work can be represented as constructions, but also morphological structures. As such, the [prefix – verb stem] structure, and more particularly the [be – verb stem] structure to be dis-

138

Peter Petr´e and Hubert Cuyckens

cussed in this paper, will be regarded as a construction to be treated on a par with argument structure constructions. In line with Goldberg (1995: 22–23), then, we would like to argue that the inseparable prefix be- can be regarded as analogous to the (skeletal) argument structure, whereby the root verb plays the role of the main verb, and that “the semantic integration of morpheme [i.e., beprefix] and verb stem is analogous to the integration of construction and verb”. The constructs resulting from this integration (or fusion) will be referred to as inseparable complex verbs (ICVs). What distinguishes these [prefix – verb stem] or prefix constructions from argument structure constructions is that the former can be situated at a level similar to Croft’s verb-class-specific constructions (cf. Croft 2003), which is a level less schematic, or more substantive, than argument structure constructions. Importantly, Goldberg’s definition of constructions, with its condition of noncompositionality or non-predictability, also applies to prefix constructions, as instantiated by ICVs. For one, the semantics of ICVs is not predictable, because adding up the meaning of the simplex/root verb (including its participant roles) and that of the prepositional cognate of the prefix (if it still exists, like OE of for of- or be ‘by’ for be-) does not yield the semantics of the composite ICV. Second, the form of the ICV is not compositional either because the prefix does not constitute an independent component: an inseparable prefix only exists by virtue of the ICV of which it forms a part. The fact that inseparable prefix constructions satisfy the defining characteristics of a construction does not imply that all of its instances belong to one type of construction only. On the contrary, it will be seen that inseparable prefixes may occur in two types of constructional patterns, each of them with different semantics and different valency frames (the concept of valency frame here refers to the pairing of arguments to syntactic functions). Even prefixes with a single phonetic form, such as be-, may belong to two types of inseparable prefix construction. The constructional patterns are those which have recently been described for prefixes in Dutch (Blom 2004), namely predicative and nonpredicative constructions. An instance of a prefix construction can be found in (3) and is schematically represented in Figure 3. (3)

Læt hine gyt þis gear, oð ic hine bedelfe & ic hine bewurpe Let him alone this year, till I him be-delve.subj and I him be-throw mid meoxe with dung ‘Let it [a fig tree] alone this year also, till I shall dig about it, and dung it’. (c1025. Lk [WSCp]: 13.8. Trans. from the King James Bible)

be-’s constructional properties in its conservation

139

In particular, the verb form bewurpe (from beweorpan ‘surround/cover by throwing’) in (2) is an instance of a non-predicative, trivalent be- construction (see also Michaelis and Ruppenhofer 2001: 61). (Other be- constructions, such as the predicative and non-predicative bivalent be- constructions, will be discussed below.) This ICV shows a number of properties which cannot be predicted from its component verb wurpe ‘throw’ (from weorpan ‘throw’). As can be seen from Figure 3, when a root verb such as wurpe is unified (integrated, fused) with the verb-headed be-construction,3 the construction may arrange the participant role array of the simplex verb in such a way that in the resulting active transitive sentence, the agent argument is realized as subject, the location argument as direct object, and the theme as an oblique object of the preposition mid ‘with’ (alternatively, the theme could also be realized in the instrumental case).4 Sem ‘around, over’ + ‘affected’

SUBJ

OBJ

OBLINST/PP(mid)

Figure 3. Unification of the trivalent be- construction with the verb wurpe

Note that in this particular example the target participant associated with the verb weorpan ‘throw’ is between square brackets. This means that, despite the fact that it is lexically profiled (cf. Goldberg 1995: 44), it can be omitted if recoverable from context. In the prefixed form beweorpan ‘surround/cover by throwing’, however, wurpe’s target participant cannot be omitted any longer, since the location argument in the be-construction lines up with the direct grammatical function of OBJ. By contrast, the thing.thrown role, though profiled as a participant role of the verb, is represented by an unprofiled argument role in the be-construction and realized as an oblique syntactic argument (cf. Goldberg 2006: 40). Importantly, the location argument syntactically realized as OBJ in the prefix construction carries with it certain characteristics typical of patienthood not associated with the corresponding target participant role of the simplex. More specifically, the fig tree in (3) is affected by the action of dung thrown around it, because it can feed on that dung. In view of the fact, then, that prefixed verbs may reveal an argument pattern and a semantics that are not directly derived from the simplex verb, we have opted not to use the label ‘derived verb’ for prefixed forms such as beweorpan, but rather ‘inseparable complex verb’ (ICV) (adopted from Lexical Functional Grammar; see for instance Van Kemenade and Los 2003). Tradition-

140

Peter Petr´e and Hubert Cuyckens

ally, derivations are seen as altering the meaning of their input (in this case, a verb). According to CxG, however, which is basically a non-derivational theory, verb meaning is constant across syntactic contexts (Goldberg 1995: 9–10, 27– 31). The difference in semantic content, then, between simplex verb and ICV is not the result of a derivational rule but rather of a fusing process, in which the meaning of the verb is lined up with that of the prefix construction, which possesses its own semantic content and (possibly) argument structure. A construction in CxG which possesses its own argument structure is called a linking construction. As has already been indicated, the mechanism which ensures the appropriate interaction between simplex/root verb and construction (and also between distinct constructions) is called unification, as described by Kay and Fillmore (1999). Michaelis and Ruppenhofer make use of the following metaphor to explain this process: Unification of constructions [in CxG, lexemes are also considered to be (lexical) constructions, PP & HC] can best be described in terms of a metaphor involving the superimposition of slides. Any slide (construction) can be superimposed upon any other as long as the semantic and syntactic specifications on each slide ‘show through’ – that is, provided there is no conflict among the specifications on the slides in the stack (Michaelis and Ruppenhofer 2001: 54).5 Thus, one can think of a prefix construction as being superimposed upon (or stacked on top of) the lexical entry of a given verb (or noun or adjective). The lexical entry contains a minimal valency, i.e., an array of thematic roles [Goldberg’s participant roles, PP & HC], whose grammatical expression is determined by the linking construction or constructions applied. A [...] lexical entry which is unified with linking constructions is said to be a fully specified lexical entry: one in which every thematic role supplied by the lexical entry is linked with a grammatical function (Michaelis and Ruppenhofer 2001: 55). An approach based on the unification of constructions has particular advantages for explaining language change. For instance, it elegantly explains the phenomenon of structural reanalysis (cf. Hopper and Traugott 2003: 40) as a reinterpretation of the constructional make-up of a certain set of constructs instantiating one or more constructions. For instance, a process of reanalysis probably caused the be- prefix construction to be verb-headed in the first place. It is commonly assumed that prefixes originated in preverbs, which were originally free adverbs (cf. Watkins 1964; Hopper 1975; Booij and Van Marle 2003;

be-’s constructional properties in its conservation

141

Bubenik 2004). Later on, due to frequent adjacency to verbs in actual constructs, preverbs became inherently associated with verbs (and became bound morphemes). On the constructional level, this shift to boundedness seems to presuppose a change in the constructional status of these preverbs, brought about by the reanalysis of the actual constructs from being instantiations of a preverb (construction) functioning as an adjunct plus a verb (construction) into ICVs, that is, instantiations of a prefix construction with a verb (Bubenik 2004 gives an idea of how this development took place in Hittite). While the development of a prefix construction out of a preverb (construction) is presupposed in this paper, and will not be pursued any further, the construct-based reanalysis of a preverb construction plus verb into a prefix construction plus verb is similar to a type of reanalysis occurring within the prefix construction that will be discussed below in more detail (see section 6.2). For now, it will be assumed that constructs provide the relevant level of language change, but that it is the constructions instantiated by them that change (see also Traugott’s conclusion on the locus of change in Traugott, this volume).

4. A description of prefix constructions 4.1.

Predicative prefix constructions

A first type of prefix construction found in the data is the predicative one. In OE, this type occurs both with intransitive and transitive verbs of motion. Sentences (4)–(6) are instances of intransitive simplex verbs unified with a predicative prefix construction; sentences (7)–(9) provide examples containing transitive simplex verbs. (4)

nu of rode. Gif godes sune siæ Astig If god’s son be of-come:IMP.PRS.2SG now from cross. ‘If you be god’s son, come down from the cross now.’ (c950. MtGl [Ru1]: 27.40)

(5)

Ic ongite þæt ealla gesceafta TOflowen swa swa I understand that all creatures to-flew:SBJV.PRS.3SG as like wæter [...] gif hi næfdon ænne God þe him eallum stiorde. water [...] if they not-had one God who them all guided. ‘I understand that all creatures would flow AWAY[i.e. perish] like water [...] if they did not have one God who guided them all.’

(c950. Bo: 34.94.8)

142

Peter Petr´e and Hubert Cuyckens

(6)

Gewat ða neosian, syþðan niht BEcom [He] went then [to] seek, when night be-came:IND.PST.3SG, hean huses. high house. ‘So he went to seek, when night came BY, the lofty house.’ (c1015. Beo: 115)

(7)

Þa Maximianus geacsade þæt his sunu feng to þæm When Maximianus discovered that his son came into the onwalde, he þa [...] þohte his sunu to beswicanne, & power, he then [...] thought his son to supplant, and him siþþan fon to þæm onwalde. Ac þa hit se sunu him after come into the power. But when it the son anfunde, þa Adræfde he þone fæder. out-found, then a-drove:IND.PST.3SG he the father. ‘When Maximianus discovered that his sone came into power, he [...] thought to supplant him, and ascend the throne after him. But when his son discovered this, he drove his father AWAY’ (c925. Or 6: 30.148.16)

(8)

Ac mid þæm þe he from þære clusan But with that that he [= Maximus] from the prisons afaren wæs wiþ þara scipa, þa com Theodosius away-travelled was with the ships, then came Theodosius þærto & funde þæræt feawa men, [...] & he hie thereto and found thereat few men, [...] and he them raðe aweg aþewde, & þa clusan TObræc quickly away away-drove, and the prisons to-broke:INDPST3SG. ‘But after [Maximus] had travelled away from the prisons with his ships, Theodosius arrived there and found there only few men, [...] and quickly drove them away, and broke the prisons ASUNDER’ (c925. Or 6: 36.154.13)

(9)

Her hiene BEstæl se here [...] to Cippanhamme Here him be-stole:IND.PST.3SG the army [...] at Chippenham & geridon Wesseaxna lond. and occupied West-Saxons’ land. ‘This year the army stealthily moved itself in [...] at Chippenham and occupied the territory of the West-Saxons.’ (c890. ChronA [Plummer]: 878.1)

be-’s constructional properties in its conservation

143

In all these examples, the prefix is functionally equivalent to a secondary predicate, as can be seen from (4 )–(9 ). Secondary predicates “are verbal constructions in which an embedded predicate denotes the result of the action of the verb” (Van Kemenade and Los 2003: 86).6 They are predicated of the theme of the clause, which, in these predicative constructions, may be the subject in the case of intransitive motion verbs or the object in the case of transitive verbs. It is precisely because of their equivalence with these secondary predicates that each of the prefixes illustrated in (4)–(9) can be said to constitute a predicative prefix construction. (4 ) (5 ) (6 ) (7 ) (8 ) (9 )

→ You come + you are down. → All creatures flow + all creatures are away. → Night came + night was by. → He drove his father + his father is away. → He broke the prisons + the prisons are asunder. → The army moved itself + it is in.

A schematic representation of sentence (8) containing a transitive ICV can be found in Figure 4 (the representation of an intransitive ICV, such as astigan ‘come down’, in (4), would link the theme to SUBJ and have neither an agent nor an OBJ). Sem R: means

Syn

to- ‘be apart’ bræc

to-bræc



SUBJ

OBJ

Figure 4. Unification of the to- construction with the verb bræc

It can be seen from Figure 4 that the argument role pattern of the prefix construction in OE is basically not different from that of the participant roles of the simplex verb. In other words, there exists an isomorphism between participant roles and argument roles such that intransitive verbs of inherent directed motion (see sentences (4)–(6)) remain intransitive and transitive verbs remain transitive (sentences (7)–(9)) when unified with the predicative prefix construction. In fact, prefix constructions which contribute a participant and link it to a grammatical function, as in Goldberg’s well-known instance of the caused-motion construction He sneezed the napkin off the table, are extremely rare in OE, if they exist at all. The random sample we used of 235 sentences containing toonly revealed one possible example:

144

Peter Petr´e and Hubert Cuyckens

(10) Seo Wisle is swyðe micel ea & tolið Witland & The wisl is very big stream and to-lies Witland and Weonodland Weonodland → causes Witland and Weonodland to be ASUNDER by lying in be(c1025. Or 1 [c]: 1.16.29) tween As we will see below, precisely because of the isomorphic line-up between a simplex verb’s participant roles and the construction’s argument roles, the predicative prefix construction can be said to be less salient than the non-predicative prefix construction. Predicative prefix constructions are not entirely redundant with, or predictable from, the meaning of their root verb: they often involve a shift in telicity if the root verb is atelic. For instance, the root verb stelan ‘move stealthily’ underlying bestæl in (9) is atelic, whereas the ICV is telic ‘stealthily move in(to some place)’, as it has as a natural endpoint the place moved into. The idea that predicative prefixes are telicizing also receives support from their frequent occurrence in bounded clauses, that is, clauses that represent the situation as reaching a goal or an endpoint (see Declerck Forthc for the relation between boundedness and telicity). This tendency is statistically highly significant, as a comparison to the frequency with which simplex verb forms occur in bounded clauses reveals (p =< 0.01). Finally, this shift in telicity also lies at the basis of aspectual extensions, as for instance adruwian ‘dry up’ or oflætan ‘give up’ (see Brinton 1988: 204–212).

4.2.

Non-predicative prefix constructions

Not all prefix usages are examples of predicative constructions or extensions from it. In particular, prefixes whose adverbial etymon denotes route paths such as around, over, or along (see Dewell 1996: 111; Blom 2004: 20ff.) are often morphological markers belonging to another type of construction. Consider in this respect the prefix be-, which frequently has the sense of ‘around, over’, as in (11), and also in (3), here repeated as (12): (11) & [Cyneheard] hine [= Cynewulf] þær berad And [Cyneheard] him [= Cynewulf] there around-rode:INDPST3SG ond þone bur utan beeode and the chamber from outside around-went ‘And [Cyneheard] surrounded him (by riding) there and surrounded the chamber from outside.’ (c890. ChronA [Plummer]: 755.10)

be-’s constructional properties in its conservation

145

(12) Læt hine gyt þis gear, oð ic hine bedelfe & ic Let him alone this year, till I him be-delve.subj and I hine bewurpe mid meoxe him be-throw with dung ‘Let it [a fig tree] alone this year also, till I shall dig about it, and dung it’. (c1025. Lk [WSCp]: 13.8. Trans. from the King James Bible) A comparison of these ICVs with the predicative construction ICVs described above reveals considerable differences. In transitive predicative ICVs, the prefix is the equivalent of a secondary predicate predicated of the object of the sentence, which, in such ICVs is the realization of the theme. However, in the kind of transitive ICV illustrated by the two occurrences of be- in (11), the prefix cannot be paraphrased as a secondary predicate predicated of the object of the transitive sentence, as can be seen from (13). (13) *He rode + the king was around. *He went + the chamber was around. Rather, the prefix expresses a prepositional relationship between Cyneheard/he (the agent/subject/trajector) and Cynewulf/the king (the location/object/landmark) – and, together with the root verbs ridan ‘ride’ and gan ‘go’, constitutes the relationship between the trajector and the landmark. As a first approximation, the prefix usages in (11) and (12) could be paraphrased as in (11 ) and (12 ), but this paraphrase will have to be substantially revised; see below. (11 ) He rode around the king. He went around the chamber. (12 ) He threw dung around the tree. Because these prefix constructions are not equivalent to constructions involving a secondary predicate – but rather to constructions involving a prepositional object – they will be called non-predicative constructions. Two types of non-predicative be- construction will be distinguished: bivalent constructions (with an intransitive simplex verb) and trivalent ones (with a transitive simplex verb). As the latter type has already been discussed in some detail in section 3, we will here give most attention to the bivalent type. This beconstruction is instantiated by He berad þone cyning ‘He surrounded the king’, or by sentence (11), which can schematically be represented as in Figure 5. There are a number of ways in which this non-predicative be-construction can be said to be non-predictable from its component parts.

146

Peter Petr´e and Hubert Cuyckens Sem R: instance

Syn

be- ‘around’ + ‘affected’ rad

be-rad



OBJ

Figure 5. Unification of the bivalent be- construction with the verb rad

First, in the two ICVs in (11), the prefix be- and the root verbs ridan and gan co-constitute the (spatial) relation between subject (trajector) and object (landmark), indicating that the direct object is subjected to the activity expressed by the verb in all the (spatial) points of the path denoted by the prefix – whereby all of these points are subjected to the verb’s activity simultaneously (that is, not sequentially). As such, the castle in (14) is subjected to the army’s ‘sitting’ (i.e., occupation) in all the points of a path surrounding the castle. This situation of complete and simultaneous surrounding gives rise to the notion of ‘total affectedness’; indeed, the castle and in particular its inhabitants are highly affected by the army’s action of ‘sitting’ (i.e., by the occupation), because in the end they are conquered and their king is slain. Importantly, this notion of ‘total affectedness’ is not part of the semantic content of the root verbs, i.e., is contributed by the construction itself. (14)

þæs ilcan sumeres gegadorode micel folc hit on Eadweardes The same summer gathered much people itself in Edward’s cynges anwalde. Of þam niehstum burgum, [...] & foron king’s power. From the nearest fortresses, [...] and travelled to Tæmeseforda. & besæton ða burg. & fuhton to Tempsford, and be-sat:IND.PST.3SG the fortress, and fought ðær on oð hi hie abræcon, & ofslogon þone cyning. there on until they them conquered, and of-slew the king. ‘The same summer many people gathered under the dominion of king Edward from the nearest fortresses, [...] and travelled to Tempsford, and besieged [i.e., sat AROUND AND THEREBY TOTALLY AFFECTED] the fortress, and fought there until they conquered them, and slew the king.’

(921. ChronA [Plummer]: 921.29) The notion of ‘total affectedness’ is also present in (11), and as such, (11 ) and (12 ) provide a more adequate rendering of (11) and (12) than did (11 ) and (12 ): (11 ) He rode AROUND [i.e., surrounded] the king, He went AROUND [i.e., surrounded] the chamber, [i.e. its WAS TOTALLY AFFECTED. inhabitants] WAS TOTALLY AFFECTED.

WHO WHICH

be-’s constructional properties in its conservation

(12 ) He

THREW

dung around the tree,

147

WHICH WAS TOTALLY AFFECTED.

This specific semantic characterization of the non-predicative prefix construction, as exemplified in (11) and (14), distinguishes it in a salient fashion from its prepositional cognate, as can be seen from the paraphrases in (11 ) and (14 ), where the ‘total affectedness’ meaning is absent. (14 ) The army sat around the castle. One could argue that this contrast results from a particular development of the prefix be-, by which it has been removed semantically from its OE prepositional cognate be ‘by, along’(PDE by). However, a similar contrast apparently holds for other prefix–preposition pairs with route path semantics. Consider the following examples, where a sentence containing OE ofer- ‘over’ is contrasted with a sentence containing its cognate preposition ofer ‘over’. (15) Þa forleton hie hie, & eodon ofer land þæt hie Then left they them, and went over land that they gedydon æt Cwatbrycge be Sæfern arrived at Bridgenorth by Severn ‘Then they left them and went over land till they arrived at Bridgenorth by Severn’. (896. ChronA [Plummer]: 896.14) (16) Her on ðissum geare com Unlaf mid þrim & hund Here in this year came Unlaf with three and hunderd nigentigon scipum to Stane, & forhergedon þæt on ninety ships to Stone, and harried:ind.pst.3pl that from ytan, & for ða ðanon to Sandwic, outside, and travelled:ind.pst.3sg then thence to Sandwich, & swa ðanon to Gipeswic, & þæt eall ofereode, and so thence to Ipswich, and that completely overran & swa to Mældune and so to Maldon ‘In this year Unlaf arrived with hunderd ninety three ships at Stone and they harried it from outside and he travelled then from that place to Sandwich, and so to Ipswich, and completely overran it, and so to Maldon’. (993. ChronA [Plummer]: 993.1) Sentences (15) and (16) show the same difference between preposition and prefix as did (11 ) and (11). While the landmark of ofer, in (15), i.e., land, is

148

Peter Petr´e and Hubert Cuyckens

left unaffected once the army has passed over it, the landmark Ipswich in (16) is completely devastated by the action denoted by the ICV ofergan ‘overrun’. In PDE, only few verbs prefixed by over- still carry this original route path semantics. One example, which is actually well-known to CxG, is the verb override. It is not only the addition of the ‘affectedness’connotation that makes the nonpredicative be- construction non-predictable. Indeed, its argument role pattern contains a ‘location’ role which is not a participant of the intransitive input verb, but is contributed by the prefix construction itself. One could say, then, that the prefix be- has a valency pattern of its own (unlike most of the predicative prefixes), or that employing a simplex in a construction with a non-predicative route path prefix entails a valency shift. More specifically, when intransitive verbs of motion such as ridan and gan are employed in a non-predicative prefix construction, they become transitive, with the construction’s location argument (i.e., the landmark around which the motion is directed) being linked with the syntactic function of direct object (the king and the chamber in (11), Ipswich in (16); see also Figure 5). Finally, just as for verbs in the predicative construction, the addition of the prefix be- turns atelic verbs in the non-predicative construction into telic ones (this can be inferred from their frequent presence in bounded clauses, with an ever higher statistical significance [p =< 0.001] than in the case of predicative prefix constructions; compare section 4.1). A second type of non-predicative be- construction is trivalent, and contains a transitive simplex. Examples are in (17) and (3), which was discussed earlier and schematically represented in Figure 3. The non-predictable nature of this type of be- construction derives (i) from its inherent affectedness meaning, and (ii) from its argument pattern which is not isomorphic with the participant roles of the simplex. In particular, a prepositional paraphrase of this type of predicative construction would not convey ‘affectedness’: when one ‘around-throws a tree with dung’ (cf. 3), the whole tree is affected, and as a result, it might grow better; throwing dung around a tree just states that the tree is the goal of the throwing event – it might just be an indeliberate act – and no affect-relationship is implied. Furthermore, the transitive verbs employed in this construction show a particular kind of valency alternation, known as the applicative alternation. In particular, the theme argument in the ‘applicative’ construction, which was syntactically realized as direct object of the transitive simplex, is now realized as a PP/instrumental adjunct:

be-’s constructional properties in its conservation

149

(17) Sume ða yða he becerð mid ðy scipe. Some those waves he be-turned with the ship. ‘Some of those waves he by-passed with his ship.’ (894. CP: 56.433.5) As a result of its specific semantic and syntactic makeup, the non-predicative construction can be regarded as a salient construction: (i) the presence of the ‘affectedness’ component makes this construction more salient than its prepositional cognate; (ii) the alternations in valency structure make it syntactically salient in contrast to simplex verbs.

5. The development of prefix constructions It is commonly known that during the OE period English shifted from an OV language, where objects tend to precede the verb they specify, to a VO language, where objects tend to follow (e.g., Fischer et al. 2000: 138–139). This shift was not restricted to nominal objects, but it also held for other types of verb specifiers, such as adverbs, or, as for instance Hiltunen (1983) and Ogura (1995: 79) argue, for Old English prefixes. Those prefix uses where the prefixes were still felt to be specifiers of the verb in Old English came under pressure to be replaced with new patterns adapted to the new VO order of English. In the case of spatial predicative prefixes, this replacement could be made without any substantial loss of the functional/conceptual qualities of the prefix construction. As examples (18) and (19) show, the construction with a phrasal verb particle retained the argument roles/syntactic functions and semantics of the predicative prefix construction: (18a) and (18b) are both intransitive, just as (19a) and (19b) are transitive; furthermore, the (b) sentences retain the telicity of the (a) sentences. Note that this type of replacement of one structure by a functionally/conceptually equivalent one falls under the rubric of what Croft (2000) calls ‘intraference’, a mechanism of language change which applies more generally in language (see for instance the replacement of the that-clause with suasive verbs by the to-infinitive; cf. Los 1999). (18) a. Gif godes sune siæ astig nu of rode. (c950. MtGl [Ru1]: 27.40) b. Gyf þu sy godes sunu, ga nyþer of þære rode. (c1025. Mt [WSCp]: 27.40) ‘If you be god’s son, come DOWN from the cross.’

150

Peter Petr´e and Hubert Cuyckens

(19) a. Theodosius [...] þa clusan TObræc ‘Theodosius broke the prisons ASUNDER’ b. He broke the cordes AL

(c925. Or 6: 36.154.13)

ASUNDER.

(Caxton. c1490. Bl: 190.14) Abstract extensions of these spatial predicative constructions, such as aspectual usages, were initially probably less easy to replace, but once the phrasal verbs had developed an aspectual system themselves, this replacement also became unproblematic (see, on this development, Brinton 1988: 212). In many cases, rather than being replaced by verb particle constructions, the prefix could also be left out entirely, because its semantics overlapped with that of the root verb. Night comes is usually equivalent to Night comes by, even though the focus on the result is somewhat lost. The same holds for break and break asunder (Bechler 1909: 76–77): (20) He brake the grete rondes (Caxton. c1490. Bl: 63.3) What about non-predicative prefix constructions? Our claim is that the nonpredicative be-construction was not replaced by a prepositional cognate because this would have meant changing its semantics as well as the overall valency pattern. Indeed, the replacement by a prepositional cognate would have meant losing the semantic component ‘total affectedness’(cf. the situation in present-day German, where the non-predicative prefix construction remains productive because the prepositional alternative is inappropriate to convey affectedness of its participants; Michaelis and Ruppenhofer 2001; Dewell 1996). In other words, the cause of the preservation of the non-predicative be- construction lies with its specific constructional properties, combining ‘path’ and ‘affectedness’, and linking the location participant to OBJ. The precise character of the conservatory influence of these constructional properties will be discussed in detail in section 7.

6. Major grammaticalized uses of the non-predicative beconstruction It has been seen in section 4.2 that, in contrast to the predicative be-constructions, the bivalent and trivalent spatial be- constructions constituted a highly salient set, semantically as well as syntactically. It is argued that this set, which will be referred to as the ‘Surrounding’ construction, served as a prototype for many extensions, where the original spatial semantics of be- ‘around’ is gradually

be-’s constructional properties in its conservation

151

lost and more grammatical meanings develop. There is one major series of nonpredicative extensions which can be said to make up, together with the prototype, the core of be-, in that they are semantically and/or syntactically related most closely to the prototype and to each other, and that they also have, overall, the highest frequency.These extensions can be labelled ‘Extensive Coverage’, ‘Total Affectedness’, and ‘Furnishing’ (this last one, being the most recent one, is also the most productive in PDE). Similar extensions have already been described by Michaelis and Ruppenhofer (2001) for German, but their importance in English has not yet been fully appreciated. Although all these extensions occur in OE, it is still possible to reconstruct their probable chronological order of appearance. An analysis of all be- verbs in Gothic only reveals instances of the ‘Extensive Coverage’ construction, which indicates that the ‘Total Affectedness’ and ‘Furnishing’ sense are of later date. Of these two, the ‘Furnishing’ sense is likely to be the later one, because it is fairly rare in OE, but becomes more frequent in ME and Modern English – the chronology here is, however, less clear. We would like to see the development of the non-predicative be-construction as an instance of the grammaticalization of a construction (cf. Hopper and Traugott 2003; for another example cf. Kemmer and Hilpert 2005).7 Several of the grammaticalization criteria, such as recently proposed by Himmelmann (2004), are satisfied: ‘host-class expansion’, ‘syntactic context expansion’, and ‘pragmatic-semantic expansion’. First, in the extensions described below, hostclass expansion is evident: while originally only spatial verbs of motion with a landmark participant could properly be unified with the non-predicative construction marked by be-, other hosts, where there is no clear sense of spatial motion, could increasingly be used as well (e.g., wyllan ‘boil’ versus bewyllan ‘boil away’). Secondly, although the non-predicative be-construction retains its valency pattern in its extensions, and as such does not prima facie show syntactic context expansion, we would like to argue that the extension through the ‘Furnishing’ construction to nominal and adjectival heads is an instance of syntactic context expansion as well (e.g., bespouse based on the noun spouse, befriend based on the noun friend). Finally, through pragmatic processes such as invited inference, the prefix has expanded to contexts where it is no longer used nonpredicatively. In this type of expansion, its equivalence with a preposition is entirely lost (e.g., beswælan ‘scorch’ versus swælan ‘burn’).

6.1.

Extensive coverage

A very early metonymic extension from the ‘Surrounding’ construction marked by be- is that of ‘Extensive Coverage’, in which be- means ‘(all) over’ and in

152

Peter Petr´e and Hubert Cuyckens

which it denotes a route path as well. The resulting polysemy of the prefix can be compared to the one we find in PDE around and New High German um/herum. These prepositions are “ambiguous between a sense of ‘surrounding an enclosed space’ and ‘being distributed over a surface area”’ (Michaelis and Ruppenhofer 2001: 89). This development is illustrated by the following bivalent (21a) and trivalent (21b) OE examples: (21) a. Wæs bi eastan þære welneah sumo cirice in are Was by east that.gen well-close some church in honour Sancti Martini geo geara geworht mid þy Romani Saint Martin’s once years’ built when Romans þa gyt Breotone beeodon then still Britain be-walked. ‘Close east of that place a church in honour of Saint Martin was built, when the Romans still occupied [i.e., walked about] Britain.’ (c900. Bede 1: 15.62.2) b. Se apostol hine begeat mid ðam wætere. The apostle him doused [lit. over-poured] with the water. (c1000. ÆCHom II, 31–32: 247.175) The extension can be accounted for as follows: when an agent or theme completely surrounds a landmark, it can be said to be spatially contiguous (and as such metonymically related) to the surface area contained within the landmark’s boundary. It is not surprising, then, that be- also comes to denote the agent/theme extending over a surface area, either by following a path in various directions across the landmark’s surface area (as in 21a) or by being distributed over it (as in 21b). In the ‘Extensive Coverage’ extension, both the ‘affectedness’ meaning characteristic of the ‘Surrounding’ prototype and its shifted valency pattern (with respect to that of the root verb) are preserved. Other OE ICVs instantiating this extension are for instance bebaðian ‘bathe (lit. put water over sb.)’, behelian ‘cover’ (cf. helian ‘conceal’), besettan ‘cover, adorn’ (cf. settan ‘set, place’), besmitan ‘soil, defile’ (cf. smitan ‘daub, smear’), bestyman ‘bedew’ (cf. styman ‘steam’), beswapan ‘clothe, cover over’ (cf. swapan ‘sweep’) (some of these may also belong to 6.2).

6.2. Total Affectedness The extension to ‘Extensive Coverage’is not only instantiated by ICVs where the notion ‘Coverage’ solely derives from the be-prefix, but it also occurs with ICVs

be-’s constructional properties in its conservation

153

where the notion of ‘coverage’ is already present in the root verb, as is illustrated by the pair bewreon and wreon, both meaning ‘cover’. In this case, then, there are two types of root verbs the prefix can combine with. The first type consists of simplex verbs whose transitive valency pattern differs from that of the beconstruction after unification with the simplex verb. Compare in this respect the valency pattern of the verb lecgan in the non-attested but plausible example in (22) with the ICV belecgan ‘cover’ in (23), which arranges the participants of lecgan differently – the unification here is similar to that represented in Figure 3. (22) Lege hatte wulle on þæt sar. ‘Put hot wool on the wound’ (23) Belege [þæt sar] æfter þære beþinge mid hatte wulle Be-put [the wound] after the heating with hot wool ‘Cover [the wound] after the heating with hot wool.’ (c950. Lch II [2]: 47.1.4.) The valency pattern in (23) comprising a location OBJ and a theme OBL is identical to the one found in some simplex verbs, such as wreon in (24). These simplex verbs constitute the second type of verbs the construction can combine with, as is illustrated in (25). (24) & wreoh [þa eagsealfe] mid brede ‘And cover [the eye-salve] with bread’

(c950. Lch II [3]: 2.1.3)

(25) & bewreoh [þæt heals] fæste ufan mid leafum. And be-cover [that neck] firmly from above with leaves. ‘And cover over [that neck] firmly with leaves.’ (c950. Lch II [1]: 4.2.3) In cases such as (25), the only semantic component added by the prefix to the verb is that of ‘affectedness’. What seems to be occurring at this stage in the development of the non-predicative be-construction is a process of semantic reanalysis (see Croft 2000: 130) in which the ‘Coverage’sense of be- gets lost and only ‘Total Affectedness’ is retained, together with the separate valency frame of the prefix. Two factors may have contributed to this semantic reanalysis. First, while in the earlier be-ICVs, the notions ‘Surrounding’ and ‘Extensive Coverage’ both enter into the semantic makeup of the be-construction, in the construct unifying wreon and the be-construction, a semantic division of labor seems to be operational whereby the ‘Covering’ semantics is uniquely assigned to the root verb – the ‘Covering’ meaning is, after all, already present in that

154

Peter Petr´e and Hubert Cuyckens

root verb – and the ‘Total Affectedness’ sense to the construction. The slide metaphor mentioned in section 3 allows us to clarify how this may happen. Assuming that a language user is at first only faced with the resulting projection of all superimposed slides, it seems plausible that he or she cannot always clearly tell from the resulting complex ‘slide’, i.e., the construct, from which slide a certain specification originates. In the construct combining wreon and the beconstruction, the semantics of coverage shows through only once. Therefore, if there is not sufficient additional information, the language user may think that this property is encoded only once, either in the simplex verb or in the be-construction, and consequently reanalyse the make-up of one of them or even sometimes both. A second factor that has led to the semantic reanalysis ‘Covering’ > ‘Total Affectedness’ is the process of invited inferencing. In a lot of cases, an event of an agent extensively covering a location will also entail that agent totally affecting that location, to the extent that the ‘Extensive Covering’ notion may become secondary or get lost. A factor that is frequently involved in this process of inferencing is the presence of the semantic component of ‘iteration’, which also leads to the sense of ‘affectedness’: in (24), repeated actions of coverage are necessary before the neck is totally covered over with leaves (or totally affected by the leaves); in (26), repeated hitting is necessary for the body to be totally affected by the whips. (26) We scylen beon on ðisse ælðeodignesse utane beheawene We shall be on this pilgrimage abroad be-hewn mid suingellan. with whips. ‘We have to be beaten all over with whips on this pilgrimage.’ (894. CP: 36.253.17) Once the sense of ‘Total Affectedness’ has been established independently from the ‘Extensive Coverage’ scenario, reanalysis is made apparent in the addition of the prefix to transitive verbs with non-location arguments. The resulting effect of the addition of be- then consists mainly of a strengthening of the transitivity (and concomitant affectedness) of the simplex: (27) Bewyl þara meolce þriddan dæl. Be-boil of the milk third part. ‘Boil away a third part of the milk.’ (c950. Lch II [3]: 22.1.3) Other examples are beswælan ‘burn, scorch’(vs. swælan ‘burn’) and behamelian ‘mutilate’ (vs. hamelian ‘hamstring’).

be-’s constructional properties in its conservation

6.3.

155

Furnishing

It has been seen that throughout the various extensions discussed so far, the valency (or linking) pattern of the prefix construction was preserved. This valency pattern of [agent SU B J be-V location O B J (theme O B L )], now, was so prevalent in the minds of the language users, that it became possible to extend the V-slot to nouns and, rarely, adjectives (valency creation). This extension, however, also meant a new development in the semantics of be-, whereby the noun or adjective had to be provided with a verbal semantics. The sense be- was able to add without the support of a root verb can roughly be captured as ‘Furnishing’ (e.g., bespectacled related to the noun spectacles), or, metaphorically, the causative sense ‘making someone into something’ (as in befool [cf. the noun fool] ‘furnish someone with foolhood’, or befoul, related to the adjective foul). Sentences (28) and (29) are some typical instances of this extension; (29) would have the following valency pattern: [be-spade location O B J (Major and Curate) theme O B L (spades)]. (28) Nis na stude to istreone bicumelic butan ða þe istreonieð beon bispused rihtliche to gedere. ‘No place to procreate is fitting except if those who procreate are be-spoused [‘married’] rightly together.’ (c1225. LambX1: 133) (29) The neighbouring Villages turn out: their able men come marching, to village fiddle or tambourine and triangle, under their Mayor, or Mayor and Curate, who also walk bespaded [‘with spades (on their shoulders)’], and in tricolor sash. (Carlyle. 1837. The French Revolution) One might object that it cannot be the original valency pattern [be-V location O B J theme O B L ] that underlies this extension, as the theme in denominal ICVs is already expressed by the source nominal of the ICV itself. However, while it is true that oblique themes are generally omitted in denominal ICVs, the reason for this omission is of a pragmatic rather than a syntactic nature. Usually, the source nominal renders the theme argument easily recoverable, and this recoverability makes the expression of the theme redundant. However, examples (30)–(32) show that this may not always be the case. (30) Min freond siteð under stanhliþe storme behrimed. My.gen friend sits under stone-cliff storm:ins be-rimed. ‘My friend sits under a cliff of stone, behoarfrosted with storm.’ (c970. Wife’s Lament [Exeter book]: 48–49)

156

Peter Petr´e and Hubert Cuyckens

(31) Too foul-mouthed I am to becollow or becollier him with such chimneysweeping attributes of smoking and parching. (Nashe. 1593(orthography 1871). Lenten stuffe) (32) Shakespeare [...] can be complemented by the sober heft of. Eliot’s Murder in the Cathedral or spritzed with My Fair Lady in an ingeniously extravagant production that bejewels the stage with chandeliers, dinner jackets and hats. (Time magazine, 22.08.1988, http://www.time.com/time/ magazine/article/0,9171,968174,00.html [19.07.2007]) These examples, and particularly the active clauses in (31)–(32), show that the source nominal of the derivation does not inherently refer to the theme as such, once it is unified with the be- construction, and that the theme has to be expressed separately if it is not recoverable from context (see Michaelis and Ruppenhofer 2001: 47). This section is also the appropriate place to discuss the productivity status of be- in PDE. As (29), (32), and the recent ICVs (un)besocked and betied given in the introduction show, it is the ‘Furnishing’ extension that provides almost all of the new ICVs from Late Modern English onwards. A glance at the lemma for the prefix be- in the OED2 suggests that this extension was fairly productive during the late Modern English period, with forms such as bedismalled (1751), be-knighted (1794), bedirtied (1803), bedoctored (1806), becowarded (1831), be-lioned (1837), be-baroned (1842) or begreened (1864). In addition, a considerable number of new forms have been attested in the Present-day English BNC corpus. At least forty-six forms occur only once or twice, and at least twentythree of these are hapax legomena which do not occur anywhere in the OED, among them be-costumed, be-hymened, be-mirrored, be-navelled, be-sticked, be-trainered, bebow-tied, beleathered, belipped, beplushed, besweatered, betweeded. One might suspect from this list that be- has become a marker of (participial) adjectives rather than of full-fledged verbs in PDE. While this is certainly a strong tendency, (33) shows that these new forms can also be used in other verb forms. (33) Lyle, as defending champion, had to be-jacket Faldo, and coincidentally it signalled the end of perhaps the finest stretch of golf played by a British golfer. (1989. The guardian [BNC, AAW]) In sum, while it is likely that many of these new forms are conscious coinings, they still seem too widespread to give up the idea of a low productivity rate completely.

be-’s constructional properties in its conservation

157

7. The conservatory influence of be-’s core Table 2 shows that during the OE period – when the grammaticalization of behad not yet reached its final stage – the types belonging to be-’s core took up a substantial proportion of the total number of types making up be-’s network (see below for the meaning of the decimals in Table 2). Table 2. Share of core of be- within 235 token sample of be- prefixed verbs Gothic Types from be-’s core

OE

ME

EModE

LModE

22 (41.5%) 44.09 (58.7%) 30.7 (42.6%) 11 (44%) 13 (50%)

Total number of types 53

75

72

25

26

Among the other types falling outside of be-’s core were: (i) Predicative types instantiated by verbs like becuman ‘come by’or belimpan ‘happen’[lit. ‘fall/limp by’], which were either replaced by phrasal particles or lexicalize (like become); (ii) Types making up the ‘privative construction’ and instantiated by such verbs as behead (OE beheafdian) or bereave (OE bereafian). This extension from the ‘Surrounding’ prototype, while very productive in OE, had gradually distanced itself in meaning and function from its prototype, and this may have influenced the consequent loss of productivity of this construction in ME. The shift in case assignment for the theme from instrumental to genitive may have further added to the distancing and consequent loss (Lenze 1909: 113); (iii) Idiomatic extensions from the prototype, with verbs of deception as a fairly productive set in late OE and early ME, and still seen in PDE beguile, betray, and bewray (see Petr´e 2006 for a detailed account). From the figures in Table 2, it can be inferred that the relatively high type frequency of the core of the be- non-predicative construction had come to play an important role in the conservation of the prefix. In other words, even when the spatial prototype, i.e., the ‘Surrounding’ construction, was gradually given up, the high type frequency of its extensions preserved the high degree of entrenchment of the non-predicative be- construction; an important reason why the high type entrenchment of these major grammaticalized usages exerted this conserving influence is that, in addition to the fact that they shared a (salient) semantics which was not predictable from the simplex verb, they preserved to a large extent the salient valency pattern of the original non-predicative construction, with a location participant construed as a direct object, and, in case of the trivalent construction, an optional theme participant as an oblique (either a mid-

158

Peter Petr´e and Hubert Cuyckens

PP ‘with ...’ or an instrumental). This valency pattern often differed from that of the simplex verb unified with an Active or Passive construction.8 By contrast, predicative constructions such as those marked by the prefix to- almost always preserved the valency patterns of the verbs they attached to. The independent valency pattern of the verb-headed be- construction is made particularly visible when it is superimposed on a non-verbal lexical entry. Examples of this kind of unification are ICVs formed with nouns like bespaded or bespectacled. In such cases, the verb-specification of the be- construction (i.e., the verbal status imposed by it on the ICV) not only overrides the nominal specification of the root noun, it also provides an entirely non-derived argument structure of its own. While one could argue that a noun like spade also has some participants associated with it in its semantic frame, such as a spader, these are completely absent in the argument structure of the resulting ICV. Hence, the contrast between root noun and ICV becomes particularly clear. Also, recall examples (23) and (22), repeated here as (34) and (35), respectively. ICV with valency pattern of be- construction: (34) Belege [þæt sar] æfter þære beþinge mid hatte wulle Be-put [the wound] after the heating with hot wool. ‘Cover [the wound] after the heating with hot wool.’ (c950. Lch II [2]: 47.1.4) Simplex verb in a default active realization with a different valency pattern: (35) Lege hatte wulle on þæt sar. ‘Put hot wool on the wound’ The example in (35) is non-attested, yet plausible and has been added, in a way similar to minimal pairs, to make the syntactic contrast with (34) particularly salient. Language users also make use of such techniques to establish the meaning of a certain word or phrase, for instance in second language acquisition. While it is not very likely that native speakers do this as consciously as linguists, frequent exposure to this type of contrast in valency pattern frequently enables them to establish the particular contribution of the prefix construction to the semantics and syntax of the verb more easily. In psycholinguistic terms, the more the language users are exposed to differences in valency pattern, the more salient the construction will be syntactically. If the construction is sufficiently salient, our prediction is that it will remain in use. How can this saliency actually be measured in our data? As we do not have access to OE native speakers, we have worked out an alternative way to measure this degree of salience. Each sentence containing an ICV containing the prefix

be-’s constructional properties in its conservation

159

be- within a representative sample of about 200 tokens was matched with another sentence from the corpus containing the simplex cognate of this ICV. For instance, we compared the three occurrences of befon to three occurrences of fon to find out if they were used with different valency patterns in their particular contexts. We then performed a similar comparison for the prefix to-. This procedure also accounts for the decimals in Table 3 below. In two out of its three occurrences, befon has a direct object and a prepositional phrase in its valency (meaning ‘surround’), and only once a prepositional phrase (meaning ‘have to do with’). By contrast, fon ‘lay hands on, take to’ always has only the prepositional phrase. Consequently, in Table 3, the type befon receives a 0.33 rating for ‘No difference in valency’ and a 0.67 rating for ‘Difference in valency’. Mutatis mutandis, the fact that different tokens of the same type can be instantiations of different constructions also accounts for the decimals in Table 2. The following are some examples of resulting excerpt pairings. The prefix beExcerpt pairings with different valency patterns. 1. The simplex smitan ‘smear’ in the following excerpt licenses a theme as direct object and a landmark as a prepositional adjunct introduced by on: (36) Wiþ gongelwæfran bite, smit on isen swat. Against spider’s bite, smear on iron sweat. ‘Smear sweat on iron against spider’s bites.’ (c950. Lch II [2]: 65.5.9) The ICV besmitan ‘soil, defile’ in (37), however, licenses a patient (equivalent to the landmark of the simplex in (36)) as a direct object and the theme (in (34) the direct object) in the instrumental case. (37) Þu ellþeodig usic woldest on þisse folcsceare You foreign us wished:ind.pst:2sg in this nation facne besyrwan, synnum besmitan. treachery:ins be-plan, sins:ins be-smear. ‘You stranger wished to deceive us within this nation with treachery and defile us with sins.’ (?c1000 Genesis [Krapp]: 79.2680ff) 2. The simplex ridan ‘ride’ in (38) is intransitive, whereas the ICV beridan ‘surround’ in (39) is transitive, the affected landmark being construed as the direct object.

160

Peter Petr´e and Hubert Cuyckens

(38) Her rad se here ofer Mierce innan East Engle. Here rode the army over Mercia into East Anglia. ‘This year the army rode through Mercia into East Anglia’ (c890. ChronA [Plummer]: 870.1) (39) & And ond and

[Cyneheard] hine [= Cynewulf] þær berad [Cyneheard] him [= Cynewulf] there around-rode:IND.PST.3SG þone bur utan beeode. the chamber from outside around-went. ‘And [Cyneheard] surrounded him there and surrounded the chamber from outside.’ (c890. ChronA [Plummer]: 755.10)

3. The simplex geotan ‘flow’ in (40) has a dative object as ‘malefactee’, whereas the ICV begeotan in (41) is transitive and construes a patient-landmark as direct object and a theme as PP introduced by mid: (40) He nawuht ne wyrcð, ac sio slæwð him giet on He naught not produces, but the sloth him.dat flows in ðone slæp. the sleep. ‘He will produce nothing, instead sloth will overwhelm him during his sleep.’ (894. CP: 39.283.6) (41) Þa yrsode he ond gebealh hyne ond het hig Then raged he and angered him and commanded them aðenian on yren bed ond hig begeotan myd out-stretch on iron bed and themacc be-pour with weallende leade boiling lead ‘Then he raged and became angry and commanded to stretch them out on the iron bed and to cover them with boiling lead’ (c1060. Mart 2.1 [Herzfeld-Kotzor]: De10, A.10.280) Excerpt pairings with identical valency patterns 1. The sentence pair (24)-(25) containing wreon and bewreon respectively, and here repeated for convenience as (42)–(43): (42) & wreoh [þa eagsealfe] mid brede. ‘And cover [the eye-salve] with bread.’

(c950. Lch II [3]: 2.1.3)

be-’s constructional properties in its conservation

161

(43) & bewreoh [þæt heals] fæste ufan mid leafum. And be-cover [that neck] firmly from above with leaves. ‘And cover over [that neck] firmly with leaves.’ (c950. Lch II [1]: 4.2.3) 2. Both drifan in (44) and bedrifan in (45) have a theme in direct object-position (and a prepositional adjunct of place). (44) Hu Orosius spræc ymbe Romano gielp, hu hie monega How Orosious spoke about Romans’ glory, how they many folc oferwunnon; & hu hie monege cyningas beforan peoples conquered; and how they many kings before hiora triumphan wið Rome weard drifon. their triumphs against-Rome-wards drove. ‘How Orosius spoke about the glory of the Romans, how they conquered many peoples; and how they drove many kings in front of their triumphal processions towards Rome.’ (c1025. OrHead [c]: 5.1.48) (45) Þæ [Ælle & his i¨u suna] ofslogon monige Wealas, & sume There [Al and his three sons] slew many Britons, and some on fleame bedrifon on þone wudu þe is genemned to flight be-drove into the wood that is called Andredes leage. Andred’s lea. ‘There Al and his three sons slew many Britons, and put some to flight into the wood that is called Andred’s lea.’ (c890. ChronA [Plummer]: 477.1) 3. Both swingan ‘beat’ and beswingan ‘flog’ in (46) and (47) have a (highly) affected patient as a direct object. (46) For ðan symle God her wundað & swingð ða þe Because always God here wounds and beats those that he wile habban & to þam ecan life gelædan. he wishes have and to the eternal life lead. ‘Because God always wounds and beats here [on earth] those whom he wishes to have and lead to eternal life.’ (c950–1000. HomU 7 [ScraggVerc 22]: 81) (47) Gif hine mon beswinge, mid XX scillingum gebete. If him people be-swing, with 20 shilling amend. ‘If people flog him, amend it with 20 shilling.’ (c1000. LawAf 1: 35.1)

162

Peter Petr´e and Hubert Cuyckens

The prefix toExcerpt pairing with different valency patterns. The simplex lætan ‘let’ in (48) has a direct object and an object complement, whereas the ICV tolætan ‘relax’ in (49) is intransitive. (48) He mid ungemetlicre grimsunge his hieremonna wunda to He with immense harshness his servants’ wounds too suiðe ne slite ne ne ice, ne eft for much neither tears nor not enlarges, nor again for ungemetlicre mildheortnesse he hie ne læte unwriðena. immense mildheartedness he them not lets unbound. ‘Neither does he, out of excessive harsness, tear apart or enlarge the wounds of his servants very much, nor does he, out of excessive mercy, let them be assuaged.’ (894. CP: 17.125.12) (49) Ond ðonne he bið utane ymbhringed mid ungemetlicre And when he is outside surrounded with excessive heringe, he bið innan aidlad ðære ryhtwisnesse, praise, he is within deprived the.gen righteousness:gen, & forgiet hine selfne ðonne he tolætt. and forgets him self when he to-lets. ‘And when he is surrounded by excessive praise from the outside, within he is deprived of his righteousness, and forgets himself when he relaxes.’ (894. CP: 17.111.8) Excerpt pairing with identical valency patterns. (50) Gif se geswollena mon on þære lifre oððe se aþundena If the swollen man on the liver and the swelling swa aswollen gebit oþ þone fif & twentigeþan dæg so swollen remains until the five and twentieth day swa se swile ne bersteþ þonne onginð sio lifer so the swelling not bursts then begins the liver heardian. harden. ‘If a man has a swollen liver and the swelling remains swollen for twenty five days and if the swelling does not burst, then the liver begins to harden.’ (c950. Lch II [2]: 19.1.3)

be-’s constructional properties in its conservation

163

(51) & on Somnia þæm londe seo eorþe tobærst And in Somnia the land the earth to-burst. ‘And in the land Somnia the earth burst asunder.’ (c925. Or 5: 10.123.18) Table 3. Comparison of valency patterns of type pairings in OE beFrom core

toOther

Total

Difference in Valency 24.54 10.46 35* 7.53 No Difference in Valency 10.8 10.2 21 30.47 No simplex found 0 1 1 2 Total amount of types 35.34 21.66 57 40 *Fisher Exact: The probability in the grey area table of a larger value for * is < 0.001

The results of this sample pairing are presented in Table 3, which makes clear that the behaviour of be- is the reverse of that of to-. For be-, 61.4% of the types found in the sample have a different valency pattern relative to their simplex counterparts (Column ‘Total’). The majority of these types derive from the core grammaticalization derived from the non-predicative prototype (Column ‘From core’), which also provided the highest number of types within be-’s lexical network (58.7 %, see Table 2). For to- this is only 18.8%. It would appear, then, that the high type entrenchment of the be- non-predicative construction can to a large extent be quantified in terms of high type frequency of the construction’s specific valency pattern. The different figures for be- and to- correlate perfectly with the different behaviour of the frequencies given in Figure 1. Be-, being syntactically highly salient, still increased in token frequency in ME, whereas to-, being low in syntactic salience, diminished. Therefore, we believe that together with the different semantics of these prefixes – non-predicative be- combining ‘path’ with ‘total affectedness’ and to- merely adding the result of the action of the verb – the factor of syntactic salience/entrenchment has played an important part in the conservation of the prefix.

8. Conclusions We have tried to show how the original non-predicative properties of certain constructions marked by be- could have played a major part in the conservation of this prefix. The spatial prototype construction, with its combination of ‘path’ and ‘affected object’, was highly salient in terms of semantics and syntax. This

164

Peter Petr´e and Hubert Cuyckens

initially slowed down a possible wearing out of the prefix and at the same time provided a good basis for useful extensions such as those of ‘Extensive Coverage’, ‘Total Affectedness’ and ‘Furnishing’. The high degree of entrenchment of these salient extensions made it possible for be- to be conserved after the storm of the shift from OV to VO had calmed down. We hope to have shown how a constructional framework can provide interesting insights in language change. Some constructions can be more salient than others and therefore resist tendencies the others cannot. We further believe that this concept of constructional salience, if properly developed and implemented, may shed light on the factors that determine the pace and character of a particular grammaticalization process as compared to the grammaticalization of other constructions.

Notes *

1.

2. 3.

4.

We would like to thank Elizabeth Traugott, Hendrik de Smet, and two anonymous reviewers for their helpful comments and constructive criticism on earlier versions of this article. Instead of the commonly used denominator of 1,000,000 words of running text (including every kind of word), we take as a reference point the number of lexical verbs. Frequencies per 1,000 lexical verbs give a more representative picture of the frequency history of verbal prefixes, because the total number of words in English has gradually increased over time as a result of its shift to an increasingly analytic language. This tendency can be observed, for instance, in the increased use of auxiliaries instead of inflectional endings (e.g., the replacement of he dyde it by he has done it) and of prepositional phrases instead of bare case endings. To control the possible distortion caused by the increase of auxiliaries, auxiliary verbs have also been excluded from the verb count. As such, with the number of lexical verbs as a denominator, ME shows an increase in be-verbs of 25.7% as compared to OE; if calculated relative to the total number of words, this increase would only have been 16.9%, which is significantly less. In her recent publications, Goldberg has also included patterns that “are fully predictable as long as they occur with sufficient frequency” (Goldberg 2006: 5). ‘Verb-headed’ here means that the construction imposes verbal status on the ICV of which it forms a part; in Construction Grammar terms, the be-construction has in its syn-component the cat-value V (compare Michaelis and Ruppenhofer 2001: 61). Our diagrammatic representations as in Figure 3 follow the Goldbergian formalism, and suggest that the prefix construction is the only linking construction at work, whereby all of the verb’s participants are linked to grammatical functions. By contrast, Michaelis and Ruppenhofer (2001: 51–63) follow Kay and Fillmore’s formalism and distinguish the prefix construction from the Subject construction, the

be-’s constructional properties in its conservation

5.

6.

7.

8.

165

Oblique theme construction, the Active construction, and the Passive construction. It should be kept in mind, then, that our diagrams represent the be-construction as co-occurring with an Active construction (with which it probably, due to its collostructional entrenchment, forms a cognitive chunk, and which, in turn, justifies Goldberg’s way of representing them). If it co-occurred with a Passive construction, there would be no Agent argument, the Location argument being linked to the syntactic function of SUBJ instead. For the sake of clarity, we have not provided representations of these Passive + prefix constructions. Still, in particular cases, a syntactic specification of an open-class (lexical) element can be overridden by that of a schematic (non-lexical) construction (a phenomenon known as coercion, see Michaelis 2005: 49). This happens for instance when a noun or adjective unifies with the be-construction, which is verb-headed. An example of such a secondary predicate is kept ... open in He kept the doors open, which is embedded within the primary predicate kept the doors in the sense that open is predicated of the direct object of this primary predicate doors. Of course, the prefix be- itself results from a more prototypical type of grammaticalization change, whereby a lexical item had developed into a (morpho)syntactic one – a development which had probably already been completed in Proto-Germanic. Within a framework that sticks to a kind of subcategorization frame of the lexical entry, such as the PAS-structure proposed by Booij (this volume), the argument structure of the be-construction can even be said to override the valency frame of the simplex verb. While we adhere to Goldberg’s distinction between participants of a verb and arguments of a construction (e.g., the transitive construction), entrenchment of verb + (in)transitive construction is likely to exert an influence on the cognitive perception of the verbal lexical entry (on similar frequency effects, see e.g. Bybee 2001; 2003). Unfortunately, this issue is beyond the scope of our present research.

References Bechler, Karl 1909 Blom, Corrien 2004

Das Pr¨afix to im Verlaufe der englischen Sprachgeschichte. K¨onigsberg: Hartungsche.

On the diachrony of complex predicates in Dutch: Predicative and non-predicative preverbs. Journal of Germanic Linguistics 16: 1–75. Booij, Geert and Jaap van Marle (eds.) 2003 Yearbook of morphology 2003. Dordrecht: Kluwer. Booij, Geert this volume Constructional idioms as products of linguistic change: the aan het + infinitive construction in Dutch.

166

Peter Petr´e and Hubert Cuyckens

Brinton, Laurel J. 1988 The development of English aspectual systems. Aspectualizers and post-verbal particles. Cambridge: Cambridge University Press. Bubenik, Vit 2004 On the evolution of the Hittite system of postpositions from Proto-Indo-European. Paper presented at the 17th International Conference of Linguists, Prague, July 24–29 (http://iling.spb.ru/pdf/des/028bub.pdf [11.01.2008]). Bybee, Joan 2001 Phonology and language use. Cambridge: Cambridge University Press. 2003 Mechanisms of change in grammaticization. In: Brian D. Joseph and Richard D. Janda (eds.), The handbook of historical linguistics, 602–623. Oxford: Blackwell. Croft, William 2000 Explaining language change: An evolutionary approach. Harlow/Essex: Longman. 2001 Radical construction grammar: Syntactic theory in typological perspective. Oxford: Oxford University Press. 2003 Lexical rules vs. constructions: A false dichotomy. In: Hubert Cuyckens, Thomas Berg, Ren´e Dirven and Klaus-Uwe Panther (eds.), Motivation in Language 49–68. Amsterdam: John Benjamins. Declerck, Renaat Forthc Distinguishing between the aspectual categories ‘(a)telic’, ‘(im)perfective’ and ‘(non)bounded. Submitted to Studia Linguistica. De la Cruz, Juan 1975 Old English pure prefixes: structure and function. Linguistics 145: 47–81. De Smet, Hendrik 2005 A corpus of Late Modern English texts. ICAME Journal 29: 69–82 (http://icame.uib.no/ij29/ij29-page69–82.pdf [19.07.2007]). Dewell, Robert B. 1996 The separability of German u¨ ber-: A cognitive approach. In: Ren´e Dirven and Martin Puetz (eds.), The construal of space in language and thought, 109–133. Berlin: Mouton de Gruyter. Fillmore, Charles J., Paul Kay and Mary Catherine O’Connor 1988 Regularity and idiomaticity in grammatical constructions: The case of let alone. Language 64: 501–538. Fischer, Olga, Ans van Kemenade, Willem Koopman and Wim van der Wurff 2000 The syntax of early English. Cambridge: Cambridge University Press. Goldberg, Adele 1995 Constructions: A construction grammar approach to argument structure. Chicago/London: University of Chicago Press.

be-’s constructional properties in its conservation 2006 Hiltunen, Risto 1983

167

Construction at work: The nature of generalization in language. Oxford: Oxford University Press.

The decline of the prefixes and the beginnings of the English phrasal verb: the evidence from some Old and early Middle English texts. Turku: Turun Yliopisto. Himmelmann, Nikolaus P. 2004 Lexicalization and grammaticization: opposite or orthogonal. In: Bj¨orn Wiener, Walter Bisang and Nikolaus Himmelmann (eds.), What makes grammaticalization? A look from its fringes and its components, 21–42. Berlin: Mouton de Gruyter. Hopper, Paul J. 1975 The syntax of the simple sentence in Proto-Germanic. The Hague: Mouton. Hopper, Paul J. and Elizabeth C. Traugott 2003 [1993] Grammaticalization. Cambridge: Cambridge University Press. Kastovsky, Dieter 1992 Semantics and vocabulary. In: Cambridge history of the English language, Volume 1: The beginning to1066, 290–408. Cambridge: Cambridge University Press. Kay, Paul and Charles Fillmore 1999 Grammatical constructions and linguistic realizations: The what’s X doing Y ? construction. Language 75. 1–33. Kemmer, Suzanne and Martin Hilpert 2005 Constructional grammaticalization in the make-causative. Paper presented at the workshop on Construction and Language Change, held at the 17th International Conference on Historical Linguistics (ICHL), Madison, WI, 31 July-5 August. Lakoff, George 1987 Women, fire, and dangerous things. What categories reveal about the mind. Chicago: University of Chicago Press. Lehmann, Winfred P. 1986 A Gothic etymological dictionary. Leiden: E.J. Brill. Lenze, Josef 1909 Das Pr¨afix bi- in der altenglischen Nominal- und Verbalkomposition mit gelegentlicher Ber¨ucksichtigung der anderen germanischen Dialekte. Kiel: Fiencke. Los, Bettelou 1999 Infinitival complementation in Old and Middle English. Den Haag: Thesus. Michaelis, Laura A. 1994 A case of constructional polysemy in Latin. Studies in language 18: 45–70.

168

Peter Petr´e and Hubert Cuyckens 2005

Entity and event coercion in a symbolic theory of syntax. In: Jan-Ola ¨ Ostman and Mirjam Fried (eds.), Construction grammars. Cognitive grounding and theoretical extensions, 45–88. Amsterdam: John Benjamins. Michaelis, Laura A. and Josef Ruppenhofer 2001 Beyond alternations. A constructional model of the German applicative pattern. Stanford: CSLI Publications. OED2: 1989 The Oxford English Dictionary. Second Edition, XX vol. Prepared by J.A. Simpson – E.S.C. Weiner. Oxford: Clarendon Press. Ogura, Michiko 1995 The interchangeability of Old English verbal prefixes. Anglo-Saxon England 24: 67–93. Petr´e, Peter 2006 The prefix be-/bi- as a marker of verbs of deception in late Old and early Middle English. Belgian Journal of English Language and Literatures. New Series 4: 109–127. Traugott, Elizabeth C. this volume The Grammaticalization of NP of NP Patterns. van Kemenade, Ans and Bettelou Los 2003 Particles and prefixes in Dutch and English. In: Geert Booij and Jaap van Marle (eds.), Yearbook of morphology 2003, 79–117. Dordrecht: Kluwer. Watkins, Calvert 1964 Preliminaries to the reconstruction of Indo-European sentence structure. In: Horace Lunt (ed.), Proceedings of the 9th international congress of linguists, 1035–1042. The Hague: Mouton.

Corpora used BNC World: The British National Corpus, version 2 2001 Distributed by Oxford University Computing Services on behalf of the BNC Consortium. (http://www.natcorp.ox.ac.uk/ [25.01.2008]). CLMET: Corpus of Late Modern English texts. Hendrik De Smet. Leuven: Linguistics Department (http://perswww.kuleuven.be/∼u0044428/ [19.07.2007]). HC: Helsinki corpus of English texts. Diachronic part (ICAME, version 2) 1999 Matti Rissanen et al. Helsinki: Department of English. PPCME2: Penn-Helsinki parsed corpus of Middle English, 2nd edition. Anthony Kroch. Pennsylvania. (http://www.ling.upenn.edu/hist-corpora/ [19.07.2007]).

be-’s constructional properties in its conservation

169

The Tagged LOB Corpus 1976 Stig Johansson, Geoffrey Leech et al. Bergen: Available for distribution through ICAME, The Norwegian Computing Centre for the Humanities (http://icame.uib.no/cd/ [17.07.2007]). YCOE: The York-Toronto-Helsinki parsed corpus of Old English prose 2003 Ann Taylor et al. York: Department of Language and Linguistic Science. Wulfila Project. Tom De Herdt. Antwerp: University of Antwerp (http://www.wulfila.be [19.07.2007]). YPC: York-Helsinki parsed corpus of Old English poetry 2001 Susan Pintzuk and Leendert Plug. York: Linguistics Department.

Negative verbal clause constructions in Puyuma: exploring constructional disharmony Malcolm Ross 1. Introduction Historical linguists are typically interested in change and in what causes change, but the advent of cognitive and usage-based approaches to language has brought with it a complementary interest in the stability which results from the “resistance to change by high frequency instances of constructions” (Bybee 2005). Constructional stability has two dimensions. A construction is a form–meaning pairing, and we may talk about stability of form and stability of meaning. Stability of meaning entails the maintenance over time of distinctions among constructions.1 Stability of form entails the maintenance of structures in circumstances where, for example, analogical levelling might otherwise be expected to bring about structural change.2 This paper is concerned with an example of stability of form in Puyuma, an Austronesian language of Taiwan.3 Out of four basic Puyuma clause constructions – intransitive, transitive, negative intransitive and negative transitive – the first three show clear constructional relationships to one another, but the last, the negative transitive construction, is ‘disharmonic’ in relation to the other three. That is, it has a structure other than the one we would predict on the basis of the first three. I explore how this disharmony arose, and show that changes in the transitive and negative intransitive constructions brought about a realignment of structural relationships among the four constructions, leaving the (conservative) negative transitive construction high and dry. There are no written records of earlier stages of Puyuma, but we can infer that both the rise of disharmony and its retention are due to frequency effects. Finally, I suggest that because of the messiness of the structural relations among constructions generally, they are better modelled as a network than a hierarchy. Within a network we can track both the semantic and the structural relationships among constructions and show where meaning and structure do not match. It is widely accepted among Austronesianist linguists that Proto Austronesian was spoken on the island of Taiwan, and that Austronesian languages were carried to their present locations across the Philippines, mainland Southeast Asia, the Indo-Malaysian archipelago, Madagascar, New Guinea, Melanesia, Micronesia and Polynesia by a series of migrations which probably originated

172

Malcolm Ross

in a single migration out ofTaiwan.The major linguistic ground for this inference is Blust’s (1977, 1999) phylogenetic tree of Austronesian, based on shared innovations, relative to a carefully reconstructed Proto Austronesian, which places Austronesian languages in ten primary subgroups, nine of them consisting of the 14 Austronesian languages still spoken on Taiwan (the ‘Formosan languages’) and one containing all the thousand or so Austronesian languages spoken outside Taiwan.4 Puyuma is the sole member of one of the nine Formosan groups and thus a first-order subgroup of Austronesian in its own right.

2. Puyuma verbal clause constructions Typologically, Puyuma is what Himmelmann (2002: 11) calls a “Philippinetype” language, i.e. it has what Philippinists refer to as a “focus” system. However, I prefer a less opaque analysis (Ross and Teng 2005b; cf also Reid and Liao 2004). The so-called ‘actor focus’ construction is here simply labelled ‘intransitive’, since it is not formally distinct from the intransitive construction. The ‘non-actor focus’ constructions (alias ‘patient focus’, ‘locative focus’ and ‘instrument/beneficiary focus’) are the basic transitive constructions, occurring in three applicative-like variants labelled TR1, TR2 and TR3. The verb is thus always marked for transitivity. The morphemes of verb forms occurring in this paper are presented in Table 1, where M-represents a morpheme which surfaces in different forms with different lexical verbs (infix , me-, m- or zero) and Ca-represents reduplication in which C is the root-initial consonant and the vowel is -a-.

Table 1. Summary of selected Puyuma verbal morphology TRANSITIVE INTRANSITIVE Realis

M -ROOT

Realis progressive M -Ca-ROOT Irrealis

Ca-ROOT

Negative

(as non-negative)

TR1 ROOT-aw

TR2 ROOT-ay

TR3 ROOT-anay

Ca-ROOT-aw Ca-ROOT-ay Ca-ROOT-anay Ca-ROOT-i ROOT-i

Ca-ROOT-an ROOT-an

Using the S, A and P labels of Comrie (1978), S and P appear in the nominative case. Thus in the intransitive clause in (1) S is in the nominative case, marked here by the case-marker i:5

Negative verbal clause constructions in Puyuma

(1)

173

sa-senay i walegan prog-sing nom:s Walegan ‘Walegan is/was singing.’

In a transitive clause the A appears in the ‘oblique’ case but is coreferenced in the verb phrase by a ‘genitive’ proclitic, while the P is nominative.6 (2)

tu=tusuk-aw na Lutung kann isaw gen:3=pierce-tr1 nom:def monkey obl:s Isaw ‘Isaw speared the monkey.’

Schematically the intransitive and transitive constructions in (1) and (2) may be represented respectively as in (3) and (4).7 The schema in (3) says that the intransitive construction consists of an intransitive verb ItrV and an optional (hence parenthesised) nominative noun phrase nomNP encoding S (superscripted). The schema in (4) says that the transitive construction consists of a transitive verb TrV which is (i) optionally followed by a nominative noun phrase nomNP encoding P and (ii) preceded by a genitive proclitic gen= encoding A and coreferential with an optional oblique noun phrase oblNP, also encoding A, which follows the verb complex. (3) (4)

ItrV (nomNP S ) genA i =TrV (nomNPP)(oblNPA i )

Examples (1) and (2) also show that a third person S or P is not encoded on the verb.8 When the S or P is first or second person, however, it is encoded by a nominative enclitic pronoun, as in (5) and (6). (5)

sa-senay=ku prog-sing=nom:1s ‘I am/was singing.’

(6)

tu=dirus-aw=ku kann walegan gen:3s=bathe-tr1=nom:1s obl:s Walegan ‘Walegan washed me.’

In schematic form, the two constructions are: (7) (8)

ItrV=nom:1/2S genA i =TrV=nom:1/2P (oblNPA i )

174

Malcolm Ross

In the negative intransitive construction, corresponding to the non-negative construction in (1) and (3), the negator aDi precedes the verb, which retains its intransitive form. If there is a nominative enclitic, it is attached to aDi, as in (10). (9)

aDi senay i walegan neg sing nom:s Walegan ‘Walegan isn’t/wasn’t singing.’

(10) aDi=ta par-ka-inaba’ neg=nom:1ip recip-ka-good ‘We won’t reconcile.’ The four intransitive constructions are tabulated in schematic form in (11). The two rows marked ‘Intransitive’ replicate the intransitive constructions in (3) and (7), with respectively a third-person and a first-or second-person S. The two rows marked ‘Negative intransitive’ give the corresponding negative constructions exemplified in (9) and (10). (11) Intransitive: ItrV (nomNPS) Negative intransitive: aDi ItrV (nomNPS) Intransitive: ItrV=nom:1/2S Negative intransitive: aDi=nom:1/2S ItrV The negative transitive constructions introduce an unexpected twist to the constructional paradigm. The nominative enclitic appears from (11) to be a second position clitic, and from these constructions and the non-negative transitive constructions in (4) and (8), replicated as the first and third rows of (12). One might expect to be able to predict the negative transitive constructions as shown in the two rows of (12) that are marked ‘Negative intransitive’ – but one can’t, as the asterisks indicate. (12) Transitive: (Negative transitive:) **aDi Transitive: (Negative transitive:) **aDi=nom:1 /2

genA i =TrV (nomNPP) genA i =TrV (nomNPP) genA i =TrV=nom:1/2 P genA i =TrV

(oblNPA i ) (oblNPA i ) (oblNPA i ) (oblNPA i )

Instead, the negative transitive constructions are as exemplified in (14) and the actual constructional paradigm is as in (13): (13) Transitive: genA i =TrV (nomNPP) Negative transitive: aDi genA i =NegTrV (nomNPP) Transitive: genA i =TrV=nom:1/2P Negative transitive: aDi genA i =NegTrV=nom:1/2P

(oblNPA i ) (oblNPA i ) (oblNPA i ) (oblNPA i )

Negative verbal clause constructions in Puyuma

175

(14) a. aDi tu=dirus-i na gung kan walegan obl:s Walegan neg gen:3s=bathe-tr1/tr2:neg nom ox ‘Walegan didn’t wash the ox.’ b. aDi tu=dirus-i=ku kan walegan neg gen:3s=bathe-tr1/tr2:neg=nom:1s obl:s Walegan ‘Walegan didn’t wash me.’ In the negative transitive the negator aDi precedes the verb as expected, but transitive negation differs from intransitive negation in two respects. First, there is a special negative transitive verb form with a suffix -i tr1/tr2 or -an tr3 which otherwise marks the irrealis, as shown in Table 1. Second, if there is a nominative enclitic marking P, it does not occur on the negator but remains on the verb, as in (14b). In order to highlight the constructional disharmony between the negative transitive construction on one hand and the non-negative intransitive, negative intransitive, and non-negative transitive on the other, (15) brings together the constructional variants with first-and second-person nominative enclitics (I ignore the third-person nominative variants, as they can be predicted from the first-/second-person variants, but not vice versa.) The upper four rows of (15) are the lower two rows of (11) (intransitive) and the lower two rows of (13) (transitive). The ‘disharmonic’ constituents of the negative transitive construction in the fourth row are shown bolded. For comparison’s sake, the double-asterisked ‘expected’(harmonic) negative transitive construction from (12) is shown below the line. (15) Intransitive: Negative intransitive: Transitive: Negative transitive: aDi ‘Expected’ negative transitive

ItrV=nom:1/2S aDi=nom:1/2S ItrV genA i = TrV=nom:1/2P (oblNPA i ) NEGTRV =NOM: 1/2P (oblNPA i ) genA i =

**aDi=nom:1/2P genA i =

TrV (oblNPA i )

3. Constructional disharmony In the various frameworks proposed for construction grammars (e.g. Fillmore, Kay and O’Connor 1988, Kay and Fillmore 1999, Croft 2001), a construction like the negative transitive is typically accounted for as the child of two parent constructions, a transitive construction and a negative construction. One would expect the negative construction to be a parent of both, the negative intransitive and the negative transitive constructions. But for Puyuma, because of the dif-

176

Malcolm Ross

ferences between them, one cannot posit a single parent construction to account for the two constructions with aDi in (15).9 This disharmony raises two questions. The first, quite simply, is: Why have Puyuma speakers retained this disharmony, rather than bringing the negative transitive construction into line with the other constructions and creating the ‘expected’ but non-occurring construction in (15)? A possible answer is offered by a number of recent studies which show that high-frequency constructions are less prone to change than low-frequency ones. Although I am not in a position to demonstrate this on the basis of a corpus study, it is close to certain that the negative transitive is a high-frequency construction in Puyuma, and has remained disharmonic because of this. This takes us to the second question: How did this disharmony arise in the first place? This is quite easy to answer, as the reconstruction of Proto Austronesian has progressed substantially in the past quarter-century. I offer an answer in § 4.

4. The history of Puyuma verbal clause constructions The disharmony of the Puyuma negative transitive is quite easily accounted for. For some pre-Puyuma stage (between Proto Austronesian and modern Puyuma) we can reconstruct the constructional paradigm shown in (16). The presence of the question-mark against the free pronoun encoding the P in the transitive construction will be addressed when the reconstruction is briefly justified in § 5. (16) Pre-Puyuma: Intransitive without auxiliary: Intransitive with auxiliary: aux=nom:1/2S Transitive without auxiliary: Transitive with auxiliary: aux=genA

ItrV=nom:1/2S atempItrV TrV=genA freeP (?) AtempTrV=nom:1/2P

Third-person enclitics do not occur in modern Puyuma and cannot be satisfactorily reconstructed in Proto Austronesian (Ross 2006: 38), implying that pre-Puyuma also had none. Hence the nominative enclitic is specified as first-/ second-person in (16). At some stage Puyuma innovated third-person genitive (but not nominative) enclitics. The negator in Proto Austronesian and presumably in pre-Puyuma, as in other Formosan and Philippine languages, was one of a set of clause-initial auxiliary morphemes which mainly encoded tense or negation. The verb which followed the auxiliary assumed a form which I have dubbed the atemporal (Ross 1995, 2002).10 In Puyuma the plain atemporal is reflected in the negative transitive

Negative verbal clause constructions in Puyuma

177

verb form (NegTrV in 15; see also Table 1), and nowhere else.11 The intransitive atemporal has been lost in Puyuma, probably because its form was simply root. I noted earlier that the realis intransitive morpheme M -has zero manifestation with certain classes of verbs (e.g. beray ‘give’), so that their realis intransitive form was also root, and this was certainly the case in Proto Austronesian and pre-Puyuma. As a result, for these classes of verbs there was never a distinction between realis and atemporal intransitives. The distinction was then lost from verbs where M -is manifested (e.g. senay‘sing’ came to serve as both realis and atemporal). When the A and P of the transitive construction were both represented by pronominals and there was no auxiliary, I assume that the A occurred as a genitive enclitic attached to the verb and the P as a free pronoun, as in Tsou. An alternative would be to assume that they occurred as an enclitic sequence after the verb (TrV=genA =nomP ). I return to this in § 5. As (16) shows, if there was an auxiliary (including the negator), an enclitic pronoun was attached to the auxiliary. If the clause was transitive and had two pronominals, then this was the genitive (A) enclitic. The P was then encoded as a nominative enclitic attached to the verb. Tracking the rise of the disharmony step by step, we find that it has arisen because of (i) the loss of the atemporal form in the negative intransitive and (ii) a change in the ordinary (non-negative) transitive construction. These have, so to speak, left the more conservative negative transitive construction stranded. (There are no differences between the pre-Puyuma and Puyuma intransitive constructions, except that Puyuma no longer has a distinct atemporal intransitive verb form.) Step 1 In (17) the Puyuma and pre-Puyuma versions of the negative transitive construction from (15) and (16) are compared. The order of elements in the Puyuma negative transitive is the same as it was in pre-Puyuma, but the genitive clitic has changed its host. In pre-Puyuma it was an enclitic on the negator; now it is a proclitic on the verb. (17) (Negative transitive): Pre-Puyuma: aux =genA ATempTrV=nom:1/2P (from 16) (oblNPA i ) (from 15) Puyuma: aDi genA i = negTrV=nom:1/2P

A second change is implicit in (17). In pre-Puyuma, as in other Formosan languages, an A noun phrase was marked with genitive case and a genitive pronominal clitic occurred only when there was no A noun phrase (the construction without an A noun phrase is shown in (17)). In modern Puyuma, the A is always coreferenced by a genitive proclitic, i.e. the clitic is there whether or not the

178

Malcolm Ross

A (oblique) noun phrase (oblNP A i ) is present. Thus the function of the clitic has changed from anaphoric pronoun to agreement marker. This change was almost certainly associated with the change in agent noun phrase case-marking from genitive to oblique, as the genitive agreement marker serves to distinguish an oblique-marked A from other oblique-marked noun phrases (Ross and Teng 2005a, 2005b). Step 2 It was perhaps the shift in the function of the clitic that led to the crucial change in the non-negative transitive construction, shown in (18). Apparently by analogy with the negative transitive, the genitive clitic also became a verbal proclitic in the non-negative transitive, leaving the postverbal slot free for occupation by a nominative enclitic encoding the P, previously encoded by a free pronoun. (18) (Transitive): (from 16) Pre-Puyuma: trv =genA freeP (?) A Puyuma: gen i = trv =nom:1/2P (oblNPA i ) (from 15)

Step 2 One might protest that the non-negative construction would have greater frequency than the negative construction, and that it is therefore unlikely that the non-negative would be remodelled on the negative. However, pre-Puyuma non-negative clauses probably occurred far more frequently with an auxiliary than without one, i.e. they occurred in the same construction as negative clauses. Indeed, in the Tsou language which neighbours Puyuma, all declarative clauses begin with an auxiliary. Hence one might suppose that the change in the Puyuma transitive construction should more accurately be modelled as in (19), i.e. the genitive clitic changed its host from auxiliary to verb, then remained in this position even when there was no auxiliary, as in (18) above. This is essentially the analysis of Starosta, Pawley and Reid (1981).13 However, the atemporal verb form was lost in the Puyuma non-negative transitive construction, suggesting that the processes in (18) and (19) both played a role in its history. (19) (Transitive): Pre-Puyuma: aux =genA atempTrV=nom:1/2P Puyuma: genA i = TrV=nom:1/2P (oblNPA i ) The Puyuma transitive constructions provide a small object lesson in how an easily explained change can give rise to constructional irregularity. The prePuyuma constructions with pronominal arguments shown in (16), repeated as (20), can be said to be harmonic with one another insofar as there are systemic relationships among their structures.

Negative verbal clause constructions in Puyuma

(20) Pre-Puyuma: Intransitive without auxiliary: Intransitive with auxiliary: aux=nom:1/2S Transitive without auxiliary: Transitive with auxiliary: aux=genA

179

ItrV=nom:1/2S atempItrV TrV=genA freeP (?) atempTrV=nom:1/2P

Setting aside generalisations which apply to both pre-Puyuma and Puyuma constructions (i.e. that they are predicate-initial, that the auxiliary/negator if any precedes the verb, and that a verbal clause always has a subject S/P pronoun and, if it is transitive, also a genitive A pronoun), the relationships among the structures in (20) can be expressed as prose statements like those in (21). These generalisations would be expressed as parent constructions in construction grammar formalism. (21) a. If there is an auxiliary, the verb has atemporal form. b. There are maximally two enclitic pronoun slots (which must be filled if there are pronouns to fill them): one after the auxiliary (if any) and one after the verb: i. if there is an auxiliary enclitic slot, it is filled first; ii. if there is a genitive A pronoun (i.e. the verb is transitive), it takes precedence over the subject pronoun and is always cliticised. The statement in (22) can be added for the sake of completeness, but it is not a generalisation across constructions: (22) If an enclitic slot remains available after (21b-ii), the subject pronoun is cliticised as a nominative enclitic (i.e. in the transitive-with-auxiliary construction). Otherwise it is manifested as a free pronoun (i.e. in the transitive-without-auxiliary construction). The Puyuma constructions in (15) are repeated as (23). (23) Intransitive: aDi=nom:1/2S

Negative intransitive: Transitive: Negative transitive: aDi

ItrV=nom:1/2S ItrV genA i = TrV=nom:1/2P(oblNPA i ) genA i = NEGTRV =NOM:1/2P (oblNPA i )

If we attempt to make a set of generalisations across these constructions parallel to the Pre-Puyuma generalisations in (21), the result is more limited: (24) There are two clitic pronoun slots: a. if there is a genitive A clitic (i.e. the verb is transitive), it is proclitic to the verb and functions as an agreement marker; b. the subject pronoun is always a nominative enclitic.

180

Malcolm Ross

To describe the constructions in (23) more accurately, we need to add the statements in (25), but these each apply only to a single construction. (25) a. In the negative transitive construction the verb has the negative transitive form. b. In the negative intransitive construction the nominative enclitic is attached to the auxiliary; otherwise it is attached to the verb. That is, any statement about verb form or clitic position applies to only one of the two negative constructions, because of the disharmony among the Puyuma constructions. We now know that the disharmony arose because the genitive clitic switched host, resulting in a change in its position in non-negative transitive clauses, and because the atemporal intransitive form was lost in Puyuma. Ironically, the negative transitive construction, which underwent minimal change, is the construction which appears disharmonic in the modern language.

5. Supporting evidence In § 4 I accounted for modern Puyuma constructions on the basis of a prePuyuma reconstruction. The objective of this section is to offer some support for that reconstruction. It does not otherwise further advance the argument of § 4. Pre-Puyuma represents an interstage sometime between Proto Austronesian and modern Puyuma. However, the pre-Puyuma reconstruction differs from Proto Austronesian at just one possible point, the positioning of enclitics in transitive clauses. As (26) indicates, it is possible that when A and P pronominals both occurred in a Proto Austronesian clause, they formed an enclitic sequence. (26) Proto Austronesian: (Intransitive) ItrV=nom:1/2S atempItrV (Negative intransitive) TrV=genA =nom:1/2P (?) (Transitive) aux=genA =nom:1/2P (?) atempTrV (Negative transitive)

aux=nom:1/2 S

The outlines of the reconstruction above were provided by Starosta, Pawley and Reid (1982), although much of the supporting data is found only in the unpublished Starosta, Pawley and Reid (1981). Proto Austronesian verbs in clauses without auxiliaries are morphologically largely identical to nominalisations, and Starosta, Pawley and Reid (1982: 148) suggest that (i) verbs in Formosan languages are derived diachronically from nominalisations and (ii) the reanalysis

Negative verbal clause constructions in Puyuma

181

of nominalisations as verbs had hardly begun in Proto Austronesian. There is no particular reason to assume (ii), and it is possible that (i) is incorrect (Ross 2002: 39–50). I will assume here that the Proto Austronesian forms were already verbs, as this makes no difference to the history of Puyuma and it simplifies presentation. Detail relevant to the morphological reconstruction of Proto Austronesian is provided by Ross (1995, 2002). Proto Austronesian is reconstructed on the basis of data from members of the ten primary subgroups of Austronesian, and it is assumed that a feature that is present in three or more non-contiguous subgroups can be reconstructed for Proto Austronesian, unless there is conflicting evidence. There is general agreement in the literature that Proto Austronesian had a set of clause-initial auxiliaries which encoded negation, tense and perhaps aspect (Starosta, Pawles and Reid 1982: 149). These are preserved in a number of modern languages, but in Puyuma only the negator aDi survives. For example, in Squliq Atayal, a Formosan language of northern Taiwan, a range of auxiliaries are found, followed by an atemporal verb form. These auxiliaries include the negator ini/ in (27a), what Egerod (1969) calls the ‘actual’ auxiliary si in (27b) and the prohibitive laxi in (27c).14 (27) Squliq Atayal: a. ini/=sami kac-i na/ mqu/ neg=nom:1ep bite-pvce2:atemp genA snake ‘We have not been bitten by snakes.’ (Egerod 1966: 354) b. si=nha/ sr/ag-i ma ai actual=gen:3p go.along-pvce2:atemp it.is.said interjection ‘They were following (the river).’ (Egerod 1969) c. laxi z/-i snonan=maku/ isu/ prohibitive forget-pvce2:atemp message=gen:1s free:2s ‘You must not forget my message.’ (Egerod 1966: 358) Tsou, a Formosan language of central Taiwan, has an auxiliary in ‘every’ independent clause, as in (28). As a consequence, there is no contrast between atemporal and independent verb forms, as the latter have been lost. (28) Tsou: a. m-i=/o mo-fi su avce-punc=nom:1s avce-give free:2s ‘I gave to you.’ (Tsuchida 1976: 97) b. os=ko eobak-a n-a/o pvce.past=gen:2s hit-pvce1 nom-free:1s ‘I was hit by you.’ (Starosta 1985)

182

Malcolm Ross

c. is=/o fi-i na-suu pvce.punc=gen:1s give-pvce2 nom-free:2s ‘You were given (it) by me.’ (Tsuchida 1976: 97) Further afield, a construction similar to that in Squliq Atayal occurs in Cebuano Bisayan in the central Philippines: (29) Cebuano Bisayan: walaq=ku hug´as-i ang manga pl´atu neg=gen:1s wash-pvce2:atemp nom p plate ‘I didn’t wash the plates.’ (Wolff 1996: 26) The question of whether a sequence of two enclitics could occur in Proto Austronesian, as shown with a query in (26), is not so easily resolved. Happily, the answers are not particularly relevant to the history of Puyuma sketched in § 4, but I will give summary answers in order to show their low relevance. In a range of Formosan and Philippine languages the enclitics are second position, i.e. Wackernagel position, clitics: they are attached to the first prosodic word in the clause, whether this is a verb, an auxiliary or an adverb, and this was probably also true in Proto Austronesian (Starosta, Pawley and Reid 1982). In the Formosan languages of the Atayalic subgroup (Atayal, Seediq; e.g. Pecoraro 1979: 68, Huang 1994: 28–36), Kavalan (Li 1978: 590), and Bunun (Huang et al. 1999), and in a number of Philippine languages (Billings and Kaufman 2004), A and P clitics may be attached in sequence. In Kavalan, Bunun, the Central Luzon subgroup (Kapampangan, Sambal, Bolinao,Aita) and Mamanwa (Central Philippines), the genitive precedes the nominative. In other languages one or the other ordering is preferred, but not obligatory, and in some languages a monosyllabic pronoun precedes a disyllabic (Billings and Kaufman 2004).15 If Proto Austronesian independent verb forms were indeed derived from nominalisations, then the Proto Austronesian use of the genitive case for the A argument probably reflects its earlier possessor relationship to the nominalisation (Starosta, Pawley and Reid 1981), and one may therefore infer that the genitive enclitic would have been bound directly to the newly reanalysed independent verb form. It would follow from this that, if there were enclitic sequences this early, the earliest order of enclitics would have been genitive–nominative. However, we do not know how long before the stage we call Proto Austronesian, this reanalysis occurred (nor with certainty whether it ever occurred), and so the genitive–nominative order was not necessarily rigidly maintained in Proto Austronesian. Despite these caveats, however, it is noteworthy that certain portmanteau enclitic forms inAtayal (Starosta 1988), Seediq (Pecoraro 1979: 67–68) and Paiwan (a Formosan language of southern Taiwan) (Egli 1990: 156–157,

Negative verbal clause constructions in Puyuma

183

296–297) reflect an earlier genitive–nominative sequence. This, combined with the evidence mentioned in the previous paragraph, suggests on balance that (i) Proto Austronesian had enclitic sequences and (ii) their ordering was genitive– nominative. And if they had become second-position clitics, then they would also have occurred in sequence after an auxiliary, as shown in (26). There are, however, a number of languages that do not allow enclitic sequences. One is Tsou, where the A is encoded as a genitive enclitic but the P appears as a free pronoun, as in (28b) and (28c). This is the construction reconstructed for the pre-Puyuma transitive in (20). As I indicated in § 4, it is tempting instead to reconstruct an enclitic sequence in the pre-Puyuma transitive. However, this would result in a mismatch between the pre-Puyuma nonnegative transitive, with an enclitic sequence, and the negative transitive, with only a single enclitic. If the enclitics are indeed second-position enclitics, then we would expect the enclitic sequence to occur both after the verb (in the non-negative transitive construction) and after the auxiliary (in the negative transitive construction). But this would be incompatible with the very probable sequence of events reconstructed in (17) and (19), which requires that there was only one enclitic attached to the auxiliary. Accordingly, I infer that no enclitic sequence occurred in either transitive construction, and that pre-Puyuma resembled Tsou in this regard. If Proto Austronesian did indeed have enclitic sequences, then they were lost in some of its daughter-languages, including pre-Puyuma and Tsou (and if Proto Austronesian did not have enclitic sequences, then it must have had a set of constructions like those reconstructed for pre-Puyuma).

6. Conclusions and concluding thoughts I return now to the matter of constructional disharmony in Puyuma. This disharmony had two contributory causes: – the genitive clitic switched host, resulting in a change in its position in nonnegative transitive clauses; – the atemporal intransitive form was lost. Both these changes can be attributed to accidental frequency effects. The frequency of pre-Puyuma transitive clauses with an aux=genA sequence was high, and this sequence often immediately preceded the verb (cf 19). This led to the re-association of the clitic with the following transitive verb and also by analogy to the loss of the post-verbal genitive enclitic from the rarer construction in which transitive clauses had no auxiliary (cf 18). The result was a disharmonic situation whereby in an intransitive clause the auxiliary took a nominative enclitic but in a transitive clause no enclitic was attached to it. The subsequent

184

Malcolm Ross

(presumably) gradual loss of auxiliaries other than the negator led to the modern Puyuma disharmony where the negator aDi takes a nominative enclitic in the intransitive, but not in the transitive. The loss of the atemporal intransitive form was due to the lack of a morphological contrast between the realis intransitive and the atemporal intransitive forms of certain high-frequency verbs like beray ‘give’, and the consequent loss of the separate atemporal intransitive form from (lower-frequency?) verbs that had one. This, and the fact that the negator remained as the only auxiliary, left a special negative verb form in the transitive. The seemingly disharmonic construction in the modern language – the negative transitive – is in fact more conservative than either the negative intransitive or the non-negative transitive. However, this fact alone is of limited interest. Of greater interest is what these developments imply about relationships among constructions. The schematic representations in this paper have employed the somewhat informal style of Croft (2001). This has been more than a matter of convenience, as I accept the basic tenets of Croft’s Radical Construction Grammar, and especially his insistence that construction grammar be non-reductionist, i.e. that the categories which occur in a construction are defined by the construction itself (Croft 2001: 54–55). This accounts for my use of, for example, the separate categories ItrV and TrV (instead of V ‘verb’), since they are in a meronomic (part–whole) relation to intransitive and transitive constructions respectively.15 SUBJ PRED

ITRSUBJ ITRV

I slept

TRSUBJ TRV TROBJ

I saw John

Figure 1. A (n incomplete and rudimentary) taxonomy of English clause constructions

Croft uses tree diagrams to represent taxonomic hierarchies among clause constructions. Figure 1, after Croft (2001: 56), is incomplete and rudimentary and intended only as a graphic representation of the principle of inheritance. Like other construction grammarians, Croft claims that a sentence may (and will normally) have multiple parent constructions in a taxonomic hierarchy, such that

Negative verbal clause constructions in Puyuma

185

each parent is a partial specification of a daughter construction, as illustrated in Figure 2 (2001: 26). ITRSUBJ ITRV

SUBJ AUX-n’t V

I didn’t work Figure 2. Multiple parents in a construction taxonomy

The architecture of this model posits a cognitive organisation in which constructions higher in the hierarchy are abstractions from those below. This architecture works well in many instances, but, as I have tried to show in § 3, it does not work for the Puyuma constructions in (15). The relationships among the intransitive and transitive constructions and their negative counterparts are more fragmentary, and at best form a rather complex cognitive network with rather messy organisation and partial, overlapping abstractions. To quote Bybee and Hopper (2001: 3): The notion of a language as a monolithic system has had to give way to that of a language as a massive collection of heterogeneous constructions, each with affinities to different contexts and in constant structural adaptation to usage. . . Relationships among constructions can thus better be modelled as a network than a hierarchy. Fragments of a network can be presented graphically in a format like that used by Bybee (1985: 124, 130 and 1995) to present morphological relationships.16 Figure 3 applies this format to the constructions represented above in Figure 2. Here, there are no parent constructions, but simply relationships among tense/aspect constructions and between negative constructions and their non-negative counterparts. If the relationship is one where the difference in meaning/function is encoded as a systematic difference in form (as with the encoding of tense/aspect differences by various auxiliaries and an appropriate verb form or the encliticisation of n’t to the auxiliary), the relationship is regarded as constructional and represented in Figure 3 by an unbroken line. Where the difference in meaning/function is encoded by a non-systematic difference in form, i.e. is disharmonic, the relationship between the two constructions is treated as one of meaning/function only and represented by a broken line. Such a relationship is the one between I worked and I didn’t work, where the systematically predicted form of the negative construction is I worked not, a form

186

Malcolm Ross

which was of course lost during the seventeenth century in all except archaising contexts (Denison 1993: 458).

Figure 3. A fragment of a constructional network (some possible links not indicated)

This representational format is applied in Figure 4 to the constructions in (16). The relationships among the four constructions are formally more complex than those among the English constructions in Figure 3 but they are systematic. Intransitive constructions have an intransitive form of the verb, transitive constructions a transitive form and a genitive enclitic encoding the agent. Negative (and other auxiliary) constructions have a negative auxiliary (which attracts an enclitic) and an atemporal form of the verb. When this format is applied to present-day Puyuma in Figure 5, however, the picture changes. Since there are only two pairs of constructions along each

Negative verbal clause constructions in Puyuma

187

Figure 4. A network representation of basic Pre-Puyuma verbal constructions

dimension (intransitive/transitive and non-negative/negative), I have made the (arbitrary) decision to treat the relationships between the two non-negative and the two intransitive constructions as systematic, reflecting (12). The structurally most complex construction, the negative transitive, is treated as non-systematic. If I had decided to treat this construction as systematic, then the negative intransitive construction would have to be treated as non-systematic. In principle the result would be the same. The non-negative intransitive construction has an intransitive form of the verb, the non-negative transitive construction a transitive form and a genitive agreement proclitic encoding the agent. The negative intransitive construction has a negative auxiliary which attracts the nominative enclitic. The negative transitive construction, however, is non-systematic, as we saw earlier. The negative auxiliary does not attract the nominative enclitic, and the verb assumes a special negative form. The most important point to be made with regard to Figure 5 is that the network format affords us a rough graphic representation of the constructional relationships in Puyuma, whereas it is very difficult to present them in the form of a taxonomic hierarchy with parent constructions of the kind represented in Figures 1 and 2. The network model thus provides a better representation of constructional relationships than the taxonomic hierarchy model, and is probably a more realistic cognitive model of these relationships. The fact that the dishar-

188

Malcolm Ross ITRV =NOM:1/2S senay =ku ‘I sang’ n tr

NEG =NOM:1/2 S ITRV aDi =ku senay ‘I didn’t sing’

A

TRV =NOM:1/2 P tu= dirus-aw =ku ‘s/he washed me’

GEN =

tr n

NEG GEN A= TRV:NEG =NOM:1/2 P aDi tu= dirus-i =ku ‘s/he didn’t wash me’ Key tr

relationship between transitive and intransitive constructions, with formal differences underscored.

Otherwise see Figure 3.

Figure 5. A network representation of basic Puyuma verbal constructions

mony among the Puyuma constructions probably has quite a long history also supports a network model. If constructions are stored as a taxonomic hierarchy, one would expect one of the negative constructions to have been brought into line with the other three constructions, but it hasn’t.

Abbreviations Pronouns are marked in interlinear glosses by an abbreviation of the form X:nY, where X is nom(inative), gen(itive) or free, n marks person (1, 2, 3), and Y is s(ingular), p(lural), ep exclusive plural or ip inclusive plural. Other abbreviations are: atemp atemporal; avce A voice; itr intransitive; neg negator, negative; obl oblique; p plural; prog progressive; punc punctiliar; pvce P voice; recip reciprocal; s singular; tr transitive. < x > indicates that x is an infix. In schematic representations of constructions, nom and gen represent nominative and genitive clitic pronouns, free a free pronoun, nomNP and oblNP

Negative verbal clause constructions in Puyuma

189

nominative and oblique noun phrases, whilst superscripted S, A and P are Comrie’s labels. Parentheses indicate elements that a noun phrase may or may not occur.

Notes 1. In Oceanic Austronesian languages, for example, the maintenance of a distinction between location constructions encoding semantic and pragmatic definiteness (Ross 2007). 2. Obviously there are other kinds of formal change in constructions, e.g. the replacement of morphemes in an existing construction (Heath 1997, 1998, Ross 2007). 3. The Puyuma data used in the paper were provided by Stacy Fang-Ching Teng. The data are drawn from the large corpus of Puyuma oral texts that she has collected, transcribed and analysed. I am very grateful to her for allowing me to use her data and analysis here. Earlier accounts of aspects of Puyuma are Cauquelin (1991), Huang (2000), Tan (1997), Tsuchida (1980). 4. Perhaps two or three times as many Austronesian languages were spoken on Taiwan before the arrival of the Spanish, Dutch and Chinese in the seventeenth century. 5. Abbreviations are at the end of the paper. 6. Given the S/P alignment of Puyuma, one might say that the language is ergatively aligned. Note, however, that the would-be antipassive verb form in the example below has no special marking and the ‘antipassive’ construction is identical to the intransitive construction in (1), apart from the fact that an oblique indefinite P occurs (unlike the oblique A, it is not coreferenced on the verb). tusuk i isaw Da Lutung pierce nom:s Isaw obl:indef monkey ‘Isaw speared a monkey.’ The terms ‘absolutive’ and ‘ergative’ may seem more appropriate than ‘nominative’ and ‘genitive’. I have retained the latter as they are commonly used in descriptions of Formosan and Philippine languages, and the genitive also encodes possessors. 7. The theory behind schemas of the kind used here is touched on briefly in § 6. 8. This is also true in the absence of a noun phrase S or P. Thus ‘S/he is/was singing’ is simply sa-senay. 9. A more detailed examination of disharmonies between clause-level constructions in Puyuma is provided in Ross and Teng (2005a). 10. In languages which preserve it, the atemporal is the verb form used (i) after an auxiliary, (ii) as a coordinate dependent verb, and (iii) as an imperative. 11. A reduplicated reflex of the atemporal serves as the Puyuma irrealis. 12. Wolff (1996) offers a somewhat different analysis to account for a parallel change in the Austronesian languages of Sulawesi. He suggests that fronting (topicalisation) of the nominative pronoun in intransitive clauses reinforced the auxiliary + pronoun pattern, and that the genitive pronoun was then fronted by analogy with the nomi-

190

13.

14.

15.

16.

Malcolm Ross native. This process seems to presuppose that the fronted pronouns were not clitics, however. The glosses ‘A voice’(Avce and ‘P voice’(Pvce) are used in these examples, rather than the itr and tr of the Puyuma examples, as the functions of these constructions with regard to transitivity are not as clear-cut as in Puyuma. In some languages non-third person precedes third person, but this is not relevant here as Proto Austronesian and pre-Puyuma both lacked third-person nominative enclitics. In this light, my representations may appear inconsistent, in that I have used the term nomNP in both constructions, rather than, say, Itrsubject and Trsubject. However, this is informality rather than inconsistency, since I have distinguished nomNPS from nomNPP , and nomNP represents the taxonomic relationship in form between the parts of the two constructions (Croft 2001: 55–56). An anonymous reviewer asks about the status of such networks and suggests that there is an element of arbitrariness in their construction. I view the network primarily as a presentational device. I also assume that it represents a fragment of a cognitive network, but that this representation is very approximate. Hudson (2007: 153–157) suggests that constructions may be modelled as hierarchies within a network, but this work came to my notice just before the deadline for the final submission of the present paper.

References Billings, Loren and Daniel Kaufman 2004 Towards a typology of Austronesian pronominal clisis. In: Paul Law (ed.), Proceedings of the Eleventh meeting of the Austronesian Formal Linguistics Association, 15–29. Berlin: Zentrum f¨ur Allgemeine Sprachwissenschaft. Blust, Robert A. 1977 The Proto-Austronesian pronouns and Austronesian subgrouping: a preliminary report. University of HawaiiWorking Papers in Linguistics 9(2): 1–15. 1999 Notes on Pazeh phonology and morphology. Oceanic Linguistics 38: 321–365. Bybee, Joan L. 1985 Morphology: a study of the relation between meaning and form. Amsterdam: John Benjamins. 1995 Regular morphology and the lexicon. Language and Cognitive Processes 10: 425–455. 2005 The impact of use on representation: grammar is usage and usage is grammar. Unpublished ms.

Negative verbal clause constructions in Puyuma

191

Bybee, Joan L. and Paul J. Hopper 2001 Introduction to frequency and the emergence of linguistic structure. In: Joan L. Bybee and Paul J. Hopper (eds.), Frequency and the emergence of linguistic structure, 1–24. Amsterdam: John Benjamins. Cauquelin, Josiane 1991 The Puyuma language. Bijdragen tot de Taal-, Land-en Volkenkunde 147: 17–60. Comrie, Bernard 1978 Ergativity. In: Winfrid Lehmann (ed.), Syntactic typology, 329–394. Austin: University of Texas Press. Croft, William 2001 Radical Construction Grammar: Syntactic theory in typological perspective. Oxford: Oxford University Press. Denison, David 1993 English historical syntax: verbal constructions. London: Longman. Egerod, S¨oren 1966 Word order and word classes in Atayal. Language 42: 346–369. 1969 The origin of headhunting: an Atayal text with vocabulary. Bulletin of the Institute of History and Philology, Academia Sinica 39: 291–325. Egli, Hans 1990 Paiwangrammatik. Wiesbaden: Otto Harrassowitz. Fillmore, Charles J., Paul Kay and Mary Catherine O’Connor 1988 Regularity and idiomaticity in grammatical constructions: the case of let alone. Language 64: 501–538. Heath, Jeffrey 1997 Lost wax: abrupt replacement of key morphemes in Australian agreement complexes. Diachronica 14: 197–232. 1998 Hermit crabs: formal renewal of morphology by phonologically mediated affix substitution. Language 74: 728–759. Himmelmann, Nikolaus P. 2002 Voice in western Austronesian: an update. In: Fay Wouk and Malcolm Ross (eds.), The history and typology of western Austronesian voice systems, 7–16. Canberra: Pacific Linguistics. Huang, Lillian M. 1994 The syntactic structure of Wulai and Mayrinax Atayal: a comparison. Paper presented to the Seventh International Conference on Austronesian Linguistics, Leiden. Huang, Lillian M. [Hu´ang M˘eij¯ın] 2000 B¯ein´anyu c¯ank˘ao y˘uf˘a [Puyuma reference grammar]. Taipei: Yuanliu. Huang, Lillian M., Elizabeth Zeitoun, Marie M. Yeh, Anna H. Chang and Joy J. Wu 1999 A typological overview of pronominal systems of some Formosan languages. In: Hsu Wang, Feng-fu Tsau and Chin-fa Lien (eds.), Pro-

192

Malcolm Ross ceedings of the Fifth International Conference on Chinese Linguistics, 165–198. Taipei: Crane Publishing Co.

Hudson, Richard 2007 Language networks: The new Word Grammar. Oxford: Oxford University Press. Kay, Paul and Charles J. Fillmore 1999 Grammatical constructions and linguistic generalizations: the What’s X doing Y? construction. Language 75(1): 1–33. Li, Paul Jen-kuei 1978 The case-marking systems of the four less-known Formosan languages. In: Lois Carrington and S.A. Wurm (eds.), Second International Conference on Austronesian Linguistics: Proceedings, 569– 615. Canberra: Pacific Linguistics. Pecoraro, Ferdinando 1979 El´ements de grammaire taroko, pr´ec´ed´es de la pr´esentation de la vie et de la culture des taroko. Paris: Association Archipel. Reid, Lawrence A. and Hsiu-chuan Liao 2004 A brief syntactic typology of Philippine languages. Language and Linguistics 5: 433–490. Ross, Malcolm 1995 Reconstructing Proto Austronesian verbal morphology: evidence from Taiwan. In: Paul Jen-kuei Li, Dah-an Ho, Ying-kuei Huang, Chenghwa Tsang and Chiu-yu Tseng (eds.), Austronesian studies relating to Taiwan, 727–791. Taipei: Institute of History and Philology,Academia Sinica. 2002 The history and transitivity of western Austronesian voice and voicemarking. In: Fay Wouk and Malcom Ross (eds.),The history and typology of western Austronesian voice systems, 17–62. Canberra: Pacific Linguistics. 2006 Reconstructing the case-marking and personal pronoun systems of Proto Austronesian. In: Henry Y. Chang, Lillian M. Huang and Dahan Ho (eds.), Streams Converging into an Ocean: Festschrift in Honor of Professor Paul Jen-kuei Li on His 70th Birthday, 521–564. Taipei: Institute of Linguistics, Academia Sinica. 2007 Two kinds of locative construction in Oceanic languages: a robust distinction. In: Jeff Siegel, John Lynch and Diane Eades (eds.), Language description, history and development: Linguistic indulgence in memory of Terry Crowley, 281–295. Amsterdam: John Benjamins. Ross, Malcolm and Stacy Fang-ching Teng 2005a Clause constructions in Nanwang Puyuma. Concentric: Studies in Linguistics 31: 119–158. 2005b Formosan languages and linguistic typology. Language and Linguistics 6: 739–781.

Negative verbal clause constructions in Puyuma

193

Starosta, Stanley 1985 Verbal inflection versus deverbal nominalisation in PAN: the evidence from Tsou. In: Andrew Pawley and Lois Carrington (eds.), Austronesian linguistics at the 15th Pacific Science Congress, 281–312. Canberra: Pacific Linguistics. 1988 The case for Lexicase: An outline of Lexicase Grammatical Theory. London and New York: Pinter. Starosta, Stanley, Andrew Pawley and Lawrence A. Reid 1981 The evolution of focus in Austronesian. Paper presented at the Third International Conference on Austronesian Linguistics. 1982 The evolution of focus in Austronesian. In: Papers from the Third International Conference on Austronesian Linguistics, 2: Tracking the Travellers, 145–170. Canberra: Pacific Linguistics. Tan, Cindy Ro-lan 1997 A study of Puyuma simple sentences. M.A. thesis, National Taiwan Normal University, Taipei. Tsuchida, Shigeru 1976 Reconstruction of proto-Tsouic phonology. Tokyo: Institute for the Study of Languages and Cultures of Asia and Africa. 1980 Puyuma-go (Tamarakao hoogen) goi: fu gohoogaisetsu oyobi tekisuto [Puyuma (Tamalakaw dialect) vocabulary: with grammatical notes and texts]. In: Kuroshio Bunka no Kai [Black Current Cultures Committee] (ed.), Kuroshio no minzoku, bunka, gengo [Ethnology, cultures and languages along the Black Current], 183–307. Tokyo: Kadokawa Shoten. Wolff, John U. 1996 The development of the passive verb with pronominal prefix in western Austronesian languages. In: Bernd Nothofer (ed.), Reconstruction, classification, description: festschrift in honor of Isidore Dyen, 15– 40. Hamburg: Abera.

Borrowed rhetorical constructions as starting points for grammaticalization Marianne Mithun Viewing constructions as a locus of grammatical change can often help us understand seemingly inexplicable synchronic structures. Here we consider a grammatical pattern that appears at first highly unmotivated, a typological peculiarity. It involves the marking of syntactic relations not on predicates (heads of clauses), nor on arguments (dependents), but on words that need bear no syntactic relation to either. Even more surprisingly, this odd pattern crops up in neighboring languages that are not genetically related. We thus have two puzzles: Why does such a structure exist in the first place, and how could it be borrowed? Answers emerge once we identify the construction from which the modern grammatical pattern is apparently descended, a common discourse pattern which apparently evolved into a conventionalized rhetorical strategy before crystallizing into the morphological structure we see today. These developments illustrate the importance of considering whole constructions as potential units of change, complete with their prosody and discourse uses. Furthermore, a dynamic view of constructions as entities that evolve over time can further our understanding of the kinds of grammatical change that may be triggered by language contact. A structure that is shared by neighboring languages need not have been borrowed in its current form; it may have developed in both the source and target languages from an earlier borrowed construction. Such processes are illustrated here with material from two language families indigenous to the Northwest Coast of North America: Wakashan and Tsimshianic.*

1. The structure In discussions of the marking of syntactic relations, a distinction has been drawn between syntactic heads and dependents (Nichols 1986). At the clause level, the head is defined as the predicate, and the dependents as other constituents such as subjects or objects. In head-marked clause structures, grammatical relations are marked on the predicate (the head), often with pronominal affixes. An example of a prototypical head-marking structure can be seen in the Mohawk sentence in (1). The roles

196

Marianne Mithun

of the woman and the boy are identified solely on the predicate by a pronominal prefix -honwa- ‘she/him’. There is no indication of syntactic function on the noun for ‘boy’. (1)

Mohawk (Iroquoian, Quebec): Mae Niioronhi`a:’a Montour, speaker p.c. head dependent Sa-honw´a-hser-e’ ne raks`a:=’a. repetitive.factual-fi.sg.agent/m.sg.patient-follow-prf the m.sg-child=dim she followed him back the boy ‘She followed the boy back.’

In dependent-marked clause structures, syntactic relations are marked on arguments (the dependents). In the Mongolian clause in (2), relations are indicated by case markers on the nouns for ‘basket’ (comitative), ‘pear’ (accusative), and ‘bicycle’ (dative/locative). (2)

Khalkha Mongolian: Erdenebaatar Erdene-Ochir, speaker p.c. dependent dependent dependent Tegeed ter xuuxed ter neg sags-tai liir-iig n’ then that child that one basket-com pear-acc poss dependent head duqui-nd-aa aˇc-aad bicycle-loc-rfl.poss load-seq ‘Then that child loaded the pears in the basket onto his bicycle and . . . ’

An intriguing pattern of marking occurs in some languages indigenous to the Northwest Coast of North America. The examples below are from a dialect of Kwakw’ala, a language of the Wakashan family spoken on northern Vancouver Island and the adjacent British Columbia mainland. The language is documented in Boas (1911a, 1947), Boas and Hunt (1902–5) (where it is called Kwakiutl, after the term for one group of its speakers), Anderson (1984, 1992, 2005), and elsewhere. The sentence in (3) suggests a typical head-marking pattern, with the predicate ‘be deep’ carrying a nominative case marker for the role of the subject ‘hole’. (3)

Kwakw’ala: Boas cited in Anderson (2005: 90) head dependent w´nq´la=ida xw ´p’a be.deep=common.nominative hole ‘The hole was deep.’

Borrowed rhetorical constructions as starting points for grammaticalization

197

The sentence in (4) appears at first to continue the head-marking pattern: the head (the predicate ‘eat raw’) carries a nominative marker indicating the role of the dependent (the subject ‘Indians’). The location of the accusative marker for the object ‘gooseberry’ is surprising, however: it is attached to the subject ‘Indians’. (4)

Kwakw’ala: Boas cited in Anderson (1992: 34) head dependent dependent k’´lxk’axa-/ax.a=ida bak’ w ´ma=x. t’´mxw alin eat.raw-also=common.nominative Indian=common.accusative gooseberry ‘The Indians also eat raw gooseberries.’

The sentence in (5) is again surprising. The oblique marker =s, which identifies the role of the instrument ‘stick’, is attached to the object noun ‘dog’. (5)

Kwakw’ala: Anderson (1984: 34, 2005: 13) head dependent w y´lk mas=ida b´g w an´ma=x.a cause.hurt=common.nominative man=common.accusative dependent dependent ’watsi=sa g w ax.λ¯ ux.w dog=common.oblique stick ‘The man hurt the dog with the stick.’

These case markers are enclitic determiners, of the type described in Klavans (1985). They are bound phonologically to whatever word precedes them, regardless of its syntactic function, but they are in construction syntactically with the nominal phrase that follows. There thus seems to be a mismatch between their phonological and syntactic properties: they lean left phonologically, but right syntactically. There is no question about the word boundaries in these sentences. As Anderson notes (2005: 15), word-internal phonological processes of epenthesis operate between the enclitics and their preceding hosts, but not between the enclitics and the following nominals. Furthermore, stress in Kwakw’ala is initial, appearing on the first full vowel of words. Enclitics never carry stress. In fact nominals never appear at the beginning of utterances in topic or focus constructions, because there would be no preceding word to host their determiners. This placement of case markers is a bit surprising: we tend to expect that grammatical markers will generally be attached to elements they pertain to. Even more surprising is the fact that a similar pattern can be seen in the languages of the Tsimshianic family, spoken immediately to the north of the Wakashan

198

Marianne Mithun

languages. No genetic relationship has ever been seriously proposed between the Wakashan and Tsimshianic families. The Tsimshianic family consists of two language groups: Maritime and Interior. The examples below are from Tsimshian (also called Tsimshian Proper or Coast Tsimshian), a variety of Maritime Tsimshian. Here, too, clause structure appears at first to be head-marking, with syntactic relations marked on the predicate. In (6), the predicate ‘blow’ (the head of the clause) is marked for the role of the absolutive ‘northwind’. (6)

Tsimshian: Boas (1912) cited in Mulder (1994: 204) head dependent Da gwaant=ga ’wii yisiyaask then blow=common.absent.absolutive great orthwind ‘Then a great northwind [absolutive] blew.’

In (7), the predicate ‘kill’ (the head) is marked for the role of the ergative ‘wolf’. (7)

Tsimshian: Boas (1911b: 356) cited in Mulder (1994: 81) dependent head Dm dzakda=sga gibaw=ga fut kill=common.absent.ergative wolf=common.absent.absolutive dependent haas=ga dog=dem ‘The wolf [ergative] will kill the dog [absolutive].’

But example (7) also provides a surprise. The role of the absolutive ‘dog’ is marked not on the predicate but on the ergative noun ‘wolf’, another dependent. Like their counterparts in Kwakw’ala, these case markers are enclitic determiners. They are bound phonologically to whatever word precedes them, regardless of its syntactic function, but they are in construction syntactically with the nominal that follows. The enclitics are not pronominals. They appear only when a lexical noun phrase follows. The language does contain pronominal clitics and affixes, but they have different shapes. (8)

Tsimshian pronominal: Mulder (1994: 182) Galksaxswoosgit gisga aks galksa-xswoosg-it gi=sga aks through-dive-3 sg dem=oblique water ‘He dove into the water.’

Borrowed rhetorical constructions as starting points for grammaticalization

199

The phonological bond between the Tsimshian enclitics (called connectives in the literature on Tsimshianic languages) and their hosts is strong. Dunn reports: In hesitating and pausing, speakers always tie the connective to the preceding word, that is, they always pause after a connective. They never continue a sentence (after a pause) by starting with a connective; they may repeat the last word before the pause but never just the connective. (Dunn 1979a: 131–132)

Stebbins (2003a: 405–406) notes that word-internal processes of lenition operate over combinations of enclitics and their preceding phonological hosts, but not the enclitics and following lexical nominals. Our first challenge is to account for the existence of a structure that differs so radically from those of most other languages and seems at odds with usual expectations of iconicity. Our second is to account for its simultaneous presence in genetically unrelated languages spoken in the same geographical area. Usual explanations for parallelism seem unpromising. The probability of chance resemblance is extremely low, given the relative rarity of the pattern. The similarity cannot be a common genetic inheritance, since the languages are not related. The most obvious explanation should be language contact, since the languages are spoken in the same geographical region, once known as a strong linguistic area. Yet patterns of head and dependent marking have been hypothesized to be highly resistant to areal influence (Nichols 1992: 181). Furthermore, it is difficult to imagine how such an abstract morphological structure, so deeply integrated into the core of the grammar, could be transferred. Finally, the categories marked by the morphological structure are not even equivalent: the Kwakw’ala system shows a nominative/accusative pattern, and the Tsimshian system shows an ergative/absolutive pattern.

2. The source construction On the surface, this locus of marking seems unmotivated. But if we enlarge our perspective from the written representation of words to speech complete with prosody, and from the confines of the clause to larger stretches of discourse, we can see how a construction might come into existence that could evolve into the morphological pattern we see today. Chafe has pointed out that “Conversational language appears subject to a constraint that limits an intonation unit to the expression of no more than one new idea” (1994: 119). Speakers do not generally introduce multiple separate elements of new information within a single, prosodically unbroken stream

200

Marianne Mithun

of speech. Such prosodic organization of information characterizes spontaneous speech in all types of languages. We will illustrate it here with material from another member of the Wakashan family, Nuuchahnulth (also called Nootka). The Wakashan family consists of two major branches. Kwakw’ala, seen above, is a member of the northern branch. Nuuchahnulth, seen below, is a member of the southern branch. The Nuuchahnulth material cited here is in the Ahousaht dialect, from the speech of George Louie, recorded, transcribed, and analyzed by Nakayama (2003a, 2003b). Nakayama has produced a rich record of Ahousaht speech and has made available copies of his audio recordings. An example of the prosodic packaging of information in speech can be seen in (9), part of an account of an early encounter between local people and the crew of an English schooner called the Kingfisher. The passage is divided into prosodic sentences, defined by a final terminal prosodic contour. (Terminal contours rise in pitch in this language.) Nakayama’s free translation of each sentence is given first. The translations are followed by the original corresponding Ahousaht material, arranged so that each line represents a separate intonation unit or prosodic phrase. As can be seen, each prosodic phrase introduces just one substantial new piece of information, one new idea. In (9a), the first line presents the action, the second introduces the couple, and the third adds their origin. In (9b), having sea otter hides is packaged as a single idea: the verb ‘have’ does not contribute substantial information of its own. (9)

Ahousaht: Nakayama (2003a: 395.31–397.37), George Louie, speaker a. ‘Then a couple that came from outside the village went out to meet them [to trade].’ cˇ uu hinaaˇciλ¯ , cˇ u: hin-a:ˇciλ¯ now.then there.momentaneous-go.out.to.meet now then go out to meet hicnup, hicnup couple λ¯ ’aa/aa λ¯ ’a:/a: outside outside

histaqˇsiλ¯ . his-taq-ˇsiλ¯ there-come.from-momentaneous came from

Borrowed rhetorical constructions as starting points for grammaticalization

201

b. ‘They had sea otter hides.’ /unaak k’ w ak’ w aλ¯ . /u-na:k k’ w ak’ w aλ¯ it-having sea.otter having sea otter c. ‘My grandmother didn’t tell me [what they had for trade].’ wiiksa/iˇs /iiqh. uk yaquk w itiis wik-sa-/i:ˇs /i:qh. -uk yaq-ukw -it-i:s not-just-indicative.3 tell-dur who-poss-past-indefinite.1sg not to me narrating who was mine naniiqsu nani:qsu grandparent grandparent d. ‘She only mentioned that there was a couple on the ship.’ hicnupista, /anuwa ka/upˇsiλ¯ /anu-wa: ka/up-ˇsiλ¯ hicnup-ista only-say mention-momentaneous couple-people.on.board only said mentioned a couple on board hicnup. hicnup. couple e. ‘[When the Captain saw the couple,] he suddenly acted strangely.’ nay’iik’atwa/iˇs k w iish.ii/at. nay’i:k-’at-wa:/i:ˇs kw is-h.i:-’at at.once-shift-quotative.3 different-durative-shift at once different f. ‘As they [white men] took them [the couple] on board the ship, they took the husband and the wife to different places.’ hinaasiph. /atał, hin-a:s-ip-h.-’at-/a:ł, there.momentaneous-on.board-momentaneous.causativesimultaneous-shift-pl as they were taken on board,

202

Marianne Mithun

łuucma/i, k w isca/ap/aλ¯ ’at kw is-ca-’ap-’aλ¯ -’at łu:cma-/i˙ different-go.to-causative-finite-shift wife-definite taken to a different place the wife, w k isca/ap’at cˇ akup/i. kw is-ca-’ap-’at cˇ akup-/i˙ different-go.to-causative-shift husband-definite taken to a different place the husband g. ‘The man was suspicious.’ cˇ ’uˇsaa quu/as/i. cˇ ’uˇs-a˙ qu:/as-/i˙ suspecting-continuous person-definite suspicious the man Among the rhetorical devices observable in these texts is a pattern found in many languages of North America and elsewhere. It involves demonstratives. Ahousaht demonstratives, like those in other languages, can be used referentially on their own or associated with another nominal. The identity of the referent might be interpreted from the extra-linguistic context, previous discourse, subsequent discourse, or from the nominal. In (10), the referent of the demonstrative ‘this’ was established by the speaker’s gesture. (10) Ahousaht: Nakayama (2003a: 533.12), George Louie, speaker ‘And he did this.’ (with gesture of looking far away) /ah.. heey qw is/aλ¯ quuˇc w /ah. heey q is-’aλ¯ -qu:-ˇc do.so-finite-conditional.e-inferential this heey he used to do so this In (11) the referent was established by the previous discourse. (11) Ahousaht: Nakayama 2003a: 420.107), George Louie, speaker [Kapˇcuk was a chief and owned everything along the mountains. Young people today call it ‘Nuu-chah-nulth’ (‘all along the mountains’)] ‘Chiefs took this [the ownership within their territories] seriously.’ /iih.camisukλ¯ a/ał /i:h.-camis-uk-λ¯ a:-/ał great-thing-possessive-also-pl they take it seriously

/ah. /ah. this this

h.aw’iih.. h.aw’i:h. chiefs chiefs

Borrowed rhetorical constructions as starting points for grammaticalization

203

The sentence in (12) was the opening of a narrative. The demonstrative ‘this’ refers cataphorically to the whole event that is about to be described. (12) Ahousaht: Nakayama (2003a: 386.1), George Louie, speaker ‘This happened a long, long time ago.’ qii/aλ¯ , qi:-’aλ¯ , for.a.long.time-finite happened long ago qiit’an’aλ¯ , qi:-t’an’a-’aλ¯ for.a.long.time-slightly-finite quite awhile ago qw iyuck w i/itq qw iyu-ckw i˙/i˙tq time-done-relative.3 when it occurred

qw is qw is happen.thus happen thus

/ah. /ah. this this

[‘They sailed into the bay of Maaqtusiis. ...’] Ahousaht demonstratives can occur on their own, as in the examples above, or with other referring expressions. The demonstrative ‘this’ in (13) is associated prosodically and referentially with the following word ‘hole’. (13) Ahousaht: Nakayama (2003a: 372.178.50), George Louie, speaker ‘He only went in and out through this hole.’ /ana/is /ana-/is only-diminutive only

/ah. /ah. this this

kuh. kuh. hole hole

hiinii/aasˇsiił hini:/as-ˇsi:ł there.momentaneous-go.out-iterative goes out repeatedly

In spontaneous speech in many languages, demonstratives are sometimes followed by a pause for lexical selection or retrieval, as speakers decide how to designate a referent. In some languages, demonstratives have even become conventionalized hesitation forms, as in Japanese (Akiyo Maruyama, p.c.). In a number

204

Marianne Mithun

of languages however, including many spoken in North America, demonstratives are more often exploited for another purpose. Special constructions built on demonstratives allow speakers to manage the flow of information, particularly in more formal speech such as oratory and narrative. To avoid introducing too many distinct ideas under a single intonation contour, demonstratives can be used cataphorically, as place holders within the phrase to stand in for larger referring expressions. The demonstrative provides a signal to the listener that further specification of the referent is to come. Such a construction occurs in Ahousaht Nuuchahnulth. The first line in (14) ends in a demonstrative which points to the place identified in the following line: the chief’s house. (14) Ahousaht: Nakayama (2003a: 378.192), George Louie, speaker [‘All the way down to Neah Bay there are our relatives. We are all related, thanks to whatever happened to the young man. And thanks to women who got married and left for these places–or the chiefs who have these daughters.’] ‘When a daughter gets her husband, she leaves the chief’s place.’ cˇ apxnaakˇsiλ¯ cˇ apx -na:-ˇsiλ¯ man-having-momentaneous get husband

/ucaˇciλ¯ /u-ca-ˇciλ¯ it-go.to-momentaneous went there

hiist’iλ¯ hist-’iλ¯ there-take taken from there

/ah., /ah. this this

h.aw’ił/i, h.aw’ił-/i chief-definite h.aak w aaλ¯ . h.a:kw a:λ¯ girl

Similar patterns can be seen with other demonstratives referring to locations. (15) Ahousaht: Nakayama (2003a: 433.154) ‘He ran over there to the foot of the mountain.’ kamatqˇsiλ¯ , kamatqw -ˇsiλ¯ running-momentaneous run

Borrowed rhetorical constructions as starting points for grammaticalization

/ucaˇciλ¯ /u-ca-ˇciλ¯ it-go.to-momentaneous went to

205

h.uuł h.uuł there there

hiłh.uu/as/i. hił-h.u˙-’as-/i˙ there-in.front-on.ground-definite the foot of the mountain These multi-phrasal demonstrative structures constitute a kind of construction in the technical sense. Goldberg has defined the term ‘construction’ as below. A distinct construction is defined to exist if one or more of its properties are not strictly predictable from knowledge of other constructions existing in the grammar ... Phrasal patterns are considered constructions if something about their form or meaning is not strictly predictable from the properties of their component parts or from other constructions. (Goldberg 1995: 4)

The demonstrative construction seen in Ahousaht and elsewhere typically consists of an initial element (usually a predicate) plus a word from a small paradigmatic set (the demonstratives) in one intonation unit, followed by a referring expression in the next. This construction is distinguished from ordinary nominal expressions containing demonstratives (like ‘this hole’ above) by both form and function. Formally, the demonstrative construction has a distinctive prosodic contour, which spans two intonation units separated by a break between the demonstrative and the referring expression. Functionally, it allows speakers to manipulate the flow of information so that substantial new ideas are introduced in separate intonation units. The existence of such constructions on the Northwest Coast and within the Wakashan family itself suggests a likely source for the enclitic structures seen in Kwakw’ala. Unfortunately, there is no written documentation of earlier stages of any of these languages (as for most languages in the world), so we cannot rely on philology for attestation of individual steps in the development of the modern enclitics. We do, however, know about certain recurring processes of language change. We know that it is common for demonstratives to develop into articles over time. The link between stressed demonstratives and unstressed definite articles in German has long been noticed, as has that between English demonstratives and the definite article, and Latin demonstratives and definite articles in modern Romance languages. In a landmark work, Greenberg (1978) outlined a process by which demonstratives can evolve into obligatory definite articles that simply

206

Marianne Mithun

label referents as identifiable (Stage I), then be extended to label all specific referents (Stage II), and finally result in general noun markers, perhaps retaining gender distinctions (Stage III). Diessel (1999: 8) provides a list of works detailing the development of demonstratives into articles in European languages. Semantic distinctions encoded in the modern Kwakw’ala enclitic system suggest that the enclitics did indeed originate as demonstratives. Boas himself noted that “the pronominal forms . . . have demonstrative significance” (1911a: 529). They distinguish not only three degrees of distance, but also common from proper nominals: one set is used before common nominals, and another before proper names and indefinites. Demonstratives in many languages show distinct patterning in just these two contexts, often not even occurring in the second at all. We also know that repeated prosodic association of linguistic material can set the stage for phonological fusion over time. Heavy use of patterns like the demonstrative construction illustrated above, in which a demonstrative is pronounced in the same intonation unit as a preceding predicate but separated prosodically from a following larger referring expression, could lead to just the kinds of enclitic structures now characteristic of Kwakw’ala. The original demonstrative would fuse with the material preceding it rather than with the following nominal. It should be noted that the exact process of development outlined here is not the only one that could lead to such enclitic structures. Bybee (2002) has suggested, for example, that English auxiliary enclitics such as the future =l (I’ll go) the perfects =v and =d (I’ve gone, I’d gone) and the conditional =d (I’d go) have fused phonologically with the preceding word (typically a subject), rather than with the following word (typically the rest of the verb phrase), purely because of frequency. In these English structures the host word is most often drawn from a very small set of words (the subject pronouns), while the following word could be any lexical verb and more. The Kwakw’ala enclitics do not appear to have developed by this route, because their phonological hosts are drawn from the full inventories of predicates and discourse markers. In some other languages, the development of enclitics of this type is purely phonologically based. Booij (1996) discusses Dutch vowel-initial determiners that syllabify with the preceding word, though they depend syntactically on the following noun. Reid (2001) describes Bontok determiners that encliticize to any vowelfinal preceding word. Again the Kwakw’ala situation is slightly different: the enclitics fuse phonologically with the preceding word whatever the phonological shapes of each element.

Borrowed rhetorical constructions as starting points for grammaticalization

207

3. The transfer An understanding of the process by which the enclitic structure could develop opens the door to an explanation of how it could be transferred across genetic lines. The Northwest Coast is a well-known linguistic area, comprising a large number of genetically unrelated language families (Thompson and Kinkade 1990; Beck 2000, 2002). The area, shaded in Figure 1 below, stretches from southeastern Alaska southward through British Columbia, Washington State, and into Oregon.

Figure 1. The Northwest Coast Linguistic Area of North America Suttles 1990:iii

Languages in the Northwest Coast area show the effects of longstanding multilingualism. They share numerous structural features, among them basic predicate-initial order. This order is conducive to the development of the cataphoric demonstrative construction seen above inAhousaht. The Wakashan and Tsimshianic languages are spoken in adjacent territories in the center of the area, shown in closer detail in Figure 2. In all of the languages at the center, speakers lay out the skeleton of the clause in an initial predicate, often containing pronominal reference to core arguments. The predicate typically introduces a substantial new idea. For this reason, heavy lexical nominals are often not included in the same prosodic phrase. If the multi-phrasal demonstrative construction did not already exist independently in the various families, it could easily be transferred from one language to another by bilinguals. The sequential and prosodic arrangement of a predicate and demonstrative in one prosodic phrase, followed by a pause and a larger referring expression in a second, would be easy to copy in the target language by exploiting material already present in that language, namely initial holophrastic predicates and demonstratives. If the demonstrative construction was already available to bilinguals as a rhetorical option in both of their languages, a propensity to favor the option

208

Marianne Mithun

Tsimshianic Family ….Maritime ……..(Tsimshian = Coast Tsimshian = Tsimshian Proper, Southern Tsimshian = Klemtu)

….Interior (Nisgha = Nass = Nisqa’a , Gitksan)

Wakashan Family North Wakashan branch …...Haisla Heiltsuk-Ooweky’ala

(Heiltsuk = Haihais, Bella Bella Ooowekyala = Oowekeeno) Kwakw’ala = Kwakiutl

South Wakashan branch Nuuchahnulth = Nootka Nitinaht = Ditidat Makah

Figure 2. The Central Northwest Coast of North America Adapted from Suttles 1990:ix

could also be carried from one language to the other. Speakers accustomed to exploiting the construction heavily in one language might match this frequency of use in the other. The increased frequency would provide a necessary prerequisite for phonological fusion. A tendency toward fusion could then develop independently in each language through regular processes of grammatical change. It could also be hastened by bilinguals carrying the stylistic option of fusion from one of their languages into the other. It is becoming ever clearer that the transfer of discourse patterns through contact is not uncommon. As early as 1966, Kaplan reported that native speak-

Borrowed rhetorical constructions as starting points for grammaticalization

209

ers of Arabic, Romance languages, and Asian languages bring rhetorical structures from their mother tongues into written English. Examining the English of Mandarin and Japanese speakers, Rutherford observes that ”the transferable typologies – topic-prominence and pragmatic word order – are discourse phenomena, whereas the untransferable S, V, and O configurations are a syntactic phenomenon. I take these observations as evidence that it is therefore discourse and not syntax that gives gross overall shape to interlanguage” (1983: 368) Bartelt documents the transfer of rhetorical structures from Apache into English. He finds that “rhetorical redundancy exists in Apachean languages as a stylistic discourse feature for the expression of emphasis . . . Redundancy in English interlanguages of Apachean speakers is a result of transfer of a similar rhetorical technique in Apachean languages” (1983: 298–299) Mithun (1992) describes certain discourse patterns in Central Pomo, a language of the Pomoan family indigenous to Northern California, that are found throughout the area and beyond. One consists of a simple clause that lays out a basic message in one prosodic phrase, followed by one or more prosodic phrases, each of which provides further elaboration. Such structures are used in formal oratory, but they also appear in conversation, as in (16). (16) Central Pomo: Mithun (1992: 112), Frances Jack, speaker p.c. M´a: el ’´elˇsi:yaw’k h e. ma: el ’´elˇsi-:ya-w=’kh e land he sell-defocus-prf=future ‘The land would be sold. Dan´oma: ’el do: ’´elˇsi:yaw’k h e. dan´o=ma: ’el =do: e´ lˇsi-:ya-w=’kh e mountain=land the copula=quotative sell-defocus-prf=future They said the mountain land would be sold.’ Interestingly, the same pattern can be seen in the English of Central Pomo speakers. The passage in (17) is also from conversation. (17) Central Pomo English: Mithun (1992: 112), Frances Jack, speaker p.c. They gave them cattle. The government gave them cattle. Nice breed of cattle. A second discourse pattern common in Central Pomo and other languages throughout the Americas involves couplet constructions. Pairs of prosodically

210

Marianne Mithun

and semantically parallel lines are used to make points of importance. The second line in the pair typically provides the same information as the first but in a slightly different way, as with a shift in word order or syntactic structure, the substitution of synonyms, etc. These couplet structures are characteristic of formal oratory, but they also appear in conversation, as in the second and third lines below. (18) Central Pomo: Mithun (1992: 112–113), Frances Jack, speaker p.c. ¸t a´ whal yh´e¸taˇc’ ’el d´a:’ˇc’iw. cˇ h o´ w ¸ta´ whal yh´e-t¸a-ˇc’ ’el d´a-:’-ˇc’-w cˇ h o´ -w work do-mult.event-imprf.pl the want-rfl-imprf.pl-prf be.absent-prf ‘They don’t want to work p´e:su cˇ h o´ w ’in; p´e:su cˇ h o´ -w ’i=n money be.absent-prf be=as because there is no money; man´a¸tayt¸ammaw’k h e ¸t h ’in ’in. man´a-t¸a-ˇc-t¸a-m-a-w=’kh e ¸t h´ı-n ’=´ın. pay-mult.event-sml-mult.event-mult.agt-defocus-prf=fut be.absent-imprf be=as because they are not going to be paid.’

A similar couplet structure appears frequently in the English of these Central Pomo speakers. (19) Central Pomo English: Mithun (1992: 113), Frances Jack, speaker p.c. When the payoff came, he got some of it too; they paid him too. Such parallels call to mind what Ross terms ‘metatypy’, processes in which speakers of neighboring languages begin to reorganize their ‘ways of saying things’, which can ultimately result in the restructuring of syntax (Ross 2001: 146). It certainly seems more plausible that an abstract discourse structure, a demonstrative construction like that illustrated in Ahousaht above, was transferred through contact than that speakers pulled a fully developed inflectional system, complete with specific enclitic markers, from one language into the other, particularly when the languages are unrelated genetically.

Borrowed rhetorical constructions as starting points for grammaticalization

211

4. The markers The alert reader may have noticed some tantalizing similarities in shape between some of the Kwakw’ala and Tsimshian markers. Common Kwakw’ala enclitics are =ida and =x.a. (20) Kwakw’ala enclitics: Boas and Hunt (1905: 8.13–14) Wt¸ l´a:lae:=da g´a:lagiwa’e: well hearsay=common.nominative leader ‘Well, then it is said, the leader d´a:x.’ida’xaase:=s a’yas´o: l´a=xa took.also=instrumental hand dem=common.accusative took hold of the p´a:q’a h¨e gwe:x.s sa´o:k w t’´e:sema. flat that like board stone board-like stone with his hand.’ Common Tsimshian enclitics are =da and =ga. (21) Tsimshian enlitics: Mulder (1994: 34) a. Smga ’woomxg-a=da very be.in.pain-epenthetic=common.present.absolutive txa’nii txamoo-t=ga. all body-3 .poss=distal ‘His whole body was really in pain.’ sm’ooygit=ga. b. T’aa=ga be.sg=common.absent.absolutive chief=distal ‘There was a chief.’ A closer look reveals that the two pairs of markers represent quite different distinctions. The Kwakw’ala enclitics =ida and =g.a distinguish nominative and accusative case (for common, distal referring expressions). The Tsimshian =da and =ga elements do not mark case; they distinguish present from absent referents. TheTsimshian connective system is described in Boas (1911b), Dunn (1979a, 1979b, 1991), Mulder (1987, 1988, 1994), Stebbins (2003a, 2003b), and elsewhere. The examples seen so far are from what Mulder identifies as the oratory

212

Marianne Mithun

register, characterized by a more elaborate enclitic system than that used in everyday speech. Enclitics in the oratory register specify three kinds of distinctions: they distinguish common and proper nouns; present, absent, or unspecified referents; and ergative and absolutive case. Table 1 shows the patterns found in independent clauses (indicatives). The patterns found in dependent clauses (subjunctives), which are actually very common, are slightly different. (Boas notes that forms for present and unspecified ergative proper nouns, which correspond to blank cells in Table 1, did not occur in his corpus.) Table 1. Table 1: Tsimshian Indicative Connectives: Oratory Register Boas (1911b: 355) cited in Mulder (1994: 33) common nouns present

absent

proper nouns

unspec

present =dat

abs

=da

=ga

=a

erg

=sda

=sga

=a

absent =gat

unspec =at

=s

In fact this =da/=ga distinction cannot be reconstructed for Proto-Tsimshianic. These elements do not appear in the Interior enclitic system, where all common nouns are preceded by the enclitic =ł. They do not even occur in the Maritime everyday register, where presence and absence or degrees of distance are not distinguished. (Everyday speech does show a common ergative enclitic =da used in certain aspects.) What can be reconstructed is the general common enclitic *=ł, which persists in irrealis contexts in Maritime Tsimshian. The distribution of the =da/=g.a and =da/=ga markers suggests that they may have been transferred through contact as well, but this transfer was apparently distinct from the development of the enclitic patterns. Among the numerous features shared by languages in this region are elaborate deictic systems (Thompson and Kinkade 1990: 46–47). A visible/invisible or present/absent distinction is made in the deictic systems of languages over a large area on the Northwest Coast, including all of the North Wakashan languages (Haisla, Heiltsuk, Kwakw’ala), both subgroups of the Tsimshian family (Maritime, Interior), and the isolate Haida to the north, as well as the neighboring Salishan languages to the east and south (Bella Coola, Comox, Sechelt, Squamish, Halkomelem, Nooksack, Northern Straits, Clallam, Lushootseed, Twana, and Tillamook) and the Chimakuan languages (Chemakum, Quileute). The North Wakashan languages (Heiltsuk, Haisla, and Kwakw’ala) and Tsimshian all distinguish proximal, middle, and distal categories as well. It is not unlikely that the da and ga elements originated as demonstratives which were borrowed widely throughout the area to add spatial distinctions to the existing deictic systems.

Borrowed rhetorical constructions as starting points for grammaticalization

213

Elements with these shapes still function as demonstratives in the modern North Wakashan and Tsimshianic languages. In this capacity they are not obligatory, unlike the enclitics. The Kwakw’ala uvular =g.a is probably not comparable to the Tsimshian velar =ga. Both languages maintain a distinction between uvulars and velars that goes back to their respective parents. A distal demonstrative =ga can be seen at the end of the Kwakw’ala sentence in (22). It marks referents or events as distant in space and/or time. (22) Kwakw’ala demonstrative: Boas cited in Anderson (1984: 36) hik’ala=gada make.that.noise=proximal.invisible.common.nominative x´sala-xd=ga’ disappear-past=distal.demonstrative ‘The ones (near me, invisible) who had disappeared made that noise. Modern Tsimshianic languages also still show phrase-final or clause-final clitics =da and =ga which indicate physical or psychological proximity and distance respectively (Tarpent 1998). The distal demonstrative =ga was seen at the end of both Tsimshian sentences in (21) above. Both =da and =ga can be seen in the Southern Tsimshian examples in (23) below. (23) Southern Tsimshian deictics: Tarpent (1998) a. Hlaa ix ts’uu k’ay daaw’ihls ła: ’ix ts’u: q’ay t´a:’wł=s=t now again trying still leave=conn=proper Scottyda’a Scotty=ta’ name=proximal.demonstrative ‘Scotty [who is living in the house] has gone out again dim si’ixsihoontksit. tm si’ixs´h´o:ntkws-t future try.to.get.fish-3 to try to fish.’ b. Oo, hlaa dii lip waalptga’a. o´ : ła: ti: lip w´a:lp-t=ka’ yes now insistant self house-e=distal.demonstrative ‘Yes, they do have their own house now (over there).’

214

Marianne Mithun

Tarpent notes that the modern Southern Tsimshian =da’a marks referents that are absent but close to the speaker, including deceased relatives, while =ga’a is used for referents that are distant from the speaker, in Klemtu especially referents on the other side of the bay, and always in referring to the Christian God. The apparent similarity in shape between some of the Kwakw’ala and Tsimshian enclitics thus does not indicate that the enclitics themselves were spread through contact. It is more likely that demonstratives were borrowed, then the demonstratives developed into elements of the enclitic systems independently in the different languages.

5. The contrasting case patterns The enclitic case systems in Kwakw’ala and Tsimshian do not express the same case patterns. Evidence of the development of the different systems can still be seen within the modern languages.

5.1.

Kwakw’ala case

Among the Wakashan languages, only Kwakw’ala distinguishes core case for lexical nominals. The other North Wakashan languages show enclitics only before obliques. Boas notes, “[In Heiltsuk], nominal subject and object are defined by their position, the subject preceding the object” (1947: 298). No markers precede ‘the man’ or ‘the dog’ in (24). Only the oblique ‘binoculars’ is marked by an enclitic. (24) Heiltsuk: Rath (1981: 1.85) D´aduqvl´a w´ısm´a-x.i w’´ac’i´a-x. hi=s d´ugv´ay´ua-x.i. watch man-dem dog-dem dem=oblique binocular-dem ‘The man watched a dog with [oblique] binoculars.’ In the Haisla sentence in (25), the core arguments ‘the king’ and ‘the bear’ are similarly unmarked for case. (25) Haisla: Bach (Forthc, p.c. 2004) Keta-th h´ımas-ax.i sak-ax.i. shoot-future king-definite.absent grizzly-definite.absent ‘The king is going to shoot the grizzly bear.’

Borrowed rhetorical constructions as starting points for grammaticalization

215

The enclitic structure does not exist at all in the South Wakashan branch of the family. In the Nuuchahnulth sentence in (26), there is no marker of the syntactic function of ‘the fog bag’ or ‘the wolf tribe’. (26) Ahousaht Nuuchahnulth: Nakayama (2003b: 142), Caroline Little, speaker suk w iλ¯ /uˇcqc’uu/i qw ayuc’iikˇstaqumł, w qw ayuc’:ik-ˇstaqumł, suk iλ¯ /uˇcq-c’u:-/i: take fog-inside-definite wolf-groups take the fog bag wolf tribe ‘The wolf tribe took the fog bag pun’isaλ¯ /ał pu-n’i-‘saλ¯ -/a:ł run.in.group-downslope-on.beach-plural and ran down the beach.’ The enclitic structure does not appear in the other South Wakashan languages, Nitinaht and Makah, either. It can therefore not be reconstructed for ProtoWakashan or even Proto-North Wakashan. It appears to be a comparatively recent development within Kwakw’ala itself. The forms of the Kwakw’ala case markers are suggestive of their sources. The subject marker =ida (perhaps with segmentable deictic element i according to Anderson (1992)) looks very much like the proximal deictic element da seen throughout the area. The Kwakw’ala object markers, which all contain a uvular element =x., look very much like distal demonstratives in all of the North Wakashan languages. Though Heiltsuk and Haisla to the north do not have grammatical object marking with lexical nominals, they do contain elaborate sets of demonstratives comparable to those in Kwakw’ala, many based on uvulars. (27) Heiltsuk: Rath (1981: 88) w´ısem ‘man’ qix.w w´ısm=´a=x.i ‘that man’ (‘over there, near neither you nor me’) (28) Haisla: Bach (Forthc) genem ‘woman’ genem=ax.i ‘that woman’ It appears that the Kwakw’ala subject marker developed from a proximal demonstrative (‘this’), and the object marker from a distal demonstrative (‘that’).

216

Marianne Mithun

5.2. Tsimshianic case The Tsimshianic enclitic systems show ergative/absolutive patterning. The development of this pattern can be traced to relatively recent changes within the family as well. We know that one common route by which ergative systems can develop is through increased use of passivization (Chung 1976). When, for one reason or another, passives come to be used more frequently than actives, they may come to be interpreted as pragmatically unmarked, basic transitive constructions. Evidence of this phenomenon can be seen elsewhere in the Northwest Coast linguistic area (Mithun 2007). In the South Wakashan languages, the Chimakuan languages immediately to the south of them, and the neighboring Salishan languages immediately to the east, a tendency to cast first and second persons as subjects (points of departure of the clause) has crystallized into a grammatical requirement. Events in which a third person acts on a first or second (‘John annoyed me’) can be expressed only with a passive construction (‘I was annoyed (by John)’). The third person agent is often not mentioned at all (‘I was annoyed’), but it may be mentioned in an oblique lexical nominal (‘by John’), particularly if it is a specific, known, topicworthy participant. If original passives with oblique agents are reinterpreted as basic transitives, the grammatical markers they contain are reinterpreted as well. The original oblique marker (such as ‘by’) can be reinterpreted as an ergative marker, identifying transitive agents. The original unmarked passive subject, a semantic patient, is reinterpreted as an unmarked absolutive (matching the subjects of intransitive clauses which remain unchanged). Original passive morphology associated with the verb (‘was —ed’) may simply be interpreted as meaningless, or it may be reinterpreted as a mark of transitivity. I subject ↓ stage ii absolutive stage i

was annoy-ed passive ↓ transitive

by John. oblique ↓ ergative

Two kinds of evidence suggest that just such a reanalysis took place in Tsimshianic. The modern ergative marker =s matches the oblique marker =s. (29)

Gitksan ergative =s: Hunt (1993: 19) hlimooyitg as Kathy t John ł´mo:-y´-t-qat=s t Kathy John Kathy pr John help-transitive-3-reportative=proper.ergative pr ‘Apparently Kathy helped John.’

Borrowed rhetorical constructions as starting points for grammaticalization

217

(30) Gitksan oblique =s: Rigsby (1986: 364) yixyugu’m ga’as Joe,... Gak’ali q´-q’ali y´x-yukw-´’m qa’=s Joe rdp-upstream rdp-follow-1.pl prep=proper.oblique Joe ‘As we were going upstream to Joe’s, [...] ’ Interestingly, in Interior Tsimshianic, this ergative marking appears only with proper nominals. Identified by a marker t, proper nominals include personal names, independent personal pronouns, and demonstrative pronouns referring to people. These are exactly the kinds of agents that are more likely to be mentioned in passive constructions than less central less individuated or unidentified ones. The marker =s was generalized to all ergatives in the Maritime formal register, but it still matches the oblique =s there. Furthermore, all transitive verbs in Interior Tsimshian carry a suffix -´-, which intransitives lack. (The schwa, which can surface as -a- or -i- is preceded by an epenthetic glide y after vowels and is automatically lost before resonants.). (31) Gitksan: Hunt (1993: 29, 30, 36) a. Intransitive bax ’nii’y pax. n’i:y’ run 1.sg.abs ‘I ran’ b. Transitive ga’a’y ’niin ka’-´-y’ n’i:n see-transitive-1sg.erg 2sg.abs ‘I saw you’ This suffix is not a transitivizer. It occurs in all transitive verbs, even when the roots are already transitive in meaning, like ‘see’ above and ‘make’ below. (32) Gitksan: Rigsby (1986: 339) Jabit. cap-´-t make-transitive-3sg ‘She made it.’ It appears even when a derivational transitivizer is present.

218

Marianne Mithun

(33) Gitksan: Rigsby (1986: 344) Didaa’whlit. t´-ta:’wł-´-t transitivizer-leave-transitive-3sg ‘He took her away.’ The very different case patterns inherent in the Kwakw’ala and Tsimshianic enclitic systems, each of which can be seen to have developed relatively recently within their families, provide further support for the hypothesis that it was the abstract demonstrative construction that was borrowed, along with a propensity for its use, rather than a fully developed enclitic system.

6. Differential entrenchment Additional evidence that it was the rhetorical strategy that was borrowed comes from a comparison of the relative entrenchment of the modern enclitic patterns in the Wakashan and Tsimshianic families. As we saw in the previous section, within the Wakashan family, enclitics appear only in languages of the northern branch and only to a limited extent in those. The structure is most extensive in Kwakw’ala, where enclitics precede all lexical nominals: subjects, objects, and obliques. In Heiltsuk and Haisla, enclitics appear only before obliques. The South Wakashan languages do not show the enclitic structure at all. By contrast, the enclitic structure is well entrenched in Tsimshianic, appearing in all members of the family. Still, the systems are not identical. As noted earlier, the Tsimshianic languages fall into two groups: Maritime and Interior. Maritime Tsimshian consists of Tsimshian (also called Coast Tsimshian or Tsimshian Proper) and Southern Tsimshian (spoken at Klemtu). The two are mutually intelligible. Interior Tsimshian (or Nass-Gitksan) consists of Nisgha (or Nisqa’a, also called Nass) and Gitksan. Nisgha and Gitksan are also mutually intelligible, but as Rigsby (1989) points out, their speakers consider themselves distinct politically and culturally and refer to their speech as separate languages. Grammatical descriptions of Tsimshian are in Boas (1911b), Dunn (1979b, 1991, 1995), Mulder (1987, 1988, 1994) and Stebbins (2003a, 2003b). Southern Tsimshian is described in Dunn (1979b, 1991, 1995) and Tarpent (1998). Grammatical descriptions of Nisgha are in Boas (1911a) and Tarpent (1987, 1989, 1991) and of Gitksan in Rigsby (1986) and Hunt (1993). Maritime Tsimshian contains two enclitic systems: an elaborate one used in oratory and a simpler one used in everyday speech. Enclitics in this system distinguish ergative and absolutive case, common and proper nominals, and

Borrowed rhetorical constructions as starting points for grammaticalization

219

present, absent, or deictically unspecified referents. The enclitic system used in everyday speech is simpler. It shows the distinction between common and proper nouns, but not the deictic distinctions among present, absent, and unspecified referents. (The systems used in indicative and subjunctive clauses differ as well.) The Interior enclitic system, like that of the Maritime everyday register, shows the common/proper distinction but no deictic distinctions. The enclitic =hl [ł] precedes common nominals in any core function. (34) Gitksan common nouns: Rigsby (1986: 256, 261) a. Saa baxhl gat=gi. sa: pax.=ł kat=k´ off ran=common man=distal ‘The man ran off.’ hlgutk’ihlxwhl logom b. Hlimooyihl łkw ´-tk’iłkw =ł luq-´m ł´mo:-y´=ł help-transitive=common small-child=common old-attr ’wiigatgi ’wi:-kat=k´ big-man=distal ‘The child helped the old man.’ The enclitics that precede proper nominals distinguish syntactic function. Proper nominals are preceded by an additional marker t. Absolutive case (as on ‘John’ and ‘Mary’ below) is unmarked. (35) Gitksan proper absolutives: Hunt (1993: 23), Rigsby (1986: 260) John a. bax t pax. t John run proper John ‘John ran.’ b. Hlimooyihl hlgutk’ihlxw t Mary w w ł´mo:-y´=ł łk ´-tk’iłk t Mary help-transitive=common small-child proper Mary ‘The child helped Mary.’ Ergative proper nominals are distinguished by the enclitic =s.

220

Marianne Mithun

(36) Gitksan proper ergatives: Hunt (1993: 23, 20) John t Peter. a. Sdilis stil-´=s t John t Peter accompany-transitive=proper.erg proper John proper Peter ‘John [ergative] accompanied Peter.’ Peterhl b. Needit gups ne:-ti:=t kw up=s t Peter=ł not-contrastive=3 eat=proper.erg proper Peter=common susiit. susi:t potato ‘Peter [ergative] didn’t eat the potatoes.’ (Tarpent (1989) shows that the singular proper marker t is lost following this =s through phonological processes of consonant cluster reduction. The plural counterpart tip remains in place in the same context.) The enclitics are not limited to marking core arguments. They appear before obliques and possessors. (Tarpent (p.c.) notes that instruments are more often identified in separate clauses in spontaneous speech.) (37) Gitksan oblique: Rigsby (1986: 425) Galxsi gahlxwit ahl t’uuts’xw. ’a=ł t’u:c’-xw galxs´ kał-xw -´=t through stab-passive-transitive=3 with=common knife ‘She stabbed him right through with a knife.’ (38) Gitksan possessor: Rigsby (1986: 396) ts’uuts’ andihlguuhlxwhl ’´nt´-łkw u:łxw =ł c’u:c’ container-child=common bird ‘bird’s nest’ They also link clauses to initial focused elements. (39) Gitksan focus constructions: Hunt (1993: 21, 24) a. t Peterhl si’moogitxwu’m. t Peter=ł sim’o:kit-xw -m’ proper Peter=common chief-passive-1pl ‘Peter is our chief.’

Borrowed rhetorical constructions as starting points for grammaticalization

221

b. t Peterhl ga’as ka’´=s t t Peter=ł proper Peter=common see-transitive=proper.erg proper John John John ‘John saw Peter.’ They link relative clauses to their heads. (40) Gitksan relative clause: Rigsby (1986: 404) bahatgi Dim ’nu’whl gathl tim ’nu’w=ł kat=ł pax.-´t=k´ future die=common man=common run-subject.relative=distal ‘The man who ran will die.’ They even link clauses to certain initial tense and aspect markers. Presumably structures like the progressive below originated as complex sentences consisting of an initial motion predicate (the ancestor of the modern progressive marker) followed by a sentential complement. (41) Gitksan aspect: Rigsby (1986: 275) yukwhl maadim. yukw =ł ma:t´m progressive=common falling.snow ‘It is snowing.’ It is clear why Boas termed the enclitics ‘connectives’ rather than simply case markers: they mark general syntactic dependency. The Wakashan and Tsimshianic languages thus differ from each other and among themselves in the degrees to which enclitic patterns have penetrated their grammars. This variation provides further evidence that what was transferred was not a single, fully-developed inflectional morphological system, but rather a rhetorical construction that served as a starting point for the development of the various inflectional systems in the modern languages.

7. The nature of the contact In the best of all possible worlds, we would have detailed information concerning the nature of social relations and interaction among the various Wakashan

222

Marianne Mithun

and Tsimshianic speaking peoples over the past millennia. Such information is of course lacking, as it is for most languages of the world. We do know of extensive multilingualism in recent times. We know that Southern Tsimshian has lost speakers to Heiltsuk, and that the last speaker of Southern Tsimshian also speaks both Heiltsuk and Coast Tsimshian (Tsimshian Proper). As early as 1916, Sapir hypothesized that certain morphological similarities such as reduplicative patterns and distributives “seem to be indicative of a much earlier contact of the Tsimshian with the Kwakiutl and Salish than with the Haida and Tlingit. Such contact need, of course, not have been in precisely the same territory as now occupied by the tribes, nor need their geographical relation have been quite the same” (1951: 450). Indeed, if their locations have remained largely unchanged, the fact that the two languages with the most fully developed enclitic systems (Kwakw’ala and Maritime Tsimshian) are not spoken in contiguous areas adds further evidence that it was not the fully-formed enclitic systems themselves that were borrowed, but rather the rhetorical strategies that were their precursors. The ethnographic accounts that do exist indicate that intense contact continued into recent times. We have descriptions of contacts between all of the adjacent communities. All mention intermarriage. Community locations at contact, shown earlier in Figure 2, are repeated below. Describing marriage practices of the Tsimshian, Halpin and Seguin mention intermarriage with their neighbors to the south, the North Wakashan Haisla and Heiltsuk. All marriages were supposed to be between social equals; the children of parents of unequal rank inherited rank no higher than that of the lower-ranked parent. The social distinction between the smkik´et ‘real people’ (singular sm’´o:ket ‘chief’), that is, the chiefly families, and the liq’akik´et ‘other people’, that is those who had names of lesser rank, was maintained through intermarriage with other chiefly families, including those from Tsimshian-speaking as well as other language groups (Tlingit, Haida, Haisla, and Heiltsuk). (Halpin and Seguin 1990: 275– 276)

Describing the Haisla, Hamori-Torok mentions intermarriage with theTsimshian. Culturally the Haisla were close to the Tsimshian, especially in technology and social organization. They were the only Wakashan-speaking people with a fully developed matrilineal clan system. On the other hand, like other Northern Wakashans, they had a well developed set of secret societies. Their clan system was almost certainly Tsimshian in origin, just as the Tsimshian secret societies were probably largely Haisla in origin. (Hamoi-Torok 1990: 306)

Hilton reports that the groups to the south of the Haisla, the Heiltsuk Haihais and Bella Bella, married their Haisla and Tsimshian neighbors.

Borrowed rhetorical constructions as starting points for grammaticalization

223

Tsimshianic Family Maritime (Tsimshian = Coast Tsimshian = Tsimshian Proper Southern Tsimshian)

Interior (Nisgha = Nass = Nisqa’a , Gitksan)

Wakashan Family North Wakashan branch Haisla Heiltsuk-Oowekyala

(Heiltsuk = Haihais, Bella Bella Ooowekyala = Oowekeeno)

Kwakw’ala = Kwakiutl

South Wakashan branch Nuuchahnulth=Nootka Nitinaht = Ditidat Makah

Figure 3. The Middle North Coast of North America Adapted from Suttles 1990:ix

The Haihais gained access to resources through intermarriage and trade with the Kitasoo Tsimshian [Southern Tsimshian now at Kelmtu] and Haisla. . . . For the Haihais and Bella Bella, the crest groups were the counterpart of the exogamous matrilineal clans of the tribes to the north. Their names were similar, and their existence no doubt made intermarriage more acceptable to the northerners. (Hilton 1990: 314, 317)

In her description of the Kwakiutl (Kwakw’ala speakers), the southernmost of the North Wakashans, Codere mentions intermarriage with all of their neighbors to the north.

224

Marianne Mithun

The most dependable of such relations [friendly relations] existed with the Haisla, Haihais, Bella Bella, Oowekeeno, and Nootkans. Communication with the Haisla and Bella Bella was mostly of a social and ceremonial character and included intermarriage. (Codere 1990: 360)

Such marriages would certainly have produced bilinguals capable of replicating rhetorical constructions from one language in another. Up into recent times there was contact of another kind that could be pertinent for the transfer of the demonstrative construction that concerns us here. There is evidence of substantial borrowing of ceremonies from the North Wakashans by the Tsimshians. The secret society dances were apparently borrowed [by the Tsimshian] from the Haisla and Heiltsuk-speaking people just before contact with Europeans; they were most fully expressed among the Southern Tsimshian, who obtained them directly from the Heiltsuk speakers, and had only partially been received by the other divisions. Most of the names for the dancers are Northern Wakashan in origin. (Halpin and Seguin 1990: 279)

These traditions were passed to the Interior Tsimshian groups, the Nishga and Gitksan, along with the ceremonial language that was an integral part of them. Many Nishga and Gitksan people once spoke CoastTsimshian [in addition to their own languages], which was more prestigious, especially for ceremonial purposes. (Halpin and Seguin 1990: 267)

Heiltsuk ceremonies were apparently passed southward to the Kwakiutl as well. The other Northern Wakashan peoples, especially the Bella Bella and Oowekeeno, seem to have been the sources for many ceremonial concepts. The Kwakiutl themselves recognize these groups as the source of many of their privileges and in fact customarily insert words and phrases from the northern languages into their dance songs. The influence of the Kwakiutl and the other Northern Wakashans has been far-reaching in the ceremonial life of the coast. The Winter ceremony performances of the Haida, Tlingit, and Tsimshian have been strongly influenced if not derived from the dances of the Haisla and Bella Bella, a diffusion most clearly demonstrated by the northern use of Wakashan terms for figures that resemble those in Bella Bella and Kwakiutl ceremonial systems. (Holm 1990: 385–386)

It thus appears that many ceremonial traditions originated with the North Wakashan Heiltsuk, who passed them northward to Haisla and Tsimshian speakers, and southward to Kwakw’ala speakers. Tarpent (2000) points out that influence from Heiltsuk and Haisla on Tsimshian can be seen in lexical borrowing but not the reverse. Among the North Wakashan terms borrowed into Tsimshian she cites names of secret societies, chiefly names, and legendary names. The transfer of

Borrowed rhetorical constructions as starting points for grammaticalization

225

ceremonies, along with the associated oratory, would also be an obvious vehicle for the transfer of rhetorical constructions, especially of the type seen here, and particularly in the context of substantial bilingualism. It is perhaps significant that Maritime Tsimshian, spoken closest to the North Wakashans, contains an additional more elaborate enclitic system used for formal oratory. Of course in the absence of detailed written records from pre-contact times, it is not possible to know just how ancient such contact is, nor how the timing of the transfer of ceremonies might correlate with the transfer of demonstratives or of the conventionalized demonstrative construction.

8. Conclusion The rise and development of a multi-phrasal construction, complete with its prosodic structure and discourse uses, provides an explanation of an apparent structural anomaly in some languages of the Northwest Coast of North America. Two languages, genetically unrelated but spoken within a well-known linguistic area, show neither classical head-marking nor classical dependent marking. Their seemingly arbitrary marking pattern makes sense, however, once we uncover a rhetorical construction from which it could have developed. An understanding of the series of processes involved in the evolution of the construction over time also helps us to explain the simultaneous presence of this unusual structure in the two languages. The enclitic structure itself was probably not borrowed after all. It was apparently the precursor to its development that was borrowed, the multi-phrasal demonstrative construction. Many languages indigenous to the Northwest Coast of North America, as elsewhere, contain a special rhetorical construction that is exploited for manipulating the flow of information. The construction consists of one prosodic phrase containing a predicate and a demonstrative, followed by a pause and a second prosodic phrase containing a larger, co-referential referring expression. The predicate in the first phrase presents the skeleton of the clause. The demonstrative provides a signal that an argument will be identified in further detail in the following prosodic phrase. It would be a simple matter for bilingual speakers to carry such a construction from one language to another, exploiting comparable grammatical units (initial holophrastic predicates and demonstratives) already present in the target language. It would also be a simple matter for bilinguals to carry heavy exploitation of the construction from one language to the other. We know that longstanding conditions of intermarriage and bilingualism were present on the Northwest Coast to set the stage for such transfer. There is

226

Marianne Mithun

also ample evidence of the transfer of ceremonies among these groups, along with the associated oratory. The transfer of ceremonial speeches would be an excellent vehicle for the transfer of certain stylistic constructions used particularly often in them, and for an increase in their use. The use of such constructions might even come to be admired as characteristic of superior speech. In this case, we have substantial evidence that the parallelism in linguistic structure came about through the borrowing of a rhetorical construction rather than any fully developed inflectional enclitic structure. The modern languages differ in the extent to which the enclitic structures have developed, both through the families and through the individual languages. Within the Wakashan family in the center of the area, the probable point of origin of the ceremonies, the Haisla and Heiltsuk enclitics are used only before obliques. To the south, in Kwakw’ala, enclitics mark subjects, objects, and obliques. To the north, in the Tsimshianic family, enclitics mark not only all arguments, but also serve as general markers of syntactic dependency. The enclitic systems also differ in the nature of the categories they mark. The Kwakw’ala enclitics distinguish subjects and objects, while the Tsimshianic enclitics distinguish ergatives and absolutives. All of these details become explicable once we focus on the demonstrative construction as the object transferred and the source of development of the modern systems.

Notes *

I am grateful to Jean Mulder, Bruce Rigsby, and Marie-Lucie Tarpent, all of whom have had extensive experience working with Tsimshianic languages, for their help. The time they spent in reading earlier drafts of this work and their generosity in sharing their expertise are very much appreciated. Their comments have proven invaluable. Geert Booij has also contributed helpful comments.

References Anderson, Stephen R. 1984 Kwakwala syntax and the government-binding theory. The syntax of native American languages. Syntax and Semantics 16: 21–75. 1992 A-morphous morphology. Cambridge: Cambridge University Press. 2005 Aspects of the Theory of Clitics. Oxford: Oxford University Press. Bach, Emmon Forthc A Haisla book. Submitted to UBC press.

Borrowed rhetorical constructions as starting points for grammaticalization

227

Bartelt, Guillermo 1983 Transfer and variability of rhetorical redundancy in Apachean English interlanguage. In: Susan Gass and Larry Selinker (eds.), Language transfer in language learning, 297–305. Rowley, Massachusetts: Newbury House. Beck, David 2000 Grammatical convergence and the genesis of diversity in the Northwest Coast Sprachbund. Anthropological Linguistics 42: 147–213. 2002 Tsimshianic from a central Northwest areal perspective. Paper presented at the 37th International Conference on Salish and Neighboring Languages. Bellingham, WA, 14–16 August. Boas, Franz 1911a Kwakiutl. Handbook ofAmerican Indian languages, 423–558. Bureau of American Ethnology Bulletin 40(1). Washington D.C.: Smithsonian. 1911b Tsimshian. Handbook of American Indian languages, 283–422. Bureau of American Ethnology Bulletin 40(1). Washington D.C.: Smithsonian. 1902 Tsimshian texts. Washington: Gov. Prin. Off. 1947 Kwakiutl grammar with a glossary of the suffixes. Philadelphia: American Philosoph. Soc. Boas, Franz and George Hunt 1902–5 Kwakiutl texts. Leiden/New York: Brill. Booij, Geert 1996 Cliticization as prosodic integration, the case of Dutch. The Linguistic Review 13: 219–242. Bybee, Joan 2002 Sequentiality as the basis of constituent structure. In: Talmy Giv´on and Bertram F. Malle (eds.), The evolution of language out of prelanguage, 109–134. Amsterdam: John Benjamins. Chafe, Wallace 1994 Discourse, consciousness, and time: The flow and displacement of conscious experience in speaking and writing. Chicago: University of Chicago. Chung, Sandra 1976 On the Gradual Nature of Syntactic Change. In: Charles N. Li (ed.), Mechanism of Syntactic Change, 3–55. Austin: University of Texas Press. Codere, Helen 1990 Kwakiutl: Traditional culture. In: Wayne Suttles (ed), Handbook of North American Indians, Volume 7, 359–77. Washington, D.C.: Smithsonian Institution Press.

228

Marianne Mithun

Diessel, Holger 1999 Dunn, John 1979a

Demonstratives: form, function, and grammaticalization. Amsterdam: John Benjamins.

Tsimshian connectives. International Journal of American Linguistics 45: 131–140. 1979b A reference grammar for the Coast Tsimshian language. Ottawa: National Museums of Canada. 1991 An autolexical sketch of Sg¨uuXs ¨ (Southern Tsimshianic) predicate adjectives and predicate nominatives. International Conference on Salish and Neighbouring Languages 26: 81–96. 1995 Sm’alguax: A reference dictionary and grammar for the Coast Tsimshian language. Seattle: University of Washington/Sealaska Heritage Foundation. Greenberg, Joseph H. 1978 How does a language acquire gender markers? In: Joseph H. Greenberg, Charles A. Ferguson and Edith A. Moravcsik (eds.), Universals of human language, Volume 3, 47–82. Stanford: Stanford University Press. Goldberg, Adele 1995 Constructions:A Construction Grammar approach to argument structure. Chicago: Chicago University Press. Halpin, Marjorie and Margaret Seguin 1990 Tsimshian peoples: Southern Tsimshian, Coast Tsimshian, Nisgha, and Gitksan. In: Wayne Suttles (ed.), Handbook of North American Indians: Northwest Coast, Volume 7, 267–284. Washington: Smithsonian Institution. Hamori-Torok, Charles 1990 Haisla. In: Wayne Suttles (ed.), Handbook of North American Indians: Northwest Coast, Volume 7, 306–311. Washington: Smithsonian Institution. Hilton, Susanne 1990 Haihais, Bella Bella, and Oowekeeno. In: Wayne Suttles (ed.), Handbook of North American Indians: Northwest Coast, Volume 7, 312– 322. Washington: Smithsonian Institution. Holm, Bill 1990 Kwakiutl: Winter ceremonies. In: Wayne Suttles (ed.), Handbook of North American Indians: Northwest Coast, Volume 7, 378–86. Washington: Smithsonian Institution. Hunt, Katharine 1993 Clause structure, agreement and case in Gitksan. Ph.D. dissertation, University of British Columbia, Vancouver.

Borrowed rhetorical constructions as starting points for grammaticalization Kaplan, Robert 1966 Klavans, Judith 1985

229

Cultural thought patterns in intercultural education. Language Learning 16(1): 1–20. The independence of syntax and phonology in cliticization. Language 61: 95–120.

Mithun, Marianne 1992 The substratum in grammar and discourse. In: Ernst Hakon Jahr (ed.), Language contact: Theoretical and empirical studies, 103–116. Berlin: Mouton de Gruyter. 2007 Integrating approaches to diversity: Argument structure on the Northwest Coast. In: Yoshiko Matsumoto, David Oshima, Orrin Robinson and Peter Sells (eds.), Diversity in Language, 1–28. Stanford: CSLI (Center for the Study of Language and Information). Mulder, Jean G. 1987 Morphological ergativity in Coast Tsimshian (Sm’algyax). In: Paul Kroeber and Roger Moore (eds.), Native American Languages and Grammatical Typology, 165–185. Bloomington: Indiana University Linguistics Club. 1988 Ergativity in Coast Tsimshian (Sm’algyax). Doctoral dissertation, University of California, Los Angeles. 1994 Ergativity in Coast Tsimshian (Sm’algyax). Berkeley: University of California Press. Nakayama, Toshihide (ed.) 2003a George Louie’s Nuu-chah-nulth (Ahousaht) texts with grammatical analysis. Kyoto: Nakanishi. 2003b Caroline Little’sNuu-chah-nulth (Ahousaht) texts with grammatical analysis. Kyoto: Nakanishi. Nichols, Johanna 1986 Head-marking and dependent-marking grammar. Language 62: 56– 119. 1992 Linguistic diversity in space and time. Chicago: University of Chicago. Rath, John C. 1981 A practical Heiltsuk-English dictionary with a grammatical introduction. Ottawa: National Museums of Kanada. Reid, Lawrance 2001 On the development of agreement markers in some northern Philippine languages. In: Joel Bradshaw and Kenneth Rehg (eds.), Issues in Austronesian morphology: A focusschrift for Byron W. Bender, 235– 257. Canberra: Pacific Linguistics. Rigsby, Bruce 1986 Gitksan grammar. MS, Prepared for the Linguistics Division, British Columbia Provincial Museum.

230

Marianne Mithun 1989

Ross, Malcolm 2001

A later view of Gitksan syntax. In: Mary R. Key and Henry M. Hoenigswald (eds.), General and Amerindian Ethnolinguistics, 245– 259. Berlin: Mouton de Gruyter.

Contact-induced change in Oceanic languages in Northwest Melanesia. In: Alexandra Aikhenvald and Robert M.W. Dixon (eds.), Areal diffusion and genetic inheritance, 134–166. Oxford: Oxford University. Rutherford, William 1983 Language typology and language transfer. In: Susan Gass and Larry Selinker (eds.), Language Transfer in Language Learning, 358–370. Ann Arbor: University of Michigan. Sapir, Edward 1916 Time perspective in aboriginal American culture: A study in method. Ottawa: Government Printing Bureau. Reprinted 1951 in: David Mandelbaum (ed.), Selected writings of Edward Sapir in language, culture, and personality, 389–462. Berkeley: University of California. Stebbins, Tonya 2003a On the status of intermediate form classes: Words, clitics, and affixes in Sm’algyax (Coast Tsimshian). Linguistic Typology 7: 383–416. 2003b Fighting language endangerment: Community directed research on Sm’algyax (CoastTsimshian). Kyoto: Nakanishi. Suttles, Wayne (ed.) 1990 Handbook of North American Indians: Northwest Coast, Volume7. Washington D.C.: Smithsonian Institution. Tarpent, Marie-Lucie 1987 Between ergative and accusative syntax: lessons from Nisgha/English syntactic interference. International Conference on Salish and Neighboring Languages 22: 149–171. 1989 A grammar of the Nisgha language. Doctoral dissertation, University of Victoria. 1991 The morpheme - and the mysteries of Nisgha syntax. International Conference on Salish and Neighboring Languages 26: 317–145. 1998 The function of the Southern Tsimshian elements =ta’and =ka’and the nature of theTsimshian “connectives”. Paper presented at the annual meeting of the American AnthropologicalAssociation, Philadelphia. Thompson, Laurence C. and Marvin Dale Kinkade 1990 Languages. In: Wayne Suttles (ed.), Handbook of North American Indians: Northwest Coast, Volume 7, 30–51. Washington: Smithsonian Institution.

(De)grammaticalisation as a source for new constructions: the case of subject doubling in Dutch Gunther De Vogelaer 1. Introduction

1

While the alleged unidirectionality of grammaticalisation is a matter of ongoing debate, most grammaticalisationists still claim that ‘true’ counterexamples are rare. For instance, Haspelmath (2004) discusses some critical voices on unidirectionality, most notably those expressed in the papers in Campbell and Janda (2001), and concludes that only eight examples of ‘antigrammaticalisation’ can be distinguished. Similarly, Hopper and Traugott (2003: 137) provide only one instance of a “mirror-image reversal” of grammaticalisation. In this paper, data will be presented from southern Dutch dialects showing the rise of a construction with two strong subject pronouns out of a former combination of a subject clitic and a strong pronoun. This development will be argued to be indicative of a degrammaticalisation process that is observed in some of the dialects under investigation. In addition, the rise of this new construction has lead to a remapping of the ‘conceptual space’ of subject marking. This paper is structured as follows: after a brief discussion of the phenomenon under investigation (section 2), the data will be put in their diachronic context. Using both diachronic data (section 3.1) and a number of dialect maps based on recent data from the Syntactic Atlas of Dutch Dialects (or ‘SAND’; see section 3.2), it will be claimed that the rise of subject doubling in Dutch must be seen as a three-stage process (section 3.3). The first two stages of the process are paradigm examples of a grammaticalisation process; the third stage will be argued to be an instance of degrammaticalisation. In section 4, it will be shown that both grammaticalisation and degrammaticalisation may cause the rise of new constructions, more specifically combinations of elements that were formerly unavailable in Dutch, which have very specific pragmatic and semantic functions. Finally, the most important points will be summarised in section 5.

232

Gunther De Vogelaer

2. Subject doubling in Dutch dialects Some examples of subject doubling are shown in (1). (1)

a. Ga=de (gij) naar Brussel? go.2SG=youclitic youstrong to Brussels ‘Are you going to Brussels?’ b. Ge=gaat (gij) naar Brussel. youclitic =go.2SG youstrong to Brussels ‘You are going to Brussels.’ c. Gij gaat gij youstrong go.2SG youstrong ‘You are going to Brussels.’

naar Brussel. to Brussels

In (1a), -de is a subject clitic that occurs obligatorily in all sentences with inverted word order (i.e. subject following inflected verb). The full pronoun gij may be inserted for pragmatic or stylistic reasons. The rise of the construction in (1a) is an obvious example of grammaticalisation (see section 3), whereby a former pronoun -de (originally -di), an enclitic variant of the 2sg. pronoun gij (‘you’), is turned into a clitic or, in some dialects, even into an agreement marker on the verb. An apparently similar situation is observed in (1b), a sentence in which an obligatory preverbal clitic is combined with an optional postverbal strong pronoun. (1c) differs from (1a) and (1b): in (1c), two strong pronouns are used. In this particular case, the strong and the weak pronoun are phonological variants (gij and ge), but the same patterns are found when morphologically distinct weak and strong pronouns are available (e.g., in dialects with 1pl. me as a weak form, and wullie as the strong one). Dialects can have one, two or all three types of doubling at their disposal (see section 3 for maps). Quite unlike (1a), constructions such as (1b) and (1c) constitute a challenge to the current literature on grammaticalisation, for different reasons. Firstly, the two diachronic scenarios that are proposed in the linguistic literature for the grammaticalisation of pronouns into clitics or agreement markers (Giv´on 1976, Ariel 2000) both seem to imply that the clitic and the full pronoun must appear on the same side of the verb. In (1b) and (1c), however, the pronominal elements appear on different sides of the inflected verb. Secondly, the use of two coreferential strong subject pronouns in (1c) is unexpected. Not only does the appearance of two equal constituents with the same syntactic function violate principles such as ‘functional uniqueness’or the ‘Theta-criterion’, constructions

Subject doubling in Dutch

233

with two coreferential strong subject pronouns also seem to be rare in the languages of the world, given their complete absence in the linguistic literature. Indeed neither diachronic accounts on pronominal clitics, such as Giv´on (1976) and Ariel (2000), nor typological overviews, such as Siewierska (2004), discuss examples resembling the construction in (1c). The examples in (1) show, among other things, the importance of word order for subject doubling: (1a) is an example of a subject doubling construction in a sentence with inverted word order; (1b,c) are sentences with ‘regular’ subjectverb word order. One other parameter needs to be taken into account here, viz. the grammatical person of the subject.2 This is shown in (2): in some of the dialects under investigation, subject doubling is only attested for a limited number of grammatical persons. A case in point is the dialect of the Belgian city of Antwerp. (2)

‘Person effects’ on subject doubling in the dialect of Antwerp (source: SAND) 1sg. Ga=k (ik) = go=Iclitic (Istrong ) 2sg. Ga=de (gij) = go=youclitic (youstrong ) 3sg.m. (Gaat=em / Gaat hij) = goes heweak / hestrong 3sg.f. (Ga=se / Ga zij) = goes sheweak / shestrong 3sg.n. (Gaat=et / Gaat dat) = go itweak / itstrong 1pl. (Gaan=me / Gaan wij) = go weweak / westrong 2pl. Ga=de (gullie) = go=youclitic (youstrong ) 3pl. (Gaan=ze / Gaan zullie) = go theyweak / theystrong

In Antwerp, subject doubling is only found in the first person singular, the second person singular, and the second person plural. In all other grammatical persons, there is a complementary distribution of weak and strong pronouns: either a weak form or a strong one is used as a subject. Probably these person effects reflect different diachronic origins of subject doubling in the different grammatical persons (see section 3; cf. De Vogelaer and Neuckermans 2002). Interestingly, the person effects correlate with word order: person effects are only found for subject doubling in inverted word order (i.e., subject following verb). In sentences with regular SV-word order, the situation is much more straightforward: a dialect either has subject doubling for all pronouns, or it has no doubling at all in sentences with regular word order. Before discussing the diachrony of subject doubling, a terminological remark is in order. This article focuses on the emergence of the use of multiple pronominal subjects in Dutch dialects, and hence it is obvious that the terminology reflects whether a given element is found in a subject doubling construction or

234

Gunther De Vogelaer

not. Therefore the term ‘clitic’is reserved for the formally reduced pronouns that can appear in subject doubling constructions. Other reduced pronouns, which cannot be doubled, are termed ‘weak pronouns’, even if these elements are typically found in the immediate adjacency of a syntactic host (usually the inflected verb). This use of terminology departs from many descriptions of Dutch and German pronouns where all reduced forms are considered clitics, but it is in line with some recent pronominal typologies, in which the possibility to have more than one pronominal subject in a clause is considered an important criterion to distinguish between clitics and weak pronouns (see, e.g., Cardinaletti and Starke 1999, Siewierska 1999, Bresnan 2001, Corbett 2003).

3. The diachrony of subject doubling 3.1.

Historical sources

Turning to the diachrony of subject doubling, one immediate observation is that surprisingly few attestations are found. Both in the primary sources (the historical texts themselves) and in the secondary sources (the linguistic literature), instances of subject doubling are hard to come by. In the secondary sources,3 by far the oldest examples of subject doubling are found for the second person, in sentences with inverted word order, where the phenomenon apparently is attested as early as the Middle Dutch period (ca.1200 – ca.1500). The phenomenon is discussed briefly in Van Helten’s (1887: 282) Middle Dutch grammar, with examples such as tempteer=de ghy (literally ‘try=youclitic youstrong ’) and sie=de ghy (lit. ‘see=youclitic youstrong ’).4 The Middle Dutch examples provided by Meert (1901: 79) and Van Loey (1964: 114) are probably taken over from Van Helten, as, like Van Helten, neither Meert nor Van Loey mention the precise date and location of the instances. Vanacker (1963: 320) does provide (slightly younger) examples that are independent of Van Helten’s, the oldest one being dated in 1496. Vanacker’s oldest example, again a 2sg.-form, is shown in (3). (3)

Middle Dutch example of subject doubling (inverted word order, 2sg.) by Vanacker (1963: 320): wil=de ghy zulcke zaken doen als... (1496) want=youclitic youstrong such things do as ... ‘If you want to do things like ...’

The same study by Vanacker also gives the oldest examples of doubling of a 1sg. pronoun and a 2pl. pronoun. The earliest instances of most other types are not found until the 18th or 19th century. The relevant examples are presented in (4).

Subject doubling in Dutch

(4)

235

Oldest examples of some types of subject doubling 1sg., inverted word order: WNT (VI: 1450): Sou ‘ck ick da niet wete. (17th century) lit. ‘Should Iclitic Istrong that not know’ 1sg., regular word order: Vanacker (1963: 321): Ic en komme-‘ric niet meer inne. (1591) lit. ‘Iclitic NEGclitic come-thereclitic Istrong not in’5 3sg., regular word order: WNT (XXVIII: 359): En ze zit zou daar thuis... (ca. 1899) lit. ‘and sheclitic sits shestrong there at home’ 1pl., regular word order: WNT (XXV: 2498): me zullen wunder van d’ occasie profiteren. (1753) lit. ‘weclitic shall westrong take advantage of the situation’// 2pl., inverted word order: Vanacker (1963: 321): had-de ghijlieden eenen botram... (1628) lit. ‘had-youclitic youstrong a sandwich’

The oldest examples of doubling of a 1sg. pronoun in inverted word order are particularly instructive, as it concerns parts of 17th-century plays in which the language of the southern, mainly Brabantic immigrants in the Dutch provinces of North and South-Holland is ridiculised.6 The fact that a geographically limited construction such as subject doubling is more easily found in parodies of a certain local variety than in texts written in the relevant variety itself, is a clear indication that even in a period in which no real standard language existed, certain linguistic phenomena were carefully kept out of the written documents. Hence it comes as no surprise that for a lot of types of subject doubling, not a single historical example is found, as becomes clear in the overview in (5) (cf. the empty cells). Apart from the fact that subject doubling for the second person singular in sentences with inverted word order indeed appears to be much older than the other types, not much can be derived from (5). For a more detailed diachronic picture, it seems then that additional information is needed. In the next section, it will be shown how current data on the geographical distribution of different types of subject doubling can be used to derive the diachronic development.

236

Gunther De Vogelaer

(5) Oldest instances of subject doubling in secondary sources on older Dutch inverted word order (1a) sg. 1

17th

C. parodies (WNT VI: 1450) 2 Middle Dutch (Van Helten 1887: 282; cf. Van Loey 1964: 114, Meert 1901: 79) 3 (20th C.)

pl. 1 2 3

3.2.

(20th C.) 1628 (Vanacker 1963: 321) (20th C.)

regular word order (1b-c) 1591 (Vanacker 1963: 321) (20th C.) before 1899 (WNT XXVIII: 359) 1753 (WNT XXV:2498) (20th C.) (20th C.)

Dialect geography

3.2.1. Methodological preliminaries: the SAND-project It is a long-standing custom in dialectology to interpret geographical variants as representing different stages in a diachronic evolution. The basic principles for the interpretation of maps were already used in early 20th-century linguistics (see Bonfante 1947, Weijnen 1977 for an overview). In a more recent discussion Chambers and Trudgill (1998: 167–168) provide the following principle: “If, of two forms, one is used over a larger area than the other, then that is the older.” Of course, this principle needs to be applied carefully. For instance, comparing the distribution of two forms, or syntactic patterns for that matter, only makes sense when they are somehow linguistically related to each other. In addition, counterexamples to this principle can be found, so, in order to obtain a reliable account, the proposed diachronic relationships need to be tested against additional data, and they need to be explained. For the present purposes, it is important to remark that the proposed diachronic analysis of the dialect geographical data is fully in line with the (sparse) historical data presented in section 3.1 (see De Vogelaer and Devos Forthc for further discussion). The database that will be used in this section is the one collected by the socalled ‘SAND’-project, i.e. the Syntactic Atlas of Dutch Dialects.7 The SANDdatabase consists (mainly) of the transcriptions of 266 interviews that were held in the entire Dutch language area, which stretches over (parts of) three different countries. Of the 266 sampling points, 157 are situated in the Netherlands,8 102 in the northern, Dutch-speaking part of Belgium, and 7 in the small ‘French-

Subject doubling in Dutch

237

Flemish’ region in the northeast of France, where the older generations still speak a dialect of Dutch. For each sampling point, at least two informants were consulted, using a questionnaire that consisted of a combination of translation tasks and elicitation tasks. In addition, some 10 minutes of spontaneous speech were recorded. For the present data, it is important to note that the fieldworkers were instructed to ask the informants explicitly whether subject doubling could occur in their respective dialects. Hence the results should provide a maximal distribution of the phenomenon. 3.2.2. The geographical distribution of subject doubling: inverted word order As stated in section 1 and 2, the most important parameter for the distribution of subject doubling is word order. Therefore, the data for inverted word order and regular word order will be shown on different maps. Map 1 shows the dialects in which subject doubling is found in inverted word order.9 To provide maximal insightfulness, the map depicts three areas rather than the separate sampling points, depending on the types of subject doubling that have been found in each of the dialects under investigation. The instances of subject doubling pattern quite neatly, in three clearly distinguished areas.

Map 1. Subject doubling, inverted word order : core (all pronouns) : transitional (2sg./pl.; 1sg.)

: peripheral (2sg./pl.)

238

Gunther De Vogelaer

The most striking observation is that there are substantial distributional differences for the different grammatical persons. For the second person, both the singular and the plural, the geographical distribution is at its largest, with attestations in French-Flanders, in the entire Belgian provinces of West Flanders, East Flanders, Flemish Brabant and Antwerp, in the west of Belgian Limburg, in the entire Dutch province of North Brabant, and in a few neighbouring sampling points in Gelderland and Dutch Limburg. Hence the dialect geographical data confirm the historical finding that subject doubling has indeed arisen in the second person, in sentences with inverted word order. Other types of subject doubling are, apart from a few instances in some Dutch border towns, only found in Belgium and French-Flanders. The most productive area, or the ‘core’, is the western, Flemish dialect area, i.e. French-Flanders, and East and West Flanders, where doubling is found for all grammatical persons. In between the core area and the peripheral area, a transitional zone is found, where both second person pronouns and first person singular pronouns can be doubled. It is unclear how this transitional zone must be interpreted, especially since the historical data for the patterns attested here are too sparse to warrant conclusions. Hence the converging evidence on the relative age of the relevant subject doubling patterns is lacking here. For the present purposes, it suffices to say that in inverted word order both doubling of 1sg. pronouns (found in the transitional zone) and doubling of the other non-second-person pronouns (found in the core area) are certainly younger than doubling of 2sg./pl. pronouns. Of course, it is possible to formulate a somewhat tentative hypothesis concerning the transitional zone. The most likely explanation is that the instances of subject doubling in the first person singular are relics of an older stage in which more extensive doubling was possible. Historically, then, in the transitional zone most or even all the patterns that are found in the core must have occurred. Support for this hypothesis comes from several data. The two main arguments are the following: first, the combination of subject doubling patterns that are found in present-day dialects in the transitional zone is also found in less proficient dialect speakers in the core area. Second, the transitional zone is part of an innovative dialect area, viz. Brabant, in which quite a number of ‘old’ dialect features have disappeared, and in which some of the remaining subject doubling patterns are indeed gradually being lost (see De Vogelaer 2005: 222–224 for a more elaborate discussion). 3.2.3. Subject doubling, regular word order Map 2 shows the geographical distribution of subject doubling in sentences with regular word order.10 In contrast to map 1, there is no need to distinguish the

Subject doubling in Dutch

239

different grammatical persons; the phenomenon has the same distribution in all cases. The only parameter that needs to be taken into account, is whether a dialect always combines a weak and a strong pronoun in a subject doubling construction (cf. (1b) above), or whether it is also possible to use two strong subject pronouns in one clause, as in (1c). Unlike map 1, the distribution on map 2 does not fall apart into three partly overlapping areas, as the attestations of subject doubling with two strong pronouns seem to be scattered almost randomly over the eastern half of the construction area.

Map 2. Subject doubling, regular word order •: clitic + Verb + strong : strong + Verb + strong

The most important observation is that the first type, in which a weak and a strong pronoun are combined, is clearly the most widespread, occurring in both the Flemish dialect area, i.e. the core on map 1, and the Belgian part of the Brabantic area, i.e. the transitional zone on map 1. This leads to the conclusion that subject doubling in regular word order has a comparable age as doubling of first-person subjects in inverted word order. The data for subject doubling with two strong pronouns are somewhat more difficult to interpret, as they do not constitute a continuous area. Rather, subject doubling with two strong

240

Gunther De Vogelaer

pronouns seems to be a contact-induced phenomenon, arising in the border zone between the Belgian part of the Brabantic dialect area on the one hand, i.e. map 1’s transitional area, and, on the other hand, the Flemish core area and the eastern area without subject doubling (see section 4.3.1 for further details). It is important to note that all dialects having doubling with two strong pronouns, also have doubling with a weak and a strong pronoun, but not vice versa. Hence, it seems a necessary condition for a dialect to have doubling with a weak and a strong pronoun before doubling with two strong pronouns can arise, indicating that the former phenomenon predates the latter.

3.3. A three-stage development Given the data in section 3.1 and 3.2, it is not clear for all types of subject doubling when they have originated. Also, a number of differences in geographical distribution cannot be easily accounted for, such as some of the person effects in inverted word order. In inverted word order, doubling of 1sg. pronouns is remarkably more widespread than doubling of 3sg., 1pl. and 3pl. pronouns, but these differences are possibly quite recent. Despite these difficulties in interpreting the maps, it does seem possible to decide which type of doubling is the oldest one, and which one the youngest, especially since the diachronic and synchronic evidence converge here. Indeed both the Middle Dutch data (section 3.1) and synchronic dialect-geographical data (section 3.2) lead us to believe that doubling of second person pronouns in sentences with inverted word order (cf. example (1a)) is the oldest type of subject doubling. In addition, among the types that have been shown on the maps, doubling in regular word order with two strong pronouns (cf. example (1c)) obviously is by far the youngest type. Geographically, this phenomenon does not seem to be diffused from the core area, but patterns as a contact phenomenon occurring at the borders of the Belgian part of the Brabantic dialect area, which is a transitional zone when it comes to subject doubling. Hence the diachrony of subject doubling can be summarised as a three-stage development, as in (6).

(6)

The diachrony of subject doubling: 3 stages Stage I Stage II Stage III inverted, 2sg./pl. > -inverted, 1sg. > regular, full + V + full -inverted, other persons -regular, 1sg. -regular, other persons (all clitic + full) example: (1a) (1b) (1c) mechanism: grammaticalisation extension through degrammaticalisation analogy

Subject doubling in Dutch

241

Having established the relative chronology of the different types of subject doubling, the question remains as to how these phenomena can be explained. Most likely, the relevant changes must be attributed to different mechanisms in each of the three stages. In section 4 a more detailed account of these different stages will be provided. The rise of subject doubling in the second person in inverted word order, as in example (1a), will be argued to be the result of grammaticalisation of the pronominal element -de into an obligatory clitic.11 In the second stage, the syntactic pattern of sentences with an obligatory second person clitic and an optional second person strong pronoun is extended to sentences with regular word order (cf. 1b) and to the other grammatical persons. This type of extension to other syntactic environments is very common in grammaticalisation processes (cf. Heine and Kuteva 2005: 50–58, see also Timberlake 1977, Andersen 2001 on ‘actualisation’). Finally, in stage three some more recent developments have caused clitics like 2sg./pl. -de and ge to enter in a complementary distribution with strong pronouns like 2sg. gij or 2pl. gullie (1c). This step is a remarkable one, as it restores to some extent the original situation, in which either a weak pronoun or a strong one was used.

4. Grammaticalisation, degrammaticalisation, and conceptual space 4.1.

Stage 1: grammaticalisation of second person pronouns

4.1.1. The rise of subject doubling Section 3 presented evidence showing that subject doubling in Dutch originates in inverted word order in the second person. This finding is completely in line with the data discussed by Haiman (1991), who observes clitisation phenomena in the second person singular in virtually all Germanic and Romance languages, which have triggered the rise of subject doubling in a number of Northern Italian and Romantsch dialects. The crucial step in the rise of doubling is the shift of a former pronoun towards an obligatory clitic or an agreement marker on the verb. At present, the typological literature provides two diachronic accounts of how this process might take place. The oldest one, by Giv´on (1976), acknowledges the role of topic-shifting constructions such as left-dislocation, which are, according to Giv´on, likely to be over-used by language users. At one point, the frequency of the topic-shifting construction is so high that it is reanalysed as the unmarked construction through a ‘markedness shift’. Accordingly, the original topic constituent is re-interpreted as the subject, whereas the erstwhile

242

Gunther De Vogelaer

pronominal subject is turned into an obligatory agreement marker on the verb. The actual source for the rise of new inflectional morphology is thus assumed to lie in certain reinforcement strategies that are used in discourse, and that push the original pronouns into grammatical service. By contrast, Ariel (2000) seeks the motive behind the formation of new agreement morphology in the tendency to minimally encode highly ‘accessible’ elements, i.e. elements that are highly salient or active in discourse. Pronouns, typically encoding given, highly accessible referents, therefore tend to be formally reduced and to be cliticised to the verb, until they are reinterpreted as an inflectional element on the verb. Rather than to be pushed into grammatical service, the former pronouns are actually drawn to the verb. Both accounts are compared in (7). (7)

Diachronic accounts for the rise of pronoun doubling Giv´on (1976: 155) vs. Ariel (2000: 207) a. left-dislocation: a. Pronoun # Verb topic, pronoun V b. Clitic + Verb b. through ‘markedness shift’: c. zero + V+inflection subject agr-V d. NP/Pronoun/zero # V+inflection = reanalysis of topic-shifting = ‘Accessibility’ account

It is unlikely that one of these accounts explains all currently attested instances of subject doubling in the languages of the world. Rather, each account is valid for a specific part of the data. In Dutch, a Giv´on-like explanation can be ruled out in favour of an account along the lines of Ariel (2000) for a number of reasons. Firstly, the strong pronoun in Dutch subject doublings always follows the inflected verb, which invariably occupies the second position in main clauses. Hence the strong pronoun always occurs somewhere in the middle of the sentence, so it cannot be an originally dislocated element, as would be implied in Giv´on’s account. Secondly, subject doubling in Dutch is, in the vast majority of the relevant dialects, restricted to pronominal subjects: strong pronouns can combine with clitics, whereas lexical subjects cannot. This, again, seems to argue strongly against left-dislocation as a source for subject doubling, or any other topic-shifting construction for that matter, because the use of topic-shifting constructions is generally not restricted to pronominal subjects (cf. Geluykens 1992: passim). By contrast, Ariel’s account seems to predict, at least in stages a-c, a situation in which only pronominal subjects can be combined with clitics, and lexical subjects cannot, as is the case in Dutch. Since sentences with lexical subjects do not contain subject pronouns, they are indeed of lesser importance for the development of clitics or agreement markers. Thirdly, the insertion of the full pronouns in subject doubling constructions seems to be triggered by

Subject doubling in Dutch

243

parameters such as empathy or information structure (Nuyts 1995), two parameters that have a bearing on the degree of ‘Accessibility’ of a certain referent (cf. Siewierska 2004: 174–177). Left-dislocation on the other hand, seems to be conditioned by completely different parameters in Dutch, such as constituent length and constituent complexity (Jansen 1981: 167). And finally, Dutch, being a non-pro-drop language, seems to be a very good candidate for an Accessibility account. The obligatory status of the subject pronouns in Dutch makes them very susceptible to formal reduction, since they are used not only more frequently than subject pronouns in pro-drop languages, but also refer to very accessible, active discourse elements that are zero-marked in pro-drop-languages. 4.1.2. Subject doubling in inversion, and conceptual space The accounts discussed in the previous section lead to a different understanding of the development of subject doubling in relation to other person marking constructions. In the original, Middle Dutch system, two different person marking constructions occurred: a verb was combined with either a weak or a strong pronoun, as in present-day Standard Dutch. A topic-shifting account like Giv´on’s (1976) implies a rise in frequency of a formerly marked construction, and thus a large restructuring of the person marking paradigm. This is not the case in the Accessibility account by Ariel (2000), the preferred account for subject doubling in Dutch. There, the grammaticalisation of pronouns into obligatory clitics or agreement markers causes the new subject doubling construction to take over the very same slot that was originally occupied by the combination of a verb and a strong pronoun. Hence, the effects on the person marking paradigm remain limited, as shown in (8). (8)

Inventory of person marking constructions in inverted word order, stage 1 originally after stage 1 inversion: Verb + Pronweak Verb=clitic Verb + Pronstrong Verb=clitic + Pronstrong

The inventory of person marking constructions in those subject doubling dialects that have only gone through stage 1, does not differ radically from the inventory of non-doubling dialects: although different constructions are used, the “conceptual space” (cf. Croft 2001) of person marking is not necessarily remapped.

244

Gunther De Vogelaer

4.2.

Stage 2: generalisation to non-second person and regular word order

4.2.1. An extension of subject doubling In a large number of dialects, subject doubling is not restricted to the second person in clauses with inverted word order. Rather, these dialects seem to have copied the subject doubling pattern from the second person in inverted word order to other syntactic environments. This extension of subject doubling to a whole range of syntactic environments can be considered analogy. The syntactic constructions to which the subject doubling pattern is extended are sentences with regular word order on the one hand, and clauses with non-second-person pronominal subjects on the other hand, as shown in (9).12 (9)

a. Analogy (1): subject doubling in sentences with regular word order (stage 2) ga=de gij > ge=ga gij go=youclitic youstrong youclitic =go youstrong ‘you go’ ‘you go’ b. Analogy (2): doubling of non-second-person pronouns, e.g. 1sg. (stage 2) ga=de gij > ga=’k ik go=youclitic youstrong go=Iclitic Istrong ge=ga gij > ’k=ga ik youclitic =go youstrong Iclitic =go Istrong ‘you go’ ‘I go’

(9a) shows the extension of subject doubling to sentences with regular word order. The existence of morphologically distinct proclitic and enclitic weak 2sg.pronouns does not block the extension of the phenomenon. Possibly, this has to do with the fact that morphologically distinct proclitic and enclitic forms are, although common, far from general in the relevant area. (9b) shows the extension to other grammatical persons, such as 1sg. The fact that the strong pronoun occurs postverbally in all sentences with subject doubling, has lead to the statement that Dutch, or at least the dialects with subject doubling, must be considered a VSO-language (De Meersman 1986, Vandeweghe 2000). The most striking consequence of the developments in (9) is that a new construction is generated in regular word order. The origin for this new construction lies in the extended combinatory possibilities of the former weak pronoun that has grammaticalised into a clitic, which can combine with strong pronouns in the same

Subject doubling in Dutch

245

clause. Hence, the genesis of a new construction is enabled by the formation of new morphology through grammaticalisation. 4.2.2. Increasing paradigmatic variation in the subject marking system The developments in stage 2 have important consequences for the person marking system as a whole. Dialects that have gone through stage 2, i.e. in which all (or most) pronominal subjects can be doubled both in inverted and in regular word order, make an extensive use of clitics. In almost every clause, the verb is accompanied by a clitic: the clitics have become obligatory elements and may be considered part of the verbal morphology. To this combination of a verb and a clitic, an optional, strong pronoun may be added for stylistic or pragmatic reasons. There is only one sentence type in which a strong pronoun can combine with a verb, without a clitic being present, i.e. in clauses with a sentence-initial strong pronoun. This type of sentence is associated with very strong emphasis. In addition, while subject doubling in inversion competes for the same slot in the paradigm as the original combination of a verb followed by a strong pronoun, the rise of subject doubling in sentences with regular word order creates a new construction, as shown in (10). (10) Inventory of person marking constructions, stage 2 originally after stage 2 inversion: Verb + Pronweak Verb=clitic Verb + Pronstrong Verb=clitic + Pronstrong regular: Pronweak + Verb clitic=Verb clitic=Verb + Pronstrong Pronstrong + Verb Pronstrong + Verb Due to the rise of subject doubling in regular word order, the relevant dialects have at their disposal three different constructions for person marking in SVsentences, while originally there were only two. As each of these constructions has its own semantic and pragmatic properties, an increase of the number of available constructions implies a partial remapping of the ‘conceptual space’ of subject marking. Although there are no attempts in the linguistic literature to operationalize true semantic or pragmatic parameters determining the use of these constructions, which is at least partly due to a lack of tagged dialect corpora, the conditions to use each of them are discussed at some length by Vandekerckhove (1993: 175–179; cf. also Nuyts 1995). Figure 1 summarises Vandekerckhove’s conclusions, and also provides a description of how the same semantic/pragmatic functions are encoded in Standard Dutch, which is taken to be representative of the varieties of Dutch without subject doubling.13

246

Gunther De Vogelaer

Figure 1. Subject marking, regular word order: from stage 1 to stage 2

The differences are clear: the use of a preverbal strong pronoun (‘Pronstrong + Verb’ in Figure 1) is associated with strong emphasis in subject doubling dialects, unlike in Standard Dutch, where the use of a preverbal strong pronoun is less pragmatically marked. Consequently, the construction is relatively rarer in these dialects. The preverbal clitics on the other hand (‘clitic + Verb’ in Figure 1), seem to refer to easily accessible referents or active topics. Finally, subject doubling (‘clitic + Verb + Pronstrong ’ in Figure 1) is used when the topic is changing, or whenever a sense of (positive or negative) empathy is involved, most notably while joking or contradicting a statement by another discourse participant. Hence, the contexts in which subject doubling is used in regular word order, correspond to different constructions in Standard Dutch: whereas the first function, i.e. reference to changing topics, is typically attributed to strong pronouns in Standard Dutch, the second function, empathy, does not trigger the use of a strong pronoun.

4.3.

Stage 3: degrammaticalisation in the Brabantic dialects

4.3.1. A ‘mirror-image reversal’ of grammaticalisation? The least widespread and the youngest type of subject doubling is the combination of a verb and two strong pronouns in clauses with regular word order, as exemplified in (1c) above. The phenomenon is attested at the borders of the Belgian part of the Brabantic dialect area (see map 2), and must be understood as a further development of ‘ordinary’ doubling in regular word order, as in (1b). The development is illustrated in (11). It is labelled a ‘degrammaticalisation’ because it results from a shift of the obligatory clitics, like ge, towards a more ‘pronoun-like’ behaviour, which is visible in two ways. Firstly, at the end of stage 3 clitics like ge no longer are obligatory elements, and, secondly,

Subject doubling in Dutch

247

the former complementary distribution between these clitics and their strong counterparts, which was lost in stage 2, is restored. (11) Degrammaticalisation of clitic pronouns in Brabantic dialects, 2sg. (stage 3) ge=ga gij > ge / gij ga gij youclitic =go youstrong youweak/strong go youstrong ‘you go’ ‘you go’ ge = clitic > ge = weak pronoun The proposed development from clitic to weak pronoun in (11) provides a counterexample to the alleged unidirectionality of grammaticalisation. Since the clitics are pronouns in origin, it is even a possible “mirror-image reversal” of grammaticalisation, of which Hopper and Traugott (2003: 137) provide only one instance. At first sight, the data in (11) can also be interpreted as a case of renewal rather than reversal, in which the clitic ge is lost and replaced by a newly formed weak variant of gij, which happens to be identical in form to the former clitic. The data in (12), however, rule out such an interpretation as renewal. In (12a), a morphologically distinct clitic, the 3sg. masculine form en, is degrammaticalised to become a weak pronoun. In addition, ‘true’ degrammaticalisation is expected to be a gradual development, as is grammaticalisation (Norde 2001: 236–237). The gradualness of the process becomes clear in (12b): most dialects with subject doubling show a tendency to develop clearly distinct weak and strong pronouns. For instance, 1pl. we and wij are replaced by me and wullie (i.e. a transition from stage 2 to stage 2’ in the schedule). In Brabantic dialects however, the degrammaticalisation in stage 3 triggers the opposite shift (to stage 3’), in that Brabantic dialects show a tendency to replace me and wullie with we and wij, undoing all phonological changes from stage 2 (see De Schutter 1989: 36, SAND: 34, De Vogelaer 2005: 121). (12) Degrammaticalisation in Brabantic dialects, 3sg.m. and 1pl. Stage 2 Brabantic (stage 3) a. 3sg.m.: gaat=en (hij) > gaat=en OR gaat hij goes=heweak goes hestrong ‘he goes’ goes=heclitic (hestrong ) b. 1pl.: st. 2: gaan=we (wij) > st. 3: ga=me OR gaan wullie ‘we go’ go=weclitic (westrong ) go=weweak go westrong st. 2’: ga=me (wullie) st. 3’:gaan=we OR gaan wij sgo=weclitic (westrong ) go=weweak go westrong

Apart from this mirror-image reversal of the grammaticalisation of subject doubling, the Brabantic dialects show data that lead us to believe that the developments as outlined in (11) and (12) are part of a larger shift in Brabantic, viz.

248

Gunther De Vogelaer

from a clitic-based person marking system to a pronoun-based system: all clitics seem to be banned in these dialects. Firstly, the area where subject doubling with two strong pronouns is found is a less productive one when it comes to subject doubling in general. In the Brabantic area, the phenomenon occurs less consistently (cf. Brabant’s transitional status on map 1), and less frequently (cf. elicitation problems that were reported by the SAND-fieldworkers for the area). Secondly, a number of more recent developments have taken place in the relevant area, showing that subject doubling, and hence the use of clitics, is indeed losing ground, as exemplified in (13). (13a) exemplifies the fusion of a former clitic ’k and the strong pronoun ik into a new strong pronoun kik, which no longer exclusively occurs in subject doubling constructions, but also elsewhere, such as in conjoined NPs. The use of kik in conjoined NPs is quite widespread in the entire Dutch part of the Brabantic dialect area, and even more to the west (for a map, see De Vogelaer 2005: 189). (13b) shows a similar phenomenon, in the second person: the combination of a clitic -de and the 2sg.-pronoun gij (or 2pl. gullie) is reinterpreted as an inflectional -t and the pronoun -egij (-egullie). This pronoun -egij (-egullie) is used, for instance, following verbs in the simple past, where the dental preterite suffix (-de) blocks the insertion of the homophonous clitic, and where the initial /e/ in 2sg. -egij (2pl. -egullie) even triggers the insertion of an intervocalic linking-n. This development is less widespread than (13a): there are only eight attestations in the SAND-corpus. But, significantly, all attestations come from Brabantic sampling points, most even from regions that have subject doubling with two full pronouns as well (for maps, see De Vogelaer 2005: 197, 228). Hence, the correlation between both phenomena is clear. Apparently the Brabantic instances of subject doubling in 1sg. (13a), and (to a lesser extent) in 2sg. and 2pl. (13b), which illustrate stage 3 of the diachronic development, must be interpreted as instances of compound pronouns, which are relics of subject doubling rather than manifestations of a productive doubling construction.14 (13) Further developments in Brabantic: clitic + strong pronoun > strong pronoun a. 1sg. ‘k + ik > kik Stage 2 Brabantic (stage 3) e.g. ga=k ik > ga kik go=Iclitic Istrong go Istrong ‘I go’ ‘I go’ Jan en ik > John and Istrong ‘John and I’

Jan en kik John and Istrong ‘John and I’

Subject doubling in Dutch

249

b. 2sg./2pl. -(d)e + gij / gullie > -egij / -egullie Brabantic (stage 3) Stage 2 e.g. ga=de gij > gaat egij go=youclitic youstrong go youstrong ‘you go’ ‘you go’ leefde gij live.pret youstrong ‘you lived’

>

leefde-n egij live.pret-linking-n youstrong ‘you lived’

A comparison of the SAND-data with older dialect data from the 1940s, as provided by Schuurmans (1975) and De Schutter (1989), shows even more manifestations of the Brabantic tendency to ban clitics. A number of formally distinct clitics, i.e. ‘special clitics’in Zwicky’s (1977) terms, have been replaced recently by ‘simple clitics’that are merely formal variants of the strong pronouns, such as 1pl. me by we (strong pronoun wij), or, in some syntactic environments, the 2sg. clitic -de by ge (strong pronoun gij) (see SAND, esp. maps 39–41, 44–45).15 This development too can be interpreted as a shift in Brabantic from a person marking system with an extensive use of clitics, as found in the contemporary southwestern, Flemish dialects, to a more pronoun-based system, as found in most northern varieties of Dutch, including Standard Dutch. It is not clear what has triggered the loss of clitics in the Brabantic dialect area. Although this question must remain largely unaddressed here, it seems that different factors may have played a role. Among the system-external factors, a different degree of standardisation may be involved, although it is unlikely that standardisation is the main factor. System-internal factors that may have influenced the Brabantic innovations include a tendency towards economy and/or simplification in Brabantic, and changes in word order (see De Vogelaer 2005: 282–284, 296–300 for discussion). 4.3.2. A further increase of constructions in the subject marking paradigm The shift of a clitic-based person marking system to a pronoun-based system in Brabantic dialects is bound to have important ramifications for the paradigm of person marking constructions in the relevant varieties of Dutch, which is given in (14). (14) shows that in Brabantic dialects the use of subject doubling is altogether quite limited, and in general restricted to sentences with regular word order. That subject doubling in sentences with regular word order is kept, provides evidence to the idea that syntax can indeed serve as a repository for older language phenomena (cf. Chafe, this volume). It is remarkable that a new type of subject

250

Gunther De Vogelaer

(14) Inventory of person marking constructions, stage 3 (Brabantic) Stage 3 Stage 2 inversion: Verb=clitic Verb + Pronweak Verb=clitic + Pronstrong Verb + Pronstrong regular: clitic=Verb clitic-Verb + Pronstrong Pronstrong + Verb

Pronweak + Verb Pronweak + Verb + Pronstrong Pronstrong + Verb + Pronstrong Pronstrong + Verb

doubling has originated in Brabantic, where the verb can combine with two strong pronouns. This shows that not only grammaticalisation but also degrammaticalisation can result in the genesis of a new construction. This new construction obviously has its own formal characteristics, and it has its own semantic and pragmatic functions as well. The semantic/pragmatic connotations of subject doubling with two strong pronouns have not received any attention yet in the linguistic literature. The construction is rare, so the conditions to use it are far more restricted than for the older types of doubling, in which a verb combines with one weak and one strong pronoun. According to Nuyts (1995: 54–57), who describes the Antwerp dialect (that allows doubling with two strong pronouns), doubling in Brabantic is triggered by similar factors as in the West-Flemish dialect described by Vandekerckhove (1993): both information structure and feelings of (positive or negative) empathy towards the subject facilitate doubling. Nuyts, however, does not provide any specific information concerning the use of two strong pronouns. Tentatively, it could be suggested that this construction is likely to be used in clauses in which both the conditions to use a preverbal strong pronoun are met, and the ones to use a postverbal strong pronoun. This would be clauses in which, on the one hand, the subject is strongly emphasised, and, on the other hand, in which there is either a changing or a non-accessible topic, or in which feelings are involved of positive or negative empathy towards the subject, as shown in figure 2. Figure 2 shows that, although the erstwhile clitics have degrammaticalised to become ‘ordinary’ weak pronouns, the Brabantic dialects have kept subject doubling and have even introduced a new variant of the construction. Crucially, the rise of a new type of doubling boils down to a simplification of the person marking system (see also De Vogelaer 2003: 192–194). Rather than having to store three different combinations of person markers, speakers of Brabantic dialects have the same two constructions at their disposal as, for instance, speakers of Standard Dutch, to which they can add a strong pronoun as a marker for changing topics (or for empathy).

Subject doubling in Dutch

251

Figure 2. Subject marking, regular word order: from stage 2 to stage 3 (Brabantic dialects)

5. Conclusion In this paper, subject doubling data are discussed from a number of southern Dutch dialects. The rise of subject doubling is accounted for as the result of grammaticalisation: former pronouns cliticise to the verb and become obligatory elements, first in the second person in inverted word order, and consequently in the other grammatical persons, and in regular word order. During this development, a new subject marking construction originates in regular word order, causing a massive remapping of the conceptual space of subject marking (figure 1). Hence, it is clear that grammaticalisation can cause the rise of new constructions. Some dialects have moved beyond this stage. For instance, in most Brabantic dialects in Belgium the use of clitics is highly regressive, as most clitics have lost their obligatory status, and the complementary distribution between the weak pronouns and the strong ones is restored. It is argued that these Brabantic data constitute an example of degrammaticalisation, and even of a “mirror-image reversal” (Hopper and Traugott 2003: 137) of grammaticalisation. In addition, the degrammaticalisation process in these Brabantic dialects, more specifically the restored complementary distribution between the weak and the strong pronouns, has triggered the rise of another new subject marking construction, viz. subject doubling with two strong pronouns. Developments such as the one proposed for Brabantic are probably quite exceptional from a typological point of view. This, of course, does not release linguistics from studying them (cf. van der Auwera 2002). A tentative explanation for the exceptional status of the Brabantic data may be that they must be seen as the result of an ‘abortive change’ from a pronoun-based subject marking system to a clitic-based one, which is bound to leave very few traces in purely synchronic data. Since synchronic facts are often used as the main data source in studying the diachrony of languages, it seems at least possible that the excep-

252

Gunther De Vogelaer

tional status of the Brabantic instances of degrammaticalisation has more to do with linguistics than with language.

Notes 1. This article is an elaborated version of a chapter from my Ph.D-dissertation (De Vogelaer 2005). I would like to thank Magda Devos and Johan van der Auwera for their comments on the particular chapter, and Evie Couss´e, Bart Defrancq and Elizabeth Traugott for reading a previous version of this article. In addition, this article has benefited from discussion with the co-authors of the SAND-atlas (Sjef Barbiers, Hans Bennis and Margreet van der Ham) and the participants to the workshop on ‘Constructions and Language Change’ at the ICHL-conference in Madison, Wisconsin. Joop van der Horst provided some very interesting Middle Dutch data. Last but not least, I would like to thank all the people that were involved in the gathering and the transcription of the SAND-data, without whose efforts this article would not have been possible. 2. Previous research on subject doubling in Dutch has found more parameters to be relevant, including the NP-type of the subject (pronoun vs. noun), information structure and empathy, and the part of speech of the element that serves as a host for the clitic (a verb or a complementiser). In addition, there is variation as to the number of clitics/pronouns in the construction (in some dialects, a third element may be inserted). For discussion, see the relevant literature (e.g. Haegeman 1992, De Geest 1995, Van Craenenbroeck and Van Koppen 2002 for accounts within a generative framework; and Vandekerckhove 1993 and Nuyts 1995 for the pragmatics of subject doubling; the most elaborate empirical descriptions of subject doubling are De Vogelaer 2005 and the SAND-atlas). 3. See the appendix for a list of the sources that were consulted. 4. At face value, some of the examples are ambiguous between a reading as an instance of subject doubling and a reading as a (non-doubled) past tense, since -de is used for marking the preterite as well. All the examples presented here, however, are glossed as present tense forms by the relevant authors, who themselves analyse them as legitimate instances of subject doubling. 5. Although both manifestations of the pronoun ik (‘I’) are identical in spelling, the first one probably must be considered a clitic pronoun (cf. also section 3.2). In general, it seems to be a spelling convention in Middle Dutch texts to avoid the use of weak forms, except when they can be spelled as elements encliticised to the verb, such as ’r (‘there’). 6. In most cases, the examples from the parodies exaggerate the doubling phenomenon and are thus not suitable to serve as an information source for the linguistic side of the facts. The examples provided by the WNT-dictionary (VI: 1450) include instances of sentences with three full pronouns, such as Ik ben ik ik noch vechterken noch smijterken, zey den braber (no date; lit. ‘Istrong am Istrong Istrong nor a fighter nor a

Subject doubling in Dutch

7.

8.

9.

10. 11.

12.

13.

253

thrower, said the brabantic man’) and ‘Ick-ick-ick weet niet wat ick seggen sal (1629; lit. ‘Istrong Istrong Istrong know not what I shall say’). Similar examples, however, are not found in any historical or contemporary variety of Dutch. The SAND-project (2000–2003) was a fieldwork project sponsored by the VNC (Flemish-Dutch commission for Dutch language and culture), carried out at several institutions in Belgium and the Netherlands, including Ghent University and the University of Antwerp in Belgium, and the University of Amsterdam, the University of Leiden, the Meertens Institute and the Fryske Akademy in the Netherlands. The methodological outlines of the project are described in detail in Cornips and Jongenburger (2001a,b). This includes the Dutch province of Friesland, where both Dutch and Frisian serve as official languages. As a matter of fact, some Frisian dialects have been included in the database, since the Frisian dialects belong to the same dialect continuum as the Dutch dialects. Map 1 contains data for 24 different test sentences per sampling point, equalling 6.384 questions. In most cases, more than 1 answer is provided, as at least 2 informants were consulted per place, and fieldworkers were instructed to test exhaustively which types of subject doubling were possible in the relevant test sentences, so the actual amount of tokens surpasses 10.000. Among these answers, about 1.000 instances of subject doubling are found. Answers to about 2.400 questions are mapped, of which more than 500 show subject doubling. There is quite some debate in the Dutch dialectological literature about the precise status of -de. Most see it as a pronoun, as -de is a descendant of the Middle Dutch pronoun -di, an enclitic variant of 2sg. ghy, in which the pronoun is fused with the inflectional ending on the verb (-t or -d). Some, e.g. Zwart (1997: 139), consider -de as an agreement marker on the verb, since it is an obligatory element that only appears encliticised to the verb. The syntactic behaviour of -de, however, differs significantly from other agreement markers in Dutch: it allows pro-drop and it cannot be followed by a weak pronoun in most dialects. In De Vogelaer (2005), it is argued that -de shares characteristics with both weak pronouns and agreement markers. Also, a more fine-grained typology is provided in which the ‘deviant’ behaviour of -de and other controversial person markers is accounted for. Possibly, a more refined chronology could be composed, in which the developments in (9a) and (9b) occupy different positions, and in which the grammatical person of the pronominal subject plays a role as well. However, the present data are ambiguous as to the relative chronology of the developments in (9a) and (9b) (see section 3). Figure 1 is only intended to take stock of the important differences in the pragmatic and semantic functions associated with the different subject marking constructions in Standard Dutch and the subject doubling dialects. Needless to say, both accounts are somewhat simplified. For instance, parameters such as person, animacy, adjacency to the verb, etc., which also influence the choice of a weak vs. a strong pronoun, are not mentioned. In addition, the precise conditions for the use of a certain construction

254

Gunther De Vogelaer

may vary among the different varieties of Dutch. This is shown by the simple fact that there are significant differences in the frequency with which they are used (see Will 2004: 206 for an illustration). 14. Hence the data in (13) also cast some light on map 1, as they explain the deviant behaviour of 1sg., 2sg. and 2pl.: a lot of the attestations of subject doubling in the transitional and peripheral zones are merely relics from subject doubling, in which, from a synchronic point of view, only a strong pronoun is found in stead of a combination of a clitic and a strong pronoun. 15. Zwicky uses the term ‘clitic’ to refer to a formal category of elements and hence his terminology differs from the syntactically motivated typology that is adopted in this article. The most important difference is that Zwicky’s use of the term ‘clitic’does not imply that clitics can be doubled.Typologically, however, it appears that special clitics are typically found in doubling constructions, whereas simple clitics are not (see Cardinaletti and Starke 1999, Siewierska 2004: 26–27,34–38 for discussion). Hence the shift from special clitics to simple clitics in Brabantic is indeed a syntactically relevant phenomenon.

References Andersen, Henning (ed.) 2001 Actualization. Linguistic Change in Progress. Amsterdam/Philadelphia: John Benjamins. Ariel, Mira 2000 The development of person agreement markers: from pronouns to higher accessibility markers. In: Michael Barlow and Suzanne Kemmer (eds.), Usage-based Models of Language, 197–260. Stanford: CSLI Publications. Bonfante, Giuliano 1947 The neolinguistic position. Language 23: 344–375. Bresnan, Joan 2001 The Emergence of the Unmarked Pronoun. In: G´eraldine Legendre, Jane Grimshaw and Sten Vikner (eds.), Optimality-Theoretic Syntax, 113–142. Cambridge, MA: MIT Press. Campbell, Lyle and Richard Janda (eds.) 2001 Theme issue on grammaticalization. Language Sciences 23(2–3): 93– 340. Cardinaletti, Anna and Michal Starke 1999 The typology of structural deficiency: A case study of the three classes of pronouns. In: Henk Van Riemsdijk (ed.), Clitics in the Languages of Europe, 145–233. Berlin/New York: Mouton de Gruyter. Chafe, Wallace This volume Syntax as a Repository of Historical Relics.

Subject doubling in Dutch

255

Chambers, Jack K. and Peter Trudgill 1998 [1980] Dialectology. Cambridge: Cambridge University Press. Corbett, Greville G. 2003 Agreement: the range of the phenomenon and the principles of the Surrey Database of Agreement. In: Dunstan Brown, Greville G. Corbett and Carole Tiberius (eds.), Agreement: a typological perspective. Special issue of Transactions of the Philological Society 101(2): 155– 202. Cornips, Leonie and Willy Jongenburger 2001a Het design en de methodologie van het SAND project. Nederlandse Taalkunde 16: 215–232. 2001b Elicitation techniques in a Dutch syntactic dialect atlas project. In: Hans Broekhuis and Ton van der Wouden (red.), Linguistics in The Netherlands 18, 57–69. Amsterdam/Philadelphia: John Benjamins. Croft, William 2001 Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford: Oxford University Press. De Geest, Wim 1995 Cliticisation and clitic doubling in East Flemish. In: Thomas F. Shannon and Johan P. Snapper (eds.), The Berkeley Conference on Dutch Linguistics 1993: Dutch Linguistics in a Changing Europe, 151–170. Lanham: University Press of America. De Meersman, Alfons 1986 EenVSO-maneuver in de Zuidnederlandse dialecten. In: Hugo Ryckeboer, Johan Taeldeman and Val`ere Frits Vanacker (eds.), Hulde-album Prof.dr. Marcel Hoebeke, 123–131. Gent: Seminarie voor Nederlandse Taalkunde RUG. De Schutter, Georges 1989 Pronominale clitica in de Nederlandse dialecten.Antwerpen:Antwerp Papers in Linguistics. De Vogelaer, Gunther 2003 Person marking in Dutch dialects. In: Bernd Kortmann (ed.), Dialectology meets Typology: Dialect grammar from a Cross-Linguistic Perspective, 181–209. Berlin/New York: Mouton de Gruyter. 2005 Subjectsmarkering in de Nederlandse en Friese dialecten. Proefschrift Universiteit Gent. De Vogelaer, Gunther and Magda Devos Forthc On geographical adequacy, or: How many types of subject doubling in Dutch dialects? In: Sjef Barbiers, Margreet van der Ham, Olaf Koeneman and Marika Lekakou (eds.), Microvariations in Syntactic Doubling. Amsterdam: Elsevier.

256

Gunther De Vogelaer

De Vogelaer, Gunther and Annemie Neuckermans 2002 Subject doubling in Dutch: a dialect phenomenon in cross-linguistic perspective. Sprachtypologie und Universalienforschung (STUF) 55: 234–258. Geluykens, Ronald 1992 From discourse process to grammatical construction: on leftdislocation in English. Amsterdam/Philadelphia: John Benjamins. Giv´on, Talmy 1976 Topic, pronoun, and grammatical agreement. In: Charles N. Li (ed.), Subject and Topic, 149–188. New York: Academic Press. Haegeman, Liliane 1992 Theory and description in generative syntax: a case study in WestFlemish. Cambridge: Cambridge University Press. Haiman, John 1991 From V/2 to subject clitics: Evidence from Northern Italian. In: Elizabeth Traugott and Bernd Heine (eds.), Approaches to Grammaticalization. Vol. 2. Focus on Types of Grammatical Markers, 135–157. Amsterdam/Philadelphia: John Benjamins. Haspelmath, Martin 2004 On directionality in language change with particular reference to grammaticalization. In: Olga Fischer, Muriel Norde and Harry Perridon (eds.), Up and down the Cline -The Nature of Grammaticalization, 17–44. Amsterdam: John Benjamins. Heine, Bernd and Tania Kuteva 2005 Language Contact and Grammatical Change. Cambridge: Cambridge University Press. Hopper, Paul and Elizabeth C. Traugott 2003 [1993] Grammaticalization. Cambridge: Cambridge University Press. Jansen, Frank 1981 Syntaktische konstrukties in gesproken taal. Amsterdam: Huis aan de drie Grachten. Norde, Muriel 2001 Deflexion as a counterdirectional factor in grammatical change. Language Sciences 23(2–3): 231–264. Nuyts, Jan 1995 Subjectspronomina en dubbele pronominale constructies in het Antwerps. Taal & Tongval 47: 43–58. SAND = Barbiers, Sjef, Hans Bennis, Gunther De Vogelaer, Magda Devos and Margreet van der Ham 2005 Syntactic Atlas of Dutch Dialects.Volume 1: Pronouns,Agreement and Dependencies. Amsterdam: Amsterdam University Press.

Subject doubling in Dutch

257

Schuurmans, N.J. 1975 Verbindingen met specifiek enclitische pronomina in het Westbrabants. Themanummer van de Mededelingen van de Nijmeegse centrale voor dialect- en naamkunde. Nijmegen: Nijmeegse centrale voor dialecten naamkunde. Siewierska, Anna 1999 From anaphoric pronoun to grammatical agreement marker: why objects don’t make it. Folia Linguistica 33(2): 225–251. 2004 Person. Cambridge: Cambridge University Press. Timberlake, Alan 1977 Reanalysis and actualization in syntactic change. In: Charles N. Li (ed.), Mechanisms of syntactic change, 141–177.Austin/London: University of Texas Press. Vanacker, V.F. 1963 Syntaxis van gesproken taal te Aalst en in het land van Aalst in de XVde, de XVIde en de XVIIde eeuw. Brussel: Belgisch interuniversitair centrum voor neerlandistiek. Van Craenenbroeck, Jeroen and Marjo Van Koppen 2002 Pronominal doubling and the structure of the left periphery in southern Dutch. In: Sjef Barbiers, Leonie Cornips and Suzanne van der Kleij (eds.), Syntactic Microvariation. Amsterdam: Meertens Institute Electronic Publications in Linguistics. (http://www.meertens.knaw.nl/books/synmic/ [17.01.2008]) Vandekerckhove, Reinhild 1993 De subjectsvorm van het pronomen van de 2e p.ev. in de Westvlaamse dialecten. Taal & Tongval 45: 173–183. van der Auwera, Johan 2002 More thoughts on degrammaticalization. In: Ilse Wischer and Gabriele Diewald (eds.), New reflections on Grammaticalization, 19–29. Amsterdam: John Benjamins. Vandeweghe, Willy 2000 Pronominale reduplicatie: sporen in het AN. In: V´eronique De Tier, Magda Devos and Jacques Van Keymeulen (eds.), Nochtans was scherp van zin. Huldealbum Hugo Ryckeboer, 439–443. Gent: Vakgroep Nederlandse Taalkunde UGent. Weijnen, Antoon A. 1977 The value of the map configuration. Nijmegen: Katholieke Universiteit Nijmegen. Will, George 2004 Zeggen k-ik da(t) zˆo? Reduplicatie van het subjectspronomen in Zeeuws-Vlaanderen. Taal & Tongval 56: 187–211.

258

Gunther De Vogelaer

Zwart, Jan-Wouter 1997 Morphosyntax ofVerb Movement: a minimalist approach to the syntax of Dutch. Dordrecht/Boston/London: Kluwer Academic Publishers. Zwicky, Arnold 1977 On Clitics. Bloomington: Indiana University Linguistics Club.

List of consulted secondary sources on Dutch Bouman, Arie C. 1934 Middelnederlandse bloemlezing met grammatika. Zutphen: Thieme. Duinhoven, Antonius M. 1988 Middelnederlandse syntaxis: synchroon en diachroon. Leiden: Martinus Nijhoff. Den Hertog, Cornelis H. 1973 Nederlandse spraakkunst. Derde druk. Amsterdam: Versluys. Hermkens, Hendrikus M. 1973 Inleiding tot het zeventiende-eeuws. Malmberg: ’s-Hertogenbosch. Le Roux, Thomas H. and Jacobus J. Le Roux 1969 Middelnederlandse grammatika. Pretoria: Van Schaik. Meert, Hippoliet 1901 Vormleer van de taal van Ruusbroec. Gent: Siffer. Overdiep, Gerrit S. 1946 Vormleer van het Middelnederlandsch der XIIIe eeuw. Antwerpen: Standaard-boekhandel. Overdiep, Gerrit S. 1949 Stilistische grammatica. Tweede druk, door G.A. van Es. Zwolle: Willink. Rijpma, Enneus and Frans G. Schuringa 1978 [1917] Nederlandse spraakkunst. Vijfentwintigste druk. Groningen: WoltersNoordhoff. Ten Kate, Lambert 2001 [1723] Aanleiding tot de kennisse van het verhevene deel der Nederduitsche sprake. facsimile-editie. Alphen aan de Rijn: Canaletto/ReproHolland. Terwey, Tijs 1903 Nederlandse spraakkunst. Groningen: Wolters. Vanacker, V.F. 1963 Syntaxis van gesproken taal te Aalst en in het land van Aalst in de XVde, de XVIde en de XVIIde eeuw. Brussel: Belgisch interuniversitair centrum voor neerlandistiek.

Subject doubling in Dutch

259

van den Berg, Berend 1971 Inleiding tot de Middelnederlandse syntaxis. Groningen: WoltersNoordhoff. Van Helten, Willem L. 1887 Middelnederlandsche spraakkunst. Groningen: Wolters. Van Loey, Adolphe 1964 Sch¨onfelds historische grammatica van het Nederlands (zevende druk). Zutphen: Thieme. WNT = Woordenboek der Nederlandsche Taal 1864–1998 Elektronische versie: Het Woordenboek der Nederlandsche Taal op CD-ROM. Den Haag: Sdu.

Syntax as a repository of historical relics Wallace Chafe My concern in this chapter is with the broad-scale design of language. I believe it is impossible to understand language design without taking account of language change, seeing historical processes as shaping the fundamental nature of language. Constructions, as I hope to illustrate, form an integral part of this broad picture. I will begin by assuming that language has the basic components shown in (1).

(1)

Language design in broad perspective Thoughts → Semantic → Syntactic → Phonological → Sounds organization organization organization

Language provides above all a way of associating thoughts with sounds. It cannot make such an association directly, however, because thoughts and sounds are very different in nature. It is necessary, therefore, for various kinds of adjustments to be made before thoughts and sounds can be related. In the first place, thoughts must be organized into what can be called semantic structures, for at least the reasons that are listed in (2). (2)

Why thoughts must be organized semantically The need to select Thoughts contain more than can be verbalized. The need to categorize The vast number of particular ideas. The need to orient Ideas are located in time, space, epistemology, emotions, etc. The need to combine Semantic elements participate in constructions.

First, thoughts contain more than can possibly be verbalized, and thus speakers must select what they will “put into words.” They must select topics to be verbalized, but also ways of organizing their thoughts within those topics. Second, thoughts usually involve many particular ideas. People talk about events and states and objects that are particular in time and space. It would obviously be impossible for each particular idea to be associated with a particular sound, and

262

Wallace Chafe

thus speakers must categorize particular ideas as instances of already familiar categories. Third, ideas of events and states and objects are located in a complex, multidimensional space involving time, space, epistemology, emotions, and so on, and each language orients ideas in its own way within that space. Fourth, those ideas and orientations must be combined within established patterns that can well be called constructions. When thoughts have been adjusted in accordance with the processes in (2), why is it that these semantic elements are not associated directly with sounds? Why should there be another intervening stage comprising what we know as syntax? As suggested in (3), the answer is that languages change, and in particular that they are subject to processes of lexicalization, grammaticalization, and the formation of new constructions within which its various elements are combined. (3)

Why is there a separate syntactic organization? Because language changes Lexicalization Grammaticalization Construction formation

In order to make this concrete, I will refer repeatedly to the example in (4). Imagine that you were observing a friend, perhaps an artist of some kind, who was preparing for a performance that would take place in front of a large audience. Imagine that you reacted by saying, “That’s gonna bring down the house.” (4)

An example That’s gonna bring down the house.

You would be expressing your prediction that the audience would respond to the performance enthusiastically. This idea was organized in accordance with the semantic resources of English in a way that is suggested in (5). Whatever may ultimately be seen as the best way to represent semantic structures, what is important about them is the fact that they contain all and only elements that are associated directly with thoughts, and also elements that will eventually enjoy some kind of phonological expression. Semantic structures organize thoughts in accordance with the semantic resources offered by the language in which they are expressed, and they are expressed in some way, sometimes quite indirectly, as we will see, by sounds.

Syntax as a repository of historical relics

(5)

263

Semantic organization

distal demonstrative oriented as given

starting point

event categorized as BDTH oriented as future new

In this outline of a semantic representation the box on the right uses the initials BDTH as an arbitrary way of representing the way in which this thought, the idea of the performance being enthusiastically received, was categorized. This BDTH category was only indirectly expressed with sounds for reasons that will become clear. In the course of that expression the fact that it was a new idea led to it being prosodically stressed. Its orientation as future was also expressed in an indirect way. The distal demonstrative in the left-hand box captured the idea of the performance, which functioned as the starting point for the expression of the new idea represented in the right-hand box. The idea of the performance was assumed to be already in the consciousness of the listener, and thus it was given and expressed with the prosodically weak word “that.” In (6) there is a summary of those elements of this semantic structure that were actually expressed directly in the syntax of this sentence. (6)

Elements of the semantic structure that were directly represented in the syntax distal demonstrative “that” given weak prosody starting point subject properties new prosody

There was the demonstrative word “that”, its weak prosody expressing its givenness, and its function as a starting point expressed with the well known syntactic properties of English subjects. The newness of the BDTH idea was expressed with strong prosody. The rest of this sentence was related to its semantic structure, and thus to the thought it conveyed, only indirectly. The question raised in (7) is really a question about the ontology of syntax. What is syntax, why does it exist, and why does it contain what it does? (7)

Where does syntax come from?

In (8) there is a suggestion that syntax is like a sailor suit, the traditional uniform worn by enlisted men in the navy.

264

Wallace Chafe

(8)

Syntax is like a sailor suit jumper flap

bell bottoms

Parts of this uniform are directly functional, in the sense that they serve to keep the sailor warm and cover parts of his body that are meant to be covered. But (8) points out two of its parts that are not directly functional, or are functional no longer. The jumper flap is present because in earlier times sailors used to grease their hair, and this flap kept the grease from soiling the jumper. The origin of bell bottom trousers is obscure. Folklore has it that sailors could easily roll them up when they swabbed the deck or climbed the rigging, or perhaps that they were easier to kick off in the water. “Whatever the real reason, reliable documentation validating any assertion was lost long ago. What is true, however, is that the bell bottom trouser has consistently remained as a part of the identifiable occupational dress of sailors throughout modern history” (Dervis 2000). The point of (8) is that this navy uniform is a mixture of elements that are directly functional and elements that are not. Syntax is the same. It mixes elements that are directly related to semantics, and thus to thoughts, and elements that are not directly related to semantics and thoughts. We can examine now the ways in which our example, “That’s gonna bring down the house,” is like a navy uniform and, like that uniform, can be satisfactorily explained only in historical terms. First we can examine the process of idiom formation that is evident in the construction “bring down the house,” as suggested in (9). (9) Idiom formation BDTH → “bring down the house” This idiom seems to have originated, perhaps in the eighteenth century, as a useful way of expressing the idea characterized by the Oxford English Dictionary as “to evoke such demonstrative applause as threatens or suggests the downfall of the building.” A quote from 1754, “His apprehension that your statues will bring the house down,” appears to be an early attestation of it. This idiom, however, made use of two earlier innovations. One of them is the metaphor set forth in (10). (10) Metaphor (“bring down” for “destroy”) Evidently by the 16th century the verb-particle combination “bring down” was already being used in the sense of destroying something, causing its downfall.

Syntax as a repository of historical relics

265

“God is purposed to brynge downe all stoute mountaynes” is an example from that century. The other earlier innovation is shown in (11): the use of “house” metonymically to express the idea of an audience. (11) Metonymy (“house” for “audience”) Thus, also from the 16th century, we have “The house, by its frequent plaudits, did show their sufficient approbation.” It was not difficult then, at some later time, to talk about “bringing down the house” as a way of expressing the idea of the enthusiastic reception of a performance. I have used this as an example of the way in which a unitary semantic categorization of a particular idea came to be expressed in this indirect way. In (5) I also included the fact that the idea of this event was oriented as something that would take place in the future. Here this future orientation was expressed syntactically with the construction “be going to,” as in (12), which has been discussed frequently in the literature on grammaticalization. Hopper and Traugott (1993) cite the 15th century attestation that appears in the OED: “Thys onhappy sowle .. was goyng to be broughte into helle for the synne and onleful lustys of her body.” (12) Grammaticalization of “be going to” as a way of expressing futurity It is of some interest also that we have the more recent phonological reduction of “going to” to “gonna,” mentioned in (13), establishing the complete grammaticalization of “be going to” as an auxiliary. (13) Phonological reduction to “gonna” This “be going to” construction could never have arisen, however, if the language did not already possess the result of an earlier grammaticalization: the use of “be ... -ing” as a way of expressing the so-called progressive aspect, as outlined in (14). The history of this construction is controversial; there is a good discussion of varying points of view in the Middle English Grammar by Mustanoja (1960). In Old English there were evidently two constructions. One was a participial construction with the “-end” ending, the use of which was discussed in great detail in Nickel (1966). The other was a gerundial construction with the “-ing” ending. This latter construction was introduced with the preposition “on”. In the 13th century the “-end” ending fell together with the “-ing” ending, and as a result the distinction between the two constructions was blurred. Meanwhile, the “on” was reduced to “a-,” as in “he was a-hunting.” The form without this

266

Wallace Chafe

“a-,” however, simply “he was hunting,” gradually took over, leaving the “be a-hunting” alternative today possessing a nonstandard or quaint flavor. (14) Grammaticalization of “be ... -ing” Old English had two constructions: “wæs ... -ende” and “wæs on → -inge” 13th century: -ende → -inge Modern English: triumph of “be ... -ing” with “on” “a” relegated to nonstandard To review what I have illustrated with this example, the syntax of “That’s gonna bring down the house” combines elements that extend all the way from Old English to the relatively recent reduction of “going to” to “gonna”, probably in the order suggested in (15). First came the progressive expressed by “be ... –ing,” then the future expressed by “be going to,” then the idea of destruction expressed by “bring down,” and more or less simultaneously the idea of an audience expressed by “house,” and then the idea of an enthusiastic reception expressed by “bring down the house.” Finally the interpretation of “going to” as an auxiliary led to its phonological reduction to “gonna”. (15) Order of innovations Semantic progressive orientation future orientation destruction idea audience idea enthusiastic reception idea “going to”

→ → → → → →

Syntactic “be ... -ing” “be going to” “bring down” “house” “bring down the house” “gonna”

A syntactic structure, to repeat, is a mixture of elements that have direct semantic relevance with other elements that may have had semantic relevance at an earlier stage of the language, but do so no longer. I have been calling such elements “quasi-semantic,” because they behave as if they were semantic although they no longer are. Just as a sailor suit mixes functional and quasi-functional elements, a syntactic structure is a mixture of semantic and quasi-semantic elements. There is one more consideration to add to this picture. The components of language design set forth in (1) are not isolated from each other. For example, the way in which thoughts are organized in semantic structures feeds back into the nature of the thoughts themselves. There is more to thinking than language alone (certainly imagery and emotions are important too), but nevertheless “talking to yourself” does play a crucial role, and to that extent thoughts and semantic

Syntax as a repository of historical relics

267

structures have much in common. I have shown this mutual interinfluence in (16) with the upward-pointing arrow from semantic organization leading back into thoughts. (16) Interaction between language components Thoughts ↓ ↑ Semantic organization ↓ Syntactic organization But it is also important to note that at least some aspects of syntactic structure that no longer express semantic elements directly may nevertheless have a residual effect on them. I have discussed this kind of interaction in terms of “shadow meanings”: the reflections of the literal meanings of idioms and grammaticalized constructions back into semantics and thought. This kind of interaction is shown with the dashed arrow in (16), leading back from syntactic organization to semantic organization. In the example before us this shadow meaning might be realized as a vague image of literally bringing down a house or destroying a building that might accompany the idea of reacting to a performance with enthusiasm. Shadow meanings can vary with particular examples, and even with different individuals. It is worth noting in this connection that Hopper and Traugott mentioned what I would call a shadow effect of “be going to” on the semantics of futurity: “The original purposive meaning continues to constrain the use of the auxiliary... This property of persistence of meaning presumably derives in part from the fact that the older be going (to ...) coexists with the newer use, and hence there is reinforcement of older meanings” (Hopper and Traugott 1993: 3). In summary, I have suggested that syntax is a hodgepodge of elements that are directly meaningful and elements that are not, analogous to a navy uniform. Those elements that are not directly meaningful are relics of historical processes that took place at various times in the history of a language. It would, I think, be helpful for linguists to view syntax in this way. In so doing they might be led to pay more attention to, first, the nature of semantic structures and, second, the processes of language change that are responsible for syntax. For further discussion and examples see Chafe (2002, 2005, 2007, in press).

268

Wallace Chafe

References Chafe, Wallace 2002 Putting grammaticalization in its place. In: Ilse Wischer and Gabriele Diewald (eds.), New Reflections on Grammaticalization: Proceedings of the International Symposium on Grammaticalization, 17–19 June 1999, at Potsdam University, Germany, 395–412. Amsterdam/Philadelphia: John Benjamins. 2005 The relation of grammar to thought. In: Christopher S. Butler, ´ Mar´ıa de los Angeles G´omez-Gonz´alez, and Susana M. DovalSu´arez (eds.), The Dynamics of Language Use: Functional and Contrastive Perspectives, 55–75.Amsterdam/Philadelphia: John Benjamins. 2007 Language and Consciousness. In: Philip D. Zelazo, Morris Moscovitch, and Evan Thompson (eds.), Cambridge Handbook of Consciousness, 355–373. Cambridge: Cambridge University Press. In press. Thoughts and sounds. In: Konrad Ehlich (ed.), Sprache –Wissen – Wissenschaft. T¨ubingen: Gunter Narr Verlag. Dervis, Peter A. 2000 Bell bottom blues. Made to Measure Magazine. Hopper, Paul J. and Elizabeth Traugott 1993 Grammaticalization. Cambridge: Cambridge University Press. Mustanoja, Tauno F. 1960 A Middle English Grammar. Part 1. Parts of Speech. Helsinki: Soci´et´e N´eophilologique. Nickel, Gerhard 1966 Die Expanded Form im Altenglischen: Vorkommon, Funktion und Herkunft der Umschreibung beon/wean + Partizip Pr a¨ sens. Habilitationsschrift. Neumu¨ nster: Karl Wachholtz.

Subject index Accessibility 242, 243 Accessible referents 242, 246 Non-accessible topic 250 Ahousaht 200, 202–205, 207, 210, 215 Analogy 3, 9, 33–37 178, 183, 189, 240, 244 Analogical extension 241 Analogical leveling 8 Animacy 8n, 63, 64, 67–69,108, 114, 253 Auxiliary 108, 109, 116, 164, 171, 177–180, 182–187, 189, 206, 265, 266 Categoriality/ 48, Categorial 49, 50, 52, 54, 57, 58, 66, 71, 73, 74 Categorization 265 Causative 87, 88, 98, 99, 115, 117, 120, 122–124, 126, 127, 134, 155, 201, 202 Central Pomo 209, 210 Clitic(s) 15, 16, 178–180, 182, 183, 187, 198, 213, 231, 243–251 Enclitic 173–177, 197, 199, 205–207, 211–216, 218–222, 225, 226 Coercion 10, 12 Coerce 34, 85, 165 Collostruction 34, 37, collostructional 110 Complementation 54, 60–62, 65–68 Complex determiner (CompDet) 23, 27–30, 32, 35, 38 (Non-)compositional 6, 13, 14, 35, 51, 52, 56, 63, 69, 71, 74, 81, 83, 84, 110, 138 Conceptual space 26, 231, 241, 243, 245, 251

Construction grammar (CxG) 1, 2, 4, 5, 7,–13, 17, 24, 25, 30, 34, 36, 37, 48, 50–52, 56, 66, 74, 75, 107, 110, 114, 126, 127, 134, 136, 164, 175, 179, 184 Constructional disharmony 15, 171, 175–177, 180, 183–185, 187, 189 Constructional idiom 13, 14, 81, 83, 84, 85, 90, 91, 96–99, 102, 103 Constructional map/ 51 Functional map 51, 72 Constructional network (Language) contact 3, 8, 10, 11, 16, 195, 199, 208, 210, 212, 214, 221, 222, 224, 225, 240 Corpus 5, 14, 49, 50, 61, 63, 64, 110, 112, 115, 116, 118, 120, 124, 126, 127, 135, 136, 156, 159, 176, 212, 248 Subcorpus 113–115, 118, 120–126 Degrammaticalisation 16, 231, 240, 241, 246, 247, 250–252 Degree modifier 11, 27–29, 33, 35, 38 Demonstratives 7, 16, 24, 67, 202–207, 212–215, 225 Dependent-marked/marking 196, 199, 225 Discourse 52, 110, 112, 113, 124; 195, 199, 202, 209, 210, 225, 242, 243, 246 Disourse pattern(s) 16, 208 Discourse markers 206 Discourse function 26, 31, 32, 34 Doubling 16, 231–239, 241–251 Emphasis 209, 245, 246, 250, 251 Empathy 243, 246, 250, 251, 252 Ergative/absolutive pattern 199, 216, 212, 218

270

Subject index

Ergative 198, 212, 216, 217, 219, 226 Absolutive 198, 212, 216, 226, Formosan 172, 181, 182 Future 4, 9, 11, 14, 35, 107–112, 115, 116, 119–125, 127, 128, 206 FUTURE (ex) 209, 213, 214, 221 Grammaticalization 4, 5, 7,8, 9, 11–14, 16, 17, 23–27, 30, 32–34, 36, 37, 47, 51, 81, 82, 84, 100, 103, 114, 115, 118, 121–127, 134, 151, 157, 163, 164, 165, 196, 231, 232, 240, 241, 243–247, 250, 251, 262, 265, 266 Haisla 208, 212, 214, 215, 218, 222–224 Head-marking 195–198, 225 Heiltsuk 208, 212, 214, 215, 218, 222–224 Hierarchical lexicon 98 Hierarchy 98 Hierarchical organization 90 Inchoative 14, 87–92, 99, 109, 116–122, 125, 127 Ingressive 109 Inheritance tree 14, 85, 98, 99, 103 Inseparable prefixes 133, 138 -be- ‘about’ (English inseparable prefix) 14, 144, 152 (ex.), 161 (ex.) -to- ‘asunder’ (English inseparable prefix) 133, 135, 142 (ex), 144 (ex), 150 Intention 14, 108, 111, 112, 114, 116–123, 125, 126, 127, 137 Kwakw’ala 196, 199, 200, 205, 206, 208, 211–215, 218, 222–224, 226 Left-dislocation 241–243 Lexicalization 25, 103, 262 Metaphor

17, 140, 154, 155, 264

Metaphorical use 84, 95 Metonymy 29, 265 Middle English 5, 28, 133, 135, 136, 138, 141, 143, 149, 151, 152, 157, 158, 163, 265 Mirror-image reversal 231, 246, 247, 251 Modification 12, 49–52, 54, 60–69, 71, 72, 74 Morphology/ 8, 10, 12, 17, 31, 33, 52, 73, 99, 102, 172, 216, 242, 244, 245 Morphosemantic 47, 49, 56, 70, 71, 73 Movement 108, 113–120, 123, 124, 126, 127 Independent movements 17 Physical movement 9 Negation 10, 175 Negative 171, 172, 176, 178, 180, 183, 185–188 Negator 176, 181, 184 Negative polarity 30, 31, 33, 34, 38 Nominative/accusative pattern 199 Incorporation 85, 97 Non-predicative prefix construction 134, 138, 144, 145, 147–150, 157 Northwest Coast 16, 195, 196, 205, 207, 208, 212, 216, 225 Nuuchahnlth 200, 208, 215, 223 Old English 5, 28, 133, 135, 136, 138, 141, 143, 147, 149, 151, 152, 157, 158, 163, 265, 266 OV to VO in English (= word order change) 15, 134, 149, 164 Partial transition(s)/change(s) 47, 49, 51, 71 Particle adjective (PA) 48–50, 52–72, 74, 76 Particle verb 91, 93, 95–97 Verb particle 6 Partitive 11, 23, 27, 29, 30, 32, 33, 36, 38

Subject index Periphrasis 13, 85, 99, 100, 101, 102 Postural 81–83 Predication 12, 49, 61–65, 68 Predicative prefix construction 134, 138, 141, 143, 144, 148, 149 Progressive 4, 11, 13, 14, 81–87, 90–93, 95, 96, 98, 100, 102, 103, 221, 265, 266 Prosody 16, 195, 199, 200, 203, 205–207, 209, 225, 263 Proto Austronesian 171, 172, 176, 177, 180–183, 190 Purposive 9, 115, 119, 120, 123, 124, 126, 267 Puyuma 11, 15, 171–180, 182–190 Quasi-incorporation

93, 96, 99

Radical Construction Grammar 11, 23, 25, 26, 30, 32, 34, 35, 37, 38, 50, 184 Reanalysis 3, 8, 9, 10, 28, 33, 36, 37, 38, 140, 141, 153, 154, 180, 182, 216, 241 Relationship among constructions 171, 184–187 Rhetorical construction 16, 197, 221, 224–226 Rhetorical strategy 16, 195, 218 Salience (constructional, semantic, syntactic) 14, 15, 134, 144, 149, 157, 158, 163, 164, 242 Salience of contrastive function 58 Diminished salience of verbal features 69 Semantics 10, 13, 14, 24, 25, 31, 63, 67, 68, 70, 72, 73; 107, 110, 121–123, 125, 127, 135, 137, 138, 149, 150, 153–155, 157, 158, 163 Semantic structure 263 Route path semantics 140, 147, 148

271

Swedish 11, 14, 107–113, 116, 120, 122, 124, 125, 127 Old Swedish 114, 118, 119 Syntagmatic (context) 23, 49, 63, 71, 72, 73, 74, 75 Syntagmatic strings 47, 69 Syntagmatic structures 3 Syntagmatic combinability 4 Syntagmatic formation 4 Syntagmatic relation 4 Syntagmatic configurations 8 Syntax 5, 16, 17, 24, 27, 31, 33, 47, 54–56, 60, 61, 67, 71, 85, 87, 158, 163, 209, 210, 249, 261–264, 266 267 Tsimshianic family 16, 197, 198, 208, 221, 223, 226 Tsimshian 198, 199, 208, 211, 212, 214, 216–218, 222–226 Topic 197, 209, 216, 242, 246, 250 Topic domain 5 Topicality 8 Active topic 246, 251 Topic-shifting 241–243 Transitivity 15, 60, 61–63, 65, 69, 154, 172, 190, 216 Unification 98, 137, 139, 140, 143, 146, 153, 158 Unification-based formalism 75 Wakashan family 16, 196, 200, 205, 208, 218, 223, 226 Word order 5, 11, 15, 16, 23, 53, 64, 66–68, 113, 209, 210, 232–241, 243–246, 249, 251 Valency frame 138, 153, 165 Valency pattern 148, 155, 157–160, 162, 163 Valency features 14, 15