Contexts and Constructions (Constructional Approaches to Language)

Contexts and Constructions

Constructional Approaches to Language The series brings together research conducted within different constructional models and makes them available to scholars and students working in this and other related fields. The topics range from descriptions of grammatical phenomena in different languages to theoretical issues concerning language acquisition, language change, and language use. The foundation of constructional research is provided by the model known as Construction Grammar (including Frame Semantics). The book series publishes studies in which this model is developed in new directions and extended through alternative approaches. Such approaches include cognitive linguistics, conceptual semantics, interaction and discourse, as well as typologically motivated alternatives, with implications both for constructional theories and for their applications in related fields such as communication studies, computational linguistics, AI, neurology, psychology, sociology, and anthropology. This peer reviewed series is committed to innovative research and will include monographs, thematic collections of articles, and introductory textbooks.

Editors Mirjam Fried

Institute for the Czech Language Czech Academy of Sciences Prague, Czech Republic

Jan-Ola Östman

Department of Scandinavian Languages & Literature, University of Helsinki Helsinki, Finland

Advisory Board Peter Auer

Seizi Iwata

Hans C. Boas

Paul Kay

William Croft

Knud Lambrecht

Charles J. Fillmore

Michael Tomasello

Adele E. Goldberg

Arnold M. Zwicky

University of Freiburg, Germany University of Texas at Austin, USA University of New Mexico, USA Int. Computer Science Institute, Berkeley, USA Princeton University, USA

Osaka City University, Japan University of California, Berkeley, USA University of Texas at Austin, USA Max Planck Institute for Evolutionary Anthropology, Germany Stanford University, USA

Volume 9 Contexts and Constructions Edited by Alexander Bergs and Gabriele Diewald

Contexts and Constructions Edited by

Alexander Bergs Universität Osnabrück

Gabriele Diewald Leibniz Universität Hannover

John Benjamins Publishing Company Amsterdamâ•›/â•›Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data Contexts and Constructions / edited by Alexander Bergs, Gabriele Diewald. p. cm. (Constructional Approaches to Language, issn 1573-594X ; v. 9) Includes bibliographical references and index. 1. Pragmatics. 2. Grammar, Comparative and general--Syntax. I. Bergs, Alexander. II. Diewald, Gabriele. P99.4.P72C64â•…â•… 2009 306.44--dc22 isbn 978 90 272 0431 8 (Hb ; alk. paper) isbn 978 90 272 8878 3 (Eb)

2009033621

© 2009 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

Table of contents Contexts and constructions Alexander Bergs and Gabriele Diewald

1

part 1. Context in construction grammar On de-limiting context Marina Terkourafi

17

Constructions and context: When a construction constructs the context Henri-José Deulofeu & Jeanne-Marie Debaisieux

43

Representing contextual factors in language change: Between frames and constructions Mirjam Fried

63

part 2. Interactional approaches Grammatical constructions in dialogue Per Linell Interactional construction grammar: Contextual features of determination in dialectal Swedish Camilla Wide

97

111

part 3. Context and grammar Contextual cues for particle placement: Multiplicity, motivation, modeling Bert Cappelle

145

Conditionals and mental space set-up: Evidence from German word order Ilona Vandergriff

193

Semantics and pragmatics in construction grammar: The case of modal verbs 213 Ronny Boogaart Index of Constructions

243

Subject index

245

Contexts and constructions Alexander Bergs and Gabriele Diewald Universität Osnabrück, Universität Hannover

1. Introduction The notion of “context” in grammar and discourse seems to be one of the most important, most mentioned, and yet most poorly understood topics in linguistics. Whenever we speak, our utterances are always situated in both a language-external context and a language-internal co-text. Both context and co-text (which are usually subsumed under one single label, i.e. ‘context’) may have some systematic impact on the form and function of our utterances, not only just in terms of style, but also with regard to systematic grammatical variability. One particularly interesting notion in the investigation of co- and contextual factors is that of constructions, both in a mostly pre-theoretical sense and in the technical framework of Construction Grammar (CxG) ‘proper’. Here, the complex notion of ‘context’ has surfaced repeatedly and scholars from a wide range of backgrounds have highlighted both the general importance of the concept and the regrettable neglect of it in mainstream linguistics, i.e. in most linguistic theorizing outside of pragmatics and discourse studies (cf. Diewald 2006; Heine 2002). Recently, studies on language change in general and on grammaticalization in particular, which for long have been struggling to come to grips with the impact of contextual factors in linguistic change, have encompassed the constructional notions and combined them with other existing concepts of contexts as “loci for change” (cf. Bergs & Diewald, eds., 2008). This volume offers one of the first concerted attempts at investigating the role of context from a constructional point of view, with a focus on, but no limitation to, Construction Grammar as one theoretical framework. 2. Construction grammar Construction Grammar is a relatively recent and increasingly popular approach in linguistic theory which postulates that language and knowledge of language consist of conventionalized symbolic form-meaning pairings (“constructions”) at all



Alexander Bergs and Gabriele Diewald

Syntactic properties Morphological properties

Construction

Phonological properties

Form symbolic correspondence (link)

Semantic properties Pragmatic properties

(conventional) meaning

Discourse-functional properties

Figure 1.╇ The symbolic structure of constructions (Croft & Cruse 2004:€258)

levels of linguistic structure: lexicon, morphology, syntax, possibly even phonology, and discourse/text. Figure 1 illustrates one possible schematic representation for a construction. Constructions are often required to have some non-compositional aspect in their meaning/function, although ‘sufficient frequency’ has recently also been acknowledged as a criterion for construction status (Goldberg & Jackendoff 2004:€533, n.1; Goldberg 2006:€5). Constructions do not exist in isolation, but form a structured inventory, e.g. through so-called inheritance relations and family resemblances (cf. Fried & Östman 2004). Another important, but mostly under-researched tenet of construction grammar is that constructions incorporate semantics, pragmatics, and discourse properties on their function, i.e. meaning side: “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). So far, many studies in construction grammar have focused on (verb) semantic aspects. Although there are studies on constructions and speech acts, i.e. illocutions (Lakoff 1987), on metalinguistic commentary (Kay 2004), on speaker attitude (Lambrecht 1990; Kay and Fillmore 1999; Kay 2004:€16–19), and on information structuring (Goldberg 1995; Michaelis and Lambrecht 1996; Lambrecht & Lemoine 2005), most have treated pragmatics and discourse more or less in passing, so that there is still no systematic survey of this domain (cf. Kay 2005). The present volume tries to fill this gap. The notion of context is located in



Contexts and constructions

the overlapping area between pragmatics and discourse. The papers presented here explore the possibilities and risks of integrating context into particular constructions and construction grammar in general.€ Topics range from particular language and construction (grammar) specific problems to general technical analyses and proposals, from the analysis of spoken language to diachronic aspects, from questions of polysemy and semantic-pragmatic issues to syntactic and typological questions like sentence subordination and particle placement. The languages treated in the papers of this volume are Dutch, English, French, German, Greek, Old Czech and Swedish. Most contributions make use of empirical corpus data from genres as different as spoken every-day language and diachronic texts. 3. Co-text and context Catford (1965:€31) introduced the distinction between context and co-text. Co-text refers to language internal factors, such as the syntactic or textual environment of a given element, context ‘proper’ refers to language external factors, such as time, place, interlocutors and manner. While this distinction will be upheld in the following for sake of clarity of exposition, note that the difference is not undisputed. Terkourafi (this volume) convincingly shows that there is strong interdependence between the two, even to such an extent that co-text and context are in effect one and the same. Co-textual factors play an important role from a constructional point of view, but also in traditional approaches to grammar. The influence of co-textual factors can be very obvious, for example in the well-established syntactic principles of government and concord, which are direct outcomes of syntagmatic, and hence, co-textual factors. In many languages verbs usually inherently determine the number and the type of their complements. Similarly, subjects and verbs usually agree in features such as person and number. These are some obvious symptoms of co-textual effects, of how some linguistic element determines the type and shape of another one in a given sentence. Other co-textual influences can also be found. Studies in collocations and collostructions (e.g. Stefanowitsch & Gries 2003, Hilpert 2008) have shown that certain elements and constructions exert more or less strong influence on the linguistic company they keep. One of the most simple cases is the well-known collostructional pattern of beautiful woman versus handsome man. Less well-known perhaps is the case of the complex construction “N waiting to happen” as in (1): (1) In short, she is a disaster waiting to happen when her father re-enters her life, (1997–3–10, Time Corpus)





Alexander Bergs and Gabriele Diewald

Stefanowitsch & Gries (2003) clearly show that there is a strong preference for accident to be used as head noun in the construction, followed by disaster. They even go so far as to suggest that this clear correlation warrants a separate entry for accident waiting to happen in the dictionary. Note, however, that despite this strong correlation, the partially filled complex concrete construction “N waiting to happen” itself does not carry any negative connotation and can also be used with positive head nouns and in utterances conveying neutral or even positive news. The negative connotation is only added through co-textual occurrence of certain nouns such as disaster or accident. Apart from traditional syntactic and more recently collostructional factors, a third type of co-textual influence can be found. In some other cases, co-textual influence can make itself felt even across clause and sentence boundaries. In his discussion of the simple present in contemporary English as a marker of futurity, Wekker (1976) mentions that [t]he general principle that seems to hold here is that the simple present can be used only if the temporal specifier is sufficiently near and explicit, so that it can serve as a point of orientation. If there is a danger of ambiguity in using the simple present for the future, the speaker will either have to repeat the temporal specifier, or, as we saw in the longer passages, reintroduce will as a marker of futurity.  (Wekker 1976:€81)

However, he also finds occurrences such as (2), in which the temporal specifier next April is quite removed from the simple present form in question.

(2) Next April, for instance, you’ll see him […] in a six-part BBC serial version of George Eliot’s ‘Daniel Deronda’. He plays the icecold aristocrat Henry Grandcourt. (Wekker 1976:€81)

Even though the two elements are not grammatically related in any way – in fact, they are not in the same clause, not even in the same sentence – the occurrence of next April warrants the use of the simple present form he plays to refer to a future event. Interestingly, will play would also be possible here, but was not used. Here in this case we see co-textual influence at some distance, on a level that is actually textual, beyond the single sentence (see Bergs, in prep.). Quite similarly, Linell (this volume) discusses the form and function of constructions in dialogues and shows how constructions are actually built to fit into particular slots in a given conversation. This, again, is a clear case of an interdependence of co-text and grammar, which is rarely ever recognized by traditional grammatical models. Thus, bridges and interesting connections become visible between construction grammar as one grammatical model and domains from ‘applied linguistics’, such as discourse and conversation analysis. Wide (this volume) discusses some concrete ways in which



Contexts and constructions

the connection between conversation analysis and construction grammar can be developed productively (cf. also the papers in Günthner and Imo 2006). Contextual, i.e. extralinguistic factors in language use, are usually easy to see but very hard to document. This means that their concrete influence is usually very hard to pinpoint and to describe. Contextual factors include time and place of the utterance, interlocutors, including their socio-cultural background, and their world knowledge, among others. Hymes (1974:€53–62), for example, lists sixteen factors which are involved in his s-p-e-a-k-i-n-g model: message form; message content; setting; scene; speaker/sender; addressor; hearer/receiver/audience; addressee; purposes (outcomes); purposes (goals); key; channels; forms of speech; norms of interaction; norms of interpretation; and genres. This is not the place to discuss the influence of each of these components. Suffice it to say that a fairly large number of extra-linguistic, or at least semi-extra-linguistic factors, can be identified which may be immensely important for successful (‘correct’) communication, though not necessarily for ‘grammaticality’ in the strict, traditional sense. Fried (this volume) for example, discusses the role of communicative context in relation to complementation patterns. Linell (this volume) focuses on the role and function of discourse, i.e. multilogueal context. Deulofeu and Debaisieux (this volume) argue that context – i.e. common ground – itself has to be treated as part of the semantics of a construction. 4. Describing and analyzing co- and context in constructions One of the most difficult problems in discussing context is the very “slippery”, “volatile” nature of the object itself. Neither co- nor context, and their corresponding effects, easily lend themselves to principled descriptions and analyses, as, for example, attribute-value matrices (see Wide, this volume). This may be partly due to the fact that traditionally contextual factors – often being deemed as random and chaotic – have been largely ignored. The reason for their apparent randomness, however, appears to be something fundamentally ontological in this case. Co- and context can be static and more or less permanent on the one hand, or dynamic, flexible and rather volatile on the other. This partly resembles what has been described as shared knowledge in functional approaches to grammar and information structuring. Participants have some permanent knowledge base plus some dynamic, active knowledge which is constantly updated and thus everchanging. Similarly, co- and contexts have both a permanent component and some ‘online’ part which is updated as we speak (see Linell and Terkourafi, this volume). The latter is, of course, very hard to capture empirically. Describing context in the narrow sense is also not without problems. Where, for example, should the line be





Alexander Bergs and Gabriele Diewald

drawn? Traditional syntactic theories recognize certain strong contextual effects within the sentence. But some cotextual effects in grammatical structure certainly reach beyond the sentence. Approaches within the construction grammar framework are currently trying to develop ways and means of capturing all these facts and ideas. One of the most prominent is probably the notion of frames and frame semantics (cf. Fried, this volume). Frame semantics essentially links traditional semantic knowledge with encyclopaedic world knowledge. A given word (e.g. flu) activates one or more frames in which it occurs and which provide encyclopaedic world knowledge associated with that particular word (e.g. doctor, sneeze, fever, chicken soup and many more). It is usually claimed that frames, which are based on recurring experiences, should be regarded as gestalt based, i.e. in order to actually understand a given concept, one needs to know essentially all concepts associated with that particular frame. The result is a coherent network of concepts and frames which in turn forms our world knowledge as such. It is not without good reason that frame semantics has been called a ‘sister theory’ to construction grammar (Fried & Östman 2004:€5). Both theories investigate meaning and cognition on the basis of empirical studies of language use and see language rather as a network of concepts and constructions than as organized on an item-plus-rule basis. Therefore, there have been growing attempts to integrate frame semantics into construction grammar, and vice versa. 5. This volume This volume presents eight original research papers which all deal with the question of context and constructions. While some of them are explicitly couched in construction grammar as such, others deal with the notion of grammatical constructions from a more pre-theoretical perspective. For coherence reasons, the papers were organized in three different sections: i. Context in construction grammar ii. Interactional approaches iii. Context and grammar i. Context in construction grammar Marina Terkourafi presents work “On De-Limiting Context”. She points out that psycholinguistic findings have shown that pragmatic consequences of the speaker’s utterance are in fact realized quite early on in utterance production and are not necessarily the endpoint of any given process of speech production. Linguistic as well as extralinguistic contexts seemingly play a crucial rule in this process. Linguistic context in this case is treated in the same way as extra-linguistic contexts, since linguistic



Contexts and constructions

context already ‘contains’, in a way, extra-linguistic context. In other words, Terkourafi argues that there is no fundamental, systemic difference between the two. This paper examines the extent to which results from the resolution of lexical and syntactic ambiguities, reference assignment, and the processing of semantic versus pragmatic anomaly, naturally extend to illocutionary force assignment, the meaning component pertaining to understanding the pragmatic consequences of the speaker’s utterance par excellence. It is proposed that illocutionary force is projected probabilistically from a limited set of contextual parameters pertaining to who is talking to whom, when, and where. Subsequent linguistic and contextual information serves to check this probabilistic prediction, either confirming it (illocutionary assignment passing unnoticed), or revising it in the light of incoming data. On the basis of detailed and comprehensive empirical data from spoken dialogical Cypriot Greek, Terkourafi shows that different minimal contexts correlate positively with different linguistic constructions. This means that a limited set of contextual parameters jointly considered can provide a pointer to the preferred linguistic construction realizing a particular speech act, rendering reference to the fully specified, nonce context of occurrence superfluous (at least at an initial stage). The steady growth of interpretation suggests that interpretation begins as soon as a speaker engages in speaking, and proceeds in parallel with the speaker’s utterance, drawing on contextual information as and when this becomes available. Contextual parameters that can be fixed early on during an exchange prime the conventionally expected illocutionary force, thereby speeding up interpretation. Through acculturation, these contextual parameters gain in salience: interlocutors seek to settle these (if only by presumption) in order just to initiate interaction. The contextual parameters jointly constituting a minimal context are motivated by Gibsonian affordances applied to physical settings. In this way, a potentially unlimited body of sociocultural knowledge is shaped into a limited set of contextual parameters available for propositional manipulation. The resulting minimal contexts are neither impoverished with respect to the original input, nor reducible to it. Furthermore, the contextual parameters jointly constituting a minimal context cannot be broken down further into more abstract ones, such as relative status or power, or social distance, but are rather mentally manipulated as primes. The second paper in this section is by Henri-José Deulofeu and Jeanne-Marie Debaisieux who investigate “Constructions and Context: When a construction constructs the context” with the aim to explore the interrelations of a particular syntactic pattern (various uses of parce que clauses in spoken French) with its communicative context. Taking Fillmore, Kay & O’Connor’s famous let alone study as their starting point, they claim that there are constructions – not unlike let alone – which bring in contextual background information to maximize the “relevance” of the utterance they are added to by introducing contextually present





Alexander Bergs and Gabriele Diewald

background information which the speaker assumes the addressee needs to know in order to make most sense of the message. The underlying assumption is that some pieces of utterances may not only be text, or parts of text, but actually more like contextual background. The authors compare the status of these pieces to that of footnotes in written language, i.e. they are, strictly speaking, marginal to the text and nearer to context status than to plain text status. Their function is to shape the relevant part of the context in a verbal construction. These subconstructions inserted in utterances have a special status that must be accounted for in linguistic theory and in construction grammar. Their special status can be revealed in that they show idiosyncratic formal as well as pragmatic properties. The exemplary case study is on constructions which are introduced by subordinators and which are thus formally similar to prototypical subordinated clauses. One of these can be parce que. The authors show that some uses of parce que are quite specific. It has neither the properties of subordinating parce que (focus of questions, of cleft sentences, etc.), nor even of the parce que / cos’ introducing supplementive clauses. The authors argue that this use of parce que forms a special construction which they call the Context Focussing Parce que Construction (CFPC). This construction inherits its formal properties both from “supplementive” constructions (resistance against all prototypical adjuncts diagnostic tests) and from a general Parenthetical Construction, the function of which is to insert extra material in a clause, as a side remark. The function of this construction is to focus not on the truth conditions of the main construction but on the “relevance” of this construction and to convey a procedural meaning like: “take in consideration this specific piece of contextual information to process the current utterance meaning and to achieve giving relevance to it”. It is claimed that CFPCs are constructions in the true sense of the word as they are specific combinations of form and meaning, whereby the meaning expressed by them transcends regular propositional meaning (or modification of that meaning) as it belongs to the domain of information processing and the structuring of common ground. The authors suggest extending the notion of linguistic construction to encompass paralinguistic signs made use of in human communication, like gestures and other non-verbal communicative behavior. Mirjam Fried is concerned with the specific question of frames and constructions in her contribution with the title “Representing Contextual Factors in Language Change”. In this corpus-based case study she explores the question of how context enters conventional linguistic patterning and becomes part of the codified relationship between meaning and form in a given expression. In particular, she looks at the question of in which ways the complex interaction between traditional semantic structure, thematic and cultural context, and systematic morphosyntactic patterns can be captured and discussed within construction grammar. Her case study deals with the semantic and morphosyntactic history of the Old Czech



Contexts and constructions

(OCz) participial adjective (PA) věřící ‘(one) believing’, in relation to the verb věřiti ‘to believe’, which, because of its flexible nature, can be located somewhere between inflection and derivation. The OCz verb věřiti developed four senses (roughly, belief, faith, credit, trust), each associated with a distinct complementation pattern and reflecting particular communicative context (e.g., Christian religion, commercial transactions, transfer of responsibility). Using an extensive corpus from OCz texts containing a wide variety of genres, Fried show that while the PA originates within the same semantic structure as part of the verbal inflectional paradigm, its functional and semantic range became fixed differently across the verb senses and different lexico-pragmatic properties correlate not only with distinct thematic contexts but also with different syntactic preferences, all of which contributes to fixing the functional status of the PA as a (quasi-actor) noun: ‘a Christian’; ‘a creditor’. The development occurs along several dimensions: conceptual shifts vis-à-vis textual distribution and contextual restrictions; the role of collocational patterns; recurring syntactic patterning; chronological layering. The result is a complex picture of semantic, pragmatic, and grammatical factors that collectively motivate the changes in the form’s usage and that must be part of any representation that aspires to be descriptively accurate and to have an explanatory value. In that respect, the present study makes a case for Frame Semantics (e.g. Fillmore 1982) and Construction Grammar (Fillmore 1989; Fried & Östman 2004) as a particularly suitable model for representing lexico-grammatical networks in which the relative stability of grammatical form does not conflict with the relative flexibility of meaning or expressive richness, and vice versa. The paper thus establishes (i) the interdependence of lexical meaning, the context in which the word is used, its illocutionary force, and the morphosyntactic form that encodes these relationships; (ii) how these factors figure in categorial shifts known as ‘transpositional morphology’, and (iii) how the relevant generalizations about context in grammar can (or cannot) be formally represented in terms of complex grammatical signs, through frames and constructions. ii. Interactional approaches In the first paper of this section, Per Linell investigates “Constructions in Dialogue”. With reference to the notion of ‘constructional schema’, his contribution is based on a distinctly ‘dialogical’, i.e. interactional and contextual, approach to language in general and to grammatical constructions in particular. Some grammatical constructions, e.g. cleft constructions and the so-called x-and-x construction are analyzed in dialogical terms. The analyses give rise to the argument that the opportunities for using a particular construction are largely built up dialogically over sequences. Conversely, constructions often have properties that show how





Alexander Bergs and Gabriele Diewald

they are built precisely to fit into sequences, with responsive properties (requiring certain structures in prior co-text) and/or projective properties (setting up conditions on possible next units). This suggests a way of developing the notion of ‘outer syntax’ in CxG. Simultaneously, the author also discusses some general questions of constructions and context, such as whether contexts should be described in the same kinds of terms as constructions themselves or whether it is at all possible to formalize relevant contexts exhaustively. In her contribution, Camilla Wide analyzes “Contextual Features from the Point of View of CxG and Interaction”. One of her primary concerns in this paper is how to formalize contextual features in terms of attributes and values. This paper explores how contextual features can be described within the CxG framework – without compromising the basic assumptions made within CA-inspired interactional linguistics. In his early work on Construction Grammar, Fillmore stressed the importance of considering communicative purposes and interactional features when describing grammatical constructions. In recent years some versions of Construction Grammar have become increasingly more formalistic and abstract; this means, for Wide, that linguists from an interactional point of view have found it increasingly difficult to relate to CxG approaches and ideas. In her contribution, Wide analyzes the demonstrative pronoun den ‘the, that’ in Swedish from both a Construction Grammar and an interactional point of view. Special attention is paid to the use of den in Swedish dialects in Eastern Nyland in Finland where the grammaticalization of den toward a definite article, supplementing the normal -(e) n/-et definite article suffix, seems to have gone further than in standard varieties of Swedish. In standard Swedish den is used as a definite article only in noun phrases including modifiers, cf. bil-en ‘the car’ versus den röda bil-en ‘the red car-the’. Thus the Eastern Nyland use of den in noun phrases without modifiers, as in den biln, has traditionally been analyzed as a case of double definiteness. However, the use of den in Eastern Nyland is closely connected to assumptions about common ground of a certain kind. Since den is a deictic pronoun, contextual information plays a crucial role in the use, interpretation and description of structures that include den. When analyzed more closely, the Eastern Nyland structures with den seem to be licensed by a construction which draws primarily on contextual information. In order to formalize this construction, tools for describing various aspects of contextual information are needed. The paper suggests a solution in which these tools are developed in terms of meaning potentials, in this case the potential of den as a (discourse) focus marker and rhetorical device. iii. Context and grammar Bert Cappelle takes a fresh look at “Contextual Cues for Particle Placement” in his contribution. This paper discusses some contextual factors that influence



Contexts and constructions

constituent ordering in the English transitive verb-particle construction. Cappelle not only recognizes and discusses traditional factors that seem to influence particle placement, such as discourse old versus discourse new information, he also introduces a number of new contextual factors. One such factor is focality, i.e. particular scenarios which put focus on one or more elements in the clause. Other factors include syntactic priming, accessibility, and encyclopaedic world knowledge. This paper thus adduces evidence that the choice of one word-order pattern over another is prompted by more than just the assessment of the relative discourse-newness of the words (or their referents) involved. At least in the case of particle placement, the selection of a particular ordering is cued by contextual kinds of information that go far beyond what could be retrieved by an automatic processor and that can only be gathered via a full linguistic and extra-linguistic understanding of the situation referred to. Ilona Vandergriff’s contribution focuses on “Conditionals and mental space setup”. She analyzes the syntactic relation between the clauses and its role in constructing meaning. German syntax, unlike English, does not offer a marker in the main clause which signals the dependency between clauses. Some spoken German ‘if ’-clauses, by contrast, are not syntactically integrated. Instead, the main clause will have the verb position of an independent main clause. Suggesting an iconic relationship between syntactic and conceptual dependency, content conditionals with their real-world relationship between the clauses are generally described with integrative word order, whereas speech-act conditionals are taken to signal their looser conceptual dependency via non-integrative word order. These correlations are relevant to research on conditionality beyond German because the different orderings appear to provide a formal marker of the different cognitive domains in which conditionals are interpreted. This paper considers clause integration as a formal aspect of conditional sentences in German within the framework of Construction Grammar. From this perspective, conditional sentences are examples of a construction. Constructions may have predictable meanings, which are mapped onto the formal and semantic parameters or their combinations. In order to see when the variant orderings do and do not occur in speech act conditionality, this paper analyzes authentic discourse data from the perspective of Mental Spaces theory and proposes an account of the syntactic variation in speech-act conditionals that is consistent with previous analyses of similar complex clause phenomena. The findings show that clause integration has a predictable meaning but what is being marked is not the cognitive domain in which the conditional is interpreted but rather the mental space. Ronny Boogart concludes the third section with his investigation of “Semantics and Pragmatics in Construction Grammar”. He presents a case study on modal verbs in which he shows that the age-old debate concerning monosemous ver-





Alexander Bergs and Gabriele Diewald

sus polysemous readings is marred with a number of theoretical problems and obstacles which can be overcome within a construction grammar approach. With the Dutch modal kunnen (‘can’), for example, it can be shown that in a network of constructions, the property of complete inheritance takes care of the monosemy of the construction. In other words, this reading forms the central, parental kunnen construction which gives its basic meaning to all its linked subconstructions. At the same time, certain other conventionalized uses, such as indirect speech acts, form separate constructions in the network. The number of new constructions in the network is restricted by the definition of constructions as such: conventionalized pairings of meaning and form. In other words, a supposedly new construction not only needs to show a new meaning, but also some new formal aspects. In some cases, Boogart argues, these formal differences can only be found in the surrounding discourse. For him, however, this also raises the important question of where to stop with the inclusion of contextual information. 6. Outlook This volume is one of the first to systematically investigate the interaction of cotext, context and grammar. In particular, the contributions are concerned with the notions of co- and context in general and especially in interaction, and how the two can be integrated in the analysis of grammatical constructions and in construction grammar. Needless to say, there are still a number of unresolved problems and open questions. While the contributions in this volume present excellent case studies and many interesting points for departure for future analysis, three important aspects will have to be discussed in the future. First, we need to look some more into questions of formalizing co- and context in grammatical analyses. The studies presented here point towards frame semantics as one possibility in this regard. Second, we need to find better, i.e. clear and convincing dividing lines between ad hoc pragmatics and general common sense reasoning, both of which are part of everyday interaction and crucial for successful communication, and semantico-pragmatic factors, including co- and context as an important part of the grammatical system, perhaps as conventionalized parts of grammatical constructions. Third, there is a need for a model which, in a systematic way, describes the processes and motives for the transition between cotext(s) and context(s) on the one hand, and idiosyncratic, individual interpretive activity versus stereotypical, i.e. conventionalized pragmatic factors on the other hand.



Contexts and constructions

References Bergs, Alexander. In prep. Expressions of futurity in contemporary English: A Construction Grammar perspective. Bergs, Alexander & Gabriele Diewald (Eds.). 2008. Constructions and language change. Berlin & New York: Mouton de Gruyter. Catford, J. C. 1965. A linguistic theory of translation. London: Oxford University Press. Croft, William. 2001. Radical Construction Grammar. Oxford: Oxford University Press Croft, William & D. Alan Cruse. 2004. Cognitive Linguistics. Cambridge: Cambridge University Press. Diewald, Gabriele. 2006. Context types in grammaticalization as constructions. In: D. Schönefeld (Ed.), Constructions all over – case studies and theoretical implications. [Special Volume 1 of Constructions]. Online publication at http://www.constructions-online.de/articles/specvol1/. Fillmore, Charles. 1982. Frame Semantics. In: The Linguistic Society of Korea (Ed.), Linguistics in the Morning Calm. (111–137). Seoul: Hanshin. Fillmore, Charles. 1989. Grammatical Construction Theory and the familiar dichotomies. In: R. Dietrich, & C.F. Graumann (Eds). Language processing in social context. (17–38). Amsterdam: North Holland/Elsevier. Fried, Mirjam & Jan-Ola Östman. 2004. Construction Grammar: A thumbnail sketch. In: M. Fried & J.-O. Östman (Eds), Construction Grammar in a cross-language perspective (11–86). Amsterdam/Philadelphia: John Benjamins. Goldberg, Adele. 1995. Constructions. A Construction Grammar approach to argument structure. Chicago: University of Chicago Press. Goldberg, Adele. 2006. Constructions at work. Oxford: Oxford University Press. Goldberg, Adele & Ray Jackendoff. 2004. The English resultative as a family of constructions. Language 8(3): 532–567. Günthner, Susanne & Wolfgang Imo (Eds). 2006. Konstruktionen in der Interaktion. Berlin & New York: Mouton de Gruyter. Heine, Bernd. 2002. On the role of context in grammaticalization. In: I. Wischer & G. Diewald (Eds.), New reflections on grammaticalization (83–101). Amsterdam & Philadelphia: John Benjamins. Hilpert, Martin. 2008. New evidence against the modularity of grammar: Constructions, collocations, and speech perception. Cognitive Linguistics 19(3): 391–411. Hymes, Dell. 1974. Foundations in sociolinguistics: An ethnographic approach. Philadelphia: University of Pennsylvania Press. Kay, Paul. 2002. English subjectless tagged sentences. Language 78(3): 453–481. Kay, Paul. 2004. Pragmatic aspects of constructions. In: L. Horn & G. Ward, (Eds.), The Handbook of Pragmatics (675–700). Oxford: Blackwell. Kay, Paul. 2005. Argument-structure constructions and the argument-adjunct distinction. In: M. Fried & H. Boas, (Eds.), Grammatical constructions: Back to the roots (71–98). Amsterdam: John Benjamins. Kay, Paul & Charles Fillmore. 1999. Grammatical constructions and linguistic generalizations: The What’s X doing Y? construction. Language 75(1): 1–33. Lakoff, George. 1987. Women, fire, and dangerous things. Chicago: University of Chicago Press. Lambrecht, Knud. 1990. ‘What me worry?’ – Mad Magazine sentences revisited. BLS 6: 215– 228.





Alexander Bergs and Gabriele Diewald Lambrecht, Knud & L. Michaelis. 1996. Toward a construction-based theory of language function: The case of nominal extraposition. Language 72(2): 215–247. Lambrecht, Knud & Kevin Lemoine. 2005. Definite null objects in (spoken) French. A Construction-Grammar account. In: Mirjam Fried & Hans Boas (Eds.), Construction Grammar: Back to the roots (13–55). Amsterdam/Philadelphia: John Benjamins. Stefanowitsch, Anatol and Stefan Th. Gries. 2003. Collostructions: Investigating the interaction of words and constructions. International Journal of Corpus Linguistics 8(2): 209–243. Wekker, Herman Chr. 1976. The expression of future time in contemporary British English. Amsterdam: North Holland.

part 1

Context in construction grammar

On de-limiting context* Marina Terkourafi

Assistant Professor, Department of Linguistics, University of Illinois at Urbana-Champaign

1. The sequentiality of linguistic analysis: A received view There is a received view within the generative tradition in linguistics that levels of analysis, starting with phonetics and ending with pragmatics, are sequentially ordered.1 Proposed if only as a methodological division — not necessarily a psychologically defensible one — this view would run roughly as follows: phonetics furnishes the input to phonology, phonology to syntax, syntax to semantics, and, crucially for our purposes, semantics to pragmatics. That is, each of these levels uses the output of the previous level and furnishes the input to the next. This is the path followed during utterance interpretation. To reverse path is followed during utterance production. The first part of this chain, up to and including semantics, corresponds to the decoding phase and draws on our knowledge of language. The tail of the chain, the passage from semantics to pragmatics, corresponds to the inferential phase. During this last phase, we gradually expand the information we take into account, to include, further to knowledge of language, other aspects of world knowledge, such as who is talking to whom, at what time and place, within which social framework, and so on. The purpose of this process is to discover the speaker’s intended meaning. To do this, we first need to find out what the sentence means (its truth-conditions),2 and, based on that, to infer what the speaker meant by using it (presumably, so that we may react accordingly). Of course, it has been known for some time now that to fix the truth-conditions of a sentence we may actually have to go beyond semantics, i.e. beyond knowledge of language, and into pragmatics, i.e. draw on the surrounding context (see the discussion in Levinson 2000:€165ff. and the references therein). Some classical cases are lexical or syntactic ambiguity resolution, and fixing the referents of referring expressions, including indexicals. In all these cases, the truth-conditions of the sentence cannot be fixed until context has been taken into account. Context thus understood includes first and foremost the surrounding linguistic context, i.e. the utterance as part of a larger discourse. Many ambiguities are



Marina Terkourafi

routinely resolved with reference to linguistic context alone. The juxtaposition of utterances can help resolve some cases of reference assignment, including anaphora resolution. Even certain non truth-conditional aspects of meaning, such as illocutionary force, can be determined by looking at the surrounding linguistic context, as when an utterance is understood as an answer to a question in virtue of being adjacent to it in the discourse. Yet other aspects of truth-conditional meaning cannot be determined by looking at linguistic context alone. Determining the referents of indexicals such as ‘this,’ or ‘that,’ ‘here,’ and ‘now’ requires a reference to the extralinguistic context of utterance. Similarly, an utterance such as di chi è il cane? (‘whose is the dog?’) shouted by an exasperated male voice in Florence during the afternoon siesta time, treated the referent of il cane as known — though there was no previous discourse in which it could have been mentioned — in virtue of the perceptual salience of a barking dog in the surrounding extra-linguistic context. We are then left with two solutions, as far as the relationship between semantics and pragmatics goes.3 One is to consider that our semantics is underspecified; that is, semantics narrows down the possibilities of interpretation to a set of alternatives, but does not pick one alternative.4 This is done by pragmatics, which is now the level where truth-conditions are assigned. The second solution is to allow for semantic and pragmatic processes to be interleaved, i.e. to run in parallel. This solution essentially amounts to abandoning the strict sequentiality of levels assumed at the outset. The constant to- and fro-ing between semantics and pragmatics is, nevertheless, still regulated by semantics: pragmatic contributions to truthconditions are made only as and when triggered by semantics.5 This of course does not include non truth-conditional aspects of meaning, such as illocutionary force assignment or other (particularized) implicatures, which by definition require a reference to the extra-linguistic context of utterance. In sum, on the received view, language processing proceeds largely independently of context. Context is brought in during the final inferential phase, or, if needed, a bit earlier, during the last stages of semantic analysis — but not before. Utterances such as “water is H2O” or “carrots are good for health” are assigned truth-conditions independently of contextual input. Other utterances, for instance those containing ambiguous or referring expressions and many others too (for a list of examples, see, e.g. the appendix in Bach 1994), need to refer to the context of utterance, moving from its linguistic to extra-linguistic components, to fix their truthconditions. However, to receive a full interpretation, one that encompasses additional, non truth-conditional aspects of speaker’s meaning such as implicatures or indirect speech acts, all utterances must refer to context, again starting from linguistic context and only secondarily taking into account extra-linguistic context.



On de-limiting context 

2. An alternative: A constitutive role for context Now consider an alternative to the received view. Assume that you are attending a Pragmatics Conference in Northern Italy. You get up in the morning, have a shower, and go downstairs to have breakfast. There, you are asked by someone (whom you take to be another guest, given that they are not wearing the hotel uniform etc.) Scusi, sa dov’é il caffè? You don’t speak Italian, but you do recognize one wordform from your interlocutor’s utterance: caffè. So you point them to the coffee pot. Was that the correct response? It turns out it was. How did you get there? Combining (at least) five pieces of information: a) a similarity in phonological form between English /'kofi:/ and Italian /kaf ’fε/; b) a perceptually given fact about the place of the exchange (the hotel’s breakfast lounge); c) a perceptually given fact about the time of the exchange (morning); d) a socio-culturally given piece of information that in the morning people have breakfast; and e) a socio-culturally given piece of information that people in the western world today often have coffee at breakfast. Only the first of these, (a), draws on knowledge of language. The remaining four pieces of information draw on knowledge of the world, specifically, on perceptually given (b, c) and sociocultural aspects (d, e) of the extra-linguistic context of utterance. Had your interlocutor approached you at a different time and in a different physical setting, say at the exit to one of the presentation rooms at the start of the coffee break, this time you would be likely to point them to the area reserved for coffee breaks. And you would not necessarily have assumed that they wanted to drink coffee. ‘Coffee’ in this case could have been used loosely to refer to activities that accompany the drinking of coffee (as when one is invited for “coffee”, but may well drink tea). Yet on another occasion, if your interlocutor had approached you in the middle of the street and asked the same question, you may not have understood him or her at all. The linguistic information — the phonological similarity between English /'kofi:/ and Italian /kaf 'fε/ — would have still been available. But on its own it would have been hardly sufficient. Lacking a facilitating extra-linguistic context, making any sense of your interlocutor’s question would have seemed virtually impossible. One may of course object that making any sense of any utterance in any unfamiliar language is pure guesswork, and that the processes involved are qualitatively very different from those involved in communicating in a familiar language, not to mention one’s mother tongue. However this objection turns out to be untenable for at least two reasons. The first has to do with the nature of the information combined to respond to the Italian guest’s utterance. Predictably enough, this includes information about the time and place of the exchange, and relevant pieces of sociocultural information activated by the former, perceptually-based data.



Marina Terkourafi

The sheer fact that it is possible to enumerate in a principled manner the pieces of information used shows that this process is not as random as it might appear at first. Rather, what seem to be involved are successively structured layers of world knowledge, much like, e.g., Schank and Abelson’s (1977) proposed continuum from general to specific world knowledge. According to Schank and Abelson, knowledge about the world ranges along a continuum from general knowledge, that is, “knowledge [which] enables a person to understand and interpret another person’s actions simply because the other person is a human being with certain standard needs who lives in a world which has certain standard methods of getting those needs fulfilled,” to specific knowledge, that is, “knowledge [we use] to interpret and participate in events we have been through many times” (1977:€37). In the example with the Italian hotel guest, determining the time of day as morning is achieved with reference to general world knowledge (including perceptual data). This in turn activates the assumption that breakfast is an activity that is relevant to the time of day. Other pieces of information are a matter of knowledge that is increasingly specific. For instance, determining the location as a hotel breakfast lounge will depend not only on perceptual data but also on one’s familiarity with hotels and the layout of rooms and furniture therein. Relating the unfamiliar signifier /kaf 'fε/ to the familiar signified ‘coffee’ will probably be mediated by socioculturally specific knowledge that in today’s western societies coffee is a staple part of breakfast (at least for many people). Since other beverages may have been common in other parts of the world, or during different historical periods, to link the phonological form /kaf 'fε/ to its intended referent, both this piece of socioculturally specific knowledge, and the more general information that Italy is a western country must be available to the understander. In other words, the more similar the speakers’ backgrounds — i.e. the more similar their knowledge hierarchies passing from general to specific world knowledge — the more successful they will be at ‘guessing’ each other’s point. The second reason why making sense of an utterance in an unfamiliar language is not qualitatively unrelated to communicating in a familiar language is that the type of informed ‘guesswork’ combining knowledge from different sources outlined above seems to be involved also in situations that are considered central to linguistic theorizing, starting with language acquisition itself. In a series of experiments pertaining to children’s early acquisition of verbs, Tomasello (1995) showed that children rely crucially on what they take to be the adult’s intention in the situation: “what the adult is doing in a given situation and why she is doing it” (1995:€ 117). Children begin to develop social-intersubjective knowledge of this kind around 9 months (Tomasello 1999:€61–77), so it is available to draw on when acquiring language later on. A particularly frequent context for early verb acquisition is that of impending action by the infant: the majority of verb forms issued by



On de-limiting context 

mothers and correctly responded to by children were “anticipations of or requests for child actions” (1995:€123). In this respect, both mothers and children seem to be utilizing “attention as intentional perception” (1995:€136), that is, to be relying on the fact that “organisms attend to things that are relevant to their current intentional states” (ibid.), the former to guide children’s actions, and the latter to carve out of mothers’ speech the intended referent actions of their verb forms (since, unlike nouns, these cannot be ostensibly taught).6 Thus, the bedrock on which the initial stages of language acquisition are built is children’s understanding of the intentional actions of other persons, including their understanding of adults’ social actions, such as making linguistic reference to their shared world (1995:€140).

Similarly, Eve Clark observes, by age two, children are already quite skilled at tracking the speaker’s intentions in a variety of settings. These intentions may be obscured for young children by language that is too complex, but in tasks where children are focused on the activity at hand, they show good understanding of the pragmatic consequences of speaker repairs [one aspect of speaker meaning: MT] (2004:€572; emphasis added).7

In other words, language acquisition draws more on informed ‘guesswork’ of a pragmatic nature than on specifically linguistic, semantic decoding to start with — and may even proceed along the lines of the former in spite of the latter! All this suggests that, contrary to the received view, rather than constituting the end-point of utterance interpretation, understanding the pragmatic consequences of the speaker’s utterance is in fact achieved early on during this process drawing on all available clues, including context. What is more, understanding these pragmatic consequences and providing an appropriate response may well be the driving force behind the acquisition of more abstract linguistic categories altogether. And, to the extent that linguistic categories are “dynamic, context-sensitive (relational), and therefore plastic [i.e., malleable; MT]” (Hawkins and Smith 2001:€27–30), their formation does not stop after the normal period of language acquisition. In fact, understanding the pragmatic consequences of the speaker’s utterance may well drive the formation of linguistic categories throughout one’s life. However, a skeptic may still wonder: can we say that when you responded correctly to your Italian interlocutor’s question in the breakfast room you knew the meaning of the sentence uttered? Or that, when a two-year-old hears a familiar expression among several new ones, and goes on to respond appropriately, he or she knows the meaning of the sentence uttered? In short, is understanding the pragmatic consequences of the speaker’s utterance the same thing as understanding the meaning of their utterance?



Marina Terkourafi

Practically, the two are indistinguishable. To the extent that knowledge of meaning is in the head, it cannot be empirically observed. What can be observed are its pragmatic consequences, and this is how knowledge of meaning is externalized. But when knowledge of meaning is externalized in this way, there are only two possibilities: providing a pragmatically appropriate response, and providing a pragmatically inappropriate response. These do not suffice to distinguish between knowing the meaning of the sentence uttered and not knowing it (“what the world would have to be like for the sentence to be true”), since, as we saw earlier, it is possible in both cases to provide a pragmatically appropriate response. So, at a practical level, there is no way to tell the difference between knowing the meaning of an expression and responding appropriately to it. At this empirically observable level, the two are equivalent.8 Yet, this could be merely a surface coincidence, obscuring fundamental differences in substance between the two. The earlier discussion of language acquisition helps to lay such worries to rest, suggesting that the difference between knowing the meaning of an expression and responding appropriately to it is more a matter of degree and less one of substance. If, as Hawkins and Smith aptly put it, “denotative meaning might lie at the intersection of all these variants on connotative meaning” (2001:€41), knowing the meaning of an expression and providing a pragmatically appropriate response cannot be radically different: the ‘generative’ relation between them guarantees that they are made of the same ‘stuff ’, so to speak. Hence, if one (providing a pragmatically appropriate response) is imbibed with context — as in the cases of responding to an utterance in an unfamiliar language and in language acquisition — so should the other (knowing the meaning of an expression). Contrary to the received view, then, understanding the pragmatic consequences of the speaker’s utterance is in fact achieved early on during utterance interpretation. Since understanding the pragmatic consequences of the speaker’s utterance relies crucially on context, contextual input must be integrated into utterance interpretation from the first. No hard and fast line can be drawn between linguistic and extra-linguistic context in this respect. Because linguistic context is actually nothing more than language already interpreted — if you prefer, ‘fixedÂ�’ — with respect to some extra-linguistic context, linguistic context already ‘contains’ extralinguistic context. In general, the more minimal (or, less explicit) the linguistic input, the greater the contextual support needed (Hawkins & Smith 2001:€41). But no matter how explicit linguistic input is, its interpretation is never context-free. The received view, then, which treats context as a supportive element to be drawn on solely as and when needed is a fiction of post facto analysis. For the language user, who necessarily produces and understands language in specific contexts, language understanding is a priori situated and inescapably contextualized.9



On de-limiting context 

3. Psycholinguistic insights In support of the constitutive role of context in linguistic meaning advocated within theoretical frameworks such as Frame Semantics and informally argued for in the previous section, psycholinguistic studies have revealed the presence of contextual input during all stages of decoding, including phonological, syntactic and semantic analysis. In this section, I review only a small sample of the related literature to showcase both the methodologies used, and the types of contextual (perceptual as well as social stereotypical) information that have been shown to affect linguistic decoding. Johnson et al. (1999) studied the resolution of phonologically ambiguous phonetic stimuli. They found that information about the gender of the speaker affected the way listeners resolved the ambiguity, revealing different category boundaries for vowels for male and female speakers. These results persisted whether the information about speaker gender was presented visually (a male or female face pictured on screen uttering the ambiguous stimulus), or merely imagined (listeners imagining a male or female speaker uttering the ambiguous stimulus). Regarding their use of the term “gender,” Johnson et al. specify that they use this “to refer to sex-based differences between people that are subjective in that “gender” refers to a socio-cultural construct based in part on sex” (1999:€365 fn. 3). Moreover, they draw a distinction between “talker” and “speaker”: We are using “talker” to refer to the listener’s perceptual impression of a person’s identity based on hearing or watching the person speak. This contrasts with our use of “speaker” in a more generic sense which includes objective facts about a person, such as vocal tract length. … Thus, when we call our theory “talker normalization” as distinct from “speaker normalization” we mean to assert that speech perception is influenced by socio-culturally based “gender” expectations. (1999:€365 fn. 3)

Based on these results, Johnson et al. argue that “at least a part of the [vowel category] boundary shift associated with “talker normalization” is due to expectations regarding male and female voices that listeners bring to speech perception, quite apart from the sensory information presented in the stimuli” (1999:€379). In other words, it is the abstract social category of gender rather than assumptions about physical attributes, such as vocal tract length, that drive the resolution of the phonological ambiguity. Spoken word recognition was also found to be mediated by context, including both visual context and visually-induced information about social categories. With respect to visual context, Tanenhaus et al. (1995) found that when temporary ambiguities regarding word identity arose intra-lexically, as between “candle” and



Marina Terkourafi

“candy,” subjects waited till the end of the word to identify the intended object when both objects were available in the visual context. However, when no competing object was available, subjects identified the intended object before the end of the corresponding word, suggesting that word recognition proceeds incrementally beginning at word onset and is guided by visual context. Visually-induced information about social categories, such as author profession, has also been shown to enhance word recognition. Fitneva and Spivey (2005) showed subjects pictures of different professionals, and then presented them with a lexical decision task asking them to imagine that the person previously pictured was the speaker of the probe word. They found that words consistent with the professional identity of the person pictured (e.g. “jury” for judge) were recognized faster, suggesting that visually-prompted information about social categories facilitates lexical access. This experiment is of course one of many variants of the semantic priming paradigm originally developed by Meyer and Schvaneveldt (1971).10 Yet what is interesting about it is that the information facilitating lexical access is now thoroughly social stereotypical in nature. Other findings suggest that, like phonological and word processing, syntactic processing is also mediated by context. Tanenhaus et al. (1995) studied the resolution of temporary syntactic ambiguities, as in “put the apple on the towel in the box” using eye-tracking techniques. In this sentence, “on the towel” is temporarily ambiguous between a destination and a modifier interpretation. Analyses prioritizing syntax claim that “on the towel” is interpreted as destination by default, and then re-interpreted as modifier when the second PP “in the box” is heard. Contrary to these analyses, Tanenhaus et al. found that whether the first PP “on the towel” is interpreted as destination or modifier depends on the visual context. If there was an apple on a towel as well as a free towel that could serve as a destination, the ambiguity remained until the second PP “in the box” was heard. If, however, there were two apples, one on a towel and one not, the first PP “on the towel” was immediately interpreted as a modifier, serving to disambiguate the intended referent, i.e. no ambiguity arose as to its syntactic role. These results also show that, similarly to what was argued earlier with respect to language acquisition, online interpretation is goal-driven: “in natural contexts, people seek to establish reference with respect to their behavioural goals during the earliest moments of linguistic processing” (Tanenhaus et al. 1995:€1634). There is no point wasting time deriving syntactic structures that are not plausible given the visual context. In other words, there is no point doing syntax independently of context. Not only does visual context drive thematic role allocation to expressions as soon as they are uttered, it also serves to pre-empt a thematic structure from only partial linguistic information (e.g. hearing the words for agent and verb), and, based on this, to identify the referents of the associated expressions before these



On de-limiting context 

have actually been uttered. Altmann and Kamide (1999) showed subjects visual scenes representing potential semantic relationships between referents, e.g. a picture of a child and a cake, a toy train, a ball, and a toy truck. Subjects then heard sentences such as “the boy will eat the cake” or “the boy will move the cake”. The more constraining the selectional restrictions of the verb as to its object NP, the faster subjects’ eye-movements settled on an intended referent (i.e. “cake” after “eat”), even before the onset of the actual NP itself. This suggests that potentialities of interaction between the objects pictured were latent in the visual context, and activated by hearing the verb, already at verb onset. Similar results were obtained when the sentences were heard after the visual stimulus had been removed, suggesting that speakers build a mental representation of the stimulus and it is this representation that they use in subsequent processing (Altmann 2004). Finally, in a study investigating the effects of semantic and pragmatic anomaly (violations of selectional restrictions vs. violations of world knowledge) using electroencephalogram (EEG) and functional Magnetic Resonance Imaging (fMRI) data, Hagoort et al. (2004) found that both types of anomaly produce patterns of brain activation that are identical in onset (i.e. they occur within the same time frame) and latency, and very similar in amplitude and topographic distribution, both engaging in activation of the left inferior pre-frontal cortex. This suggests that semantic and world knowledge are drawn on in parallel during sentence processing, dealing a direct blow to the presumed sequentiality of semantic and pragmatic information. Differences in the processing of semantically and pragmatically anomalous sentences emerge only later, suggesting that, although semantic and world knowledge are drawn on in parallel, the mind does keep track of the source of the anomaly. One might venture that this is because of the different long-term consequences of each type of anomaly: semantic anomaly perhaps ought to lead eventually to a revision of word meaning, whereas pragmatic anomalies of the type studied merely result in sentence falsity. Taken together, the results of the psycholinguistic studies cited above give rise to the following conclusions: 1. Interpretation does not begin after the end of the utterance, but starts at speech onset and proceeds incrementally as the phonetic signal unfolds (Tanenhaus et al. 1995; Altmann & Kamide 1999). 2. Interpretation aims to establish plausible ways of interacting with the world, or, in other words, a pragmatically appropriate response; as such, it is driven by the speaker’s goals (Tanenhaus et al. 1995; Altmann & Kamide 1999) including the speaker’s projected goals when the speaker is different from the self (Fitneva & Spivey 2005). 3. Linguistic and world knowledge are integrated from the earliest moments of interpretation (Tanenhaus et al. 1995; Altmann & Kamide 1999).



Marina Terkourafi

4. Linguistic and world knowledge are integrated in parallel rather than sequentially (Tanenhaus et al. 1995; Altmann & Kamide 1999, Altmann 2004, Hagoort et al. 2004). 5. Contextual input to interpretation includes both perceptually-given context (Tanenhaus 1995; Altmann & Kamide 1999; Altmann 2004) and socio-cultural expectations (Johnson et al. 1999; Fitneva & Spivey 2005). 6. Contextual input affects all levels of linguistic analysis. 4. Extending the scope of the phenomena: Illocutionary force resolution The experimental studies cited above document input by both visual context and socio-cultural expectations to interpretation from its earliest moments and at all levels of linguistic analysis. The phenomena studied concern the resolution of lexical and syntactic ambiguities, reference assignment, and the processing of semantic versus pragmatic anomaly, all phenomena falling squarely within the tradition of truth-conditional semantic analysis (see Section 1). At the same time, these studies document the influence of speaker goals on the determination of meaning, supporting our earlier claim that knowing the meaning of an expression is not radically different from understanding its pragmatic consequences (see Section 2). If, as was argued earlier, knowing the meaning of an expression and understanding its pragmatic consequences are indeed linked by a generative relationship in which the former dynamically emerges from the latter, then the above results ought to naturally extend into illocutionary force assignment, the meaning component pertaining to understanding the pragmatic consequences of the speaker’s utterance par excellence. In other words, if context contributes to truth-conditional aspects of meaning in order to understand the pragmatic consequences of the speaker’s utterance, could it also contribute to these pragmatic consequences directly? Indeed, illocutionary force assignment itself may be viewed as a special case of contextually-driven pragmatic ambiguity resolution. This provides us with an extra reason to suspect that, like with other types of ambiguity, context enters into illocutionary force resolution from the first, i.e. as soon as contextual information is available. My proposal is that illocutionary force is projected probabilistically from a limited set of contextual parameters pertaining to who is talking to whom, when, and where. Subsequent linguistic and contextual information serves to check this probabilistic prediction, either confirming it (illocutionary assignment passing unnoticed), or revising it in the light of incoming data. This section briefly outlines the evidence for this proposal (cf. also Terkourafi 2002, 2005), while the following two sections explore why this is a sensible view of how illocutionary force assignment might proceed.



On de-limiting context 

4.1

Empirical motivation

The empirical findings motivating this proposal originate in spoken dialogical corpus data from Cypriot Greek, tape-recorded in various settings and later transcribed (approx. 60,000 words). Since conversations were recorded in the actual settings where they occurred, several physical or otherwise extra-linguistic attributes of the situation, such as the gender, age, social class of interlocutors, the relationship between them, and the setting of the exchange were noted and later fed into the analysis. During the analysis, utterances realizing offers or requests were selected and classified as such depending on the addressee’s uptake (Austin 1962). Then, the extra-linguistic attributes of the situation surrounding each utterance, which had been noted during the collection of the data, served to break down actual contexts of occurrence to schematic or otherwise ‘minimal’ contexts, consisting of the values of a limited set of contextual parameters (e.g., men, aged 31–50, of middle class, addressing women, aged 31–50, of middle class, for the first time, in a relationship of new customer to salesperson in a shop) but with all other specificities removed.11 An investigation of the linguistic means used to perform offers or requests in each such minimal context revealed that different minimal contexts correlated positively with different linguistic constructions. In other words, out of a set of semantically equivalent expressions conveying the same illocutionary force, a particular construction was preferred in a particular minimal context.12 The constructions that emerged from this analysis ranged from more to less abstract ones, the most specific cases resembling Pawley and Syder’s (1983) “lexicalized sentence stems.” In these, “the sentence structure is fully specified along with a nucleus of lexical and grammatical morphemes which normally include the verb and certain of its arguments; however, one or more structural elements is a class, represented by a category symbol such as tense, np, or pro” (1983:€210). A default intonation was also found to be part of the constructions analysed (cf. Cruttenden 1986/1997:€164). 4.2

An inheritance hierarchy of request constructions

Three types of constructions at different degrees of schematicity and context dependence were revealed. At the most schematic level, requests over a large range of minimal contexts were performed using a verb denoting a verbal or nonverbal act in the second person singular imperative.13 The form of this construction may be



Marina Terkourafi

specified as action.verb/speech.act.verb — imperative — 2ndsingular and examples include:14

(1) [At home; Speaker: female, 18–30, working-class; Addressee: male, 18–30, working-class; Relationship: family] vasili ∫epase t∫e ton artemi vassilis cover.imp.2sg and the artemis ‘Vassili, tuck in Artemis too.’

(2) [At work; Speaker: female, 31–50, middle-class; Addressee: female, 31–50, working-class; Relationship: friends] fkar’ ta friθca su. pluck.imp.2sg the eyebrows your.2sg ‘Pluck your eyebrows.’

(3) [At work; Speaker: male, 31–50, middle-class; Addressee: male, 31–50, middle-class; Relationship: old colleagues] anikse tin porta re rikho open.imp.2sg the door hey rikos ‘Hey Rick open the door’

(4) [At work; Speaker: female, 31–50, middle-class; Addressee: male, 31–50, middle-class; Relationship: salesperson to old customer] pien:e pano en’ o kostas. go.imp.2sg up is the kostas. ‘Go upstairs, Kostas is {there}.’

(5) [At work; Speaker: female, 31–50, working-class; Addressee: male, 18–30, working-class; Relationship: old customer to salesperson] andoni (.) ela pjas’ ta misa andonis (.) come.imp.2sg get.imp.2sg the half. ‘Andoni. Come and get half {of them}.’

(6) [At work; Speaker: male, over 51, middle-class; Addressee: female, over 51, middle-class; Relationship: employer to employee] kane tilefonima esi marula mu make.imp.2sg call you.2sg marula my ‘Make the call, Marula dear.’

(7) [At work; Speaker: female, 18–30, middle-class; Addressee: male, 31–50, middle-class; Relationship: employee to employer] var’ tes mes to folter su put.imp.2sg them in the folder your.2sg ‘Put them {sheets of paper}in your folder’



On de-limiting context 

As can be seen from these examples, the construction action.verb/speech.act. verb — imperative — 2ndsingular is used to perform requests across a wide range of minimal contexts encompassing interlocutors of both genders, all age groups, and both middle and working classes interacting in various roles, including family, friends, old colleagues, salesperson and customer, and employee and employer, at home and at work. This construction thus exhibits at once the least degree of context-dependence, and the highest degree of schematicity, since lexical items that can fill the verb slot are minimally constrained as to their semantics. Contrary to the construction action.verb/speech.act.verb — imperative — 2ndsingular which is characterized by a high degree of type frequency, three tokens of this construction provide examples of high token frequency (Bybee 2006). The expressions ase (leave — imperative — 2ndsingular = ‘let’), aku (listen — imperative — 2ndsingular = ‘listen’), and el’a su po (come — imperative — 2ndsingular to you i.tell =‘let me tell you’), although falling under the general action.verb/ speech.act.verb — imperative — 2ndsingular construction, appear to constitute lexically-specific constructions with particular discourse functions that override the request function of the more general construction. In the data collected, their very high individual frequency goes hand in hand with their being morphologically truncated (ase or ais < afise; aku < akuse; el’a su po < ela na su po) and occurring utterance-initially, both of which provide structural indications that these forms are in fact frozen in Cypriot Greek. Indeed, when these expressions are used, the act to be performed is typically expressed by a further verb, such as na rotisumen in (8) (cf. Klamer 2000:€76).

(8) [At work; Speaker: male, aged 31–50, middle-class; Addressee: male, aged 31–50, middle-class; Relationship: salesperson to old customer] endaksi. ase na rotisumen ton andrea OK let.imp.2sg to ask.1pl the andreas ‘O.K. Let me ask Andreas.’ The way they combine with other elements in discourse provides further evidence of their frozenness. Whereas el’a su po is literally a request, i.e. the first part of an adjacency pair, the speaker seldom waits for the hearer’s uptake before continuing (and one is hardly ever provided). Paralleling Klamer’s (2000:€76) hypothesis that the absence of an intonational break after a report verb marks its function as a complementizer, we can interpret a lack of speaker change at the end of this intonational unit as signalling the non-requestive function of el’a su po. In the case of aku and ase, it is their argument structure that has been affected: aku occurs without a complement, whereas ase occurs with a clausal complement, and is consequently not felt to have the full semantic content of ‘leaving (something)’, which it has when followed by an NP. Both processes (loss of arguments and acquisition of



Marina Terkourafi

new ones) have been associated with an expression’s becoming frozen/fixed as “the result of frequently occurring in appropriate surface contexts” (Klamer 2000:€96). Rather than requests, then, the three expressions ase, aku, and el’a su po appear to function as ‘prefaces’ (Nofsinger 1991:€134–5): they serve to attract or ensure the addressee’s continued attention, while contributing little, if anything, to the proposition expressed by the utterance. While the general action.verb/speech.act.verb — imperative — 2ndsingular construction constitutes the most frequent realization for requests overall, other lexically-specific constructions are preferred in specific minimal contexts. Exchanges in shops and formal settings, such as public discussions and interviews on the radio/TV, provide such minimal contexts. In the former, two constructions involving the verb exo, ‘to have,’ take the lead, while in the latter, another construction, this time involving the verb θelo, ‘to want,’ prevails. In requests exchanged between middle-class interlocutors in transactional settings exete NP? (have — indicative.rising.intonation — 2ndplural = ‘do. you-pl have NP?’) ranked first (Example 9).

(9) [At work; Speaker: male, 31–50, middle-class; Addressee: female, 18–30, working-class; Relationship: new customer to salesperson] eçete: phinats? have.ind?.2pl peanuts? ‘Do you have some peanuts?’ On the other hand, in requests exchanged between working-class interlocutors in the same settings exi NP? (have — indicative.rising.intonation — 3rdsingular = ‘is/are there NP?’) ranked first (Example 10). (10) [At work; Speaker: female, 31–50, working-class; Addressee: male, 31–50, working-class; Relationship: new customer to salesperson] eçi mikres pu na min exun sçeδia pano? aspro have.ind?,3sg small that subj neg have.ind.3pl patterns on? white ‘Are there any small plain {socks}? In white.’ In other words, exete NP? (have — indicative.rising.intonation — 2ndplural = ‘do.you-pl have NP?’) appears to be associated with the following minimal context:15 – Gender of speaker: {any} – Age of speaker: {any} – Social class of speaker: middle – Gender of addressee: {any} – Age of addressee: {any} – Social class of addressee: middle, working – Relationship: new customer to salesperson



On de-limiting context

– Setting: at work – Speech act occurs for the: 1st time A slightly different minimal context is associated with use of exi NP? (have — indicative.rising.intonation — 3rdsingular = ‘is/are.there NP?’). – Gender of speaker: {any} – Age of speaker: {any} – Social class of speaker: working – Gender of addressee: {any} – Age of addressee: {any} – Social class of addressee: working, middle – Relationship: new customer to salesperson – Setting: at work – Speech act occurs for the: 1st time Conversely, in addressing requests to interviewers in formal settings, women and interviewees preferred yet a third construction, θa iθela VP (fut want — past — 1stsingular = ‘I would.like VP’; Example 11). (11) [On TV: Speaker: female, 31–50, middle-class; Addressee: male, 31–50, middle-class; Relationship: interviewee to interviewer] θa iθela na prosθeso kati eδo omos. fut want.past.1sg subj add.1sg something here though. ‘I would like to add something here though.’ The relevant minimal context this time is: – Gender of speaker: female, male – Age of speaker: {any} – Social class of speaker: (middle) – Gender of addressee: {any} – Age of addressee: {any} – Social class of addressee: (middle) – Relationship: interviewee to interviewer, interviewer to interviewee – Setting: formal discussions – Speech act occurs for the: 1st time What is interesting when comparing the two exo/have-constructions with the θelo/want-construction is that semantically there is nothing stopping θa iθela VP (fut want — past — 1stsingular = ‘I would.like VP’) from performing a successful request in transactional settings too. It constitutes an indirect request by affirming the sincerity condition in much the same way as the two exo/have-constructions constitute indirect requests by questioning the preparatory condition





Marina Terkourafi

concerning the existence of goods sought (Brown & Levinson 1987:€ 137). And while affirming one’s desire for some thing should function equally well as a request as questioning its availability (compare English ‘Do you have NP?’ and ‘I’d like NP’), the preferences in the Cypriot Greek data are clear-cut: out of 61 requests by a new customer addressing a salesperson, only one (1.64%) was found using θa iθela (and then, complemented by an NP). On the other hand, of 28 occurrences of θa iθela in the data, 26 are followed by a VP and 24 of these are in formal settings. In other words, though they all fall under Brown and Levinson’s strategy of negative politeness, these constructions are not interchangeable. While the three token-frequency constructions analysed earlier are linked to the more general action.verb/speech.act.verb — imperative — 2ndsingular construction in virtue of their form, the last three constructions are linked to it in virtue of their function. In both cases, the additional degrees of lexical specification come with added pragmatic effects. In the case of the ase, aku and el’a su po constructions, which are highly frequent across contexts, these added pragmatic effects amount to overriding the requestive function of the general construction, which is now replaced by their function as prefaces. However, in the case of the exete NP?, exi NP?, and θa iθela VP constructions, these further pragmatic effects seem to follow directly from the specific values of the associated minimal contexts (italicised in the diagrammatic representations of these minimal contexts above) that override the default (underspecified) context specifications of the more general construction. These specific contextual values motivate assumptions about additional (non-illocutionary) goals that participants may plausibly be pursuing in these minimal contexts. The further pragmatic effects afforded by these lexicallyspecific constructions, then, take the form of particular identity displays, above and beyond the constructions’ illocutionary function as requests.16 The close association between minimal contexts and particular constructions introduced at the beginning of this section is most clearly seen in constructions of this last type and allows us to formulate the following hypothesis: as soon as a construction is uttered in the minimal context with which it is associated, a particular illocutionary force — that of request, in the cases reviewed above — is probabilistically assigned to it without further ado. This illocutionary force remains operative until it is contradicted by incoming information. In other words, a limited set of contextual parameters jointly provide a pointer to the intended illocutionary force, rendering reference to the fully specified, nonce context of occurrence superfluous — at least at an initial stage (see Section 5). Assuming this hypothesis is in the right direction — subject to confirmation by future studies of a similar design — two questions must be answered next. First, what is the advantage of relying — at least at an initial stage — on a limited set of contextual



On de-limiting context 

parameters to infer illocutionary force? Second, why should precisely these contextual parameters be appealed to and no others? 5. Processing minimal contexts In answer to the first question, the incrementality of interpretation provides some clues. What the Cypriot Greek findings suggest is that the incrementality of interpretation concerns not only the way in which linguistic input temporally unfolds through time and is received and processed sequentially, but also the way in which context is perceived and updated with new information. Information about the gender, age, and social class of the speaker and addressee, the relationship between them, the setting of the exchange, and the sequential order of realization of the speech act is available at speech onset. In other words, it forms part of the interlocutors’ common ground up to and including the moment when the fact that an utterance is being uttered is added to it, i.e. just before that utterance has been evaluated (cf. Stalnaker 1999:€ 102). However, this is not the case with the surrounding linguistic context (consider the first move in a dialogical game), paralinguistic and other idiosyncratic clues, which unfold in parallel with the utterance (cf. Fitneva & Spivey 2005:€318). As a result, these two sets of contextual information can take on different roles during the interpretation process. Contextual information available at speech onset can begin to play a part before the contextual information unfolding in parallel with the utterance even becomes available. In this way, it can serve to maximize the speed of interpretation by channelling the interpretation process toward conventionally expected outcomes, acting as a short-cut to the intended illocutionary force. Contextual information becoming available as the utterance itself unfolds checks the presumed illocutionary force and updates the context with the perlocutionary effects actually achieved by the speaker’s utterance (this is one way in which linguistic context, i.e. interpreted prior discourse, ‘contains’ extra-linguistic context, as was claimed earlier). The temporal priority of the first set of contextual parameters is due to the fact that they are typically fixed on the basis of background knowledge or otherwise perceptually (e.g. visually) available evidence. That is, they are largely settled automatically, i.e. by reasoning pre-emptively on whatever data are available, even to just initiate interaction. To see that this is so, consider the awkwardness of realizing that the person one is talking to on the phone is an elderly lady, not a young child as one had originally surmised judging by the high pitched voice at the other end of the line; or, of realizing that one’s addressee is a customer in a shop rather than a salesperson only after making a goods-related query. The reason why these



Marina Terkourafi

mundane situations typically prompt an apology is that settling these contextual features one way biases the interpretation process in a certain direction. Once one realizes one’s mistake, it is necessary not only to re-set the contextual parameters to their correct values, but also to re-interpret previous discourse, withdrawing all those inferences derived on mistaken premises. In other words, what such incidents suggest is that we ‘hear’ different speakers differently, depending on default assumptions that we have made about them. Acknowledging the automaticity of assumptions about these contextual parameters is not to deny their socio-cultural basis. How individuals are assigned to genders and age-groups, and even more so, to social classes and roles, how many social classes, potential relationships between interlocutors, or interactional settings are distinguished, are irreducibly social matters. As such, they rely on a process of acculturation, akin to developing the set of “nonrepresentational mental capacities that enable all representing to take place” which Searle refers to as “the background” (1994:€143). In this sense, a minimal context comprising the gender, age, social class of interlocutors, the relationship between them, and the setting of the exchange constitutes convenient shorthand for an accumulated body of sociocultural knowledge. The relationship between an utterance and its surrounding context has been likened to that of figure and ground (Goodwin & Duranti 1992:€9). However, if the above description of the relationship between a minimal context and the background is at all accurate, it seems more appropriate to conceive of this relationship as a three-way one, between an utterance, a minimal context, and a background, where continuous interaction occurs not just between utterance (figure) and minimal context (ground), but also between minimal context (ground) and background, the former simultaneously constrained and invested with meaning by the latter. The background, in other words, will always ‘colour’ minimal context, and similar minimal contexts will presuppose similar backgrounds. Goodwin (1996) offers a case in point when he describes how what appears to be “transparent vision,” i.e. obvious, unmediated understanding of scenes and events, can actually go through several stages of “seeing,” all equally “thoroughly social,” such that, in the end, “transparent vision is very much a crafted object” (1996:€382). Barsalou (1999) outlines the cognitive underpinnings of this process, whereby symbolic representations can emerge from perceptual input. As a result of this process, frequently repeated patterns of interaction give rise to symbolic representations available for propositional manipulation. Arising in this way, the contextual parameters discussed above are at once constructed and objective (Sbisà 2002): they are not subjective individual mental constructs; rather, they both arise with reference to perceptual data, and are constrained by social feedback.



On de-limiting context 

In answer, then, to the question of what is the advantage of relying (at least at an initial stage) on a limited set of contextual parameters to infer illocutionary force, the incrementality of interpretation suggests that interpretation begins as soon as a speaker starts speaking, and proceeds in parallel with the speaker’s utterance, drawing on contextual information as and when this becomes available. Contextual parameters that can be fixed early on in an exchange prime the conventionally expected illocutionary force, thereby speeding up interpretation. Through acculturation, these contextual parameters gain in salience: interlocutors seek to settle these — if only by presumption — in order just to initiate interaction. It is telling in this respect that ambiguities concerning these contextual parameters are rather badly tolerated — and, in some cultures at least, may actively deter one from engaging in interaction until they are resolved. 6. Motivating minimal contexts Let us come now to the second question posed at the end of Section 4: why should precisely these contextual parameters be appealed to, and no others? The answer, in a nutshell, is that the contextual parameters jointly constituting a minimal context are motivated rather than arbitrary. Their jointly constituting a minimal context can be predicted by the organization of human action. Some useful insights into this organization are provided by Gibson’s (1977) notion of ‘affordances’ of the physical environment, according to which the structure of an object embodies potentialities of interaction with this object (cf. Mace 1977). Drawing a parallel between objects and physical settings, physical settings may also be said to embody potentialities of interaction with them. The physical layout of a shop is different to that of a bank, or to that of one’s living room. The process by which physical settings afford us with specific potentialities of interaction is best thought of as a discursive one, resulting from the mutual interaction of an agent with his/her environment (cf. Mace 1977:€46). In this sense, once again we are concerned with the kind of transparent vision mentioned earlier. However, what is of interest now is that through such socio-culturally mediated vision, mere perception of a certain physical layout can set in motion a whole chain of inferences regarding the prototypical social roles of participants one may expect to find therein, and prototypical individuals fulfilling these roles (for instance, at a company’s reception desk (location), one might expect to find a secretary (role) and she will stereotypically be female (gender)).17 Because human action exploits the affordances of physical settings, participants’ presence in a certain setting can act as a pointer to their behavioural goals (Lindsay & Gorayska 1994:€37). This inference is clearly probabilistic and will vary



Marina Terkourafi

for different goals. Some goals can be pursued in several different settings (e.g. socializing with friends), while others cannot (e.g. trying a case in court). As a result, different settings will constrain to different extents the sets of goals one may plausibly pursue therein. This prediction is confirmed by the Cypriot Greek corpus data. Preferences for lexicalized, as opposed to merely schematic, constructions were revealed in those settings that most rigidly constrain the goals one may plausibly pursue therein: when one enters a shop as a customer for the first time, and during interviews on radio/TV (see Section 4). The notion of the affordances of physical settings may thus provide the missing link between the limited set of contextual parameters we have been considering (interlocutors’ gender, age, and social class, the relationship between them, and the setting of the exchange) and the kinds of goals that the speaker is presumed to be pursuing at any one time. By motivating the relationship between the contextual parameters available since the early stages of the interpretation process, affordances also limit their number. A potentially all-inclusive notion of context is practically unmanageable and of little empirical usefulness. What is needed, in this respect, is a notion of context that is bounded, i.e. that distinguishes context from the totality of the interlocutors’ common ground. The emergence of symbolically represented contextual parameters out of perceptual input along the lines proposed by Barsalou (1999) shapes a potentially unlimited body of socio-cultural knowledge into a limited set of contextual parameters available for propositional manipulation. Because of this, the resulting minimal contexts are neither impoverished with respect to the original input, nor reducible to it. At the same time, the contextual parameters jointly constituting a minimal context cannot be further broken down to more abstract ones, such as relative status or power, or social distance. Rather, they are mentally manipulated as primes. Interviews with informants suggest that, unlike the more abstract categories of power and social distance, the contextual parameters proposed above are intuitively graspable and correspond to participants’ own internalized categories. The primacy of the contextual parameters proposed above also receives support from the literature on language acquisition, which shows that, by age four, children have begun to pay attention to the categories of gender, age, and profession (the latter being one determinant of social class), talking differently to mothers and fathers, adults or peers and younger addressees, nurses and doctors (reported in E. Clark 2004:€573–576). That the proposed analysis uses the same categories internalized by speakers satisfies not only Occam’s injunction not to multiply entities beyond necessity, but also the ethnomethodological requirement to take “as a starting point of departure for the analysis of context the perspective of the participant(s) whose behaviour is being analyzed” (Goodwin & Duranti 1992:€4).



On de-limiting context 

7. Conclusion: Some consequences for grammar construction My aim in this paper has been to advance a theoretically motivated and empirically backed argument for the context-dependent nature of linguistic interpretation at all levels of analysis. At a theoretical level, the argument is based on the primacy of understanding the pragmatic consequences of the speaker’s utterance, from which knowledge of meaning arises by a process of abstraction over time. Examples of pragmatic understanding achieved with very little — or even in spite of — linguistic input available go to show just how far interpretation can proceed drawing on contextual information alone. In other words, context emerges as a powerful resource that speakers interpreting language have at their disposal. Psycholinguistic studies show that speakers systematically exploit this resource not just to interpret linguistic input as this becomes perceptually available, but also to anticipate future linguistic input before this has even materialized physically. Based on these experimental results and on original corpus data from Cypriot Greek, I further proposed that illocutionary force assignment as a central part of understanding the pragmatic consequences of the speaker’s utterance begins from the earliest moments of utterance interpretation and proceeds incrementally alongside other aspects of linguistic analysis. Moreover, I suggested that speakers exploit the sequential order in which contextual information becomes available. Two types of contextual information may be distinguished in this respect. Information pertaining to socio-cultural categories such as the gender, age, and social class of interlocutors, the relationship between them and the setting of the exchange is available at speech onset, and serves to set up expectations about the intended illocutionary force. Further paralinguistic and intonational cues become available as the speech signal unfolds, and serve to confirm or revise the projected illocutionary force. It is possible to project illocutionary force from this limited set of contextual parameters, because human experience is structured by exploiting the affordances of physical settings, making it possible to reason pre-emptively from one’s presence in a physical setting to one’s goals. This proposal remains to be experimentally assessed. One way of testing it is measuring the speed of interpretation of linguistic stimuli containing, e.g., referring expressions, where reference assignment would vary according to illocutionary force assignment. In this way, it ought to be possible to establish with some certainty how soon after speech onset illocutionary force is resolved, and, potentially, by appropriately manipulating contextual variables, with respect to which pieces of contextual information it is resolved. While experimental confirmation of the present proposal must await future research, there is enough in the available experimental and corpus results to suggest that continuing to abide by grammatical frameworks that adopt an essentially



Marina Terkourafi

sequential view of linguistic analysis is no longer viable. Especially if our criteria for grammar evaluation include explanatory adequacy — paying particular attention to how linguistic knowledge is acquired by children rather than how it is actually processed by adults (cf. Chomsky 1995:€32) — it is clear that contextual input, and, in particular, socially defined categories such as gender, age, and profession which are interpreted as linguistically relevant by children at an early age, cannot be excluded from our representations of grammar. The closest parallel to the integration of phonological, syntactic, semantic and pragmatic information revealed by the experimental studies is offered by signbased formalisms such as Construction Grammar, which allow the concurrent specification of phonological, syntactic, semantic and pragmatic constraints on the processing of linguistic constructions at various levels of abstraction. One advantage of Construction Grammar vis-á-vis related lexicalist formalisms is that it allows for the specification of inheritance hierarchies of constructions, which appear to correlate with increasing levels of context specificity in the Cypriot Greek corpus data (see Section 4.2). Working out how to incorporate pragmatic, and in particular, extra-linguistic, considerations into our feature structures remains a task for the future. At this stage, my aim has been simply to suggest the urgent need for such an endeavour, by substantiating this both theoretically and empirically. Notes * I would like to thank Mirjam Fried, Roger Lindsay, and an anonymous reviewer for valuable comments on a previous draft. 1. The strong sequentialist view outlined in this section is admittedly a straw-man put up for the sake of argument for at least two reasons. First, it is presented in its simplest, bare-bones form: more sophisticated versions of this view are also available. However, when considering those more sophisticated versions, it is worth bearing in mind that the alternative view developed in this paper abandons not only the traditional sequentiality of levels of analysis but also their informational encapsulation. As such, a more radical departure than merely admitting the possibility of parallel distributed processing is required to accommodate this alternative view. Second, the strong sequentialist view is contra-indicated by a mass of evidence from experimental psychology — only a fraction of which is reviewed in Section 3 below — and as such probably no-one today would defend it as a psychologically plausible view of language understanding. Nevertheless, as an idealised abstraction enabling theory development it underlies several distinctions, such as, e.g., between the lexicon and the grammar, or between the lexicon and the encyclopaedia, which, in turn, often find their way into analyses of a more formalist persuasion. As a result, although sequentiality as such may no longer be defensible or seriously defended, it seems to be embedded deeply enough in our linguistic theorising to warrant presenting it as the thesis against which to set up the antithesis argued for in this paper.



On de-limiting context  2. It is an oft-neglected aspect of this truth-conditional stance that it brings into focus the relationship of language to the world (recall the truth-conditional motto “to know the meaning of a sentence is to know what the world would have to be like for the sentence to be true”). What this essentially suggests is that there is no use (or: meaning) to language other than the information that it conveys about the world (or: some world). In this sense, the ‘purpose’ of language is to improve the alignment between what goes on inside our heads and what goes on outside of them, so to speak; to give us a grasp onto the outside world; and, ultimately, to enhance our performance as we act in this world. Seen from this perspective, the truth-conditional tradition is not quite as antithetical to the ensuing language-as-action tradition started by ordinary language philosophers as we may generally tend to believe. 3. Actually, these two solutions are not necessarily alternatives: one constitutes a point about the informational role of semantic vs. pragmatic contributions, while the other relates to how the computational processes employing this information are organized (I thank Roger Lindsay for this observation). This is understandable given the different emphases of the respective fields advocating them: on the informational import of linguistic expressions within semantics, and, increasingly, on psychological plausibility within pragmatics. 4. This view is advocated within dynamic semantic theories such as Discourse Representation Theory (Kamp & Reyle 1993) and Game-Theoretic Semantics (Hand 1999). 5. The latter solution is advocated within post-Gricean pragmatics including both Relevance Theory and Neo-Gricean approaches. 6. Pre-empting these experimental results, Fillmore claimed that the labeling of whole scenes or experiences precedes the labeling of individual, isolable parts of these, to finally produce “a repertory of labels for schematic or abstract scenes and a repertory of labels for entities perceived independently of the scenes in which they were first encountered” (1977:€62). To support this last claim, he also presented preliminary psychological evidence of the greater cognitive complexity of handling decontextualized behaviours and objects of perception (1976:€24). 7. The grounding of language acquisition in pragmatics, out of which, through successive abstractions, semantics is built, is of course one of psychology’s earliest tenets (e.g., Vygotsky 1986[1934]) to have been confirmed by modern psycholinguistics (e.g., Maratsos 1977; Schank & Abelson 1977:€222). 8. Such a verificationist view of meaning has previously been advocated by philosophers such as Dummett or Brandom. I thank Michiel Leezenberg for this observation. 9. Those of a more cognitivist persuasion will be well acquainted with similar claims, most notably through the work of Charles Fillmore on Frame Semantics (1976, 1977, 1982, 1984, 1986). Especially pertinent is his distinction between T-semantics (the semantics of truth) and U-semantics (the semantics of understanding; Fillmore 1984, 1986), which parallels the distinction between the received view and the alternative view advocated in this paper. I thank Mirjam Fried for bringing these works to my attention. 10. I thank Roger Lindsay for contributing this information. 11. For fuller description of the sampling and corpus collection and analysis procedures see Terkourafi 2001, Chapter 2. 12. A theory such as Brown and Levinson’s (1978/1987) theory of politeness, which correlates pragmatic strategies — rather than constructions — with underlying social variables such as Power, Distance and Ranking of the imposition — rather than with directly observable attributes



Marina Terkourafi of interlocutors — fails to make this more finely grained prediction. The problem is that in their theory, each pragmatic strategy can be realized by several linguistic expressions, whereas the direct association between linguistic constructions and minimal contexts revealed by the analysis of the Cypriot Greek data turned out to be conventional, i.e. not mediated by a general rationality principle such as Brown and Levinson’s Wx=D(S,H)+P(H,S)+Rx (1978/1987:€76). 13. There are several advantages to claiming that the relationship between imperatives and the illocutionary force of request is conventionalized, i.e. defeasible, rather than a semantic, encoded one, including the possibility to account more economically for imperatives performing offers, expletives, and instructions. 14. Cypriot Greek examples are rendered semi-phonologically in the IPA (post-lexical processes are not represented), followed by a word-per-word translation, and an English gloss (‘ ’). Constructions discussed in the text are underlined in all three versions of the examples. Transcription conventions and abbreviations used: 2sg=2nd person singular; IMP=imperative; FUT=future particle; NEG=negative particle; SUBJ=subjunctive particle; CG=Cypriot Greek dialectal wordform; (.) =brief pause; ? =rising intonation; , =flat intonation; . =falling intonation; : =elongated sound (also used for Cypriot Greek geminates); { }=words added to improve translation. 15. Contextual parameters appearing as {any} had no significant impact on the use of the expression in the data. Values divided by commas appear in order of frequency. Values are indicated in parenthesis when there was not enough evidence in the data to ascertain their impact on use of the expression. 16. On the function of these constructions to assert a ‘middle-class customer’ identity desirable in transactional settings, and a ‘familiar with an interviewing register’ identity desirable in formal discussions, see Terkourafi (2002:€186–187 and 191–194 respectively). 17. The notion of affordances I am operating with in this section is clearly an extension of the original Gibsonian term. Gibson specifically intended affordances as an alternative to cognitive processes, arguing that “everything necessary to explain behaviour could be recovered from the ‘flux of information at the receptor surfaces’” instead (Roger Lindsay, personal communication). A more appropriate term for the aspects of the physical environment I am interested in is perhaps “plan enablement,” a term referring to “the subset of plans enabled by the fact that their activation conditions are met by a particular physical setting” (Roger Lindsay, personal communication). However, using the term ‘affordances’ brings with it some advantages that a shift in terminology may not afford. These include recognizability of the general idea behind the term — if not of its full complexity — across disciplinary boundaries, and most importantly the hand/ glove relationship of behaviour to the environment that prompted Gibson to coin the term affordances in the first place, which provides an apt metaphor for the discursive relationship between an agent and his/her environment by which the affordances of physical settings emerge. At the risk, then, of stretching this term beyond the dictates of scientific perspicacity, I continue to refer to these visually grounded, socially-constructed (and hence distinctly cognitive) aspects of physical settings as ‘affordances.’

References Altmann, Gerry (2004). Language mediated eye-movements in the absence of a visual world: The ‘blank screen paradigm’. Cognition, 93, B79–B87.



On de-limiting context  Altmann, Gerry & Yuki Kamide (1999). Incremental interpretation at verbs: Restricting the domain of subsequent reference. Cognition, 73, 247–264. Austin, John (1962). How to do things with words. Oxford: Clarendon Press. Bach, Kent (1994). Semantic slack: What is said and more. In S. Tsohatzidis (Ed.), Foundations of speech act theory: Philosophical and linguistic perspectives (267–291). London: Routledge. Barsalou, Lawrence (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577–660. Brown, Penelope & Stephen C. Levinson (1978/1987). Politeness: Some universals in language usage. Cambridge: Cambridge University Press. Bybee, Joan (2006). From usage to grammar: The mind’s response to repetition. Language, 82, 711–733. Chomsky, Noam (1995). The minimalist program. Cambridge, MA.: MIT Press. Clark, Eve (2004). Pragmatics and language acquisition. In L. Horn and G. Ward (Eds.), The handbook of pragmatics (562–577). Oxford: Blackwell. Cruttenden, Andrew (1986/1997). Intonation. Cambridge: Cambridge University Press. Fitneva, Stanka & Michael Spivey (2005). Context and language processing: The effect of authorship. In J. Trueswell & M. Tanenhaus (Eds.), Approaches to studying world-situated language use: Bridging the language-as-product and language-as-action traditions (317–328). Cambridge, MA: MIT Press. Fillmore, Charles (1976/1981). Pragmatics and the description of discourse. In P. Cole (Ed.), Radical pragmatics (143–166). New York: Academic Press. Fillmore, Charles (1982). Frame Semantics. In The Linguistic Society of Korea (Eds.), Linguistics in the Morning Calm (111–137). Seoul: Hanshin Publishing Company. Fillmore, Charles (1984). Frames and the semantics of understanding. Quaderni di Semantica, 6, 222–254. Fillmore, Charles (1986). U-semantics, second round. Quaderni di Semantica, 8, 49–58. Gibson, James (1977). The theory of affordances. In R. Shaw & J. Bransford (Eds.) Perceiving, acting and knowing (67–82). Hillsdale, NJ: Erlbaum. Goodwin, Charles & Alessandro Duranti (1992). Rethinking context: Introduction. In A. Duranti & C. Goodwin (Eds.), Rethinking context: Language as an interactive phenomenon (1–42). Cambridge: Cambridge University Press. Goodwin, Charles (1996). Transparent vision. In E. Ochs, E. Schegloff & S. Thompson (eds.), Interaction and grammar (370–404). Cambridge: Cambridge University Press. Hagoort, Peter, Lea Hald, Marcel Bastiaansen, & Karl Magnus Petersson (2004). Integration of word meaning and world knowledge in language comprehension. Science, 304, 438–441. Hand, Michael (1999). Semantics vs. pragmatics: ANY in game-theoretical semantics. In K. Turner (Ed.), The semantics/pragmatics interface from different points of view (179–198). Oxford: Elsevier Science. Hawkins, Sarah & Rachel Smith (2001). Polysp: A polysystemic, phonetically-rich approach to speech understanding. Rivista di Linguistica, 13, 99–188. Johnson, Keith, Elisabeth Strand & Mariapaola D’Imperio (1999). Auditory-visual integration of talker gender in vowel perception. Journal of Phonetics, 17, 359–384. Kamp, Hans & Uwe Reyle (1993). From discourse to logic. Dordrecht: Kluwer. Klamer, Marian (2000). How report verbs become quote markers and complementizers. Lingua, 110, 69–98. Lindsay, Roger & Barbara Gorayska (1994). Towards a general theory of cognition. MS. Oxford Brookes University & City of Hong Kong Polytechnic.



Marina Terkourafi Mace, William (1977). James J. Gibson’s strategy for perceiving: Ask not what’s inside your head, but what your head’s inside of. In R. Shaw & J. Bransford (Eds.) Perceiving, acting and knowing (43–65). Hillsdale, NJ.: Erlbaum. Maratsos, Michael (1977). Disorganization in thought and word. In R. Shaw & J. Bransford (Eds.), Perceiving, acting and knowing (347–363). Hillsdale, NJ.: Erlbaum. Meyer, David & Roger Schvaneveldt (1971). Facilitation in recognizing pairs of words: Evidence of a dependence between retrieval operations. Journal of Experimental Psychology, 90, 227–234. Nofsinger, Robert (1991). Everyday conversation. London: Sage. Pawley, Andrew & Frances Syder (1983). Two puzzles for linguistic theory: nativelike selection and nativelike fluency. In J.C. Richards & R.W. Schmidt (Eds.), Language and communication (191–226). London: Longman. Sbisà, Marina (2002). Speech acts in context. Language and Communication, 22, 421–436. Schank, Roger & Robert Abelson (1977). Scripts, plans, goals and understanding: An inquiry into human knowledge structures. New Jersey: Lawrence Erlbaum. Searle, John (1994). Intentionality: An essay in the philosophy of mind. Cambridge: Cambridge University Press. Stalnaker, Robert (1999). On the representation of context. In R. Stalnaker (Ed.), Context and content: Essays on intentionality in speech and thought (96–113). Oxford: Oxford University Press. Tanenhaus, Michael, Michael Spivey-Knowlton, Kathleen Eberhard, & Julie Sedivy (1995). Integration of visual and linguistic information in spoken language comprehension. Science, 268, 1632–1634. Terkourafi, Marina (2001). Politeness in Cypriot Greek: A frame-based approach. PhD diss. University of Cambridge, Department of Linguistics. Terkourafi, Marina (2002). Politeness and formulaicity: Evidence from Cypriot Greek. Journal of Greek Linguistics, 3, 179–201. Terkourafi, Marina (2005). Pragmatic correlates of frequency of use: The case for a notion of ‘minimal context’. In K. Nikiforidou, S. Marmaridou, & E. Antonopoulou (Eds.), Reviewing linguistic thought: Converging trends for the 21st century (209–233). Berlin: de Gruyter. Tomasello, Michael. (1995). Pragmatic contexts for early verb learning. In M. Tomasello & W. E. Merriman (Eds.), Beyond names for things: Young children’s acquisition of verbs (115–146). Lawrence Erlbaum Associates. Tomasello, Michael (1999). The cultural origins of human cognition. Cambridge, MA: Harvard University Press. Vygotsky, Leo (1934/1986). Thought and language. Cambridge, Mass.: MIT Press.

Constructions and context When a construction constructs the context* Henri-José Deulofeu and Jeanne-Marie Debaisieux

The role of context in construction grammar has been up to now viewed from a restricted point of view. That is, context is used to help in analysing the meaning and the pragmatic aspects of constructions, in keeping with the following assumption: 1. Context (common ground) forms part of the construction’s semantics It has been observed that, in many cases, constructions have a non-compositional meaning. A sub-class of semantically non-compositional constructions is characterized by the fact that its interpretation involves the incorporation of pieces of information derived by taking into account, broadly speaking, the context of the utterance effected through the construction. The introduction of these contextual pieces of information is achieved in most cases by means of lexemes carrying an instructional meaning: “look at this aspect of context to process the meaning of the construction.” This situation is best described by Kay: A number of grammatical constructions have been described in which Part or all of the meaning of the construction is […] a virtual instruction to the addressee to examine the common ground of the conversation (along with the other interpretive content of the sentence) to fill in some partially specified part of the intended interpretation. An example involving the construction employing the expression let alone is given in (a). (a) Fred won’t order shrimp, let alone Louise, squid. The addressee of an utterance of (a) can only interpret it successfully if he can find in, or construct from, the conversational common ground a set of assumptions according to which Louise’s willingness to order squid unilaterally entails Fred’s willingness to order shrimp. (Kay 1997:€3)

In addition, (a) shows how such complex interpretational phenomena can be conventionally associated with a particular morphosyntax, that is, they are encoded in the construction via the expression let alone, serving as an instruction marker.



Henri-José Deulofeu and Jeanne-Marie Debaisieux

The aim of our paper is to address another case of a syntactic pattern involving context, which can be summed up by assumption 2: 2. A construction provides pieces of context that the speaker considers to be necessary for the addressee to get a relevant interpretation of the current utterance We claim that alongside constructions, like the let alone one, which need context to be interpreted, there exist constructions which bring in contextual background information to maximize the relevance of the utterance they are added to. This assertion seems at the beginning of the sentence quite paradoxical.€Any piece of an utterance is usually regarded as text, not context. We would like to show that some pieces of utterances are in fact not part of the ongoing text, but have some kind of background status which makes them part of the context. At first glance, these pieces of utterances have, in the domain of speech, the same status as footnotes have in written uses of language. They are, strictly speaking, marginal to the text and nearer to context status than to plain text status. In any case, these subconstructs inserted in utterances have a special status that must be accounted for both in linguistic theory and in construction grammar. Their special status can be revealed in the fact that they show idiosyncratic formal as well as pragmatic properties. In order to show this, we will focus on a subgroup of these constructions, namely those which are introduced by grammatical devices (so-called subordinators) and which appear as superficially synonymous with prototypical subordinate clauses. Consider this example of a parce que clause in spoken French translated with bold ‘cos (Debaisieux 2002): (1) le roi parce que le roi est très aimé en Thaïlande je crois que j’en ai parlé avant le roi est arrivé[…]... à leur faire comprendre que cette culture était très en dent de scie (LOU. Cas.) ‘The king ‘cos the king is very much loved in Thailand – I think I have said that before – the king succeeded in making them understand that this culture was very uneven.’

The French version is perfectly acceptable. The parce que clause reintroduces a piece of relevant common ground (as underlined by the speaker’s metacommunicative comment I think I have said that before) by interrupting the processing of the main line of discourse. We will show that this use of parce que is specific. It has neither the properties of subordinating parce que, nor even exactly those of the



Constructions and context 

parce que / cos’ introducing “supplementive” clauses according to the terminology of Huddleston & Pullum (2002) and evidenced in the well known examples:

(2) Jean est certainement à la Fac parce que j’ai vu qu’il avait laissé sa voiture dans le parking ‘Sure John is at the University ‘cos I noticed that he left his car in the parking lot ’ Parce que in (1) shares the non-subordinate status with parce que in (2), but it differs in many respects, as we will show later on. For instance it allows strict lexical repetition (see bolded words in 1), which are very awkward in (3): (3) Jean est certainement à la Fac parce que j’ai vu qu’ il (?? Jean) avait laissé sa voiture dans le parking ‘Sure John is at the University ‘cos I noticed that he (?? John) left his car in the parking lot’

Our claim is that the parce que construct in (1) belongs to a special type of syntactic construction which we will call the context focussing parce que construction (cfpc). The main goal of our paper is to provide a description of the use of parce que, introducing cfpc based on syntactic contexts as well as pragmatic functions. Besides this descriptive goal, we would like to address some theoretical issues. What is the exact status of this construction? In which kind of inheritance network is it involved? The paper is organized in the following way: in Section 1, we investigate the properties of cfpc, concluding that it is a specific syntactic pattern, in which the parce que clause is not licensed by any grammatical category or construction1. In Section 2, we distinguish two pragmatic subtypes of this pattern. In Section 3, we show that this pattern is specific to spoken language in that it cannot be syntactically reduced to the insertion of a footnote, as found in written styles. We conclude in Section 4 by addressing theoretical issues. This study is based on examples extracted from an authentic corpus of spoken French. The corpus consists of a 250,000 word transcription of recordings of spontaneous language in various natural environments (interviews, dialogs, narratives, etc.). More than 200 examples of cfpc constructs have been extracted.



Henri-José Deulofeu and Jeanne-Marie Debaisieux

1. cfpc as a specific syntactic pattern 1.1

cfpc is an instance of non subordinate conjunction + clause

Let us consider as a prototypical example of CFPC the parce que clause in example (1) repeated below as (4):

(4) le roi parce que le roi est très aimé en Thaïlande je crois que j’en ai parlé avant le roi est arrivé[…]... à leur faire comprendre que cette culture était très en dent de scie (Thailande)

As mentioned above, in this use, the parce que clause conveys backgrounded information, out of the main line of the discourse, by means of which the speaker builds a relevant “common ground” for the addressee. We can first show this CFPC as an instance of non subordinate parce que clause. The non subordinative use of clauses introduced by conjunctions has been observed in situations where the conjoined clause follows the main clause: (5) c’est une prof de SVT parce que l’an dernier j’ai fait un stage avec elle ‘She is a teacher of Natural science (I can assert that) ‘cos last year I had a training session with her.’ In this pattern, the parce que clause does not express a cause of the fact asserted in the main clause, but an argument in favor of the assertion made by the main clause. One can verify that this construct has no properties characteristic of the subordinate use, and thus cannot be analyzed as a syntactic adjunct in the main clause (Debaisieux 2002). a. This clause cannot be clefted:

(6) * c’est parce que l’an dernier j’ai fait un stage avec elle que c’est une prof de svt

b. It cannot be an answer to a why question:

(7) Pourquoi c’est une prof de svt? * parce que l’an dernier j’ai fait un stage avec elle

c. It cannot be modified by a scope adverbial:

(8) * c’est une prof de svt surtout parce que l’an dernier j’ai fait un stage avec elle

d. It cannot be preposed to the main clause:

(9) * parce que l’an dernier j’ai fait un stage avec elle c’est une prof de svt

e. It cannot be conjoined with the main clause with anaphoric cela ‘that’: (10) * c’est un prof de svt et cela parce que j’ai fait un stage avec elle



Constructions and context 

It is easy to check that the inserted parce que clause in (4) has the same behavior and thus cannot be analyzed as an adjunct subordinate clause. But other properties show that the pattern is a special case of insubordinate parce que clause. 1.2

Specific properties of CFPC

1.2.1 Formal properties a. Free insertion within the “main clause” The non subordinate parce que clause generally follows the main clause. It is unusual to find it inserted in the main clause: (11) ?? c’est une prof parce que l’an dernier j’ai fait un stage avec elle une prof de svt On the contrary, the backgrounded CFPC can be freely inserted in the main clause (12), where the insertion occurs after a fragmented word: (12) il a dit maintenant tu vas mettre la mitr- parce que tu sais c’est pas tout le monde qui sait mettre une mitrailleuse en marche euh tu vas mettre la mitrailleuse là-bas (BIR. 7, 5) ‘He says now you are to set the mach- (ine gun) ‘cos you know not everybody knows how to set up a machine gun hum you are to set the machine gun over there.’ We know that “random” insertions of adverbial adjuncts functioning as epistemic parentheticals or metalinguistic comments are also possible, but interestingly enough, these insertions are not entirely free but grammatically constrained (Espinal 1991; Marandin, 1999): they occur roughly at the boundary of any major constituent. No such constraints govern our CFPC insertions. They occur as soon as the speaker considers that they are pragmatically required for the processing of the message. b. Prosodic pattern Martin (1978) and Morel & Danon-Boileau (1998) describe in detail the specific prosodic contour which is observed in parenthetical constructs inserted in a “main” construction, whether they are bare or introduced by a grammatical device. We should point out first that the parce que clause in (5) shows the regular assertive contour of main clauses with Low F0 on the last accented syllable. On the contrary, CFPC has a specific F0 high, spread over the structure (level 2 on a scale of 4), and shows a flattened shape and fastened speech flow at the onset. All these features iconically contribute a backgrounding effect. The CFPC in our examples follows this general pattern. In cases where CFPC is postponed to and not inserted



Henri-José Deulofeu and Jeanne-Marie Debaisieux

in the main clause (see Example 16), it shows the same prosodic pattern, and the effect will be one of afterthought. c. Specific syntactic form The iconic backgrounding effect of the prosodic pattern is reinforced in many cases by other segmental iconic devices. We observe – in bold in examples (13) and (14) – lexical reiterations with extra syntactic position in the main clause. Within CFPC these elements are not perceived as disfluencies2: they help to establish a kind of natural bracketing of the parenthetical insertion. If we drop the inserted CFPC, as in (13′) or (14′), the repeated bold lexical items would appear as mere performance phenomena without any structural function. As a consequence, the syntactic form of the whole pattern is not exactly the sum of two well formed sentences, one being inserted in the other. (13) au moment où il arrive donc / ce qui est marrant bon le directeur / parce que c’est le directeur qui fait passer tout ça c’est lui qui supervise tout / le (ALI. Cas.) directeur il lui dit / vous êtes prêt ‘At the time he arrives thus what is funny (is that) well the director – ‘cos it is the director who checks all this it is him who supervises everything – the director says to him: are you ready? (13′) ? bon le directeur le directeur il lui dit / vous êtes prêt (14) bien entendu dans son système il fallait que les masses jouissent d’avantages qu’elles n’avaient pas eus auparavant/ donc le problème du logement / parce que l’Empereur Napoléon III est un empereur de la vie quotidienne / un homme qui s’intéresse de très près aux réalités de la vie de tous les jours/ (GUE. Cas.) ce problème donc il fallait qu’il le prenne à bras le corps ‘ Needless to say in his system the population had to benefit of privileges which they didn’t have before so the issue of accommodation –‘cos the Emperor Napoleon the IIId is an Emperor of everyday life a man who is deeply concerned with true everyday life (problems) – this problem thus he wanted to fix it (14′) ? bien entendu dans son système il fallait que les masses jouissent d’avantages qu’elles n’avaient pas eus auparavant/ donc le problème du logement ce problème donc il fallait qu’il le prenne à bras le corps d. Lack of isotopy constraints The utterances forming the main line of a discourse are integrated in larger discourse units like narrative or argumentative patterns. Such discursive pragmatic organizations can be considered as particular instances of a more general pragmatic pattern: communicative action + possible continuation. The continuation



Constructions and context 

link has some formal counterparts in what can be called default discourse agreement rules. Two well known instances of these rules are tense concord and default pronominalisation in the case of coreference. The CFPC does not conform to these tendencies: cases of free tense sequence (present tense sait inserted in a main line pattern with past tense in [15]) or absence of default pronominalisation (repetition of le directeur in (13) or of problème in [14]) can be observed: (15) donc en Allemagne j’ai rencontré des gens qui travaillaient pour les paléontologues/ qui étaient des amateurs mais qui savaient très bien travailler/ parce que le paléontologue il sait très bien étudier une pièce mais il ne sait pas l’extraire et la restaurer et la remettre sur gange si elle a été abîmée tout (PAUL. Cas.) ça / c’est comme ça que j’ai connu le gisement de X ‘Thus in Germany I met some people who were working for paleontologists who where amateurs but who knew very well how to do the things because the paleontologist he knows very well how to study a sample but he doesn’t know how to extract it and to restore it if it has been damaged all that it is like this that I got acquainted with the excavations of X.’ This situation is best understood if we assume that the parce que clause is not part of the current discourse sequence and functions at another level that can be named the regulative level. Utterances at this level have the function of regulating the main organization of the ongoing discourse and not of pushing it up. Consequently, we can assume that the grounds for establishing reference are different at the two levels. In (15), there is no possible coreference between the two occurrences of the NP based on the lexical head paléontologue. The first occurrence refers to paleontologists involved in the excavations which took place in Germany mentioned in the narrative main line of discourse, whereas the second occurrence refers to the general class of paleontologists and is used to provide encyclopedic information at the regulative level. e. Autonomous illocutionary force and specific groundedness of information This difference of levels can be noticed in the enunciative domain as well as in the discursive one. The main clause and the parce que clause belong to two distinct enunciative grounds as exemplified in (16) below. The enunciator of the main clause (elle disait…) is a character of the narrative (the grandmother), whereas the enunciative ground of the inserted parce que clause is the speaker itself: the CFPC provides a piece of cultural context ignored by the addressee and necessary for the correct interpretation of the default invited inference of the question: the granddaughter is not a good pupil – as one could infer from “pass the examination at twelve” – but a bad one (twelve was indeed the default age for the Certificate in those times).



Henri-José Deulofeu and Jeanne-Marie Debaisieux

(16) alors ma grand mère elle allait voir l’institutrice et elle disait /est-ce que vous pensez que elle pourra passer le certificat d’études à 12 ans/ parce que c’était pas à 14 ans c’était à 12 ans et tous les soirs elle regardait tu as bien fait tes devoirs tout (MICH. Cas.) ‘Then my grandma she went to see the school teacher and she said: do you think she will be able to pass the Certificate at 12 ‘cos it was not at 14 (as now) but at 12 and every evening she was checking: have you done your homework everything?’ f. No grammatical relation nor category involved All these formal properties could be summed up in saying that the CFPC is a clause which is inserted in an utterance as an utterance on its own, with its specific illocutionary force and communicative grounding, without any grammatical link or constraint regarding the main utterance. Apparently there is a dependency relation between the main and the inserted utterance, in the sense that the CFPC needs a main structure to appear. But what is the exact nature of this dependency? Consider for instance example (16). How is the CFPC licensed? If we take as the main clause the first part of the utterance, all the syntactic tests used to demonstrate syntactic dependency will fail. The only argument for making elle disait the main clause is that, if we drop it, the sequence becomes unacceptable: (17) ?? alors ma grand mère elle allait voir l’institutrice parce que c’était pas à 14 ans c’était à 12 ans et tous les soirs elle regardait tu as bien fait tes devoirs tout Does this test prove that the CFPC is syntactically licensed by the verb or the CP elle disait…? In fact, there is an alternative way to describe the situation. We could say that the CFPC is pragmatically or discursively licensed by the previous discourse. CFPC is not dependent on the main clause as a grammatical structure, but as a meaningful piece of discourse. One could argue against this idea of “discourse dependency” by saying that in all the examples above there are arguments in favour of grouping the parce que clause with a specific utterance. For example, in (16), as the et prohibits the grouping of the parce que clause with the following one, it seems that the only possible grouping is with elle disait est-ce que vous pensez qu’elle va pouvoir passer le certificat à 12 ans? Consequently, in such cases, the licensing pragmatic category would not be a discourse pattern but a specific illocutionary act in this pattern. We assume, nevertheless, that “discourse” licensing is the right analysis. There are indeed examples in which there is no mark of grouping with the preceding or the following clause:



Constructions and context

(18) S1 oui parce qu’il suffit qu’un bouton soit défectueux et puis ça marche pas S2 oui tu peux avoir une panne d’électrique parce que tout est commandé électriquement maintenant ou avec des motorisations des télécommandes un tas de trucs il peut y avoir des systèmes qui fonctionnent pas ‘Yes you can have an electricity failure – ‘cos it is all electrically controlled now or with additional engines or remote control a lot of devices – it can happen that some systems do not operate In such cases the CFPC should be analysed as floating above the whole discourse and not attached to a specific utterance. Even pragmatically, the grouping appears impossible: the content of the parce que clause provides a relevant background for the first utterance as well as for the second one. What CFPC needs in order to be licensed is the production of a piece of discourse, whatever its grammatical form, or even a mere communicative action on the part of one participant in the interaction, for which it will provide, in a backgrounded position, required contextual information. This communicative action can be an illocutionary act conveyed by any kind of linguistic form or even a significant communicative behavior accomplished by the participant. What licenses CFPC is the discourse as a whole and not a grammatical category. A good example for what we have called pragmatic or discursive dependency can be given by an occurrence of CFPC without any possible grammatical basis of insertion. The situation is an interaction between three speakers. L1 is the owner of a farm who speaks only French. L2 and L3 are farm workers and native speakers of Portuguese. L2 is almost bilingual, L3 almost monolingual.€L1 explains the tasks to be done in French to L2 and L3. L2, without having said anything in French to L1 to confirm that he has understood his message, speaks in Portuguese to L3, translating the message of L1, then he turns to L1 and utters: (19) parce qu’en français il comprend encore pas tout ‘cos in French he still does not understand everything The parce que clause inserted in the interaction cannot be related to any grammatical structure: it is inserted in the interaction and depends, as an explanatory background, on the communicative behavior of translation. This description allows for a coherent analysis of the formal properties of CFPC: the lack of grammatical dependency explains the failure of syntactic dependency tests, the pragmatic dependency implied by the foreground- background relationship explains the need for a previous piece of information for CFPC to be licensed.





Henri-José Deulofeu and Jeanne-Marie Debaisieux

1.2.2 Pragmatic function CFPC as a specific form of “discourse” dependent clause is associated with an instructional pragmatic function that can be informally phrased as this: “Take into account preferably the piece of contextual information conveyed by CFPC in computing the meaning of the main illocutionary act (or communicative behavior) and in making the relevant discourse inferences.”

“Preferably” means that this piece of information is focused on within all the possible contextual information available to the addressee. In many cases, it appears that CFPC provides a verbalization of missing contextual information, which is often underlined by metacommunicative comments by the speaker (in bold)3: (20) elle était venue donc au clavier parce qu’elle voulait chanter / parce que c’est une fana du chant j’ai oublié de vous le dire / seulement le problème c’est que c’est une fille qui a une voix un peu d’opéra (ALI. Cas.) ‘She turned to the piano because she wanted to sing ‘cos she is a fan of singing I forgot to tell you…’ but the problem is that she has a voice for opera.’ It is worth pointing out that the conventionalized semantic value of parce que as a lexical unit can be maintained in this use. We are not to consider that parce que has been reanalyzed as a kind of discourse particle. The role of parce que as a link between main and backgrounded utterances parallels its grammatical role relating two facts conveyed respectively by a main and a subordinate clause as in the first occurrence of the parce que clause in (20). The basic conventionalized meaning of the preposition par: à travers ‘through’ can be used to describe it in both cases. Informally, we can phrase it as: “The speech (or communicative) action associated with the main utterance is fully understandable through (à travers) the speech act introduced by parce que.” The causal value of subordinating parce que in [P (main clause) parce que Q Clause] can in turn be obtained through the basic meaning, bearing in mind that the effect of the subordinating relation is to restrict the semantic scope of the parce que clause to the interpretation as a fact of the content of the main clause: P is a fact through fact Q (= fact Q is the cause of fact P). The à travers value is conventionally associated to parce que, as shown by the fact that the CFPC introduced by other conjunctions has a different behavior. For instance, bien que introduces a clause expressing a correction to a preceding assertion as a backgrounded afterthought: (21) j’étais pas trop j’ai jamais été trop système scolaire + euh bien que je j’aime bien quand il y a des règles quand même hein – (BOR-R00PRI003) ‘I was not I never was too (favorable to) strict school regulations –well although I like it when there are rules anyway.’



Constructions and context 

This mitigating function can be linked to the conventionalized meaning of the conjunction bien que ‘in spite of ’ applied to the relationship between the two illocutionary forces of the utterances, that is informally: je suis pas scolaire du tout is a true assertion “in spite” of the truth of j’aime bien quand il y a des règles. As we may expect, the formal properties confirm this pragmatic function. A parenthetic insertion of this bien que clause sounds very strange: (22) ? j’étais pas trop euh bien que je j’aime bien quand il y a des règles quand même hein – j’ai jamais été trop système scolaire + je suis pas scolaire du tout + On the contrary, a CFPC introduced by parce que would be perfectly possible in this position: (23) j’étais pas trop euh parce que je j’aime pas quand il y a trop de règles hein – j’ai jamais été trop système scolaire + There could be a pragmatic explanation for this distribution. In order to provide a correction with bien que, the speaker needs the corrected assertion to be completed. On the other hand, contextual background information can be introduced by parce que as soon as the ongoing communicative act has begun to be processed. The conclusion is that, if the syntax of the pattern is kind of iconic, the morphological marking is achieved through lexemes with conventionalized meanings. 2. Pragmatic subtypes of CFPC We can distinguish two main subtypes of CFPC according to their pragmatic function. a. Regulation of the interaction by framing The CFPC focuses or makes explicit some pieces of background information, thanks to which the addressee will be able to make the required inferences to reach a coherent interpretation of the whole message. In interactions about commonplace topics, the inserted piece is simply missing from the common ground. Consider for instance (20) above, where the speaker explicitly points it out (‘I have forgotten to tell you that’). In more technical discussions involving experts and non-experts, the experts use CFPC in order to bring encyclopedic knowledge in the common ground. For instance, in example (24), the expert must first emphasize the status of patient for a person who has no sensation of pain, which is against the audience’s commonsense expectations. Then, the use of pauvres ‘unfortunate’ to qualify them is also justified by the medical pessimistic prognosis conveyed by the second CFPC.



Henri-José Deulofeu and Jeanne-Marie Debaisieux

(24) il y a bon de très rares cas mais ça existe quand même de malades parce que ce sont vraiment des malades qui n’ont dès la naissance aucune sensation de douleur et ces pauvres enfin ces pauvres enfants souvent parce qu ‘ils ne vivent pas très longtemps dans la plupart des cas sont complètement déformés (PRIE.7, 5) ’Well there are very rare cases – but it happens anyway – of ill people – ‘cos they are actually ill people – who from their birth on have no sensation of pain and these unfortunate –well these unfortunate children –’cos they do not live very long in most of the cases – are totally crippled.’ b. Regulation by reframing In those cases, a coherent and apparently relevant interpretation can be reached by the addressee within the frame of what he thinks is the current common ground. But this is not the interpretation intended by the speaker, whose ground is different. So, in order to guide the addressee to this interpretation, the speaker, aware of the gap between the “grounds,” must block the undesirable inferences which could lead to the wrong interpretation. The reframing is achieved by means of the CFPC, which reestablishes a suitable common ground. This situation is found in (16) above: the wrong default ground is that the examination is taken at age 14, the reframed ground is that the examination is taken at age 12. In (25), the default ground is that you do not buy a driving license but that you have to take (and pass) a test to get your driving license. If the speaker wants to block the unintended inference that the father acted against the law, he must reframe the narrative by adding in the CFPC: “there (in that country) you do not pass the license, you buy it” to get the proper ground. (25) je pense que mon père a su me donner une certaine conscience ce qui fait que dès que je suis retourné j’ai eu mon permis /enfin mon père m’a acheté le permis / parce que là-bas tu passes pas le permis tu l’achètes / donc mon père m’a acheté mon permis de voiture il m’a prêté une voiture (JAL. Cas.) ‘I think that my father succeeded in giving me moral concerns. It is the reason why as soon as I went back home I got my driving license well my father bought me the license –‘cos there you do not need to take a driving test you just buy the licence so my father bought the license for me he lent me a car.’ 3. CFPC does not have the structure of an inserted footnote We said in the introduction that patterns like CFPC play the same role in spontaneous speech as footnotes in written text. This is roughly true but there are some important differences between the two devices that are to be pointed out. Footnotes,



Constructions and context 

and even notes inserted in brackets or balloons in electronic texts, are autonomous semantic and syntactic units inserted in the main text. The main text remains a sequence of well formed sentences if the notes are removed. From a semantic point of view, the notes help to understand the main text, but even without the notes, it is a coherent piece of discourse. The parenthetical oral structures we are studying conform in many cases to this pattern but they may also interact more closely with either the syntax or the semantics of the main discourse, as already mentioned in Section 1. We will address in this section further cases where the CFPC contributes either to the syntactic cohesion or to the coherence of the utterance it is inserted in. (26) below is perfectly acceptable as a discourse pattern. But if we remove the inserted CFPC, as in (26′) it appears that the result is agrammatical, due to the presence of et alors. Whereas the whole discourse pattern is widely attested in spontaneous speech, the reduced one sounds very strange and unusual. (26) les Espagnols alors quand on voulait les faire enrager parce qu’alors les Espagnols eux ils étaient flegmatiques ils restaient des journées entières allongés dans des salles de classes à rien faire et alors pour les faire enrager on leur disait Arriba España Viva Franco (RAIM. Cas.) The Spanish then when we wanted to upset them –‘cos you know the Spanish were very phlegmatic they stayed the whole day long lying in classrooms, without doing anything – and then to upset them we said to them: Arriba España, Viva Franco.’ (26′) ? les Espagnols alors quand on voulait les faire enrager et alors pour les faire enrager on leur disait Arriba España Viva Franco One way to describe this situation is to say that the alors and the et in (26) are not licensed by any category in the construction underlying the main utterance, but by the discourse CFPC pattern as a whole (the repeated alors function as iconic brackets and the et (alors) marks the transition between the background and the main line of the discourse). It is then expected that, in the absence of one essential part of the pattern, those marks should be absent. We arrive at the conclusion that the syntax of a discourse pattern may interact with the syntactic patterns or the constructions included in it to create new kinds of syntactic forms. These blended structures are certainly ungrammatical as clause constructions but perfectly acceptable and functional as discourse patterns. Another example of this phenomenon is provided by (27): (27) en Algérie par exemple les panneaux publicitaires /rares panneaux parce que c’est un pays où la publicité n’est pas aussi dominante qu’au Maroc mais les rares panneaux publicitaires sont écrits en arabe (PHU. 8, 5)



Henri-José Deulofeu and Jeanne-Marie Debaisieux

‘In Algeria for example the very few advertisement posts very few ‘cos it is a country where advertising is not so developed as in Morocco but the very few ad posts are written in Arab.’ (27′) ? en Algérie par exemple les panneaux publicitaires rares panneaux mais les rares panneaux publicitaires sont écrits en arabe (PHU. 8, 5) Here the reduced version is also agrammatical.€But its syntax can be explained at the discourse pattern level in the full version. The lexical repetition of panneaux marks the bracketing of the CFPC and the mais is licensed as a mark of a “rhetorical” relation of contrast between the content of the parenthetical and an element of the main line (there is less advertisement in Algeria than in Morocco but it is more remarkable, because it is written in Arabic). Example (28) shows a more complex case of interaction between background and main line. The parenthetical contributes not only to the syntax of the whole pattern, but also to its coherence. There is no means in the reduced version to reach a coherent interpretation. In the full version it is clear that S1 wants to say that, for the Mexicans, French people were perhaps not heroes, but at least much more popular than Americans. In the reduced version it is impossible to obtain this interpretation, as the only possible coreference for the pronoun ils is with les français, (the preposition vis a vis blocks the coreference with les americains). So we get the wrong interpretation that in comparison with Americans, the French were not very popular. What allows the correct interpretation in the full version is that ils can be coreferent with les américains, introduced in rhematic position by the CFPC. (28) S1 mais là vraiment les français pour les mexicains c’était S2 hum des héros S1 non peut-être pas jusque là mais bon / vis à vis des américains parce qu’on avait les américains aussi ben ils ont pas ramassé un tel succès ‘but then frankly the French for the Mexicans they were… well heroes … non I wont go so far but well compared to the Americans –‘cos we had Americans also – well they didn’t get so much popularity ? S1 mais là vraiment les français pour les mexicains c’était S2 hum des héros S1 non peut-être pas jusque là mais bon / vis à vis des américains ben ils ont pas ramassé un tel succès



Constructions and context 

4. The status of CFPC in the CG framework Our description leads to the following conclusions: a. CFPC is a construction in the broad sense, that is a conventional association of form and meaning. b. It appears to be licensed by a communicative pattern: foreground –background as a whole. As it is impossible to predict the grammatical or even linguistic composition of the foreground, which can be any kind of communicative behavior, we are led to the conclusion that CFPC is pragmatically licensed by the communicative pattern, as a device allowing the speaker to focus on the contextual pieces of information he thinks the addressee needs to get the more relevant interpretation of what is being communicated. This leads us to address two questions: what is the exact nature of such a discourse pattern and what kind of inheritance network is it involved in? The basic components of the form of such patterns are not limited to linguistic elements: categories or constructions, but extend to “communicative behavior” recognized as such by the participants in the interaction (Berrendonner 2004). This behavior may be performed through utterances or significant attitudes. The utterances may be based on grammatical constructions or merely on paralinguistic forms (interjections, discourse particles, onomatopoeia). The constructions that put these forms in relation can be analyzed as functionally distinct groupings of communicative or discourse units. The kind of grouping we have analyzed in this paper can be labeled as a [discourse – background] pattern, or even more generally [communicative behavior – background]. What should be pointed out is that grammatical devices are used to specify the relationship between the background and the main line of discourse. These grammatical devices such as the conjunction parce que are therefore multifunctional: on the one hand, they can be involved in specifying grammatical relations in constructions headed by a grammatical category. For instance, parce que specifies a grammatical relation verbadjunct as a cause-consequence one. On the other hand, they specify, with the same conventionalized meaning, the relation between two communicative behaviors in an interaction. As we have seen, parce que instructs the addressee to use the information conveyed by the communicative unit it introduces as a “path” through which the meaning of the “nucleus” communicative behavior is to be reached. Such a situation can be generalized to many grammatical morphemes. Insights in this direction can be found in the literature. CFPC recalls the discourse patterns pointed out by Mithun (2005), rejoining observations by other linguists



Henri-José Deulofeu and Jeanne-Marie Debaisieux

(Goethals 2002; Miller 1998; Deulofeu 1999; Debaisieux 2004; Debaisieux & Deulofeu 2004; Hopper 1988; Iwasaki & Ono 2002; Ono & Thompson 1995): In a number of languages the prosodic, morphological, and semantic cues that might characterize the sentence do not always converge. Dependent clause markers appear pervasively in what seem, on prosodic and semantic grounds, to be independent sentences. A closer look shows that these markers are being used to signal pragmatic dependency among larger elements in discourse. The markers of dependency serve several recurring functions in discourse. The Yup’ik Participial and Barbareno nominalized sentences contribute background, descriptive, subsidiary, explanatory, or evaluative information, information that does not move narrative forward. The Yup’ik Subordinative and the Hualapai switch-reference markers signal textual cohesion, marking statements that together compose a larger discourse unit. (Mithun 2005)

The common observation is that grammatical units can be involved in patterns that go beyond the scope of sentence grammar. But the case of CFPC helps us to generalize Mithun’s conclusions: not only discourse but interaction patterns are involved in CFPC. Our conclusion will be that these constructs support the broad conception of constructions currently assumed, that any kind of pairing between form and meaning should be considered as a construction, even if the form reduces to a prosodic pattern and the meaning to a broad pragmatic function (Croft 2001). Our claim is that this comprehensive view of construction must even be extended to non verbal communicative behavior if we want to encompass all the possible forms of CFPC. One strong result that emerges from these conclusions is that grammatical signs like conjunctions are not only used to establish relationships between other signs but between signs and indexical aspects of human communication: gestures, inferences that are drawn from indexes. The challenge will be to find an adequate formalism to integrate the grammar of indexes into the grammar of signs. As for the inheritance networks, CFPC inherits its formal properties from two more general patterns. First, as it responds negatively to all diagnostic tests which could be used to show its syntactic integration within the main clause, it must be analysed as an instance of “supplementive clause construction” and not as an instance of Head-Adjunct pattern. Second, it also inherits properties from a more general bare parenthesis construction (bpc), exemplified in the italics part of the following utterance: (29) et après tu ouvres les feuillets comme un livre et avec la lumière du soleil il faut absolument qu’il y ait le soleil + il y a une ombre portée de la colonne vertébrale qui te dit que là il y a un poisson ou autre chose



Constructions and context 

‘And after you open the layers like book pages and with (the help of) the sun light –it is absolutely necessary that there be sun – there appears the cast shadow of a spinal column, which tells you that there is a (fossil) fish or something.’ CFPC seems to fulfill the same metacommunicative function as the bpc parenthetical pattern in (29). Indeed a parce que could very well be inserted in (29). The main difference is that what is achieved by means of the iconic device of mere insertion of an independent utterance in another as in (29) is achieved in CFPC through a conventionalized meaning of the morpheme parce que related to its canonical use as a causal subordinator. As a consequence, the presence of parce que, as we have seen above, narrows the scope of pragmatic functions that the bpc can fulfill. The question remains whether cfpc could be put together with the large class of parentheticals including epistemic parentheticals (Thompson & Mulac, 1991), or apparent dependant clause constructions which acquire interpersonal meaning (Traugott 1989). This important question goes beyond the scope of our paper but we want to point out differences supporting the analysis of parentheticals and parenthesis like constructions (CFPC and BPC) as distinct structures. As mentioned above, on formal grounds, parentheticals are clearly related to a specific clause in which they are inserted at a constituent boundary (Marandin 1999). CFPC and BPC are inserted more freely inside an utterance, but cannot appear in front of it. On semantic grounds, parentheticals elaborate the content of the proposition conveyed by the utterance, modifying its epistemic status or adding subjective values to it. On the contrary, CFPC is totally external to this content. The CFPC of example (15) is a good case in point. It doesn’t modify in any way the semantics of the utterance it is inserted in, but contributes to providing the common ground with which the addressee can elaborate an appropriate contextual interpretation of it and specially of the word travailler: it helps anticipate a request of clarification from the addressee about the exact role of the paleontologist “assistants” (des gens qui travaillaient pour les paléontologues). The distinction between parentheticals and parenthesis insertions, of which CFPC is an instance, seems descriptively justified, even if the line between the discourse and metadiscourse levels is not always easy to draw. Notes * We are very grateful to the anonymous reviewer whose comments helped us to improve a first version of this paper.

 Henri-José Deulofeu and Jeanne-Marie Debaisieux 1. On the contrary, the internal shape of the parce que clause is more grammatically constrained. Even if we can find more complex patterns than in adjunct parce que clauses, they are all based on a tensed clause. In (13) and (14) for instance, the CFPC is composed of a juxtaposition of IPs, instead of the coordination of CPs found in subordinate parce que clauses. We will not address the analysis of the internal structure of the parce que clause (Debaisieux 2002). 2. The same kind of pattern has been found in other languages. For example in spoken German (Stoltenburg 2003) or Finnish (Duvallon 2006). 3. Notice that in this piece of discourse a parenthetical parce que clause is combined with a preceeding subordinate adjunct parce que clause (parce qu’elle voulait chanter).

References Berrendonner, Alain (2004). Grammaire de l’écrit vs Grammaire de l’oral: le jeu des composantes micro- et macro-syntaxiques. In A. Rabatel, (Ed.) Interactions orales en contexte didactique (249–264). Lyon: Presses universitaires de Lyon. Croft, William (2001). Radical Construction Grammar. Oxford: Oxford University Press. Debaisieux, Jeanne-Marie (2002). Le fonctionnement de parce que en français contemporain: étude quantitative. In C. Pusch & W. Raible (Eds.), Romanistische Korpuslinguistik – Romance Corpus linguistics, (349–362). Tübingen: Gunter Narr Verlag. Debaisieux, Jeanne-Marie (2004). Les conjonctions de subordination: mots de grammaire ou mots du discours? Le cas de parce que. Revue de Sémantique et de Pragmatique, 15/16, 51–67. Debaisieux, Jeanne-Marie & José Deulofeu (2004). Fonctionnement microsyntaxique de modifieur et fonctionnement macrosyntaxique en parataxe des constructions introduites par que et parce que en français parlé, avec extension au cas de perché et che en italien parlé. In F. Leoni, F. Cutugno, M. Pettorino, & R. Savy (Eds.), Parlato Italiano in cd-rom, Atti del Convegno nazionale di Napoli, 13–15 febbraio 2003, CD multimediale (isbn 88–7092– 238–3). Napoli: M. D’Auria Editore. Deulofeu, José (1999). Questions de méthode dans la description morphosyntaxique de l’élément que en français contemporain. Recherches sur le français parlé, 15, 77–101. Espinal, Teresa (1991). The representation of disjunct constituents. Language, 67–4, 726–762. Goethals, Pierre (2002). Las conjunciones causales explicativas españolas como, ya que, pues, y porque. Un estudio semiótico-linguístico. Paris€& Leuven: Peeters. Hopper, Paul (1988). Emergent Grammar and the a Priori Grammar Postulate. In Deborah Tannen (Ed.), Linguistics in context: Connecting observation and understanding (117–134). Norwood NJ: Ablex. Huddleston, Rodney &. Geoffrey K. Pullum (2002). The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. Iwasaki, Shoichi & Tsuyoshi Ono (2001). Sentence in Spontaneous Spoken Japanese Language. In J. Bybee & M. Noonan (Eds.), Complex sentences in Grammar and Discourse (176–202). Amsterdam & Philadelphia: John Benjamins. Kay, Paul (1997). Pragmatic Aspects of Grammatical Constructions. Online: http://www.icsi.berkeley.edu/~kay/cg.prag.pdf, p.€3. Marandin, Jean-Marie (1998). Grammaire de l’incidence. Online: http://www.llf.cnrs.fr/Gens/ Marandin/GrammaireDeL_incidence.htm.



Constructions and context  Martin, Philippe (1978). Questions de phonosyntaxe et de phonosémantique en français. Linguisticae investigationes, 2:1, 93–126. Amsterdam & Philadelphia: John Benjamins. Miller, Jim & Regina Weinert (1998). Spontaneous spoken language: syntax and discourse. Oxford: Clarendon Press. Mithun, Marianne (2005). On the assumption of the sentence as the basic unit of syntactic structure, Linguistic diversity and language theories. Unpublished manuscript. Morel, Mary-Annick & Laurent Danon-Boileau (1998). Grammaire de l’intonation, Paris-Gap: Ophrys. Ono, Tsuyoshi & Sandra A. Thompson (1995). What can conversation tell us about syntax? In Philip W. Davis (Ed.), Alternative linguistics (213–272). Amsterdam & Philadelphia: Benjamins. Stoltenburg, Benjamin (2003). Parenthesen im gesprochenen Deutsch. InList, 34. Online: http:// www.unipotsdam.de/u/inlist/issues/34/index.htm. Thompson, Sandra & Anthony Mullac (1991). A quantitative perspective on the grammaticalization of Epistemic Parentheticals in English. In E. Traugott & B. Heine (Eds.), Approaches to grammaticalization, Vol. II, (313–329). Amsterdam & Philadelphia: John Benjamins. Traugott, Elisabeth & Bernd Heine (Eds.) (1991). Approaches to grammaticalization, Vol. II. Amsterdam & Philadelphia: John Benjamins.

Representing contextual factors in language change Between frames and constructions Mirjam Fried

Czech Academy of Sciences, Prague

1. Introduction It is accepted both within cognitively based approaches to semantic structure (e.g. Fillmore 1982, 1984, 1986; Geeraerts 1992, 1993; Cuyckens, Dirven & Taylor 2003) and within dialogical approaches to language analysis (e.g. Linell 1998) that speakers’ expressive needs, together with general cultural understanding and specific contextual clues, participate in negotiating the meaning of a given piece of discourse. The dynamic character of language use also presents this question necessarily as one of variation and change; this dimension has been explored in a particular strand of grammaticalization studies that focus on various aspects of the interaction between codified semantic structure (‘lexical meaning’), broad thematic and cultural context, and recurring morphosyntactic patterns as giving shape, collectively, to a newly emerging grammatical structure (e.g. Bybee, Perkins & Pagliuca 1994; Bisang 1998; Hopper 1998; Traugott 2003, 2008; Harris 2003; Wiemer 2004; Wiemer & Bisang 2004). A question that is yet to be taken up concerns the way(s) in which those complex interactions and their dynamic nature can be systematically captured and further illuminated by applying a particular grammatical model. With this general goal in mind, I will present a small-scale corpus-based case study of a lexico-grammatical change that will illustrate what types of context may enter conventional linguistic patterning. By examining specific aspects of language change, my focus will be on representing a shift in speakers’ grammatical knowledge, its relationship to recurring semantic and pragmatic constraints and their gradual reorganization. On the surface, this seems to be a straightforward enough goal: in functionally oriented approaches, it is taken as self-evident that linguistic expressions are used in ‘context’ and that certain aspects of the ‘context’ may find their way into



Mirjam Fried

the grammatical properties of a linguistic expression. The task for a grammarian, then, is to articulate such a connection precisely and explicitly. Upon further reflection, though, it becomes far from clear what exactly we may mean by ‘context’ and what exactly it is we can or ought to be explicit about. A glance at the rich literature on pragmatics in general (traditionally a primary domain of addressing contextual issues) and on the connection between pragmatics and grammar suggests that there are several ways in which the notion of context is understood and incorporated in linguistic analysis. One set of contextual factors revolves around issues of usage, i.e. the broadly conceived pragmatic conditions, including the question of textual distribution, socio-cultural grounding, genres, etc. This understanding of context is traditionally associated with the field of pragmatics and usually has paid little attention to grammatical structure. For grammarians, on the other hand, reference to context has typically meant concerns with managing information flow and the configuration of discourse participants, whether within a clause or within a larger stretch of utterances. Contextual issues are thus discussed in terms of regular correlations between grammatical structure and information structure, the latter articulated in terms of specific discourse roles. More recently, though, the notion of context has also been used to mean the grammatical environment in which a given word or expression occurs, either preferentially or exclusively. This concept of linguistic ‘context’ goes back to Fillmore’s observations about the combinatorial restrictions on individual lexical items when they enter syntactic phrases and sentences (Fillmore 1982, 1989; Fillmore, Kay & O’Connor 1988) and has been also the cornerstone of Croft’s approach to defining grammatical categories (Croft 1991, 2001). Since the present study takes the grammarian’s perspective, the point of departure will be the syntagmatic context. However, it will quickly become apparent that other contextual factors must be worked in as well, echoing Fillmore’s early arguments for incorporating discourse-based descriptions in our generalizations about individual lexico-grammatical patterns (Fillmore 1974/1981, 1982). I will document and argue that all three types of context enumerated above must be taken into account if we wish to provide an adequate representation of linguistic structure and its changes over time. Following the general strategy proposed by Östman (2005), I will suggest a way of incorporating the general insights found in the extensive research on genres and text typology (Werlich 1976; Biber & Finegan 1989; Swales 1990; Bhatia 1993; Biber 1994; Halmari & Virtanen 2005; Bergs 2005a, among many others). Specifically, Östman’s (2005:€130) notion of “discourse pattern” will provide the necessary starting point for formalizing the textually motivated constraints on grammatical structure, as they emerge from the data under discussion. This general approach will lead to more comprehensive conclusions



Representing contextual factors in language change 

about how the different types of context may become distributed over different parts of language structure over time. The main questions motivating this study are thus the following: how do contextual and grammatical factors interact in language variation and change, to what extent may their interaction become conventionalized, and can any general tendencies be detected in the conventionalization process? In answering these questions, the paper also touches on the elusive nature of lexicalization as a distinct diachronic process. While it is not my goal in this study to address the lexicalization puzzle directly, the analysis will lead to some observations about the promise of a constructional approach for illuminating the complexities involved in solving the lexicalization problem. 1.1

Case study

The material for the case study comes from a particular set of expressions in Old Czech (OCz). I will trace the development of the word form věřící ‘(one) believing’, which in many ways illustrates the rich semantic and morphosyntactic history of the so-called ‘long’ present active participles, a special participial form characteristic of Slavic and Baltic languages (with some parallels in German as well). I will refer to this form as ‘participial adjective’ (PA), which is misleading in terms of the form’s functions, but it reflects its morphologically mixed-category status: an adjectival inflectional case/number/gender suffix is attached to a verbal stem, turning the result into a morphologically non-verbal entity. A preliminary illustration of this morphosemantic structure is in (1). The label part stands for ‘present active participle’, which is formed by the -nt- suffix (the -nt- label indicates the participle’s relationship to its Latin cognate).

(1) [[Vroot – theme – NT ]part – c/n/g]PA

[[ věř – í – c ]part – í ]PA ‘(the one) believing’ The shift in lexical category places the PA in the gray area between inflection and derivation, which alone poses a challenge for linguistic theory. Moreover, we will see that the functional and categorical status of the c/n/g suffix is distributed between nominal and adjectival properties and, therefore, cannot by itself fix the category for the whole word form, thus contradicting the received view that participles turn into adjectives (Haspelmath 1996; Schenker 1995:€106 on Slavic; Short 1993:€487 specifically on Czech). As a result, the PA provides a particularly illustrative case for studying the relationship between meaning, form, and context, since the way in which the competition between its verbal, nominal, and adjectival potential was resolved cannot be adequately described without considering the ways in which this form was used in actual discourse.



Mirjam Fried

To start with, we need to capture the fact that the PA věřící developed from a functionally unspecified, fully context-dependent, and morphosemantically transparent form that was part of the regular verbal inflectional paradigm and that occurred in all major syntactic functions (reference, modification, and predication), into primarily an actor noun meaning ‘practicing Christian’ and marginally a quasi-adjectival modifier with the same meaning, while also maintaining its fully inflectional and compositional status that can be glossed as ‘[who] believes sth. to be true’. Furthermore, the form is simultaneously attested in a number of other interpretations (‘believable’, ‘pious’, ‘creditor’, ‘trustworthy’, ‘authorizing’, etc.) that semantically correspond to the polysemous structure of the verb root věř- ‘believe’ and that were present throughout the OCz period. Sweeping generalizations concerning just the word form itself are thus impossible even on lexical semantic grounds, let alone in its syntactic behavior. Since the case study focuses on a single word only, it would perhaps be attractive to consider its diachronic development a case of lexicalization (to the extent that there even is any consensus about what we mean by lexicalization processes, cf. recent debates in Wischer 2000; Himmelmann 2004; Traugott 2008; Haas 2005). I will suggest that such a treatment would be too simplistic; it would render the change as essentially random and unrelated to the fact that the same kind of change concerns a whole class of word forms, not just this one item. Before we can attach any classification to the type of change involved (say, V → N or V → Adj), we must consider all the relevant details of the gradual (re)shaping of multiple associations between form and function, which in this case show great interdependence between verb meaning, context in which the word form is used, and the morphosemantic structure that mediates this relationship. The analysis will demonstrate that specific lexico-pragmatic properties correlate with different functional outcomes that are not predictable simply on the basis of assigning lexical categories in the abstract but, rather, depend to a great extent on different syntagmatic preferences. The development thus will be analyzed along several dimensions: semantic shifts vis-à-vis textual distribution and contextual preferences (in the sense of conditions on usage); role of collocational patterns (syntagmatic context); potential categorial status (N, Adj, V) vis-à-vis syntactic behavior; and chronological order. The result is a complex picture of all the factors that collectively motivate the changes in the form’s usage and that must be part of any representation that aspires to be descriptively accurate and to offer an explanatory value. The examples I use are based on an extensive corpus of excerpts from OCz texts; the corpus contains a wide variety of genres and spans the period of about 250 years.1 This chronological stretch allows us to see both the synchronic variation for any given generation of speakers and the gradual conventionalization of specific patterns, i.e., the resulting change.



Representing contextual factors in language change 

1.2

Theoretical background

Given the central role of syntagmatic context in the present study, it is only natural to carry out the analysis within the theoretical apparatus of Construction Grammar (Fillmore 1988, 1989; Fillmore, Kay & O’Connor 1988; Croft 2001; Fried & Östman 2004) and Frame Semantics (Fillmore 1982; Fillmore & Atkins 1992; Atkins 1994; Atkins et al. 2003; Fillmore et al. 2003; Fried & Östman 2003; Fried 2004, 2005), which together offer a particularly attractive model for representing lexico-grammatical networks in which the relative stability of grammatical form does not conflict with the relative flexibility of meaning and expressive richness, and vice versa. Construction Grammar (CxG) does not make a sharp distinction between grammar and lexicon, thus allowing for systematic incorporation of gradient phenomena, such as is inherent in mixed-category morphology. The model’s fundamental assumption that the basic unit of grammatical structure is a construction, i.e. a conventional function-form pairing, naturally sets up grammatical constructions as the domain of language change, which is also the view increasingly argued for in grammaticalization research. Finally, both of these features ensure that we can capture the gradualness characteristic of the slow process of losing internal structure and compositional meaning, resulting in new linguistic patterns with noncompositional properties. In fact, it is one of the defining features of grammatical constructions that they are never just the sum of their parts; mixed-category morphology thus constitutes a prime candidate for testing a constructional approach. The framework known as Frame Semantics is concerned with the ‘semantics of understanding’ as the appropriate model of linguistic semantics. Linguistically relevant semantic information is organized and structured in “interpretive frames” (Fillmore 1982:€124), which represent the complete background scene associated with a given linguistic expression: the scene’s participants, settings, and any other unique semantic features (collectively referred to as ‘frame elements’) that are necessary for speakers’ native understanding of what the lexical item means and how it can be used in context. A single linguistic expression may be (and often is) associated with multiple frames and, conversely, a single frame may be shared by multiple expressions; each such expression, then, represents a particular conceptualization of certain parts of the larger background scene. In the case of predicates, i.e. argument-taking lexemes, the frame also carries information about the conventional expression of the syntactically relevant participants as they manifest themselves in the syntactic organization of sentences. This is a unique feature of Frame Semantics as a lexical semantic model: the built-in connection between lexical meaning of an item and the canonical morphosyntactic expression of its frame elements. The connection, which also plays a prominent role in the present



Mirjam Fried

analysis, is expressed in the form of a valence, which thus functions as an explicit link between meaning and form in the domain of argument expression. The paper is organized as follows. Section 2 introduces the morphosemantic structure of the PA, identifying the (potential) syntactic and semantic contribution of the PA’s components. Section 3 is devoted to the details of the diachronic development of the form věřící, identifying the factors that were instrumental in the gradual reorganization of the PA’s morphosemantic structure due to specific contextual pressures. Section 4 addresses the representational issues concerning the gradual loss of compositionality and proposes ways of capturing contextual effects in frames and constructions. Section 5 offers general conclusions about contextual representation in Construction Grammar and some questions left for future research. 2. Morphosemantic structure of věřící Before we can address the form věřící itself, we must consider the semantic structure of the verb root věř- from which the PA is derived. For the purposes of this study, it is sufficient to present a brief overview, which is based on a more detailed analysis of the verb semantics in Fried (2008a). The OCz verb věřiti ‘to believe’ is attested in at least four major senses (roughly, belief, faith, credit, trust); each is associated with a distinct complementation pattern and some senses reflect a particular communicative context (e.g., religious faith, various types of commercial transactions, transfer of legal responsibility). The overall structure is presented in the frame-based semantic network in Diagram 1. The rectangles represent the frames, each of which corresponds to a distinct sense within the prototype structure. For now, the valence information is represented in a much abbreviated, simplified format: in each frame, the list on the left represents the linguistically required frame elements and the list on the right gives the corresponding morphological expressions as they conventionally occur in active sentences. Maintaining the same indices (#1–3) across the frames is intended to capture the fact that the same set of event participants, common to the shared background scene of believing, takes on different frame-specific roles in the individual senses, each of which corresponds to a distinct conceptualization of holding a belief. (A more precise formalism will be used in Section 4, when we try to capture the PA’s place in this semantic network.) In Diagram 1, the conceptually simplest sense is in the frame labeled belief, meaning roughly ‘believe that [what somebody says] is true’ (literally, ‘believe somebody in some matter’), and as such it can be posited as the prototype sense (indicated by the thick-line rectangle). We also note that this sense requires minimally three participant roles, labeled Believer, Recipient of Trust, and Content.



Representing contextual factors in language change  BELIEF (content of thought) #1 Believer – NOM #2 Recip. of Trust – DAT #3 Content – v LOC before 1350 TRUST (expectation) #1 Dependent – NOM #2 Caretaker – DAT #3 Valued Entity – GEN around 1410

context: religious faith FAITH (content of thought) #1 Believer – NOM #2 Deity – (v) ACC before 1350 context: commercial transaction

before 1350 CREDIT (expectation) Frame: TRANSFER #1 Giver #1 Creditor – NOM #2 Recip. #2 Debtor – DAT #3 Gift #3 Valued En. – ACC Frame: COMMERCIAL TRANSACTION FEs: Buyer, Seller, Goods, Cost, Tender,...

Diagram 1.╇ (Partial) frame-based semantic network of the OCz verb věřiti

The faith sense can be seen as a narrower reading of belief, motivated by the common usage of this verb in the context of religious faith. This sense has only two participants expressed in syntax: a Believer and a Deity; the latter perhaps fuses the Recipient of Trust and the Content elements of the prototype. Both belief and faith also share the basic semantic feature of relating to the content of someone’s thought. Of about the same age is also the sense credit, in which the act of believing is cast in the context of a commercial transaction, thus sharing also the semantics (and complementation pattern) with verbs of commercial transaction (buying, selling, paying, etc.) and of transfer (giving, taking) in general; these connections are indicated by the rounded rectangles representing these additional frames, which are inherently unrelated to verbs of believing but overlap with one of the senses. The meaning of this usage is literally ‘believe something to somebody’, where ‘something’ is a Valued Entity (as a special interpretation of the Content role in the prototype) and ‘somebody’ is the buyer (Debtor), corresponding to the Recipient of Trust in the prototype. Idiomatically, the meaning can be glossed



Mirjam Fried

as ‘sell something to somebody on credit (i.e., on the belief that the buyer will pay later)’, illustrated in (2): {jal se předivně kupčiti, bera draze na úvěrky...} což kto jedno chtěl věřiti what.acc who.nom one want.ppl.sg.m credit.inf3 {všecko pobral} ‘{he traded in the strangest way, taking [things] on expensive credit},whatever who[ever] wanted to offer on credit {he [=king] took it all}’ [Budyš 63; 1420; social satire]2 (2)

Note that this sense is not about the content of someone’s thought, but expresses an expectation with respect to a particular behavior. This semantic feature is shared also by the youngest sense, trust, which means ‘believe that a (protective) action will be taken’ or, more literally, ‘believe someonei with respect to doingi X’. The trust sense is attested only marginally and will not be relevant to the issues discussed in this study since there is no attestation of a corresponding PA. Nor will I be concerned with the belief sense, whose PA did not develop into an actor noun (cf. Fried 2008a). The PA form věřící originated within the same semantic structure, as part of the verbal inflectional paradigm. However, it developed differently within individual senses and also added meanings that are not attested with other forms of the verb, including the ‘short’ participle that forms the PA’s stem. It is therefore necessary to examine more closely the morphosemantic structure of the PA itself. In addition to the lexical meaning and the valence structure contributed by the verb root, we have to consider the remaining formatives and their own contribution to the meaning of the PA as a whole. As already mentioned, the PA is derived from the present active participle (the ‘short’ form, sometimes also called, misleadingly, ‘gerund’), which was used primarily as a non-finite predicate expressing various dependent adverbial clauses, such as is shown in (3); the participle is underlined. (Throughout the paper, I will use curly brackets {} to enclose additional context surrounding the token in question.) {přidávám … múdré a opatrné lidi} věřě jim mimo všecky přátely believe.nt.sg m 3pl.dat besides all.acc.pl friend.acc.pl ‘{I’m sending … [these] wise and cautious people}, because I trust them more than any of [my] other friends’ [ArchČ 15, 551; 1455; correspondence] (3)

The PA’s participial stem marks explicitly several verbal categories. Present tense (interpreted as contemporaneousness, as is typical of present-tense depictive secondary predicates, cf. Schultze-Berndt & Himmelmann 2004:€102) is a property



Representing contextual factors in language change 

of the present stem from which the participle is derived (in contrast to a distinct past-tense stem). Active voice or ‘orientation’ (Haspelmath 1994:€153) is associated with the participial suffix -c- itself (in contrast to other types of participles, such as past or passive). Finally, the root brings along a valence and the participle normally expresses, through verbal government, the non-agent arguments in that valence (i.e. those arguments that would not be linked to the subject function in an active sentence). This feature follows the behavior of other non-finite verb forms (infinitive, gerund, active participles) and in (3) is illustrated by the dative jim ‘to them’ (= Recipient of Trust). All of this constitutes the internal morphosemantic structure of the PA and represents the PA’s verbal potential, predisposing it, at least in principle, toward uses expressing secondary predication, comparable to the short participle in (3). And indeed, such examples are attested, as we shall see shortly. The verbal potential conflicts with the external c/n/g suffix, which is clearly non-verbal. Its categorial status, however, is open: it is adjectival in its morphological form, but referential in its function; the suffix is etymologically a demonstrative pronoun and its contrastive function was not yet fully neutralized in early OCz (Kurz 1958). The development of the PA reflects these open-ended functional possibilities. Throughout OCz, the PA věřící is attested in all three syntactic functions – reference, modification, and predication – that are a priori available. In fact, it is often difficult and sometimes impossible to determine unambiguously the function of a given token. For the purpose of illustration, a straightforward and uncontroversial example of each function is given in (4): an actor noun in (4a), which reinforces the referential potential of the external morphology (‘the one believing’); an adnominal modifier in (4b), which capitalizes on the adjectival form of the c/n/g suffix (‘which/that believes’); and a predicate heading a non-finite adverbial clause (here in the form of a genitive absolute) in (4c), which builds on the verbal potential of the internal morphology (‘[while] not believing’) in spite of the external suffix. The relevant phrases are enclosed in brackets <> for easier identification. (4) a.

Kristusi Ducha Svatého dává v Christ.nom Spirit.acc Holy.acc gives.pres.3sg in nich < svými věřícím> 3pl.loc his.dat.pl believe.pa.dat.pl ‘Christi gives in them [=his gifts] the Holy Spirit to ’ [ŠtítVyš 84b2; 1396; religious tract]



Mirjam Fried

b.

lid < v€ buoh věřící > people.nom.sgm in god.acc.sg believe.pa.nom.sg spasen byl save.pass.sg m be.ppl.sg.m ‘the people < that believed in God > were saved’ [PasMuzA 488a; mid 14th cent.; legend]

c. < ješče jich nevěřících >... vece jim still 3pl.gen neg.believe.pa.gen.pl say.pres.3sg 3pl.dat ‘< as/because they still don’t believe [him] >…, he says to them’ [EvZimn L 24,41, late 14th cent.; biblical] The fluid functional distribution gradually gave way to the more restricted distribution we know from ModCz, as described in Section 1.1. In the rest of the paper, I will examine the contextual details that led to fixing the actor noun status in particular. I will take a close look at how this form was used, in what kinds of context (syntagmatic, pragmatic, semantic), and how the usage affected the functional outcome. 3. Factors in fixing the noun potential of věřící As a syntactic noun, the PA is attested in only two senses of the verb meaning: faith and credit. I will turn to the faith sense first. 3.1

věřící in the sense of faith

This PA usage appears early (by 1350 in my corpus), together with the finite forms of the verb, and it is attested in all kinds of texts. It is mostly accompanied by its non-subject argument (i.e., the Deity participant), which always retains the same form it would have in a finite clause: the accusative-marked NP, in older examples as a plain accusative, later on with the preposition v ‘in’; this is also the diachronic pattern we find with the finite forms. The verbal properties of the PA are thus preserved in terms of its internal syntax and also in its active orientation: the PA’s referent is the one who believes.



Representing contextual factors in language change 

At the same time, the syntactically verbal character is undermined in several ways. Consider the set of examples in (5): (5) a.

bieše anděl zemský … všech be.pst.3sg angel.nom.sg.m earthly.nom.sg m all.gen.pl < v buoh věřících > chvála in God.acc.sg believe.pa.gen.pl praise.nom.sg.f ‘[St. Paul] was an angel on earth…, the praise of all those ’ [PasMuzA 303a; late 1300s; legend]

b.

{Jezukriste,...jsi…} všech < v€ tě věřících > útěcha all.gen.pl in 2sg.acc believe.pa.gen.pl solace.nom.sg.f ‘{Jesus Christ, you are} … the solace of all those ’ [LegKat 65a; late 14th cent.; spiritual poetry]

c.

{boha jezukristai jenž všudy} pomáhá < v€ sěi věřícím > help.pres.3sg in self.acc believe.pa.dat.pl ‘{of God Jesus Christi who in all places} helps those ’ [OtcB 196b; late 14th cent.; legend]

First off, the syntactically present Deity argument does not provide a highly informative or novel contribution: the referent is always God/Christ, expressed sometimes directly (5a) but most often by a pronoun, either personal (5b) or reflexive (5c), that invariably refers to Christian God or Christ. The pattern suggests pragmatic ‘emptying’ of the complement, whose presence serves merely as a placeholder in the syntactic valence of the stem, not as an expression identifying a pragmatically unpredictable participant in the reported event. In fact, there are texts (e.g. VýklŠal from early 15th century, which is a religious tract interpreting the Song of Songs as a simile for proper Christian faith and the love of Christ) in which the bare form věřící consistently alternates with the phrase věřící v€Krista ‘believer in Christ’, with no detectable difference in meaning or communicative effect. In both this and other texts, the phrase was clearly used as a fixed collocation that was not intended to deliver any novel content. It must also be noted that casting the religious faith in terms of Christian faith in particular is a feature specific to the PA, not carried over from the verb semantics. There the Deity truly can be any object of worship, such as modly ‘pagan idols’ or gods of other religions (cf. Fried 2008a), both in OCz and ModCz. The PA thus narrows down the semantic range of the verb root and this is corroborated by the fact that the noun křest’an ‘Christian’ is sometimes used explicitly as this PA’s synonym (e.g. HusBlud 293).



Mirjam Fried

The low informativeness of the complement necessarily weakens the verbal potential of the PA, despite the full expression of the root’s valence. By mid-15th century, we indeed see the crystallization of the PA’s meaning into denoting ‘a Christian’, i.e. a member of a particular social and religious group rather than the more literal ‘one who believes in God/Christ’. Put differently, the compositional meaning provided by the morphosemantic structure is replaced by a more specialized meaning that is not directly predictable from the PA’s morphology, including its tense marking: having an identity based on a particular trait (here, Christian faith) obliterates the contemporaneousness that the internal morphology suggests. Moreover, once the atemporal usage based on group-identity is well established, we can expect additional semantic shifts that can only be motivated by their relationship to a semantically true noun, not by any relationship to the verb. And indeed, we find examples such as (6), where the meaning of věřící is generalized to refer to church members, both alive and dead. Evidently, group membership is the salient feature here, no just the individually assigned property of believing in something, since the act of believing presupposes a conscious mind, while being counted as a member of a class of individuals need not. Notice that the PA in (6) does not (indeed, cannot) have any complement. {žádajíc...spasení duše své a svejch starších a předchůdcích} ...i <jiných všech mrtvých also other.gen.pl.m all.gen.pl.m dead.gen.pl.m věřících > {voltář ke cti svaté Dorotě...založil a nadal} believe.pa.gen.pl ‘{desiring…the salvation of his own soul and [the souls] of his elders and predecessors…} and also of all other deceased members of his congregation,{he founded and endowed an altar to honor Saint Dorothea}’ [ArchČ 9, 243; 1400; admin. record of dedicating a new altar] (6)

However, complete absence of the complement is not always a measure of a nonverbal interpretation of the PA either, as demonstrated by the example in (7). This is one of the functionally ambiguous cases, where the PA can be either interpreted as a fully compositional usage that is externally nominal but preserves all the internally marked verbal categories, including relative tense (as a temporal or conditional circumstantial clause), or it can be interpreted as referring to a member of a particular class of people: (7) {Apolonius…nauči jej nasledovati přiebytka věčného a trpělivě hledati božieho milosrdie... řka jemu} že sě muož všecko dokonati < věříciemu > that rf can.pres.3sg everything.nom.sg.n complete.inf believe.pa.dat.sg



Representing contextual factors in language change 

‘{Apolonius…taught him to aim for the eternal dwelling and patiently seek God’s mercy…, telling him} that everything is possible ’ [OtcB 19a; late 14th cent.; legend] The example is taken from a popular historical narrative reporting a sequence of events and their consequences for the protagonist. This would support the predicative, event-reporting interpretation, still expressing relative tense, along the lines of ‘when/if you have faith, everything is possible’, since such usage is generally well-documented in event-reporting narratives (chronicles, exempla, legends, personal stories, etc.). At the same time, the religious framing of this particular narrative – the text is a compendium of the lives of Saints – equally invites the referential interpretation of ‘believers [in God]’, i.e. people who are defined by this property independently of the events in the immediate context of the story. The potential for ambiguity with respect to the relative strength of the verbal character persists through the first part of the 15th century, and not just as a special feature of this particular PA but as a general feature of OCz PAs (Fried 2003). What is thus perhaps more important than internal syntax in the shift toward a full noun status is the collocational pattern associated with this PA usage. It is conspicuous that the PA frequently co-occurs with a universal quantifier, such as všichni ‘all’, shown in (5a-b, 6) or každý ‘every’, regardless of complementation. Semantically, this reinforces the generic, class-membership interpretation of věřící, at the expense of a temporally restricted reading that applies only at a given moment. Put differently, in these quantified phrases, ‘having Christian faith’ can easily be understood not as a temporally grounded eventuality ascribed to an individual entity (the non-restrictive relative clause interpretation) but as a classification that is independent of the event expressed by the main predicate. The referential potential is simultaneously strengthened by the external syntax, by giving it the status of a head in a noun phrase. We may futher note that the quantifier slot gradually expands to other types of modifiers, including possessives and adjectives, such as we see in (4a), where the verbal government ‘believe(r) in him(self)’ is completely replaced by a nominal pattern ‘his believer’ and the Deity referent is now indicated by the NP-internal possessive modifier. When we organize the attested patterning chronologically, as in Diagram 2, it becomes evident that the syntagmatic environment, i.e. the collocational preference, plays a significant role in fixing the actor noun status.4 The external syntax (noun slots and nominal government) together with the external and internal semantics (quantification, predictable valence content) were only gradually, though steadily, followed by internal syntax (complete valence suppression). The arrows indicate persistence into Modern Czech and the dotted lines reflect relatively lower or sporadic incidence of a given feature.



Mirjam Fried 1350

1380

1410

1450

1500

... ModCz

INTERNAL: complementation (v ACC) fixed reference for Deity complement EXTERNAL: modifiers quantifiers other believer in God meaning Christian

church member

Diagram 2.╇ Chronology of the noun věřící in the faith sense

To summarize, this PA underwent a shift from a relatively transparent pattern resembling a headless relative clause (predominantly non-restrictive) with a compositional semantic structure, to a full actor noun without any temporal grounding: from ‘(the) one who believes in God/Christ’ to ‘Christian/man of the one-andonly-presupposed-as-culturally-acceptable religion’ and eventually to ‘church member’. This shift is due to several types of contextual factors. A broad communicative context is relevant and perhaps decisive in interpreting the older examples: as we see in (7), it is possible to interpret the form in an atemporal sense as well as in temporally grounded predicative function, regardless of the externally non-verbal syntax. In other words, nothing in that narrative (and other cases like these) forces a particular inference and instead, it is left to the reader/hearer to work out a plausible understanding. Gradually, though, the usage clearly consolidates the referential potential of the PA suffix, highlighting the actor of the event denoted by the root, rather than the event itself; the latter would be expected in a true participle. And this is facilitated by, or at least correlates with, two contextual factors. One is the low pragmatic and semantic salience of the referent expressed by the non-subject complement, thus futher ‘demoting’ the verbal properties of the form. The other is the syntagmatic context in which this PA is used: in syntactically nominal slots which increasingly take on overt features of a clear noun phrase structure and semantics. In Section 4, we shall return to the question of how these different contextual factors entered the conventional understanding of the PA in its fixed categorial status.



Representing contextual factors in language change 

3.2

Věřící in the sense of credit

The PA used in the credit sense presents a distinctly different picture, although the differences are consistent with the general tendencies displayed in the faith sense as well as in all other cases of PAs that turned into actor nouns. To begin with, the corpus does not contain any examples of credit-based PAs until the early 1400s, even though the corresponding verb was commonly used in this meaning as early as the faith sense (mid 1300s). Moreover, the verb occurs in all kinds of genres (poetry, popular stories, legal documents, religious tracts, legends, etc.), while its PA is attested exclusively in administrative and legal texts. This distribution suggests that věřící in the credit sense, unlike its faith counterpart, developed as a technical term and was not part of the common vocabulary. The oldest example in the corpus is shown in (8) and can be taken as fully representative of this usage. (8) {Pakli bychom toho neučinili… mámy … odtud nikam nevyjéžděti tak dlúho, až} bychom < našim věřícím > svrchupsané cond.1pl our.dat.pl believe.pa.dat.pl above.written.acc.sg.f penieze úplně zaplatili money.acc.pl entirely pay.ppl.pl ‘{If we didn’t do that… we have no permission … to leave this town} until we’ve paid the above mentioned money in full’ [ArchČ 6,459; 1410; private record of a financial obligation]

Two syntactic features stand out prominently. One concerns the internal syntax of the participial stem: without exception, this PA always appears bare, that is without showing any verbal government for expressing the non-subject arguments (potentially two – the Debtor and the Valued Entity) contained in the root’s valence. Related to this fact is the other striking feature, namely, the syntagmatic patterning. With the exception of one token (out of 20 in the corpus), to be commented on shortly, this PA is always accompanied by a possessive modifier, which is to say, it always occurs in a noun phrase. And as is also illustrated in (8), the possessive invariably refers to the first person of either number, which means that the referent of the PA věřící is always the speaker’s negotiating partner. Thus in pragmatic terms, the possessive modifier actually instantiates one of the valence participants (Debtor), but using purely nominal, rather than verbal syntax, and its referent is highly restricted, as it can only be the speaker; this constraint certainly does not follow from the meaning of the root or from any pragmatic condition associated with its use.



Mirjam Fried

All these features strongly favor nominal interpretation, in which the PA denotes a person that plays a particular role in a commercial transaction; in (8), it is the role of the Creditor, the party that expects a mutually agreed-upon payment from the Debtor-speaker. Such an interpretation is further corroborated by later texts, in which the PA věřící is often used (especially in the late 1400s) interchangeably with the morphologically true actor noun, věřitel ‘creditor’ (-tel corresponds to the English actor noun suffix –er). An illustration is in (9); the noun věřitel and its corresponding translation are underlined. (9) {a tak vždy máme vydávati na každý rok věřitelóm našim…, dokudž bychom svrchupsaných zlatých jistinných i s úrokem} < našim věřícím > svrchupsaným nedali our.dat.pl believe.pa.dat.pl above.written.dat.pl neg.give.ppl.pl {a nezaplatili úplně} ‘{and so we must pay each year to our creditors...until we might give and pay in its entirety the above mentioned principal and interest} ’

We could thus summarize that in contrast to the faith PA, this usage was from the very beginning dedicated to its potential as an actor noun, which is unambiguously signaled by its syntax, both external and internal: no verbal government, consistently a full NP structure, and consequently we can also infer a weak sense of temporal grounding. On the one hand, it is true that the attested data are mostly records of specific eventualities, in which the roles within a debt-paying transaction are assigned within a particular, sometimes explicitly established time frame. In that sense, the assignation of being a věřící need not be taken as indicating an atemporal interpretation that defines a type of participant based on general class membership. At the same time, the semantic overlap with the true, atemporal actor noun věřitel was clearly salient enough in the minds of the speakers if the forms were so easily used as synonyms. This relationship, together with the syntactic patterning, invites a shift toward an atemporal understanding, with the PA then denoting a conventional role in a particular type of business relations. This conclusion is consistent with the one exception to the preferred syntactic pattern just described, certainly so on pragmatic grounds. The one instance of the PA without a modifier is not a record of an actual contract in which the speaker enumerates his or her obligation toward a concrete partner. Instead, the example in (10), chronologically relatively late, comes from a manual for composing contracts and other official documents, where there is no need to be explicit about the remaining participants; it only offers a general advice on what conventionally accepted options a creditor has toward his or her negotiating partner(s). No



Representing contextual factors in language change 

temporally restricted reading is even possible here, the PA can only be understood as a classificatory label. (10)

na vuoli < věřícieho > jest on will.loc.sg.f believe.pa.gen.sg aux.3sg {diel od jednoho spolurukojmí vzieti} ‘it is [left] to the discretion of {to take a down payment from one of the guarantors} [ProkArs 168; late 1400s; manual for legal writing]

Thus in the credit PA, the only verbal feature left appears to be the PA’s active orientation: it is an actor noun, after all. This conclusion, however, requires some elaboration; it still remains to be determined what exactly this PA meant, particularly with respect to the general question that frames this study: how can we best capture the PA’s conventional meaning and which, if any, interpretations have to be left to broader thematic context, which is to say, to speakers’ understanding of the situation at hand vis-à-vis the morphosemantic structure of the form. The reason for posing this question are examples such as in (11) below, which present an interesting twist on how the valence of the root may be manipulated in different situational contexts. So far, all the uses have involved the most transparent distribution of roles as it follows from the semantic structure of the verb root organized around the prototype. Recall the relationship between the participants across different senses and particularly the correspondence between belief and credit, the latter also with the overlapping frame of commercial transaction (ct), repeated in Diagram 3. For reasons that will become clear shortly, the ct frame is presented in one of its particular subframes, which can be labeled as selling. The distribution depicted in Diagram 3 is reflected in (8–10). The ‘believer’ in those examples is a seller or money lender, i.e. the person who believes that the buyer or borrower will pay his/her debt. The contexts all involve paying a debt and frame [BELIEF ] #1 Believer #2 Recip. of Trust #3 Content

frame [CREDIT] #1 Creditor #2 Debtor #3 Valued Entity

frame [SELLINGCT] #1 Seller #2 Buyer #3 Valued Entity frame [TRANSFER] #1 Giver #2 Recipient #3 Gift

Diagram 3.╇ Věřící as a trusting seller/lender



Mirjam Fried

the referent of věřící is cast in the Creditor role. But the examples in (11) are different; the parentheses in (11a) reflect the actual orthography: (11) a.

b.

{pakli bychom...zbožie... v€zemské dsky...nemohli vložiti..., tehdy} slibujeme... < našim věřícím > (straně promise.pres.1pl our believe.pa.dat.pl (party.dat.sg kupující) ... napředpsanú jistinu buy.pa.dat.sg) above.written.acc.sg.f capital.acc.sg.f {peněz a k€tomu třetinu výše} ‘{should we fail to enter the [transfer of the] estate into the land register, we hereby} promise [to pay] (the buying party) the abovementioned capital {and one third of its value on top of that} [ProkArs 156b; late 1400s; legal writing manual] {Pakli bychom toho všeho neučinili, tehdá } < náš věřící > moci bude nám láti our believe.pa.nom.sg can.inf be.fut.3sg 1pl.dat rail.inf ‘{Should we fail to do all that, then} our trusting [buyer] will be allowed to rail at us’ [ArchČ 15, 305; 1455; record of the sale of a village]

In (11a) the role of the PA’s referent is explicitly identified as ‘the buyer’, while in (11b) that same role follows from the context: the speaker is the one selling the village and he is laying down the conditions of a successful sale, which includes his writing the deed into the land records. Evidently, the ‘believer’ need not always be the one who is giving anything on credit (lending money or selling property), but can be also the one who is buying on the belief that the goods (Valued Entity) are truly available or have the promised value. In (11), the distribution of roles thus reflects a shift in perspective from which the transaction is viewed: while in (8–10) the usage takes the perspective of the seller (hence the overlap with the selling subframe of the ct scene), in (11) the buyer’s perspective is taken, which means a different conceptualization of the CT background frame, namely, buying. This switch, in turn, yields a slightly different interpretation of the roles involved in the act of believing: instead of the correspondences in Diagram 3, we get the configuration in Diagram 4. To be sure, the context in (11) is still a commercial transaction and the issue at hand is to ensure that the buyer and the seller do not cheat each other, but the status of being the one who ‘believes’ is attached to the buyer rather than seller. The meaning of věřící in these cases is thus more accurately paraphrased as ‘the one who believes [the integrity and honesty of] the seller’, not as ‘the one who sells X to the buyer on credit’, as would be expected in an actor noun derived from the credit sense of the verb věřiti. The latter interpretation is provided explicitly in (11a), also through the use of the parenthetical clarification strana kupující ‘the buying party’.



Representing contextual factors in language change 

frame [BELIEF ] #1 Believer #2 Recip. of Trust #3 Content

frame [CREDIT] #1 Beneficiary #2 Guarantor #3 Valued Entity

frame [BUYINGCT] #1 Buyer #2 Seller #3 Valued Entity frame [TRANSFER] #1 Recipient #2 Giver #3 Gift

Diagram 4.╇ Věřící as the a trusting buyer

While examples such as (11) are very infrequent and seem to appear only in the later chronological layers of my corpus, they cannot be simply dismissed as errors or an individual peculiarity of a single writer; they come from different sources and it is also significant that the buyer’s interpretation is specifically noted in some of them. Instead, the ‘fuzziness’ only underscores the fluid interaction between different frames that are evoked by a single word and whose activation may not be a priori restricted to a specific type of context. It seems that the exact interpretation of the PA in the credit sense was shaped by two different thematic contexts: one involving pure money transfer (as in lending, dowry payments, or any transaction treated simply as a debt of one party toward another), illustrated in (8–10), and the other involving the purchase of a property. At the same time, the fact that an author felt the need to explicate what he means by věřící (11a) in a context in which the expected reading could be ‘creditor’ suggests that particular frames are preferentially evoked in particular contexts (or, that there are certain conventionally expected associations between a type of cultural context and a particular semantic frame) and while such associations can be violated, the unexpected framing may have to be signaled explicitly. It is also important to add that I have not found any attestation of other forms of the verb věřiti (finite or non-finite) where its subject would be interpreted as a ‘beneficiary/buyer’; it seems to be only the PA form that developed this additional interpretation. A chronological summary is in Diagram 5, where we can see that syntactically, both internally and externally, this PA usage was purely and consistently nominal from the start; note also that it did not survive into ModCz (nor did this verb sense as a whole). To conclude, the conventionally expected meaning of věřící in the credit sense seems to have been ‘one believing in his commercial partner’s honesty’. The prototype structure associated with the meaning of the verb root motivates the transparent interpretation of věřící as ‘trusting seller/lender‘, i.e. ‘creditor’, which is indeed found in all kinds of contexts. However, in the context of property buying, as a specific type of transaction, we find, at least marginally, a switch to the



Mirjam Fried 1350

1380

1410

1450

1500

... MCz

INTERNAL: no complementation EXTERNAL: modifiers: 1st pers. possess. meaning

creditor trusting buyer

Diagram 5.╇ Chronology of the noun věřící in the credit sense

perspective of the buyer, yielding the interpretation of věřící as ‘trusting buyer‘, in a shift away from the prototypical distribution of the roles maintained by the rest of the verbal paradigm. 4. Context and grammar in a network model of language change The discussion in the preceding sections makes several things evident. First off, it would be an oversimplification and misrepresentation of historical reality if we limited ourselves to noting that the form věřící turned (among other things) into actor nouns. At a minimum, it did not turn into such nouns across the full semantic spectrum provided by the meaning of the root. Furthermore, the factors that were instrumental in developing the actor noun conventionalization were not quite uniform across the relevant semantic subdomains (faith vs. credit). And finally, a sweeping generalization that does not go beyond the level of lexical categories (a V → N change, or even if seen as a Participle → N change) misses any kind of insight into the cognitive and communicative mechanisms that are involved in grammatical reorganization and that can advance our understanding of the incremental nature of such changes in general, not just in this particular case. In order to understand the gradual shift in the conventional interpretations of the form in question, we must first consider the PA’s place in the semantic network provided by the root from which the form is derived. Without this connection, we could at best provide a disjointed list of uses without any apparent motivation. Yet, we can see that the different senses and uses of věřící form coherent sub-networks within the semantic space associated with the verb. These sub-networks, in turn, represent different clusters of essentially the same syntactic, functional, and pragmatic properties and, crucially, the clustering is not necessarily predictable from anything in the morphosemantic structure per se. In fact, the overall shift can be best described as a gradual erosion of compositional structure. This means that the



Representing contextual factors in language change 

PA goes from being a transparent sum of its parts (with all the attendant flexibility in use and functional status in larger context, due to the indeterminate nature of the external c/n/g suffix) to a functionally fixed category with a meaning that is no longer fully derivable from the form’s morphological make-up. The task that remains, then, is to capture this shift in a way that properly includes the systematic participation of the relevant contextual factors and that also allows us to make generalizations about the role different types of context played in the process. Diagram 7 shows a frame-semantic/constructional network in which all the semantic, syntactic, and contextual properties can be organized. First, however, a few notes on the formalism; a slightly simplified notational practice is shown in the generic constructional representation in Diagram 6 (the interested reader is invited to consult Fried & Östman 2004 for all the technical details, none of which are crucial to the concerns of this paper). The nested boxes reflect constituent structure, but also allow us to make a principled and systematic distinction between two domains of representation: the ‘external’ properties of a construction in the outside box (e.g. the c/n/g suffix that forms the PA) and its ‘internal’, constituent-level properties (e.g. the participial stem) in the inside box. This distinction is essential for capturing the fact that a complex expression (morphological or syntactic) as a whole may have its own idiosyncratic features that do not follow from the internal composition. Individual features are represented as attribute-value pairs, in Diagram 6 labeled only as attr followed by a variable (x, y, etc.). Notice that the internal and external levels may share certain attributes (co-indexed) and/or values, but need not share all of them; hence the non-compositionality effects. Turning now to the representations of the PAs, we start by noting that the frames presented in Diagram 1 above correspond to those boxes in Diagram 7 below that contain the attribute frame in them, followed by the name of the appropriate semantic frame (belief, credit, etc.). Notice that most of these frames are now inside other boxes, indicating the fact that the frame information, along with various other features, is contributed by the PA’s stem, which, in turn, is a constituent of another form, namely, the PA itself (labeled as lform); this arrangement reflects the attri attrj attrk

x yy z

attri attrj

x y

Diagram 6.╇ CxG formalism in the abstract



Mirjam Fried

hierarchical morphological structure introduced in (1), although the representation in Diagram 7 ignores other structural details that are not crucial to the present purposes. In an example of the external/internal distinction, the features listed at the top of the outer box of the faith PA represent the constructional properties of this use of věřící that do not simply follow from adding up the properties of its constituent morphemes. (A bit more will have to be said about the representation of the credit PA.) For easier orientation, all the features that are not predictable from the morphosemantic structure of the PA are in italicized boldface. The attribute-value formalism can be quite elaborate and detailed, but I prefer to simplify the notation somewhat, in the interest of general accessibility. For our purposes, we only need to register several such attributes; some are self-explanatory, such as cat(egory), sem(antics), (discourse) role, modality, val(ence), while others may require a brief commentary: dp is an abbreviation for ‘discourse pattern’, which I use for marking textual/stylistic constraints (to be discussed in more detail shortly), lform indicates a specific, physical word-form, while lxm stands for ‘lexeme’; FE stands for ‘frame element’. The picture summarizes our observations about the faith and credit PAs, also in relation to the general background frame belief, to which they still belong. The path of the faith usage is relatively straightforward. This PA maintained the more or less transparent structure of an inflectional form and only very slowly shifted toward becoming a full-fledged actor noun. In particular, the valence of the root, represented by the val statement inside the stem box, remained a property of the PA as well: the val statement at the external level specifies that the non-subject dependent can be expressed by the PA, albeit only optionally, as indicated by the parentheses. The rest of the external properties, however, are idiosyncratic to the PA: its external syntax is unambiguously that of a noun (in spite of the adjectival morphology of the c/n/g suffix, not explicitly included in Diagram 7) and its meaning is ‘Christian’, which follows from the narrowing of the referential potential of the FE Deity (recall Diagram 1) to Christ. As the chronological summary in Diagram 2 above shows, it took at least two generations of speakers before the internal structure was sufficiently eroded and the atemporal interpretation became fully established and the only one available. It was only when this shift was fully in place that the ‘church member’ reading could arise, as an extension of the PA, without any direct relationship to the verbal origin. Notice the complete absence of any valence or, for that matter, any other reference to the participial stem in its representation; there is no motivation for positing any derivational link between the two.



Representing contextual factors in language change 

cat n sem ['Christian']

lform věřící

val {(#2 [Acc(v)])} lform věřící cat n sem ['church member']

Frame BELIEF (content of thought) FEs ...

cat vNT-part Frame

lxm věř--

FAITH FEs: #1 Believer #2 Christ

val {#1 [Nom], #2 [Acc(v)]} COMMERCIAL TRANSACTION

context: property buying cat dp

context: money lending

n admin./legal genre binding contract

inherit Modification

modality obligation #2 cat role

poss. Speaker

cat n sem ['creditor'] cat vNT-part

lform věřící lxm věř-

Frame CREDIT FEs: #1 Creditor #2 Debtor #3 Valued-Entity val {#1 [Nom], #2 [Dat], #3 [Acc]}

Diagram 7.╇ Frame-semantic/constructional network of OCz věřící as a noun

As we already mentioned, the crucial factor in the shift appears to be the syntagmatic context, i.e. the collocational pattern [Quant + PA] in which this form routinely occurred; the collocational preference itself, though, did not become conventionalized as such, it only facilitated the fixing of the PA’s categorial status. On the other hand, the communicative context kept reinforcing certain interpretive clues about the meaning of věřící that gradually did become part of the PA’s conventional meaning. Specific contextual features, namely, promotion of Christian faith, strongly favored a particular interpretation of the PA’s referent, which resulted both in the semantic specialization of the PA (Christian faith) as compared



Mirjam Fried

to the verb itself (religious faith) and in the PA’s functional specialization (identification of group membership based on certain properties), yielding a semantic and syntactic (though not morphological) noun. In contrast, the case of the credit PA is more involved, its departure from its inflectional origin much more complete and seemingly abrupt. The effects of the syntagmatic context were directly and visibly incorporated in the conventional usage from the start. For one thing, the verbal potential was completely suppressed, even beyond the issue of valence expression; this ‘fading away’ of the internal structure is indicated by the gray color in the diagram: the participial properties of the stem, the valence requirement of the root, and even the conventional configuration of the frame participants associated with the verbal usage clearly lost their significance. At the same time, the PA typically occurs in a modification structure [Poss + PA], i.e., as the head of a phrasal construction that can be generalized as [Modifier + N]. This fact is captured by enclosing the PA together with its possessive modifier in a larger box, representing them as a single, conventional phrasal unit. The features listed in the left-daughter box express the constraints associated with the modifier and the right-daughter box represents the PA, as the head of the whole Modification construction. This [Poss + PA] unit also carries all the properties of other modification patterns (such as case/number/gender agreement between the constituents, the distributional properties of the whole phrase in sentences, etc.) and this relationship needs to be captured as well, through the inherit statement at the external level. However, we know that other types of contextual factors also played a role and that those must be reflected in the representation as well. First of all, the PA was restricted to a specific textual/stylistic context, in Diagram 7 indicated by the dp (Östman’s 2005 “discourse pattern”) attribute: this PA only appears in a formulaic administrative usage and was most likely coined specifically as a ‘technical’ term. We can think of the notion of discourse patterns as conventional schematizations of the speakers’ knowledge about distinct discourse situations, which may impose specific constraints on grammatical patterning. In the case of our [Poss + PA] collocation, the relevant discourse pattern can be conceptualized as a type of legal and administrative discourse, which came in various subtypes (e.g. summonses, sentencing, contracts, testaments, etc.). I label these provisionally as genres but the meaning and use of these categories clearly need additional study. The attributes dp and genre thus serve to invoke a general frame of understanding of a discourse situation in which this PA naturally fits. Notice also that no such reference is necessary in representing the faith PA since its textual distribution remains unconstrained. Second, the communicative context in which this PA was used seems to be restricted to a particular modality: with the one motivated exception shown in (11), the context always involves an expression of obligation with respect to



Representing contextual factors in language change 

maintaining trust between commercial partners. The modality is often expressed directly by a verb of obligation, such as mít ‘ought to’ in (8–9), but can be also more subtle, as in (11). Again, this restriction is not a property of the verb itself, nor the participial stem but, rather, correlates with the kind of discourse it occurs in (a binding contract).5 And finally, full understanding of the conventional meaning of this PA must include reference to discourse roles: one of the participants provided by the verb root is necessarily the speaker. This is consistent with the performative flavor of the situation in which this PA was used: declaring contractual obligation between the speaker and his/her contractual partner. All of these context-sensitive features became part of the conventional meaning of the credit PA, turning it into an actor noun distinct from what the verbal paradigm would predict, including the fact that it prototypically occurs only as the head of a modification construction. The representation also leaves room for the development of the emerging polysemy documented in (11). The possibility of reconfiguring the participant roles indicates that the credit PA’s link to the verb’s semantic structure had been sufficiently loose by the late 1400s (represented by the gray color for the indices #1–2 in Diagram 7), thereby inviting an alternative perspective, which results in a different hierarchical organization of the roles. At the same time, it is interesting to note that this shift is found always in the same semantic context (property buying). This may mean nothing, of course, since we are dealing with an incomplete record, but it may also be indicative of the tendency toward isomorphism as a powerful motivating force in balancing the tension between the need to express new meanings (polysemy) and maintaining relative clarity through one-to-one mappings. However, we do not have strong evidence for concluding that this marginal, though motivated, expansion ever became a conventionally established distinction. The attestations suggest the potential for such a development and our representation captures where and why this kind of ‘stretching’ of the conventional pattern can occur, but the sporadic record does not justify positing a distinct construction or a frame. To summarize, once we attempt to formulate explicit generalizations about the relevant empirical observations, at least two broad conclusions become apparent. One concerns the level of descriptive detail. Although on the surface the PA development may seem primarily an issue of morphology (generating a hybrid morphological form within an inflectional paradigm), in reality it involves a number of additional dimensions that have to be integrated in the analysis if we are to arrive at an accurate picture of these forms and the changes in their use and categorial status over time. The frame- and construction-based approach provides useful tools for incorporating all the necessary detail in a principled and systematic way. Second, the network in Diagram 7 embodies the gradient nature of the distinction between lexicon and grammar. Frames are a tool for structuring lexical



Mirjam Fried

meaning (in the sense of speakers’ conventional understanding), but an accurate representation of the meaning and usage of a lexeme or its word-forms may involve a lot more than just an inventory of frames or, for that matter, a list of single words associated with that frame. As we see with the credit PA, certain generalizations about the properties of a single word (form) can easily go beyond the domain of a word and instead take the form of a phrasal construction. This particular construction has fully transparent syntactic structure and yet, in order to articulate the shifts in the speakers’ understanding of this PA, we cannot keep the external syntax separate from the semantic and pragmatic constraints, all of which incorporate various kinds of contextual clues. That is, the phrasal construction constitutes one segment of the lexico-grammatical network associated with the noun věřící, thus blurring the imaginary boundary between grammar and lexicon. 5. Conclusions The main goal of this study was to examine the ways in which different kinds of contextual information may play a role in the reorganization of specific formmeaning associations, and how this reorganization can be represented by the tools of a synchronic model of language that tries to give a cognitively plausible representation of speakers’ linguistic knowledge. I approached the problem from a diachronic perspective, studying the interaction between context and grammar as it is reflected in the incremental accumulation of the features that collectively amount to an observable shift in a form’s usage. Overall, the change under investigation can be best described as a gradual conventionalization of a pattern of understanding, in which lexical meaning, syntactic function, and communicative function form an integrated whole. The choice of the OCz PA as illustrative material is motivated by the fact that the PA is a categorially hybrid form that developed in several different directions (functionally and categorially, cf. also Fried 2003, 2008b, and in press), each direction correlating with a distinct cluster of various contextual factors. The PA’s structure and the patterns of its usage speak to many issues – semantic, syntactic, morphological – and pose challenges both for accurate analysis and, perhaps even more so, for finding an adequate way of representing the generalizations that can be made about them. The way different contextual factors entered the conventional understanding of the PA věřící as an actor noun can be summarized as follows. (i) To the extent that semantic/pragmatic context (i.e. the cultural grounding of speakers’ understanding) gets incorporated into the lexico-grammatical make-up of the PA, it affects the frame structure, i.e. the lexical meaning (cf. the narrowing of the faith



Representing contextual factors in language change 

sense or the emerging split in the credit sense). (ii) The syntagmatic context indicates structural and functional preferences of the item in question and if incorporated into the conventionalization of an expression, it may give rise either to new grammatical constructions (e.g. the PA itself, as a morphological construction whose internal structure became more opaque) or to a fixed phrasal expression based on an independently existing grammatical construction (cf. the credit PA’s modificational pattern). (iii) Finally, the fixing of a form’s usage may arise in particular textual settings, and such constraints again become part of the constructional specifications (cf. the stylistic restrictions on the credit pa). The syntagmatic context has turned out to be particularly relevant in shaping the PA’s development, whether the patterning became part of the conventional expression (credit) or just facilitated the gradual fixing of the categorial status (faith). This observation underscores two things. One is the mutually reinforcing effect of form and meaning/function: neither dimension can be strictly predicted from the other, but each influences the speaker’s understanding and interpretation. Related to this is the importance of tracking collocational patterns and dealing with them in a systematic manner, so that we can capture what is stable and predictable about them without having to deny their idiosyncratic properties. The collocations that played a role in the development of the PAs all share regular, productive features with various general syntagmatic patterns (such as the grammatical properties of the abstract modification construction), but they are much less transparent with respect to other types of context: textual, pragmatic, and semantic. These features are unpredictable and must be ‘known by native speakers outright’, i.e., must be specified explicitly in the representation, since they cannot follow automatically from the knowledge of other linguistic patterns (cf. Goldberg’s 2002 notion of “surface generalizations”, which applies here as well). In order to establish shifts within the lexico-grammatical network associated with the PA, we have to first study the details of how the network’s (potential) members are used in context and what are the factors that contribute to the sharpening of the PA’s inherently available but vaguely delimited and initially contextdependent syntactic function. One has to consider the way semantic space is divided between different uses in different contexts, even within the same potential category (in our case, nouns). We have seen that different senses within a polysemy network shape the superficially identical form quite differently. Consequently, the process of incorporating contextual information into grammatical structure requires a network-based model of language that allows multidimensional representations of the interdependencies between lexical meaning, morphosyntactic structure, and pragmatic constraints both within and across individual constructions. With respect to the representational potential of CxG vis-à-vis ‘context’, it is clear that the model is equipped, conceptually, to accommodate the pragmatic and

 Mirjam Fried

general socio-cultural dimension of grammatical patterns, when needed. What remains to be worked out more carefully, though, is a consistent set of relevant categories and the notational convention for the contextual dimension. The present analysis has taken a rather brute-force approach by simply invoking attributes that were necessary for addressing the data at hand (dp, genre, modality, etc.). It is possible and, perhaps, desirable to treat the contextual issues as a distinctly recognizable layer within constructional representations, along the lines of several recent proposals (Östman 2000, 2005, Fried & Östman 2005, Bergs 2005b) which call for organizing the contextual features and categories into coherent representational domains, akin to the notion of frame as used and applied in Frame Semantics. Finally, as suggested in the introduction, the analysis has consequences for addressing the grammaticalization/lexicalization dilemma and the PAs provide material for further exploration of this actively debated question. One possibility in addressing the PA’s conventionalization patterns is to emphasize the observation that the development may be lexically restricted (different verbs or verb senses predispose their PAs toward different contextual and structural preferences), in which case we could simply approach them one at a time, as independent cases of lexicalization in the sense of turning an inflectional word form into an independent lexeme. At the same time, all the tokens share the same morphosemantic structure and differ only in selecting which particular subsets of those initially provided features will become highlighted, or obscured, in their use. It thus follows, that the development bears signs of grammaticalization as well, through the generalization of certain semantic features and through the gradual enlisting of the new usage (motivated by various contextual factors) for different syntactic functions, in the sense of Traugott’s metonymy-based definition of grammaticalization (Traugott 1992; Traugott & Dasher 2002). The constructional approach seems to offer a way of reconciling the two sides of the problem, minimally by allowing us to frame the distinction and its manifestations as a continuum, rather than a strict opposition. Notes 1. The genre selection includes historical, biblical, administrative, expository, and didactic texts, legal documents, poetry (both spiritual and secular), popular entertainment, correspondence, drama, instruction manuals, etc. The chronology covers the full OCz period, from the first attestations well before 1300 until the early 1500s. The corpus contains over 1200 tokens of the PA form, which represents more than 240 distinct verb roots. The form věřící is commonly attested throughout the OCz period and its relatively high frequency across all kinds of texts can be taken as evidence that the form was well entrenched, including in the genres that are known to reflect spoken language fairly closely.



Representing contextual factors in language change  2. The OCz texts are identified by the abbreviations established in the Old Czech Dictionary (Staročeský slovník 1968). 3. Abbreviations in the glosses: NOM ‘nominative’, GEN ‘genitive’, ACC ‘accusative’, DAT ‘dative’, INS ‘instrumental’, M/F/N ‘masculine/feminine/neuter’, SG/PL ‘singular/plural’, PA ‘participial adjective’, PRES ‘present tense’, PST ‘past’, FUT ‘future’, NT ‘pres. active participle, PPL ‘past participle’, PASS ‘passive participle’, INF ‘infinitive’, RF ‘reflexive particle’, NEG ‘negative’, AUX ‘auxiliary’, COND ‘conditional’. 4. While many of the manuscripts in the corpus can be dated quite precisely, many others can only be placed within an estimated time period (a decade or more). For the present purposes, it is sufficient to follow the practice established by the Old Czech Dictionary (Staročeský slovník 1968), which recognizes six chronological layers, each identified by reference to its upper boundary ([up until] 1300, [up until] 1350, and so on). 5. The semantically unrestricted distribution of the verb correlates with the fact that it was used in all kinds of genres, not just a particular type of administrative texts. Recall the example in (2), which illustrates the contextually neutral nature of the verb.

References Atkins, B..T. S. (1994). Analyzing the verbs of seeing: a frame semantics approach to corpus lexicography. Berkeley Linguistic Society, 20, 42–56. Atkins, Sue, Charles J. Fillmore, & Christopher R. Johnson (2003). ‘Lexicographic relevance: selecting information from corpus evidence.’ International Journal of Lexicography, 16, 251–280. Bergs, Alexander (2005a). Social networks and historical sociolinguistics. Berlin: Mouton de Gruyter. Bergs, Alexander (2005b). The future in context: a constructional approach. Paper presented in The 9th International Pragmatics Conference, Riva del Garda, Italy. Bhatia, Vijay K. (1993). Analyzing genre: language use in professional settings. London: Longman. Biber, Douglas (1994). Dimensions of register variation. Cambridge: Cambridge University Press. Biber, Douglas & Edward Finegan (1989). Drift and the evolution of English style: a history of three genres. Language, 65, 487–517. Bisang, Walter (1998). Grammaticalization and language contact, constructions, and positions. In A. Giacalone Ramat & P. Hopper (Eds.), The limits of grammaticalization (13–58). Amsterdam: John Benjamins. Bybee, Joan, Rever Perkins & William Pagliuca (1994). The evolution of grammar: Tense, aspect and modality in the languages of the world. Chicago & London: University of Chicago Press. Croft, William (1991). Syntactic Categories and Grammatical relations. Chicago: The University of Chicago Press. Croft, William (2001). Radical Construction Grammar: Syntactic theory in typological perspective. Oxford: Oxford University Press. Cuyckens, Hubert, René Dirven & John R. Taylor (Eds.) (2003). Cognitive approaches to lexical semantics. Berlin: Mouton de Gruyter. Fillmore, Charles J. (1974/1981). Pragmatics and the description of discourse. In Ch. J. Fillmore, G. Lakoff, & R. Lakoff (Eds.), Berkeley studies in syntax and semantics I (Chapter V). Dept.



Mirjam Fried of Linguistics, UC Berkeley. [(1981) In P. Cole (Ed.) Radical pragmatics (143–166). New York: Academic Press.] Fillmore, Charles J. (1982). Frame Semantics. In The Linguistic Society of Korea (Eds.), Linguistics in the Morning Calm (111–137). Seoul: Hanshin Publishing Company. Fillmore, Charles J. (1984). Frames and the semantics of understanding. Quaderni di Semantica, 6, 222–254. Fillmore, Charles J. (1986). U-semantics, second round. Quaderni di Semantica, 8, 49–58. Fillmore, Charles J. (1988). The mechanisms of ‘Construction Grammar’. Berkeley Linguistic Society, 14, 35–55. Fillmore, Charles J. (1989). Grammatical Construction theory and the familiar dichotomies. In R. Dietrich & C. F. Graumann (Eds.), Language processing in social context (17–38). Amsterdam: North-Holland/Elsevier. Fillmore, Charles J. & B.T.S. Atkins (1992). Toward a frame-based lexicon: The semantics of RISK and its neighbors. In A. Lehrer & E. Feder Kittay (Eds.), Frames, fields, and contrasts: New essays in semantic and lexical organization (75–102). Rahway, NJ: Lawrence Erlbaum Associates. Fillmore, Charles J., Paul Kay, & Mary Catherine O’Connor (1988). Regularity and idiomaticity in grammatical constructions. Language, 64, 501–538. Fillmore, Charles J., Miriam R. L. Petruck, Josef Ruppenhofer, & Abby Wright (2003). Frame Net in action: The case of attaching. International Journal of Lexicography, 16, 297–332. Fried, Mirjam (2003). Dimensions of syntactic change: Evidence from the long -nt- participle in Old Czech texts. In R.A. Maguire & A. Timberlake (Eds.), American Contributions to the 13th International Congress of Slavists in Ljubljana (79–92). Slavica. Fried, Mirjam (2004). Predicate semantics and event construal in Czech case marking. In M. Fried & J-O. Östman (Eds.), Construction Grammar in a cross-language perspective (87–120). Amsterdam & Philadelphia, PA: John Benjamins. Fried, Mirjam (2005). A frame-based approach to case alternations: The swarm-class verbs in Czech. Cognitive Linguistics, 16, 475–512. Fried, Mirjam (2008a). A Frame Semantic account of morphosemantic change: the case of Old Czech věřící. In D. Divjak & A. Kochanska (Eds.), Cognitive paths into the Slavic domain (283–315). Berlin: Mouton de Gruyter. Fried, Mirjam (2008b). Constructions and constructs: Mapping a shift between predication and attribution. In A. Bergs & G. Diewald (Eds.), Constructions and language change (47–79). Berlin: Mouton de Gruyter. Fried, Mirjam (In press). Grammaticalization and lexicalization effects in participial morphology: A Construction Grammar approach to language change. In H. Cuyckens, K. Davidse, & J-Ch. Verstraete (Eds.), Grammaticalization and grammar. Amsterdam & Philadelphia: John Benjamins. Fried, Mirjam & Jan-Ola Östman (2003). The explicit and the implicit in the Suasion Frame. In E. Hajičová, A. Kotěšovcová & Jiří Mírovský (Eds.), Proceedings of CIL 17, 1–22, CD-ROM. Prague: Matfyzpress. Fried, Mirjam & Jan-Ola Östman (2004). Construction Grammar: A thumbnail sketch. In M. Fried & J-O. Östman (Eds.), Construction Grammar in a cross-language perspective (11–86). Amsterdam & Philadelphia, PA: John Benjamins. Fried, Mirjam & Jan-Ola Östman (2005). Construction Grammar and spoken language: The case of pragmatic particles. Journal of Pragmatics, 37, 1752–1778.



Representing contextual factors in language change  Geeraerts, Dirk (1992). The return of hermeneutics to lexical semantics. In M. Putz (Ed.), Thirty years of linguistic evolution: Studies in honor of René Dirven (257–282). Amsterdan & Philadelphia: John Benjamins. Geeraerts, Dirk (1993). Vagueness’s puzzles, polysemy’s vagaries. Cognitive Linguistics, 4, 223– 272. Goldberg, Adele E. (2002). Surface generalizations: An alternative to alternations. Cognitive Linguistics, 13, 327–356. Haas, Florian (2005). The development of English each other – grammaticalization, lexicalisation, or both? Paper presented at the conference From Ideational to Interpersonal: Perspectives from Grammaticalization, Leuven, Belgium. Halmari, Helena & Tuija Virtanen (Eds.) (2005). Persuasion across genres. Amsterdam & Philadelphia: John Benjamins. Harris, Alice C. (2003). Cross-linguistic perspectives on syntactic change. In B.D. Joseph & R.D. Janda (Eds.), The handbook of historical linguistics (529–551). Blackwell. Haspelmath, Martin (1994). Passive participles across languages. In B. Fox & P. J. Hopper (Eds.), Voice: Form and function (151–177). Amsterdam: John Benjamins. Haspelmath, Martin (1996). Word-class-changing inflection and morphological theory. In G. Booij & J. van Marle (Eds.), Yearbook of Morphology 1995 (43–66). Kluwer Academic Publishers. Himmelmann, Nikolaus P. (2004). Lexicalization and grammaticization: Opposite or orthogonal? In B. Wiemer & W. Bisang (Eds.). What makes grammaticalization? (21–42). Berlin: Mouton de Gruyter. Hopper, Paul J. (1998). Emergent grammar. In M. Tomasello (Ed.), The New Psychology of Language (155–175). New Jersey: Lawrence Erlbaum associates. Kurz, Josef (1958). K€ otázce doby vzniku slovanských adjektiv složených a jejich původního významu. In Studie ze slovanské jazykovědy, sborník k€ 70. narozeninám akademika Fr. Trávníčka (211–219). Praha: SPN. Linell, Per (1998). Approaching dialogue. Talk, interaction and contexts in dialogical perspectives. Amsterdam & Philadelphia, PA: John Benjamins. Östman, Jan-Ola (2000). On the outskirts of ‘the periphery’: Construction Grammar meets the Postcard Frame. Paper presented at the The 7th International Pragmatics Conference, Budapest, Hungary. Schenker, Alexander M. (1995). The Dawn of Slavic: An introduction to Slavic philology. New Haven: Yale University Press. Schultze-Berndt, Eva & Nikolaus P. Himmelmann (2004). Depictive secondary predicates in crosslinguistic perspective. Linguistic Typology, 8, 59–131. Short, David (1993). Czech. In B. Comrie & G.G. Corbett (Eds.) The Slavonic languages (455–532). London & New York: Routledge. Staročeský slovník [Old Czech Dictionary] (1968). Praha: Academia. Swales, John (1990). Genre analysis: English in academic and research settings. Cambridge: Cambridge University Press. Traugott, Elizabeth Closs (1992). From less to more situated in language: The unidirectionality of semantic change. In M. Kefer & J. van der Auwera (Eds.), Meaning and grammar: Crosslinguistic perspectives (497–517). Berlin: Mouton de Gruyter. Traugott, Elizabeth Closs (2003). Constructions in grammaticalization. In B.D. Joseph & R.D. Janda (Eds.), The handbook of historical linguistics (624–647). Oxford: Blackwell. Traugott, Elizabeth Closs (2008). The gammaticalization of N.P. of NP patterns. In A. Bergs & G. Diewald (Eds.) Constructions and language change (23–45). Berlin: Walter de Gruyter.



Mirjam Fried Traugott, Elizabeth Closs & Richard B. Dasher (2002). Regularity in semantic change. Cambridge: Cambridge University Press. Wiemer, Björn (2004). The evolution of passives as grammatical constructions in Northern Slavic and Baltic languages. In B. Wiemer & W. Bisang (Eds.). What makes grammaticalization? (271–331). Berlin: Mouton de Gruyter. Wiemer, Björn & Walter Bisang (2004). What makes grammaticalization? An appraisal of its components and its fringes. In B. Wiemer & W. Bisang (Eds.). What makes grammaticalization? (3–20). Berlin: Mouton de Gruyter. Wischer, Ilse (2000). Grammaticalization versus lexicalization: ‘Methinks’ there is some confusion. In O. Fischer, A. Rosenbach, & D. Stein (Eds.), In pathways of change: grammaticalization in English (355–370). Amsterdam: John Benjamins.

part 2

Interactional approaches

Grammatical constructions in dialogue Per Linell

Department of Culture and Communication, Linköping University

1. Introduction1 Construction grammars take many shapes, from more formalistic ones to more cognitive ones (Goldberg 1995; Fried & Östman, 2005; Croft & Cruse 2004). This paper takes its point of departure in the analysis of talk-in-interaction and in dialogical theory. Such a theoretical framework will still share many assumptions with Construction Grammar (CxG; Fried & Östman 2005), including of course the insistence that a theory of grammar must be usage-based. Yet it is undeniable that many variants of CxG suffer from an interactional deficit. I will therefore argue for a dialogical – a more interactional and contextual – theory and interpretation of grammatical constructions and lexical resources (Linell 2006a, 2009). 2. A dialogical conception of grammatical constructions I take dialogical theory (or ‘dialogism’) (Linell 1998, 2009, and references there) to be a general theoretical framework for the description and explanation of most kinds of human action, cognition, communication and semiotic practices. It stresses the dynamic use of linguistic resources in the construal of meaning in interaction and contexts, whether the medium is spoken, written or electronic. That is, the terms ‘dialogue’ and ‘dialogical’ are not used here exclusively with reference to direct interactions between two or more mutually co-present individuals or communicative systems, e.g. face-to-face interactions based on spoken language. On the other hand, this is not to deny the significance of the metaphor of conversation within many applications of dialogical theory. And indeed, I shall deal in this paper with linguistic resources as used in talk-in-interaction. One of the credos of dialogical theory is that, as regards language and language use, we should always prefer explanations in terms of cognitive processing and/or social action and interaction (e.g. constraints on turn design conditioned by sequential contexts), and let purely structural (language-internal) accounts



Per Linell

only stand proxy for such explanations. As applied to grammatical constructions, this view is more in line with the interactional variant of cognitive grammar of Ono & Thompson (1995), following Langacker (1987), than with formal CxG. Ono & Thompson (1995) introduce the notion of ‘constructional schema’. Below, I summarise their most important points: 1. Language users’ grammatical knowledge is organised in terms of ‘constructional schemas’ that are relatively stable and have emerged from language users’ cumulative experiences of utterances. They are “schematizations over sets of expressions parallel in formation, which are their instantiations” (p. 220). 2. Having once emerged in some form, a schema constitutes a pattern, according to which speakers configure linguistic elements when they are faced with the task of having to solve a particular communicative problem in a new situation. 3. Schemas are abstract prototypes (p. 217) with fuzzy boundaries (p. 220); “[they] do not form a finite set, but are best understood as a dynamic and constantly changing inventory” (p. 220). 4. From a formal point of view, constructional schemas can be phrases, clauses, or multi-clause structures (p. 220). Schemas can be combined, linked to and embedded within each other in different ways (p. 220). 5. The structure of each single schema is linked to principles such as “one clause at a time” (Pawley & Syder, 1977) and “one new idea per intonation unit” (Chafe, 1994). 6. Instantiations of schemas are negotiated locally, and are strongly influenced by cognitive constraints, and by social constraints that emerge in and through the situated interaction (p. 217). Another interactionist account of constructions is that of Anward (1999). Anward formulates nine theses, which I here collapse into five: Constructions are (i) utterance types, i.e. abstractions from concrete utterances, or aspects of networks of related such utterances, but (ii) they are also lexicalised items with variable parts, (iii) which are open and dynamic, both paradigmatically and syntagmatically. (iv) They (or their instantiations/PL) are context-dependent as well as context-inducing, but (v) at the same time, they have also been abstracted, “torn from contexts”, i.e. largely decontextualised, although they still bring with them some residue of their former contextual embeddedness when they become employed in new usage events. Anward’s and Ono & Thompson’s points are obviously very closely related. In principle, I would endorse all of them, but I will proceed beyond them in some



Grammatical constructions in dialogue 

respects, in particular in developing aspects of the external (or outer) syntax of constructions, with three subaspects: a. what conditions on prior sequences does (the occurrence of) a specific grammatical construction set up (these conditions being either specific requirements or merely preferences), b. what conditions on subsequent sequences does it set up, c. what are its preferred co-occurring resources? This implies that we will be concerned with some aspects of sequential context(s) when we consider grammatical constructions (cf. Anward’s point (iv)). I will argue that some conditions of the kinds suggested by (a–c) can be part of the language user’s knowledge of each and every grammatical construction. Language users also have pragma-semantic knowledge linked to specific grammatical constructions. At some level, they ‘know’ a great deal about what kinds of meanings and functions constructions have and what sorts of pragmatic effects one can achieve with them. Accordingly, this theory assumes that constructional schemas, alias grammatical constructions, have integral meaning potentials tied to themselves. Neither situated interpretations of utterances nor the meaning potentials of the resources of language can be fully compositionally derived from their constituent elements. The central idea behind the notion of meaning potential is that no word or grammatical construction has a completely fixed meaning (or function) that is actualised in all its usage events, but instead, the meaning potential is a (structured) set of affordances2 for sense-making that combines with contextual factors to yield situated meanings in different ways in different situations (Norén & Linell 2007). I shall use the term ‘functional potential’ for the meaning potential of a grammatical construction (as opposed to a lexical item). 3. X-och-x – a responsive construction in Swedish Research in interactional linguistics has demonstrated that talk in real interaction is incrementally produced and in various senses co-constructed by several participants (e.g. Auer 2007; Lindström 2006). Utterances are often co-constructed, and they are always dependent on sequential links within connected discourse. These links are either backward-pointing (‘responsive’) or forward-pointing (‘projective’) (or both). If the concrete constructs, i.e. the situated utterances, have these properties, then it seems reasonable to assume that at least some of the more abstract constructions (schemas) used, i.e. the resources of language, have been designed to be used in interaction and that they too include responsive and

 Per Linell

projective properties. The definition of such a ‘responsive construction’ is that its linguistic form encodes (local) responsivity. This means that one can tell from the morpho-syntactic (and prosodic) form of any instantiation that it cannot be the first contribution in the discursive sequence (episode) where it occurs; instead, it is responsive to some prior contribution, and sometimes, this prior contribution too must exhibit a particular form. Many constructions, and some of their instantiations, are of course not responsive in this narrow sense. For example, an utterance like It’s raining can occur as the first (and even only) contribution to a conversational episode. On the other hand, there are also many other responsive constructions in various (and arguably all) languages. Some English examples are: – various elliptical constructions, – many question types (responsive questions), such as echo-questions: “am I sure?” “do I read?”, and wh-questions with declarative word order: “you live where?”, – the “incredulity response construction” (Lambrecht, 1990): “(what?) him wear a tuxedo?”, – more well-known constructions, such as the it-cleft, – various contribution types (with a grammatical side), many of which have been explored in Conversation Analysis, for example, and-prefaced questions (Heritage & Sorjonen 1994) and so-initiated formulations (Drew 1998). Yet, explicitly dialogical analyses of grammatical constructions have seldom been made, but Wide (2002) provides one devoted to the responsive construction vera búinn aδ in Icelandic. For reasons of space, I shall focus on only one grammatical construction here, that of x-och-x (i.e. ‘x-and-x’) in Swedish. This construction belongs to a subcategory of responsive constructions that we may call ‘reactive’ constructions. In using x-och-x, the speaker reacts to (or against) the use of a word or expression x occurring in a prior utterance, and by repeating it (in this case: twice, with an interjacent och ‘and’) in the beginning of his/her response, he or she reaccentuates it, negotiating and respecifying its situated meaning. Let us cite an example before going into further details.3

(1) FLYTTA Å FLYTTA (sam:V1: 989ff) (there is ongoing talk in a dinnertable conversation about a German family that was forced to leave Finland after WW II) 1. G: sen så beslagtos huse å (0.5) dom flytta tilbaka 2. ti (0.7) ti Hamburg (å) 3. M: nå flytta å flytta men ja menar va (.) fan kan du 4. göra



Grammatical constructions in dialogue 



G: “then the house was confiscated and (0.5) they moved back to (0.7) to Hamburg (and)” M: “well (nå), moved and moved but I mean what (.) the hell can you do”

The speaker uses the x-och-x construction to comment on the situated use of a particular expression x, suggesting that x is not quite situationally appropriate, although not completely misplaced either. In (1), x is flytta ‘move (house)’, and speaker M seems to cancel the sense aspect of ‘moving voluntarily’, whereas he might still endorse another aspect, such as the more neutral semantic aspect of ‘changing one’s dwelling-place’. In the terminology of Fillmore et al.€(1988), x-och-x is a formal idiom. A better term might be ‘schematic idiom’ (Croft & Cruse 2004:€234). The x-och-x segment itself appears as a syntactically non-integrated element in the ‘pre-front field’ (cf. Auer 1996) of a turn or turn constructional unit (TCU). However, it is usually prosodically integrated, rather than marked off (separated) (in the terms of Selting 2005:€21) with the following clausal segment. It contains two copies of an expression x (a morphological form of a lexeme) that has occurred in a prior utterance. These two x’s are conjoined by och ‘and’, in conversational language almost always in the phonological form /o/, usually rendered as å in spelling. What the speaker does with x-och-x is to take an expression x from an immediately prior utterance, place a reduplicated copy of it (x och x) in the pre-front field of a new turn or turn-constructional unit, and follow this up with an utterance in which the situated appropriacy of x in relation to the current topic is negotiated. Thus, the construction has a semantic-pragmatic function, or rather functional potential that cannot be entirely derived from its constituents, i.e. ‘x’ and ‘and’, and their meanings. I shall go into more pragmatic-semantic details below. X-och-x occurs mainly in conversational language, that is, in actual talk-ininteraction as well as in certain written and electronic genres (e.g. chat, weblogs) in Swedish, and similarly at least in Danish, Norwegian, and Finnish. There is no direct counterpart of x-och-x in English; in our example (1), we could render the meaning of “moved and moved” approximately as “moved?, it depends on what you mean by that”. In German, you may compare the formal idiom Was heisst schon x? (or x oder nicht x). The source of x is usually in the interlocutor’s prior contribution, as in (1), but it can also occur in the speaker’s own utterance. There are many similarities between the other-responsive and self-responsive uses, but also some formal and pragma-semantic differences (for details, see Lindström & Linell 2007). Here, I shall cite one more other-responsive example (2), and then provide a self-responsive example (3).4

 Per Linell



(2) MÅNGA Å MÅNGA (TemaK:B9:4:6) (from the very start of a talk at a maternity clinic between a doctor (D) and a pregnant woman (W); D wonders if W is used to being tape-recorded in connection with her visits to the clinic) 1. D: jaha du e van vi de här å-eh (.) bli inspelade¿ 2. W: ja= 3. D: =du har (.) haft många inspelade samtal eller¿ 4. W: ja-eh (.) många å många men de e nåra stycken 5. så de-

D: W: D: W:

“okay (jaha), you are used to this ah (.) being tape-recorded?” “yeah” “you have (.) had many conversations recorded, or...¿” “well (ja-eh) (.), many and many, but there are quite a few, so it’s --”

In (2), the speaker objects to the use of många ‘many’, without downright denying its relevance. X-och-x accomplishes some kind of devaluation or downgrading of x, that is, from, say, ‘a large number’ to ‘some’ or ‘quite a few’.

(3) NORMAL Å NORMAL (Swedish Radio; a H(istorian) is being interviewed by an I(interviewer) on the topic of elite schools, so-called Napola schools, in Nazi Germany and their possible role in promoting Aryan elitism)



1. H: 2. 3. 4. 5. I: 6.





[…] eh (.) °ja° de va (.) delvis normal skolgång men-eh (.) eller ja, normal å normal °men°Â€(.) man ägnade mycke tid åt fysisk fostran å gymnastik å sedan så småningom också (.) militära övningar. så man höll på å kasta handgranater å skjuta me gevär?

H: “ah (.) well (ja) it was (.) a partly normal schooling but ah or well (ja), normal and normal (.) but one devoted much time to physical education and exercises and then gradually also (.) to military practices.” I: “so one was busy throwing hand-grenades and shooting with rifles?”

What the speaker does in (3) is to question in retrospect his own use of the term ‘normal’ (in line 1). However, he does not take it back completely. Rather he seems to suggest that some military education was ‘normal’ in the specific historical context, but that military exercises would not be considered ‘normal’ in today’s comprehensive schools. One may perhaps see this as a downgrading on a scale of ‘normality’. When x-och-x is self-responsive, it is partly reminiscent of what Couper-Kuhlen & Thompson (2005) have descibed as ‘concessive repair’ in English



Grammatical constructions in dialogue 

data. An example would be the following (2005: 263): a school teacher has complained about the unruliness of the children in class, saying that she needs to get it all off her mind, and then mentions that she will have an in-service training the day after: so I can switch off. well, not really switch off but you know, relax. This concessive repair involves a response to a prior description (here: “switch off ”), treating it as “partially unjustified”, as an “overstatement” (she did not mean “really switch off ”), and then following it up with a “revised statement”, which does not, however, imply a complete “backing down” (“you know: relax”). In contrast to x-och-x, however, ‘concessive repair’ is always self-responsive, and is not as tightly grammaticalised. While concessive repair is a conversational practice, x-och-x is clearly a grammatical construction. Furthermore, x-och-x is probably less of a repair; it treats the deployment of the expression x as in fact partially justified in the context at hand. 4. Conditions on x-och-x as a construction Space restrictions exclude both further examples and a detailed analysis of, for instance, conversational (impromptu) vs. written (and therefore sometimes edited) usages. (Lindström & Linell (2007) provide a comprehensive analysis.). Instead, I will consider here what properties of x-och-x one might want to specify in a more precise dialogical analysis of the construction. I shall distinguish between formalgrammatical and semantic-pragmatic aspects. Among the former, I shall talk about conditions of internal structure, and conditions on ‘outer syntax’ (co-text) of three kinds: conditions on antecedent segments (i.e. on prior contributions to the local sequence), subsequent segments, and co-occurring linguistic resources. 4.1

Formal-grammatical aspects

The primary condition on antecedent strings is of course the following: i. If an expression x (a morphological form of a lexical item x) occurs in the preceding turn constructional unit (TCU) or turn, x-och-x may be used. This rule articulates a necessary condition in the sense that x-och-x is an option if and only if x does occur in the prior discourse. This is also what makes x-och-x into a responsive (and reactive) construction. (In very exceptional cases, this rule can be flouted when a speaker begins a new communicative episode with an instance of x-och-x. But then, a prior occurrence of x will be tacitly inferred and imagined by the hearer.) However, (i) is not a sufficient condition (nor are the

 Per Linell

other conditions below sufficient conditions); if an expression x occurs in an utterance, it is of course not necessary to follow it up with x-och-x. Other, non-obligatory but enabling conditions are: X-och-x is more probable, ii. if x is focally stressed in the prior (source) utterance, iii. if x is rhematic in that utterance, and iv. if the source utterance is interrogative (this applies only when x-och-x is otherresponsive). The interpretation of each such condition is that if the condition is satisfied, the probability that x-och-x will be used in the following utterance is increased (presupposing of course that condition (i) is satisfied in the first place). In our example (1), (ii) and (iii) are satisfied, but not (iv). In (2), (iv) too is partially satisfied. As we have seen (cf. also condition (x) below), the speaker of x-och-x questions the situational appropriacy of x. If the source x is already placed in an interrogative utterance (condition (iv)), we can, in some instances, talk about a collaborative construction of uncertainty and its elimination (example (2)). As regards the internal structure of the x-och-x segment itself, we might posit the following obligatory conditions: v. x-och-x occurs in the ‘pre-front field’ of a new turn or TCU, vi. x in x-och-x is repeated (twice) in the same morphological form as in the source utterance,5 vii. both x’s (or at least the second one) are focally stressed, viii. the x-och-x segment itself is prosodically (but not syntactically!) integrated with the subsequent utterance. Co-occurring constructions are other linguistic devices and constructions that speakers tend to ‘co-select’. In traditional terms, x-och-x is preferably “constructed with” certain other resources. Thus, the following condition is optional: ix. x-och-x often co-occurs with distancing responsive particles (in the examples here: nå, ja, ja-eh) and concessive markers, especially in self-responsive cases (eller ‘or’ (cf. example (3): line 2), i alla fall ‘anyway’, etc.). In coherent discourse, especially within the bounds of local contributions, speakers routinely make semantic specifications that involve the use of several (usually convergent) semiotic resources (Deppermann 2005: ‘co-selection’). Co-selected resources “reciprocally constrain and specify each other’s local interpretation” (Deppermann, 2005: 306). Therefore, such resources often become associated with each other. In this case, the lexicalised grammatical construction x-och-x is (paradigmatically and syntagmatically) associated with distancing particles and concessive markers, as a sedimented result of speakers’ co-selection routines.



Grammatical constructions in dialogue 

Finally, moving on to the subsequent contribution, we find: x. after the x-och-x segment, it is obligatory to proceed with an utterance that – for current communicative purposes – confirms (or foregrounds) some aspects and simultaneously cancels (or backgrounds) other aspects of x’s meaning potential. Note that unlike the antecedent condition, this subsequent part is necessarily present: x-och-x does not occur without this kind of continuation.6 This fact (together with condition (viii)) may be taken as an argument for regarding this subsequent segment as part of (the internal structure of) the grammatical construction itself. 4.2

Semantic-pragmatic aspects

The subsequent segment is closely related to the semantic-pragmatic functions of x-och-x. As we have seen, what the x-och-x construction does is usually to problematise one (central) sub-sense of the meaning of x, and to enhance another as situationally appropriate. One might say that the construction is a grammaticalised method of initiating a local meta-linguistic discussion and semantic analysis of the situated use of x, and this deliberation necessarily involves consideration of x’s lexical meaning. That is, in using the construction, the speaker exploits the meaning potential of x as a communicative resource. In fact, precisely this – negotiating which semantic aspects of x will apply – can be seen as the core of the construction’s own functional potential. This amounts to another argument for including (what I just called) the subsequent segment (cf. condition (x)) within the grammatical construction itself. In our example (1), as we noted, what is questioned is the appropriacy of using flytta ‘move (house)’ in a situation in which the people moving had no other choice; it seems that flytta is often taken to mean ‘move voluntarily’, and this is the aspect of the meaning potential that is being cancelled here. At the same time, other aspects of the meaning potential are possibly supported in (1), such as ‘changing one’s dwelling place’. The analysis of x-och-x usage events involving different lexemes can be used as an illuminating case of probing the interaction between grammar and lexis in authentic language use. Accordingly, Norén & Linell (2007) analyse the interaction of x-och-x with different kinds of lexemes, and demonstrate that the construction has somewhat different semantic-pragmatic effects depending on the meaning potential of the lexical item (x) with which it interacts. X-och-x is often applied to items with a quantitative or scalar semantics, and it downgrades or contextualises the situated value of x. In this respect, a quantifier like många ‘many’ in (2) clearly belongs to another semantic category than flytta ‘move’ in (1). A word like normal ‘normal’ is not unambiguously scalar in its semantics, but Norén & Linell (2007)

 Per Linell

argue that the construction may sometimes foreground or actually construct a scale-dependent interpretation of x. In (3), the military exercises were ‘normal’ only in the particular socio-historical environment talked about, that is, the conversationalists construe it as a case placed somewhere towards the lower end of a scale of (general) ‘normality’. 5. General conclusions: Constructions in dialogue My topic in this paper is not x-och-x per se (see Lindström & Linell 2007), nor dialogical linguistics in general (see Linell 2006a, 2009). Rather, I am interested in what the study of responsive constructions can teach us, when it comes to remedy the interactional deficit in Construction Grammar. By way of summary, I shall draw attention to three or four points in a dialogical (contextual, interactional) conception of a grammatical construction. Contexts, including linguistic co-texts, sanction specific linguistic resources, and at the same time, contexts are themselves evoked or made relevant by the latter (Anward; see above). The first point is therefore that the situated occurrrences and interpretations of particular utterances involve sequential dependencies; utterances have backward-pointing (‘responsive’) and forward-pointing (‘projective’) properties. In other words, grammatical constructions are made to fit into sequences of real coherent sequences of sense-making in talk (or text). Some constructions have actually encoded such relations. X-och-x is such a responsive construction; language users’ knowledge about this linguistic resource includes knowledge about antecedent and subsequent strings, as well as about co-occurring resources. Accordingly, such constructions do not only have an internal structure; they also have an external syntax with (at most) three parts: antecedent segments, subsequent segments, and co-occurring resources. Secondly, grammatical constructions are concerned with actions and doings; they should be thought of in terms of methods, procedures or operations. Communicatively, they make a change, a ‘difference that makes a difference’, to the micro-situations in which they occur. The use of x-och-x changes, however little, the understanding of the current situation and its discoursal topic. Linguistically, constructions are operations on form and meaning, operating on some linguistic material and returning something else. On the structural, product-oriented side, they are oriented to the (speaker’s) design and production of utterances or utterance parts with a particular linguistic structure. In CxG jargon, the abstract ‘construction’ licenses or authorises the ‘construct’ instantiated. Within an overall dialogical framework, we would like to capture methods, procedures and operations, and more generally: actions and processes. If, then, we



Grammatical constructions in dialogue 

try to account formally for grammatical constructions, this is not exclusively a matter of dealing with constituent structure, as in most grammatical theories, including CxG (despite its basis in e.g. Fillmore’s semantics of understanding), but with methods used by the parties to the interaction. Furthermore, while syntacticians (and compositionalists in semantics) usually deal only with the internal structure, a dialogical linguistics must also take external syntax into consideration. The third point, too, is related to the preceding considerations. A grammatical construction has pragma-semantic properties that cannot be exhaustively derived from its constituent parts. We must assume, of course, that some loose form of compositionality is needed for assigning interpretations to new complex expressions, but strict compositionality is impossible. First, we can of course not compute interpretations of situated utterance tokens without contextual information (and relevant contexts cannot be specified in advance). Secondly, as several scholars in cognitive linguistics and construction grammar (e.g. Goldberg 1995:€13ff.) have argued, strict compositionality does not hold on an abstract ‘structural’ level either; specific constructions have semantic properties that cannot be compositionally derived. A grammatical construction involves a specific semantic operation that is tied to the construction as a whole. In the case of x-och-x, the construction is mainly reactive: it negotiates the meaning of an expression x occurring in the prior discourse, and modifies or revises the interpretation to be accepted locally. This in turn implies that the grammar as a whole will register similar information at many places. Such a ‘maximalistic’ assumption, which is opposed to the dominant minimalism of many models of theoretical linguistics (Cappelen & Lepore 2005), is nowadays commonplace in both usage-based approaches, such as construction grammar (Fried & Östman 2005), and in some more theoretical approaches to semantics (Recanati 2004). Specific constructions may inherit properties from more abstract constructions. For example, x-och-x undoubtedly belongs to a family of reactive constructions with several properties in common (Lindström & Linell 2007). Certain linguistic resources are frequently co-selected (see above), and such resources are arguably associated with similar information at several places in the language user’s knowledge of his or her language. Fourth, I have been reasonably specific about various conditions associated with x-och-x in this paper. Such constraints on the use of x-och-x could arguably be formalised in a suitable notation. However, we must be aware that a complete formalisation is not necessarily the ultimate goal.€This is argued in Linell (2006b) with particular reference to x-och-x. Formalisation is not a neutral or “innocent” meta-linguistic description, but it transforms its own subject matter. If we think of the subject matter as the dynamic practices of sense-making, and the resources used in these practices, formal accounts tend to respecify these in terms of static semantic frames and syntactic structures that are assumed to exist somehow

 Per Linell

abstracted from language use. If we think of language dialogically, as involving actions in situations in the world, we can not simply equate situated meanings with tokens of abstract fixed types. While I would definitely insist that grammatical constructions and meaning potentials do exist as such, in addition to and in one sense partly independently on the situated usage events, as accumulated resources in the situation-transcending practices of language use; these resources are, after all, only made manifest, actualised, in these real usage events. Notes 1. This paper builds upon joint work with Jan Lindström (Helsinki) and Kerstin Norén (Karlstad). The paper was first read in a workshop on “Context in Construction Grammar” organised by Jan-Ola Östman and Mirjam Fried at IPrA IX, Riva del Garda, Italy, July 14, 2005. I am grateful to all these colleagues, as well as to other participants in the workshop. My own work was supported by a grant from the Swedish Research Council (No. 421–2004–1087). The author can be contacted at the following e-mail address: [email protected] 2. The notion of affordance comes from Gibson’s (1979) theory of perception. For the application to language, see e.g. Thibault (2005). 3. The examples used here are drawn from a corpus of tape-recorded cases (Lindström & Linell 2007). Here I have simplified transcripts. The particular x-och-x segment and its source in a prior utterance are given in bold-face. In this paper, each example is followed by an approximate rendition in English, in which the x-och-x phrase has been translated word-by-word despite the fact that it lacks an idiomatic English counterpart. 4. (.) marks a micro-pause. Underlining (of a syllable nucleus, i.e. a vowel sign) indicates focal stress. 5.

There are some (non-random) exceptions to this. See Lindström & Linell (2007).

6. Just like in the case of the necessary condition (i), condition (x) can occasionally be flouted, with particular rhetorical effects.

References Anward, Jan (1999). Constructing an interactionist account of constructions. Paper presented at Workshop on Construction and Interaction, Helsinki, September 1999. Auer, Peter (1996). The pre-front field in spoken German and its relevance as a grammaticalization position. Pragmatics, 6(3), 295–322. Auer, Peter (2007). Syntax als Prozess. In H. Hausendorf (Ed.), Gespräch als Prozess. Tübingen: Narr. Cappelen, Herman & Ernie Lepore (2005). Insensitive semantics. A defense of semantic minimalism and speech act pluralism. Malden, MA: Blackwell Publishing. Chafe, Wallace (1994). Discourse, consciousness and time. Chicago: The University of Chicago Press.



Grammatical constructions in dialogue  Couper-Kuhlen, Elizabeth & Sandra A. Thompson (2005). A linguistic practice for retracting overstatements: ‘Concessive repair’. In A. Hakulinen & M. Selting (Eds.), Syntax and lexis in conversation (257–288). Amsterdam/Philadelphia: John Benjamins. Croft, William & D. Alan Cruse (2004). Cognitive linguistics. Cambridge: Cambridge University Press. Deppermann, Arnulf (2005). Conversational interpretation of lexical items and conversational contrasting. In A. Hakulinen & M. Selting (Eds.), Syntax and lexis in conversation (289–317). Amsterdam/Philadelphia: John Benjamins. Drew, Paul (1998). An exercise in the comparative analysis of talk-in-interaction in different (institutional) settings: the case of ‘formulations’. In H. Lehti-Eklund (Ed.). Samtalsstudier [‘Conversational studies’] (29–42). Helsinki: Department of Scandinavian Languages and Literature. Fillmore, Charles, Paul Kay & Mary K. O’Connor (1988). Regularity and idiomaticity in grammatical constructions: the case of let alone. Language, 64, 501–538. Fried, Mirjam & Jan-Ola Östman (2005). Construction Grammar: A thumbnail sketch. In M. Fried & J.-O. Östman (Eds.), Construction Grammar in a cross-language perspective (11–86). Amsterdam & Philadelphia: John Benjamins. Gibson, James (1979). The ecological approach to visual perception. Boston: Houghton-Mifflin. Goldberg, Adele (1995). Constructions: A Construction Grammar approach to argument structure. Chicago: The University of Chicago Press. Heritage, John & Marja-Leena Sorjonen (1994). Constituting and maintaining activities across sequences: and-prefacing as a feature of question-design. Language in Society, 23, 1–29. Lambrecht, Knud (1990). What, me worry: Mad Magazine sentences revisited. In D. J. Costa (Ed.), Proceedings from the Sixteenth Annual Meeting of the Berkeley Linguistic Society (215–228). Berkeley, CA.: Berkeley Linguistic Society. Langacker, Ronald (1987). Foundations of Cognitive Grammar, Vol. 1: Theoretical prerequisites. Stanford: Stanford University Press. Lindström, Jan (2006). Interactional linguistics. In J.-O. Östman & J. Verschueren (Eds.), Handbook of Pragmatics 2006. Amsterdam & Philadelphia: John Benjamins. Lindström, Jan & Per Linell (2007). X-och-x som samtalspraktik och grammatisk konstruktion. [‘X-och-x as a conversational practice and grammatical construction’]. In E. Engdahl & A.-M. Londen (Eds.), Interaktion och kontext (19–89). Lund: Studentlitteratur. Linell, Per (1998). Approaching dialogue: Talk, interaction and contexts in dialogical perspectives. Amsterdam & Philadelphia: John Benjamins. Linell, Per (2006a). Towards a dialogical linguistics. In M. Lähteenmäki, H. Dufva, S. Leppänen, & P. Varis (Eds.), The XIIth International Bakhtin Conference: Proceedings (152–167). Jyväskylä, Finland: Department of Languages. Linell, Per (2006b). Can we formalise the linguistic resources of interactional language? In M. Suominen et al.€(Eds.), A Man of Measure: Festschrift in Honour of Fred Karlsson on his 60th Birthday. Special Supplement to SKY Journal of Linguistics, 19, 56–65. Turku: The Linguistic Association of Finland. Linell, Per (2009). Rethinking language, mind and world dialogically: Interactional and contextual theories of human sense-making. Charlotte, NC: Information Age Publishing. Norén, Kerstin & Per Linell (2007). Meaning potentials and the interaction between lexis and grammar: Some empirical substantiations. Pragmatics, 17, 387–416.

 Per Linell Ono, Tsuyoshi & Sandra A. Thompson (1995). What can conversation tell us about syntax? In Ph. Davis (Ed.), Descriptive and theoretical modes in alternative linguistics (213–271). Amsterdam & Philadelphia: John Benjamins. Pawley, Andrew & Frances Syder (1977). The one clause at a time hypothesis. Unpublished manuscript. Recanati, François (2004). Literal meaning. Cambridge: Cambridge University Press. Selting, Margret (2005). Syntax and prosody as methods for the construction and identification of turn-constructional units in conversation. In A. Hakulinen & M. Selting (Eds.), Syntax and lexis in conversation (17–44). Amsterdam & Philadelphia: John Benjamins. Thibault, Paul (2005). The interpersonal gateway to the meaning of mind: the inter- and intraorganism perspective on language. In R. Hasan, C. Matthiessen & J. Webster (Eds.), Continuing discourse on language: A functional perspective (117–156). London: Equinox. Wide, Camilla (2002). Perfect in dialogue: Form and functional potential of the vera búinn a∂ + inf. construction in contemporary Icelandic (PIC Monographs, 3). Helsinki: Department of Scandinavian Languages and Literature.

Interactional Construction Grammar Contextual features of determination in dialectal Swedish Camilla Wide 1. Introduction1 This paper discusses the role of context in Construction Grammar (CxG) and the question of how contextual features can and should be included in the Construction Grammar formalism. I will show that certain contextual features need to be included in the formalism if we want to describe spoken language constructions properly. More specifically, I will suggest a solution that builds upon a combination of Construction Grammar with Interactional Linguistics (IL), which is an interdisciplinary and a cross-linguistic approach to language. The methodological foundation of IL is Conversation Analysis (CA). The other disciplines that have contributed to IL are discourse-functional linguistics and linguistic anthropology (see Schegloff, Ochs & Thompson 1996; Couper-Kuhlen & Selting 2001). As Fillmore points out (1988:€35), the framework of Construction Grammar is a moving target. There are thus several ‘construction grammars’ available (cf. Östman & Fried 2005). When I talk about Construction Grammar (CxG) in this paper, I refer to the original approach initiated by Charles J. Fillmore (e.g. 1988, 1989) and further developed by, among others, Fried and Östman (e.g. 2004). The pragmatic dimension has always played an important role in CxG (see in particular Fillmore 1996). Fillmore (1989:€32) stresses the role of interaction and points out that in CxG “numerous grammatical constructions are described as instruments of interaction”. However, in practice, interactional features have not been described or investigated to any greater extent in work within CxG. Fried & Östman (2004:€24) point out that CxG until very recently has had a fairly narrow view of pragmatics: “pragmatic force and effect have been recognized primarily as conveyed through conventions of language, not in terms of conversational reasoning or socio-cultural constraints and possibilities”. In addition to this, spoken

 Camilla Wide

language has, according to Fried & Östman (2005:€1753), played a fairly modest role within CxG: the bulk of constructional research has not been particularly concerned with analyzing spoken data beyond noting, occasionally, that a specific syntactic pattern may have different manifestations in spoken language from what it has in ‘standard’ grammar [...].

The purpose of this paper is to contribute to some of these less investigated areas of CxG, specifically spoken language and interaction. In addition to this, my study2 deals with another aspect of language that has traditionally not been in the center of grammatical investigation: grammatical constructions in regional non-standard dialects. In the empirical part of the article, I will discuss an adnominal demonstrative construction in the Swedish dialects of eastern Nyland in Finland (the area to the east of the capital Helsinki). An instantiation of the construction is shown in (1) below. (1) va må de ‘klåkka-n ska vara? what may that/the clock [=time]-the shall be ‘what time is it?’ (Natural observation, 2004) The construction, which consists of the pronoun de(n) ‘that’ and a noun with a suffix of definiteness (de ‘klåkka-n ‘that/the clock-the’), has syntactic parallels but no direct communicative-functional parallels in the standard Swedish language. As indicated in the glosses to example (1), the form de (also pronounced te) could be interpreted as the preposed definite article den ‘the’, which in standard Swedish is only used in front of modified nouns (cf. bil-en ‘car-the’ vs. den röda bil-en ‘the red car-the’). Forms with the demonstrative pronoun den ‘that’ do occur in front of nouns that are not modified in standard Swedish. However, in these cases the function is to point out a referent or underline a contrast (cf. Ge mig boken ‘Give me the book’ vs. Ge mig den boken ‘Give me that book’), which is obviously not the case in example (1). In order to describe the function of the eastern Nyland construction from a CxG point of view it is thus necessary to capture the contextual features of the construction in an adequate way. The paper is organized as follows. In the next section, I will compare the notions of ‘construction’ in CxG and in IL. Sections 3–5 present my empirical analysis of the eastern Nyland construction and make a comparison with other constructions in the eastern Nyland dialects as well as in standard Swedish. In Section 6, the results of the empirical analysis are discussed from a theoretical point of view in terms of how the contextual constraints on the eastern Nyland construction can be described using CxG formalism, without contradicting the findings about spoken language and interaction arrived at through IL.



Interactional construction grammar 

2. The notion of ‘construction’ In their comparison of Construction Grammar and Interactional Linguistics Fried & Östman (2005) stress the similarities between the two approaches, for example, the fact that both approaches take all ‘chunks’ of language as equally important for analysis and description. Furthermore, Fried & Östman conclude that among dialogical approaches to language (see Linell 1998), Conversation Analysis (CA) “– and especially its further development of Interactional Linguistics – is the closest to CxG in its interactional objectives and interests” (2005:€1754). This closeness is also reflected in works on Interactional Linguistics, among others, Ono & Thompson (1995) and Linell (2004, 2005, this volume), who refer explicitly to Construction Grammar (see also Günthner & Imo 2006, Imo 2007). However, Fried & Östman (2005:€1756) point out some differences between the approaches, such as the focus on the outcome in CxG vs. on process in CA, and suggest that a fertile starting point is to “accept that CxG and CA are compatible approaches that do different things and aspire to do different things, but they accomplish their goals in a complementary manner”. Both CxG and IL are usage-based approaches. One of the main points made about IL is that it takes non-experimental, everyday performance as the basis for linguistic description (Steensig 2001:€ 12, see also Couper-Kuhlen & Selting 2001:€ 4–5). In contrast to many paradigms within generative linguistics (the Chomskyan paradigm in particular) linguistic performance plays an important role also in CxG since grammatical constructions should be cognitively and communicatively plausible per se (cf. Fillmore 1989; Fried & Östman 2004:€76). However, the use of data has different implications for the analyses conducted in IL and CxG respectively. In IL, the data consist of recorded and transcribed naturally-occurring conversations. The goal of the analysis is to capture what patterns the participants of the conversation orient towards and how they make this orientation explicit. Special attention is paid to factors like sequentiality, co-construction and act-activity interdependence (Linell 2004:€8). These factors are analyzed in depth by reference to often rather long extracts. By contrast, in CxG, the data generally consist of isolated sentences, which provide a limited amount of contextual information (for some noteworthy exceptions, see Fujii 2004; Fried & Östman 2005). In CxG a construction is an abstract representational entity, that is, “a conventional pattern of linguistic structure that provides a general blueprint for licensing well-formed linguistic expressions” (Fried & Östman 2004:€18). Actually occurring linguistic expressions are not called constructions but rather constructs. Ono & Thompson (1995) and Linell (2004, 2005) also point out that grammatical patterns are to be seen as schematic rather than specific. According to Ono & Thompson, constructions consist of “schematizations over sets of expressions parallel in forma-

 Camilla Wide

tion, which are their instantiations” (1995:€220). A ‘constructional schema’, which is the term Ono & Thompson use (following the cognitive grammar tradition, see Langacker 2008:€215–227), thus serves as a template for analogous expressions. A major tenet of CxG is that form and meaning (or function) form an integrated whole in grammatical constructions (Fillmore 1989:€ 19, Kay 1995:€ 171, Fried & Östman 2004:€12). This goes well with an interactional perspective of language, where linguistic forms are seen primarily in the light of what function they have or acquire in interaction (cf. Couper-Kuhlen & Selting 2001:€6–7). However, although construction grammarians like Fillmore (1989:€32) stress the role of interaction, concepts such as ‘meaning’, ‘function’ and ‘interaction’ are applied in different ways in CxG and IL. In IL taking interactional features into account means showing and describing explicitly how participants in a conversation use linguistic resources or patterns in order to do things in the interaction. Meaning and function are treated as conventionalized but partly open and negotiable aspects of linguistic items, that is, as potentials rather than as fixed values (see Norén & Linell 2007). As Linell (1998:€113) points out, these potentials, however, “are never sufficient for determining communicative meaning”. Thus, even though constructions are based on speakers’ and listeners’ earlier experiences of situations and contexts they can be locally managed, negotiated and adapted to new situations. It is therefore not possible to predict exactly what function a specific construct will acquire in situated use. In addition to this, speakers and listeners, as Ono & Thompson (1995:€217) point out, seem to be “highly tolerant of utterances which, because of such phenomena as reformulations and repetitions, only roughly approximate any ‘constructional schema’, or which blend more than one schema”. Treating meaning and function as potentialities requires contextual features to be accounted for in descriptions – and formalizations – of constructions. This is in theory not a problem for CxG. Fillmore (1989:€33) explicitly stresses the role of context: the abstraction of linguistic knowledge away from context is rejected outright in the constructionist view, since a part of the understanding of grammatical constructions is knowing, one at a time, what they are for and what principles guide their choice.

Fried & Östman (2005) even mention that constructions can be seen as potentialities. However, contextual features are included to a very limited extent in the actual formalization of constructions in CxG studies. When they are formalized, the features are analyzed in different terms than in IL. For example, the concept of the speech act, which is one of the pragmatic attributes Fried & Östman (2005) use, is considered problematic in IL since it has traditionally been used to analyze sentences in isolation (see Linell 1998:€ 208–212). This reflects a more general difference between CxG and IL. Whereas studies on IL argue that language is



Interactional construction grammar 

dynamic and has to be treated as such, studies on CxG stress the fact that all parts of language, including contextual and interactional features, can be described in terms of discrete feature structures (cf. Fried & Östman 2004:€30). Anward (1999), however, points out that constructions should be seen as types that originally emerged out of typified utterances (see also Linell, this volume). Constructions are thus affected by cognitive and social constraints arising from interaction (Ono & Thompson 1995:€ 217) and should be analyzed and explained in relation to the situated discursive and interpretive activities themselves. Linell (2005:€281) points out that certain constructions may be restricted to or at least overrepresented in certain activities or communicative genres. Furthermore, many constructions have responsive or projective features (cf. Wide 2002), which may constrain the preceding and following turns, that is, the ‘external context’ in a wider sense than what is usually accounted for with the concept in CxG (cf. Fillmore 1988:€36). In the following sections, I will illustrate empirically the relevance of accounting for contextual and interactional features in a dynamic way when describing spoken language constructions. I start with a presentation of the demonstrative construction in eastern Nyland, and a discussion of why the construction is interesting from the point of view of Swedish grammar and CxG. After that, I present the major functions of the construction in authentic conversation. The empirical part of the study ends with a formalization of the construction in which contextual and interactional features are taken into account. My argument for why a formalization should be done is the following. If contextual and interactional features are, as I argue, part of constructions, such features should also be shown as explicitly as possible; and the most explicit way of accounting for features is to suggest a formalization in terms of attribute-value matrices. If such an attempt is not made, we cannot be sure that CxG gives an adequate picture of language and language use. 3. Structural aspects of the eastern Nyland demonstrative construction The demonstrative construction in eastern Nyland contains the following components: (2) den/det (sg.), de (pl.) + noun + suffix of definiteness ‘the/that’ ‘the/those’ ‘the’ In (2), the construction is presented in standard (written) Swedish, as opposed to the eastern Nyland dialect in which den would be te/de. In the remainder of the paper, I will use the conventions of written Swedish only when I discuss examples from standard Swedish.3 In examples from eastern Nyland dialects, I use the Swedish alphabet to represent the eastern Nyland pronunciation. Thus, long

 Camilla Wide

sounds are marked with double vowels or consonants (e.g. frååga ‘ask’). Furthermore, the quality of vowels and consonants is spelled as pronounced. Thus, the plural form [tom] ‘they’, for example, is spelled tåm/dåm (not tom/dom) since the representation of [o] in Swedish is generally å (cf. example (9) below). As examples (3)–(4) below show, the pronoun/article den ‘that, the’ is unstressed and pronounced [te] or [de] in the singular in the eastern Nyland dialects. The variation between the t- and d-forms is related to dialect leveling; t-forms occur in rural dialects and d-forms are used in town dialects. Both forms are used with all types of nouns. The eastern Nyland dialects thus do not make the standard Swedish gender distinction between den – det for singular non-neuter and neuter words, respectively. (Cf. standard Swedish den klocka-n ‘that clock-the’ (non-neuter) and det jordbruk-et ‘that farm-the’ (neuter) with de klåkka-n ‘that clock-the’ and te joordbruuk-i ‘the farm-the’ in examples (3) and (4)). (3)

Så möött an ein bisi. Te ‘bisi-n frååga [...] then met he a man that/the man-the asked Så gav te ‘bisi-n åt ånån ein tjepp. then gave that/the man-the to him a stick ‘Then he met a man. The man asked... Then the man gave him a stick.’ (NFOS 1896:€17)

(4) va må de ‘klåkka-n ska vara? what may that/the clock [=time]-the shall be ‘what time is it?’ (Natural observation, 2004) In example (3) and example (4), which is identical to example (1), a bare noun in the definite form (gubb-en ‘man-the’, klock-an ‘clock-the’) would be used in standard Swedish, since the referent is unique and the noun is syntactically unmodified. As example (5) shows, the suffix of definiteness is not present in constructs where the noun is a proper name. (5)

Men te ‘Alda-Ø, hon va så skåjjigär, hon but that/the Alda she was so funny she vild bara fråån te ‘joordbruuk-i. wanted only from that/the farm-the ‘But Alda, she was so funny, she just wanted to get away from the farm.’ (Lundström 1939:€146)

The different forms of the eastern Nyland construction in the singular and plural are summarized in Table 1.



Interactional construction grammar 

Table 1.╇ Forms of the eastern Nyland construction in the singular and plural Singular, non-neuter

Singular, neuter

Plural

te/de noun-(i/e)n ‘that/the noun-the’

te/de noun-e/i ‘that/the noun-the’

tåm/dåm noun-na ‘those/the noun-the’ tåm/dåm noun-en(a) ‘those/the noun-the’

te/de proper name-Ø ‘that/the proper name’

In example (5) above the proper name Alda would either be used alone or together with the demonstrative pronoun den där ‘that’ (den där Alda ‘that Alda’) in standard Swedish. The standard Swedish equivalent of the second occurrence of the eastern Nyland construction in example (5), the expression te joordbruuki ‘the farm’, would be det där jordbruket ‘that farm’. From a standard Swedish perspective, the use of the pattern [te/de ‘that/the’ + noun + suffix of definiteness] in examples (3)–(5) is thus unexpected for two reasons: a) den seems to be used as a definite article in contexts where it cannot be used in standard varieties (i.e., with nouns without premodifiers) and b) den is used as a demonstrative pronoun in contexts where another demonstrative pronoun, for example den där ‘that’, would be used. I will therefore argue that the eastern Nyland pattern with te/de and unmodified nouns in the definite form should be analyzed as a construction of its own, which lacks direct parallels in standard Swedish. 4. Comparison with standard Swedish The premodifying demonstrative pronoun in the eastern Nyland construction has sometimes been analyzed in the literature as a definite article (Dictionary of Swedish Dialects in Finland (FO), den 8, Reinhammar 1975:€79). This suggests that the preposed definite article could have been grammaticalized further in the eastern Nyland dialects than in standard Swedish. As Table 2 shows, the preposed definite article den is not used together with non-modified nouns in the definite form in standard Swedish. The preposed and postposed articles occur together only when nouns are premodified, for example with an adjective.4 As in many other languages (cf. Diessel 1999:€128–132), the definite article den in Swedish has developed from a demonstrative pronoun, the pronoun den (the Swedish Academy Ordbok (SAOB), den XIV.4). At the same time, den has retained some of its older functions and is still also used as a demonstrative pronoun. In present-day standard Swedish there are four demonstrative pronouns, the simple forms

 Camilla Wide

Table 2.╇ Indefinite and definite articles in standard Swedish Singular, non-neuter

Singular, neuter

Plural

indefinite forms

en bok ‘a book’

ett hus ‘a house’

definite forms, unmodified nouns

bok-en ‘book-the’

hus-et ‘house-the’

definite forms, modified nouns

den fina bok-en ‘the nice books-the’

det stora hus-et ‘the big house-the’

böcker ‘books’ hus ‘houses’ böcker-na ‘books-the’ hus-en ‘houses-the’ de fina böcker-na ‘the nice book-the’ de stora hus-en ‘the big houses-the’

denna ‘this’ and den ‘that’ and the compound forms den här ‘this’ and den där ‘that’. Except for the pronoun denna ‘this’ all the demonstrative pronouns in standard Swedish combine with nouns in the definite form, that is, they are used together with the suffix of definiteness, corresponding roughly to the use of the eastern Nyland te/de ‘that/the’. Table 3 shows an overview of the adnominal demonstrative pronouns in standard Swedish. The simple forms denna ‘this’ and den ‘that’ are older than the combined forms den här lit. ‘that here’ and den där lit. ‘that there’, which have been attested in written Swedish since the 16th century (Hirvonen 1987:€ 412–419). In writing, den and Table 3.╇ Adnominal demonstrative pronouns in standard Swedish Sing., non-neuter

Sing., neuter

Plural

denna/detta ‘this’ dessa ‘these’

denna bok ‘this book’

detta hus ‘this house’

den/det ‘that’ de ‘those’

den bok-en ‘that book-the’

det hus-et ‘that house-the’

den/det här ‘this’ de här ‘these’

den här bok-en ‘this book-the’

det här hus-et ‘this house-the’

den/det där ‘that’ de här ‘those’

den där bok-en ‘that book-the’

det där hus-et ‘that house-the’

dessa böcker ‘these books’ dessa hus ‘these houses’ de böcker-na ‘those books-the’ de hus-en ‘those houses-the’ de här böcker-na, ‘these books-the de här hus-en ‘these houses-the’ de där böcker-na, ‘those books-the’ de där hus-en ‘those houses-the’



Interactional construction grammar 

especially denna are still preferred (Fraurud 2000:€10–11, Lindström 2000:€109). In many varieties of spoken language, however, den här and den där are highly grammaticalized and seem to gain ground at the expense of the older forms denna and den5. In most varieties of Finland Swedish, the simple form denna is hardly used at all. In eastern Nyland dialects denna is used to some extent but is losing ground to the combined form den här. Both the simple form den and the combined form den där, however, are used frequently in the eastern Nyland dialects (for a quantitative comparison of different forms, see Wide 2008a). The pronoun den, which is prosodically more stressed than the definite article den, has lost most of its original dimension of distality in standard Swedish (Lindström 2000:€108; however, cf. the Swedish Academy Grammar (SAG) 2: 324). The compound pronoun den där is used almost exclusively for referring to entities further away from the speaker in a physical or psychological sense. While the form den has become highly grammaticalized as a definite article, it has at the same time lost ground as a demonstrative pronoun. This is the case in adnominal contexts in particular. Den can still be used deictically in similar contexts as the pronoun den där, e.g. Hela 'den (≈ den där) 'tunnan är 'tom ‘That whole barrel is empty’ (SAG 2: 322). However, when den is used for pointing out referents some kind of comparison is usually included, such as in example (6) below. In (6a) where the speaker refers to a referent in the physical surrounding by using a noun in the definite form (gård-en ‘farm-the’) only one referent of the kind is available. By contrast, in (6b) where the speaker refers to a referent by using den and a noun in the definite form (den gård-en ‘that farm-the’), there are several referents of the same kind to choose from. (6) a. Det var 'gården hon ville köpa. it was farm-the she wanted buy ‘It was the farm she wanted to buy.’ b. Det var 'den 'gården hon ville köpa. it was that farm-the she wanted buy ‘It was that farm she wanted to buy.’

(SAG 2: 323)

In standard Swedish adnominal den typically occurs in anaphoric contexts when the referent has been mentioned explicitly or implicitly in the immediately preceding context. An example of the latter case when the noun phrase with den is used discourse-deictically to refer to an event is given in (7) below. (7) Han gick till Canossa. Den vandringen tog två veckor. he walked to Canossa that hike took two weeks ‘He walked to Canossa. That walking tour took two weeks.’ (SAG 2: 325)

 Camilla Wide

In examples (6)–(7), the pronoun den is stressed along with the noun. Similar occurrences of the eastern Nyland pronoun te/de can also be found. However, in expressions such as the ones shown in (3)–(5) above te/de is not stressed (de ‘bisin ‘that/the ‘man’, te ‘Alda ‘that/the ‘Alda’, te ‘joordbruuki ‘that/the farm’). Prosodically, the constructs of the eastern Nyland construction thus differ from syntactically similar structures with the pronoun den in standard Swedish. In the following section, I will show that this difference in form is also accompanied by a difference in function. To summarize, the eastern Nyland demonstrative construction differs from standard Swedish in the following ways. The pronoun te/de is unstressed, which makes it prosodically reminiscent of the preposed definite article den in standard varieties. Furthermore, te/de is used with bare nouns in the definite form, which suggests that the preposed article has grammaticalized further in the eastern Nyland dialects than in standard Swedish. In examples such (3) and (4), only a noun in the definite form would be used in standard varieties. However, the eastern Nyland construction is also used in contexts, such as in example (5), where structures with other demonstrative pronouns, for example den där ‘that’, would be used in standard Swedish. The construction can therefore not be explained merely as a further development of the definiteness marking system. 5. Functions in interaction The main purpose of the examples I have discussed so far has been to show that the eastern Nyland demonstrative construction is used in contexts where the pronoun/article den is not used in standard Swedish. In this section I will focus on describing the construction from an “inside” perspective, that is, on how it is used by native speakers in authentic extracts from spoken language. I will argue that the use of the construction is highly dependent on shared knowledge. This knowledge makes it possible to use the construction in contexts that appear strange from a standard Swedish perspective. My main data consist of three everyday conversations, one institutional conversation and four dialect recordings (in total 5 hours of transcribed data, for details see Wide 2008a). In addition to this, I have also collected some examples by natural observation (cf. example (4) above) and analyzed examples presented in older literature (cf. examples (3) and (5) above). The dialect recordings in my main data are not “authentic” in the sense that they would have taken place irrespective of whether the recording was made or not. Two of the dialect recordings are reminiscent of conversations; the interviewer holds a low profile and the informants talk primarily to each other about their everyday life and old times. The other two recordings are more interview-like since the informants are, among other things,



Interactional construction grammar 

asked to talk about a film they have seen. Nonetheless, both of these recordings are comparatively informal and do not follow an explicit agenda. My method of analysis is Conversation Analysis. I have analyzed a large number of examples qualitatively in order to find typical uses of the eastern Nyland construction. Due to limited space as well as the scope of this paper, it is not possible to present as many examples as is usually done in CA studies. Hence, I will show only a few examples of the typical features I have found among the totally 155 examples of the eastern Nyland construction in my main data. In line with the theoretical and methodological background assumptions of CA I will not present any quantitative evidence for my generalizations about the construction. Extensive qualitative evidence for the generalizations presented in this paper can be found in Wide (2008a, 2008b). In order to describe and characterize the situations where the eastern Nyland construction is used, it is necessary to consider more context in the examples than was done in (3)–(5) above. Let us start by looking at example (8) from a dialect interview. The speaker in the extract is talking about a film she has seen. In this 7-minute long film (A strawberry-shaped day, Kangas et al.€2003; Östman 2008), a person is biking in the countryside in Finland and various incidents connected to strawberries take place. During the extract, the speaker refers to the bike in the film four times: with the expression me velå ‘by bike’ in line 1, with the expression me sen velå ‘with her bike’ in line 8, with the expression umpå velån ‘on bicycle-the’ in line 13 and with the expression te velån ‘that/the bike-the’ in line 16. (In order to shorten the extract I have left out the backchanneling signals uttered by the researcher.) (8) A strawberry-shaped day (Dialect recording, Liljendal, 2003) 1 I: så foor hon ti postlåådån me velå å (.) å så fick then went she to mail-box-the by bike and and then got 2 hon någa koort å he va antaagligen någa såm she some card and it was likely something which 3 hon tykkt att va roolit för hon såå nu uut she thought that was funny because she looked prt out 4 (å tykk åm de nu sku ha vari) ((laughs)) å så to like if it prt should have been and then 5 när hon tåå trampa viidare (.) från påstlåådån when she then peddled further from mail-box-the ((4 lines omitted)) 6 så va däär een.ssh- ja kåmmär ihååg en väägsjylt then was there a s- I remember a road sign 7 me en stoorer joordgubb *å så* (.) å så (.) pajna with a big strawberry and then and then went

 Camilla Wide

8 hon ivääg ååt te hålle me sen velå å så va she off to that way-the with her bike and then was 9 däär enn (.) kaar elär påjtji kva man nu ska sej there a man or boy what one prt shall say 10 såm rusta til ett stånd.hh (.) å- men så å who prepared to a stand and but then and 11 sku sälj joordgubbar intended sell strawberries ((2 lines omitted)) 12 en låådo me joordgubbar såm hon ladd i [...] kårjen a box with strawberries which she put in basket-the 13 umpå velån å så hadd ja no ett liitet avbrått on bike-the and then had I prt a little break 14 imilan att tankana måst no ha vandra between that thougths-the must prt have wandered 15 för att ja så hadd ja *så ja kåmär ihåg* så va cause that I then had I as I remember then was 16 att te velån låå ikull å å joordgubbana [...] that that/the bike-the lay over and and strawberries-the

‘then she went to the mail box by bike and then she got a card or something and it was probably something she thought was funny because she seemed to like it, if it would have been, and then she drove away from the mail box... then there was a, I remember a sign with a big strawberry, and then and then she drove in that direction with her bike and then there was a man or boy, or whatever you should say, who was putting up a stand, but then, and was about to sell strawberries... a box with strawberries which she put in... the basket on her bike, and then I had a break, my thoughts must have wandered because then I had, as I remember, that the bike was on the ground and the strawberries...’

The extract shown in (8) can be characterized as a narrative. In the sequence following the extract, the speaker finishes her story about the film by telling the interviewer about a little strawberry-shaped fan that the main character in the film picks up from a small box. The construct of the eastern Nyland construction, te velån ‘that/the bike-the’, thus occurs towards the end of a narrative. When the speaker says that the bike had fallen on the ground in line 16, she has already talked about the bike three times. The bike is hence an activated referent that has received a certain position in the common ground built up by the speaker and the listener. In this particular example, the bike has been mentioned explicitly. The



Interactional construction grammar 

reference could thus be characterized as anaphoric. However, when the special demonstrative construction occurs the bike has not been mentioned for a while in the narrative. In standard Swedish, neither a noun in the definite form (cykeln ‘bike-the’) nor an expression with the demonstrative pronoun den (den cykeln ‘that bike-the’) would therefore be used here. (In standard Swedish, the pronoun den is typically used to refer to entities or events mentioned in the immediately preceding context. See example (7) above.) An expression with the pronoun den där would be used instead, since this demonstrative pronoun is typically used in standard Swedish to reactivate referents that have not been mentioned for a while (Lindström 2000:€117–119; SAG 2: 321)). One of the main characteristics of the eastern Nyland demonstrative construction in narratives seems to be that it refers to entities that have already received a certain status in the common ground of the speakers and listeners. At the same time, the referents in question are clearly prominent in the narratives. The bike in extract (8) is shown in almost all scenes of the film. It is the bike that takes the main character through the main plot of the film. Hence, the bike plays a foregrounded role. In the particular scene that the speaker describes in line 16 in extract (8), the bike has turned over and the strawberries that were in a basket on the bike are smashed on the ground. The speaker describes this scene as a climax in the film; she directs attention to and focuses on the accident with the strawberries. Rather than just referring to it as a chronological event in the narrative, she “slows down” her story and points out that the event is special.€As seen in the extract, the narrative line has broken down just before the event with the strawberries on the ground is presented; in lines 13–15, the speaker makes several meta-comments about how her thoughts must have wandered when she was watching the film. In extract (8) the eastern Nyland te/de is thus not merely used as a pleonastic definite article in order to add an (unspecified) emphatic sense to the event described, as sometimes suggested in the literature (cf. Reinhammar 1975:€ 79). Neither is it used to point out a bike in a contrastive sense, as a structure with den would do in standard Swedish. Rather, it is used in order to refer to an entity that has been made prominent in the discourse and in order to bring a dramatic climax into focus. In narratives, constructs of the eastern Nyland demonstrative construction tend to be found towards the end when a dramatic climax is being described or built up, such as in example (8) (for further examples, see Wide 2008b). However, instances licensed by the construction can also be found in the beginnings of stories when a referent from the preceding discourse is being reintroduced. In example (9), two informants are talking about cats in their vicinity. This topic has been introduced by M earlier in the conversation. In the sequence shown in (9), K is talking about how she lets the cats eat the remains when she is gutting herring (lines 4–8). She then starts to talk about how the waste disposal is taken care of in

 Camilla Wide

the area where she and M live (lines 10–14). However, M wants to continue to talk about cats and reintroduces the topic in line 15 by using, among other things, the eastern Nyland demonstrative construction: men just täär ennu åm- åm dåm kattåna just ‘but still about about those cats’. (9) About cats (Dialect recording, Pernå, 2003) 1 K: fast vi int sjölv haar någå kattår så vi haar though we not ourselves have any cats then we have 2 åtta kattår i kring ås såm all andra haar eigth cats in around us like all others have 3 M: jåo (0.4) ºjåoº yes yes 4 K: at så täär när ja (0.4) rensar strömingana (.) [...] that so there when I gut herrings-the ((three lines omitted)) 5 så leggär ja uut täär å (0.4) så fåår dåm då it e then put I out there and then may they then eat it 6 iblaand it(är) åm de int rååkar kåm någån sometimes eat(s) if it not happens come any 7 kattå så (0.9) så fåår tåh (0.4) sjååråna (.) *åm int cat then then may then magpies-the if not 8 aadär (.) it opp tåm rääså[na* others eat up those rests-the 9 M: [men (---) ] but (---) 10 K: så bihööv(är) man] then needs one 11 int sjikk dåm ti Bårgå e de int [ti Bårgå] not send them to Borgå is it not to Borgå 12 M: [jåo ] yes 13 K: vi sjikkar (0.5) aavfalle våårt nu (0.6) we send waste-the ours now 14 e de int så is it not so ((four turns about waste disposal omitted)) 15 M: men just täär ennu åm- åm dåm kattåna but just there still about about those/the cats-the 16 just de här me våår eh Fäärdinand (.) den (1.1) precisely this with our Ferdinand it



Interactional construction grammar 

17 18

när vi far i- (0.6) i bastån så e an diit å when we go i- in sauna-the then is he there and (.) så näär [...] then when



‘K: even though we do not have any cats ourselves we have eight cats around us as everybody else has



M: yes yes



K: so when I gut herrings... I put out [the remains] and then they may eat it, sometimes they [the remains] are eaten, if there is no cat around, the magpies, if there is no one else, may eat those remains M: but (unhearable, overlapping speech) K: so one does not need to send them to Borgå, is it not to Borgå M: yes M: we send our waste these days? is it not so? M: but still about those cats, precisely this with our Ferdinand, it, when we go to the sauna, he is there, and then when...’



In standard Swedish, structures with the pronoun den där ‘that’ (or den här ‘this’) are used in contexts where a topic is reintroduced (Lindström 2000:€117). As we have seen, structures with the pronoun den ‘that’, by contrast, are used to refer to referents in the immediately preceding context, not to activate referents that are more distant. In the eastern Nyland dialects, structures with te/de ‘that’ as well as structures with de där/te tär lit. ‘that there’ can be used to activate distant referents (see Wide 2008a). Whereas structures with de där/te tär may be used to activate referents that are potentially not salient in the situation in a similar way to structures with den där in standard Swedish (cf. SAG 2: 321), structures with te/de specifically re-activate referents that are salient, such as in extract (9). In other words, when structures such as dåm kattåna ‘those/the cats-the’ are used to activate referents, these referents have already been discussed to some extent or are otherwise well known to the participants in the conversation. In examples (8) and (9), constructs of the eastern Nyland construction are used in anaphoric contexts to direct the focus to familiar or special entities that have been mentioned explicitly in the preceding context. In everyday conversations among people who know each other well, instantiations of the construction can often be found in first mentions, that is, when the referents have not been mentioned or discussed earlier in the conversation. This is the case in line 5 in extract (10) below, which is from a dinner conversation. E talks about a car that he and his son (speaker L) have been working on together. Before the sequence in the extract starts, the participants of the conversation have been discussing whether E, who is fatally ill, should travel abroad to visit his sister during his son’s winter

 Camilla Wide

holidays. E, however, also wants to finish the car for his son L. When he suggests, in lines 4–6, that he and L would use L’s vacation for working on the car instead, he is directing the attention to something that needs to be solved or taken care of, that is, a “problem”. Even though he has said in the context preceding the sequence shown in the extract that he would really like to visit his sister abroad, he seems unsure of whether he has time to do it because of the car. (10) Fixing the car (Dinner conversation, Lovisa, 2004) 1 E: de lååtär braa it sounds good 2 (0.9) 3 E: ja saa noo ååt mamma häär redan een aftå att I said prt to mum here already one evening that 4 (0.7) åm man i stelle sku åfra de vekkån if one in stead would sacrifice the week-the 5 Lasse e leedi så sku man laag de biiln Lasse is on vacation then would one fix that/the car-the 6 i sjikk hit å [(---) ] in order here and 7 M: [nå nää] nu ska [ni fa ti Irland] prt no prt shall you go to Ireland 8 L: [men ja haar ] but I have 9 tvåå vekkår (0.4) leedit sidu (0.4) ja e- ja kan two weeks free prt I am I can 10 [taa de (andra vekkan)] take the other week-the 11 M: [nää nu e de ] väärt å faar(a) ti Irland= no prt is it worth to go to Ireland 12 E: =joo [joo e så ] yes yes is so 13 L: [(---) ja tyk]kär de å (.) åm du årkkar bara (---) I think it also if you are able prt ‘E: that sounds good E: I said to mum here one evening that what if one would sacrifice the week Lasse is on vacation instead, then we could fix that car here and (---) L: but no you should go to Ireland E: but I am on vacation two weeks, I can use the other week L: no it is worth it to go to Ireland



Interactional construction grammar 



E: yes, yes it is L: I think so, too, if you just have the strength to do it’

The sequence shown in extract (10) can be described as a negotiation. E puts forth a competing solution to how he and L should spend L’s winter holidays. As seen in lines 7–13, however, his son L and M, who is his daughter-in-law, reject this solution. When E refers to the car in line 5 in extract (10), it is the first time it is mentioned in the conversation. He is not pointing out a car that can be seen from the dinner table. Neither is he contrasting the car to any other cars. Yet, it is completely evident to all the participants in the conversation what car E is talking about. E and his son have been working on an old Volvo for quite a while and are occupied with doing this on the weekend when the conversation takes place. The car in question is thus highly salient in the situation, the main object of an ongoing activity that not only E and L, but also M are involved in. When E mentions the car in line 5, he thus brings into focus an entity that is already activated, even though it has not been mentioned in the preceding conversation. Rather than bringing into focus a distant referent, he thus connects the ongoing discussion to an ongoing activity in a wider context (cf. Lindström 2000:€113). In the sequence following extract (10), the car is not discussed further. Hence, the car does not become topicalized in the same way as the cats in extract (9). During the following three minutes when the participants continue to talk about the trip to Ireland and then move on to discuss coffee making, two additional comments are made about the car. In both cases the speakers, L and E, respectively, refer to the car by using the eastern Nyland demonstrative construction. As in example (10), these comments are neither preceded nor followed by any longer discussion about the car. The main function of the eastern Nyland construction in example (10) thus seems to be to relate the ongoing discussion to a wider context that the participants share. This function is typical of the construction more generally (for further examples, see Wide 2008a, 2008b). However, the context in question must either be common to the participants of the conversation due to some shared activities or experience, as in (10), or made common to the participants through the conversation itself as in (8). In extracts (8)–(10), instances of the eastern Nyland construction appear in contexts where structures with the distal demonstrative pronoun den där ‘that’ would be used in standard Swedish. However, constructs of the construction may also occur in contexts where the proximal pronoun den här ‘this’ would be used. When the construct shown in example (11), de tyyge ‘that/the material’, was uttered, the speaker was holding up the sleeve of his jacket to the listener. Before he made the utterance, he had been talking about the new jacket he was wearing and telling the listener how fantastic it was. The jacket had in other words been made known and prominent in the preceding context.

 Camilla Wide

(11)

The fantastic jacket (Natural observation, 2005) de e så starkt de tyyge it is so strong that/the material-the ‘this material is so strong’

Constructs of the eastern Nyland demonstrative construction hence show similarities to structures with both the pronouns den där ‘that’ and den här ‘this’ in standard Swedish. The constructs are used for focusing on known or prominent referents in the communicative situation. This suggests that the te/de element in the construction is not to be analyzed as a definite article but rather as a demonstrative pronoun. Let us therefore now shortly return to examples (3) and (4), where instantiations of the construction occur in contexts that are typical of structures with definite articles rather than structures with demonstrative pronouns. In both examples (3) and (4), the referent is unique. In example (3), which is from a folk tale, the referent bisin ‘the man’ is introduced with a noun in the indefinite form (ein bisi ‘a man’) and then referred to twice with a construct of the eastern Nyland construction (te bisin ‘that/the man’). Example (4) thus resembles schoolbook examples of how the definite form of nouns is used in Swedish: first, you introduce a referent with a noun in the indefinite form and then you refer anaphorically to this referent with the noun in the definite form. This suggests that the te/de element in example (3) could be seen as a definite article rather than as a demonstrative pronoun. However, in the folk tale in question, not all referents are talked about in this way. As in the other examples discussed, the eastern Nyland construction is used to focus or point out a referent which is special or prominent, that is, plays an important or major role in the situation, in this case a narrative. In example (4), the referent is unique in a broader sense. The speaker uses the expression de klåkkan lit. ‘that/the clock-the’ to ask what time is it. This could be seen as a case of larger situation use, which is a typical use of definite articles rather than demonstrative pronouns (Hawkins 1978:€ 115–123; Himmelmann 1996:€210). However, the expression was uttered in a situation when a clock that had been turned off during the night needed to be set. The speaker was standing next to the clock and addressed the other persons present in the situation (his family). He was facing a problem that needed to be solved: the clock had to be set and he did not know the exact time. As the speaker in example (10), who talks about the car that needs to be fixed, the speaker in example (11) was thus directing the listeners’ attention to something that needed to be taken care of, a problem. Thus, if examples (3) and (4) are analyzed in isolation, it is possible to categorize the te/de element as a definite article. When these examples are investigated in a wider context, however, it turns out that the structures with te/de are used in order to perform specific discourse functions. They direct the attention to a refer-



Interactional construction grammar 

ent which is (made) prominent in the situation. I therefore see no reason to draw a distinction between occurrences such as examples (3)–(4) and (8)–(10) as regards the te/de element. Based on the communicative functions that the structures with te/de and a noun in the definite form acquire in their contexts, it is thus possible to conclude that te/de functions as a demonstrative pronoun rather than as a definite article. Primarily, however, te/de is part of the particular syntactic pattern that I have called the eastern Nyland demonstrative construction. It is the combination of te/de with a noun in the definite form with certain types of contextual features that creates the special meaning and function of the construction. To summarize the discussion in this section, the entities referred to with the eastern Nyland construction have to be part of the shared knowledge of the participants of the conversation. Either the entities must have been made known and prominent through a narrative or story as in example (8) or through an ongoing activity as in examples (4) and (10). Alternatively, the entities are prominent in discourse on a more global scale than the ongoing conversation or in some nonverbal activity that the participants of the conversation are involved in. Furthermore, the entities referred to are unique or self-evident either in the situation, as in example (10), or more generally, as in example (4); no aspect of contrast as in the use of den in standard Swedish (cf. example (6) above) can be discerned. The function of the eastern Nyland construction on a global level seems to be to activate referents from larger contexts, that is, to anchor the ongoing conversation to discussions or activities outside the conversation. On a more local level the construction tends to occur either in situations where a point is being made or prepared for, or in situations where tasks need to be checked or performed. In the following section, I will discuss how these features can be included in a formalization of the construction. As I have pointed out above, I find it necessary to include interactional and contextual features in the formalization of constructions in order to present an adequate picture of language and language use in CxG. 6. Formalization The CxG notation is based on boxes-within-boxes diagrams that show the constituent structure. Within the boxes, detailed grammatical information is provided in terms of feature structures, and co-indexication is used to mark the unification principles, that is, to show how constructions and specifications ‘fit together’ in a non-derivational fashion (see e.g., Michaelis & Lambrecht 1996:€225–228; Fried & Östman 2004:€25–29). The eastern Nyland construction is a phrasal construction, and its constituents may be represented with a box diagram as shown in Figure 1.

 Camilla Wide

The eastern Nyland demonstrative construction #1 role p-determiner lfm te/de/tåm/dåm stress null

role cat val stress

head noun {#1, #2} primary

#2 role cat

s-determiner [suffix of definitiness]

Figure 1.╇ The constituents of the eastern Nyland construction

In the diagram in Figure 1, the noun is marked as the head of the construction. The obligatory complements of the noun are represented with valence relations. The unification indices #1 and #2 show that the noun is necessarily modified by a preposed (p-) determiner and a suffixed (s-) determiner (the omission of the sdeterminer with proper names is licensed by a special lexical construction for proper names). The p-determiner is specified in terms of lexical forms (lfm) since it is represented by a special set of demonstrative pronouns that are included in the construction. The suffixed determiner, by contrast, is specified as a category (cat) since it relates to a more general pattern in Swedish nominal inflection. Finally, the values ‘null’ and ‘primary’ of the prosodic attribute stress for the p-determiner and noun, respectively, mark that the forms te/de/tåm/dåm are unstressed in the construction and the main prosodic stress is on the noun. The formalization shown in Figure 1 specifies the syntactic form of the eastern Nyland construction. However, it tells us nothing about the meaning or function of the construction. Since constructions are constellations of form and meaning/ function, this side of the construction ought by default to be taken into account in the formalization. As discussed in Sections 3–5, the eastern Nyland construction performs communicative tasks taken care of by several different constructions in standard Swedish. Hence, the functional side of the construction cannot be left to implicit shared knowledge about standard language use. When faced with the construction as a non-native speaker of eastern Nyland dialects, one typically tries to define the construction in terms of standard Swedish distinctions. However, dialects form language systems of their own and should not be described in terms of other varieties which might have a prestige status for political or socio-historical reasons. Similar syntactic forms in different varieties may have different functions and thus represent different types of constructions. Hence, in order for us to be able to explain the eastern Nyland construction properly, the communicative function of it has to be explicated and included in the formalization. However, grammatical constructions do not usually have just one single communicative function. As discussed in Section 2, they rather have a functional



Interactional construction grammar 

potential, which specifies how they can be used in order to perform communicative tasks (Wide 2002; Linell 2004, 2005). Since grammatical constructions get their final meaning or function in situated interaction, it is often difficult to analyze and pinpoint exactly how they perform certain communicative tasks. Formalizing communicative and interactional features in a neat and concise way is thus in many ways problematic (cf. Linell 2006, this volume). In order to explore the communicative-functional aspects of grammatical constructions more thoroughly than has been done so far, it is necessary to begin with tentative solutions. In Figure 2 below, I present a solution for how the eastern Nyland construction can be represented (cf. Auer 2006; Imo 2006 on the formalization of interactional features). To start with, the functional aspects are represented in the outermost box, which means that they are features of the construction as a whole. In order to capture the special combination of functional features that the eastern Nyland construction represents, I have used the attributes sequential position, reference, activity, coherence and functional potential. Since my analysis is preliminary and has to be worked out in more detail, the values of the features are presented as text within square brackets and single quotes (cf. Fried & Östman 2004:€30). The attribute sequential position is presented first in the formalization. As discussed, sequentiality plays an important role in IL. My analysis of the eastern Nyland construction so far has shown that the construction tends to occur at the boundaries of episodes, for example, at the beginnings and ends of narratives. The eastern Nyland demonstrative construction sequential position reference activity coherence functional potential

#1 role p-determiner lfm te/de/tåm/dåm stress null

[‘beginning/end of episodes, inserted comments’] [‘shared, specific knowledge, non-contrastiveness, discourse focus’] [‘narrative, global on-going discourse, talk about shared non-linguistic activity’] [‘anchoring to larger (on-going) context’] [‘creates focus on a familiar prominent referent to make a point in a narrative, check or solve a task in the situation, remind listeners of or bring into focus larger contexts’] role cat val stress

head noun {#1, #2} primary

#2 role cat

s-determiner [suffix of definitiness]

Figure 2.╇ The form and communicative function of the eastern Nyland construction

 Camilla Wide

In addition to this, the construction occurs in isolated comments referring to larger discourses than the ongoing conversation (cf. example (10) above); in order to capture this feature of the construction I have included the value ‘inserted comments’ in the sequential position attribute. In the remainder of this section, I will discuss the other communicative-functional attributes included in Figure 2 in more detail. I will start by comparing my set of attributes to some concepts of nominal reference used in earlier studies on CxG, specifically the concepts ‘topic’/‘focus’ and ‘active’/ ‘semi-active’/‘inactive’. Michaelis & Lambrecht (1996:€226–227) introduce two pragmatic values into the CxG formalization: act (activation) and prag (pragmatic role). They state that the attribute act “in those boxes which represent referential arguments denotes the activation status of the NP referent, that is the assumed status of this referent in the consciousness of the addressee at the time of utterance”. Thus, act can take two other attributes as its value, active and accessible. Accessible referents are those which are identifiable but not fully active, and yet recoverable from context. An active referent, by contrast, is “also necessarily an accessible referent, one that is currently ‘lit up’ in the shared consciousness of the speaker and addressee”. Michaelis & Lambrecht point out that the activation attribute should be distinguished from the pragmatic role attribute, which has two possible terminal values, ‘topic’ and ‘focus’. A constituent is a topic expression, according to Lambrecht (1994:€131), if “the proposition expressed by the clause with which it is associated is pragmatically construed as being about the referent of this constituent”. Focus, on the other hand, is “the semantic component of a pragmatically structured proposition whereby the assertion differs from the presupposition” (Lambrecht 1994:€213; see also Lambrecht 2004:€162–164). The attributes active and accessible discussed by Michaelis & Lambrecht (1996) are largely based on Wallace Chafe’s studies on the activation state of concepts. According to Chafe (1987), particular concepts at particular times may be in one of three different states: ‘active’, ‘semi-active’ or ‘inactive’. Chafe’s definition of an active concept is basically the same as Michaelis and Lambrecht’s definition of active referents, i.e. “a person’s focus of consciousness”. A semi-active concept is “one that is in a person’s peripheral consciousness, a concept of which a person has a background awareness, but which is not being directly focused on” and an inactive concept is “one that is currently in a person’s long-term memory, neither focally nor peripherally active” (Chafe 1987:€25). Active concepts have been mentioned in the previous context or have a naturally given status in the conversational context, such as ‘I’ and ‘you’. Hence, they correspond roughly to ‘given information’ and are typically spoken with attenuated pronunciation. Semiactive concepts, in turn, correspond to ‘accessible information’ (p. 26–28). Chafe (p. 29) points out that concepts do not remain active very long unless their



Interactional construction grammar 

activation is refreshed. Nonetheless, concepts do not become fully inactive immediately, rather they “may remain in the semi-active state for some time”. Could the eastern Nyland construction be described or explained in terms of the features ‘topic’/‘focus’ and ‘active’/‘semi-active’/‘inactive’? To start with, these concepts are problematic per se since they focus on rather small units of language. Michaelis & Lambrecht (1996) analyze sentences and Chafe (1987) investigates the representation of information in narratives from an intonation unit perspective. The values ‘topic’ and ‘focus’ are especially difficult to apply in my analysis, which aims at combining CxG with IL without sacrificing any important tenets of IL. In some of the examples presented in sections 3–5, such as (3) and (8), the instantiations of the eastern Nyland construction do refer to someone or something being talked about, that is, they have the function of being a topical expression. However, this most likely is a general feature of demonstrative constructions; they explicitly pick out referents to be discussed. In examples such as (5) and (10), the situation is not as straightforward. The first occurrence of the eastern Nyland construction in example (5) occurs in a left-dislocation, and could indeed be described as a topical expression (cf. Lambrecht 1994:€ 181–183). The second occurrence, however, can be found to the far right in the sentence, that is, in the position where new information is typically presented in Swedish. This occurrence of the construction is thus focal rather than topical.€What unites the uses of the two expressions in (5) is not focality or topicality in Lambrecht’s sense. Neither can the use of the construction be explained with reference to the notion of ‘discourse topic’, which Ochs, Keenan & Schieffelin (1976:€338) define as “the proposition (or sets of propositions) about which the speaker is either providing or requesting new information”. As example (5) shows, the eastern Nyland construction can be used both for referring to propositions (the person Alda) and the information provided (‘wanted [to go] away from the farm’). What unites the two expressions in (5) is that they refer to and bring into focus shared background knowledge of the speaker and addressee. This is also the case in example (10). Here the car, which is being referred to by the eastern Nyland demonstrative construction, is introduced without having been mentioned previously. It is thus difficult to see how the car could be topical on the sentence level. As the expression te ‘joordbruuki ‘that farm-the’ in example (5), the expression de ‘biiln ‘that car-the’ in example (10) occurs to the far right, rather than to the left in the sentence. It thereby introduces new information both on the discourse and on the sentence level and is focal rather than topical.€However, as discussed in Section 5, the car is part of a common activity of the speakers and listeners in example (10). Thus, the car could be seen as a given referent from a broader discourse point of view. By focusing on the car, the speaker directs the attention of the listeners to a problem which everybody knows that needs to be fixed.

 Camilla Wide

Investigating the topical and focal features of the eastern Nyland construction on a sentence or intonation unit level in more detail would be most interesting. However, my analysis of the construction so far indicates that the special use of the construction is not explainable on this level of language. Rather, the construction must be investigated in broader contexts, including contexts outside the ongoing conversation. As shown in examples (8)–(10), instantiations of the construction are used in interaction in order to direct the listeners’ attention to certain referents. In some cases, such as in example (9) about the cats, these referents are clearly topical.€However, in examples such as (8) about the bike, the use of the construction does not refer to the topic of the narrative, but rather to an important “contracting party”. Furthermore, one of the main purposes of the construction seems to be to anchor an ongoing discussion to a broader context, such as in example (10) about the car. In order to capture the information structure features of the eastern Nyland construction it is thus necessary to use concepts such as activity and coherence in the formalization. By ‘activity’ I refer to all types of social activities, that is, what Couch (1986:€ 113) describes with the following words: “Human beings routinely contact, interact with and leave each other’s presence; in the process they initiate, construct and terminate various forms of social activity”. Following Levinson (1979) who uses the term ‘activity type’ to refer to any culturally recognized activity, I thus include both linguistic and non-linguistic activities in the concept. For the concept of coherence, on the other hand, I rely on the hermeneutic view as expressed for example by Bublitz (1998): Coherence is the outcome of the language user’s gestalt creating power. People are driven by a strong desire to identify forms, relations, connections which they can maximize in order to turn fragments into whole gestalts, i.e. to ‘see’ coherence in strings of utterances. Coherence is also a cooperative achievement (in ongoing discourse more than in ‘petrified’ text) because it depends on both the speaker’s/ writer’s and the hearer’s/reader’s willingness to negotiate coherence. Mutual understanding not only rests on the participants sharing the same socio-cultural background, the same range of knowledge and communicative assumptions, but also on their ability to figure out unshared experience, i.e. to adjust their own world-view to that of their interlocutors’. Hearers/readers are constantly engaged in trying to re-create coherence as an equivalent of the speaker’s/writer’s coherence, but despite their efforts they can never succeed in coming up with an exact copy. Coherence is only approximate and a matter of degree and best described as a scalar notion. Any interpretation of coherence is restricted and, accordingly, partial to different degrees.

In Figure 2 above, I have given the eastern Nyland construction the coherence value ‘anchor to larger (ongoing) context’. The activity value is stated as ‘narrative, global ongoing discourse, talk about shared non-linguistic activity’. The purpose of



Interactional construction grammar 

both values is to pinpoint the function of the eastern Nyland construction as explicitly as possible. However, the values differ from the values presented in most CxG studies since they are not seen in relation to a fixed set of values. Due to the IL perspective of my study, I find it necessary to treat the list of values as open (cf. the discussion in Section 2). Let us now turn to the attribute reference. As shown in Figure 2 I have given the eastern Nyland construction the reference value ‘shared, specific knowledge, non-contrastiveness, discourse focus6’. These values are based on the results of the empirical analysis presented in Sections 3–5. Would it have been possible to use the concepts of ‘active’, ‘semi-active’ and ‘in-active’ here instead? As I have shown, the pronoun den ‘that’ is unstressed in the eastern Nyland construction. The prosody of the demonstrative element of the construction thus corresponds to Chafe’s (1987) characterization of active referents. The noun in the construction, by contrast, is stressed, which indicates that the construction does not refer to active referents. Furthermore, since the construction contains a demonstrative pronoun and a noun, it is not an entity which is typically used for referring to active referents (this is usually done with pronouns and similar expressions, cf. Chafe 1987:€26–28). As discussed, a general feature of the eastern Nyland construction is that it refers to ongoing activities of different kinds. Thus, the construction can hardly be seen as referring to inactive referents. What remains is the value ‘semi-active’, which refers to accessible information. Expressions with demonstrative pronouns fit nicely into this category because they are often used to activate referents that are known but have not been discussed for a while. According to Chafe (1987:€29), concepts become semi-active in two ways: by not having been talked about for a while and by having been implicitly introduced through a narrative schema. Thus, expressions used for semi-active concepts cover a great variety of different constructions, such as nouns in the definite or indefinite form, possessive constructions and demonstrative constructions. As regards the eastern Nyland dialects, this means that ‘semi-active’ is a feature not only of the demonstrative construction discussed in this study but also of other demonstrative constructions. My argument is therefore that concepts such as ‘active’ and ‘accessible’ have to broken down into more specific features. By using the attributes activity, reference and coherence it is possible to pinpoint the special characteristics of the eastern Nyland construction in a more precise way than by using concepts such as ‘topicality’ and ‘activation’, which were originally developed for purposes other than for describing demonstrative constructions per se. The values I have presented in Figure 2 above can be seen as contextual constraints on the eastern Nyland construction. However, in order to bring together these communicative-functional features of the construction, the functional potential of the construction needs to be explicated. In Figure 2, I

 Camilla Wide

have specified this potential as follows: ‘creates a focus on a familiar prominent referent to make a point in a narrative, check or solve a task in the situation, remind the listeners of or bring into focus larger contexts’. As with the other features presented in Figure 2, the value is tentative. It will probably be possible to describe it in more detail when the communicative-functional aspects of demonstrative and other determination constructions have been worked out more generally for eastern Nyland dialects, Swedish and other languages. Even then, the value will probably have to be stated as a rather free textual description rather than as a selection from a limited set of possibilities. As discussed in Section 2, the functional potential of grammatical constructions specifies in what way the constructions can be used for performing communicative tasks. The concept of potentiality in itself implies that the functional contents cannot be described with a finite set of values. To summarize, in this section I have presented a tentative solution for how the contextual features of the eastern Nyland demonstrative construction can be formalized in a CxG fashion. In my solution, I have followed the basic principles of the CxG notation. However, due to my aim to combine CxG with Interactional Linguistics, I have introduced some new concepts to the notations. These concepts of sequential position, reference, activity, coherence and functional potential do not just differ from pragmatic concepts used in previous studies on CxG. They also differ in the way they are treated in the formalism, that is, as features which partly overlap with each other. 7. Conclusion The purpose of this paper has been to discuss a demonstrative construction in a variety of spoken Swedish (eastern Nyland dialects in Finland) in order to explore how Construction Grammar could be combined with Interactional Linguistics. My main aim of the study has been to show that it is possible to combine these frameworks. However, when communicative-functional aspects are added to CxG formalizations, this has to be done in accordance with the findings and principles of IL. Hence, I have introduced some concepts that are used in interactionally oriented studies (sequential position, reference, activity, coherence, functional potential) to describe the eastern Nyland construction. Some of these concepts overlap each other and they generally cannot be presented in binary or explicit terms in the way syntactic features of constructions are typically presented. Nonetheless, these features describe the eastern Nyland construction in a more precise way than concepts such as ‘topic’/‘focus’ and ‘active’/‘semi-active’/‘non-active’, which do not distinguish the construction sufficiently from other determinative constructions. Even though the concepts I have used and the way they are presented



Interactional construction grammar 

in the formalism obviously have to be studied and refined further, they open up new possibilities to explore the functional sides of demonstrative constructions in a more detailed way. Notes 1 This article is based on a paper presented at IPrA, Riva del Garda, Italy, in July 2005. For valuable comments on earlier drafts of the article, I thank Jan-Ola Östman, Mona Forsskåhl, Benjamin Lyngfelt, Diana ben-Aaron and the anonymous referees. 2 The study is a contribution to the project Svenskan i Finland. Syntaktiska drag i ett jämförande perspektiv ‘Swedish in Finland. Syntactic features in a comparative perspective’ at the Society for Swedish literature in Finland 2003–2006 (see Lindström et. al. 2002; Wide & Lyngfelt 2009). 3 The main differences between spoken and written standard Swedish in the use of definite articles and suffixes are the following. The neuter singular form det is generally pronounced [de] and the plural forms de (nom.) and dem (obl.) are pronounced [dom] (or in some varieties [di] in the nominative case and [dem] in the oblique case). 4 Standard Swedish allows double definiteness marking in noun phrases where a modifier (consisting of a subordinate clause) does not precede but rather follows the noun, e.g., det (viktiga) villkoret att räntan skulle vara betald före årsskiftet ‘the (important) condition-the that interest-the should be paid before year turn-the’, den tunnan som står i garaget ‘the barrel-the that is in garage-the’ (the Swedish Academy Grammar (SAG) 2: 301, 303). This is also the case in eastern Nyland dialects. 5 However, denna is used commonly in spoken language in southern and south-western Sweden; in contrast to what is the case in standard Swedish the pronoun is here also combined with nouns in the definite form (denna bok-en ‘this book-the’). 6 With ‘discourse focus’ I do not mean that the information is new. Rather, the term describes that the speaker “zooms” in on the referent in question (cf. Lindström 2000:€119) in order to structure the discourse and create coherence between different layers of contextual knowledge.

References: Anward, Jan (1999). Interaction and Constructions. Paper presented at workshop in Helsinki, September 1999. Revised version to appear in J.-O. Östman & M. Fried (Eds.), The SocioPragmatics of Frame Semantics and Construction Grammar. Amsterdam & Philadelphia, PA: John Benjamins. Auer, Peter (2006). Construction Grammar meets Conversation: Einige Überlegungen am Beispiel von „so“-Konstruktionen. In S. Günthner & W. Imo (Eds.), Konstruktionen in der Interaktion (291–314). Berlin: Mouton de Gruyter. Bublitz, Wolfram (1998). Cohesion and Coherence. In J. Verschueren, J.-O. Östman, J. Blommaert & C. Bulcaen (Eds), Handbook of Pragmatics 1998 (1–15). Amsterdam & Philadelphia, PA: John Benjamins.

 Camilla Wide Chafe, Wallace (1987). Cognitive constraints on information flow. In R.S. Tomlin (Ed.), Coherence and Grounding in Discourse (21–51). Amsterdam & Philadelphia, PA: John Benjamins. Couch, Carl (1986). Elementary forms of social activity. In C. Couch, S. Saxton & M. Katovich (Eds.), Studies in Symbolic Interaction: The Iowa School. Parts A and B (113–129). Greenwich: JAI Press. Couper-Kuhlen, Elizabeth & Margret Selting (2001). Introducing Interactional Linguistics. In M. Selting & E. Couper-Kuhlen (Eds.), Studies in Interactional Linguistics (1–21). Amsterdam & Philadelphia, PA: John Benjamins. Diessel, Holger (1999). Demonstratives: Form, Function, and Grammaticalization. Amsterdam & Philadelphia, PA: John Benjamins. Fillmore, Charles J. (1988). The Mechanisms of “Construction Grammar”. BLS, 14, 35–55. Fillmore, Charles J. (1989). Grammatical Construction Theory and the Familiar Dichotomies. In R. Dietrich & C.F. Graumann (Eds.), Language Processing in Social Context (17–38). Amsterdam: North-Holland/Elsevier. Fillmore, Charles J. (1996). The pragmatics of constructions. In D.I. Slobin, J. Gerhardt, A. Kyratzis & J. Guo (Eds.), Social Interaction, Social Context, and Language. Essays in Honor of Susan Ervin-Tripp (53–69). Mahwah, NJ: Lawrence Erlbaum. FO = Ordbok över Finlands svenska folkmål. 1976–. Helsingfors: Forskningscentralen för de inhemska språken. Fraurud, Kari (2000). Demonstrativer i svensk sakprosa. In DENNA – DEN HÄR – DEN DÄR. Om demonstrativer i tvärspråklig belysning. En minnesskrift till Elsie Wijk-Andersson [ASLA INFORMATION, 26:2 2000] (5–28). Uppsala: Universitetstryckeriet. Fried, Mirjam & Jan-Ola Östman (2004). Construction Grammar: A thumbnail sketch. In M. Fried & J.-O. Östman (Eds.), Construction Grammar in a Cross-Language Perspective (11– 86). Amsterdam & Philadelphia, PA: John Benjamins. Fried, Mirjam & Jan-Ola Östman (2005). Construction Grammar and spoken language: The case of pragmatic particles. Journal of Pragmatics, 37, 1752–1778. Fujii, Seiko (2004). Lexically (un)filled constructional schemes and construction types: The case of Japanese modal conditional constructions. In M. Fried & J.-O. Östman (Eds.), Construction Grammar in a Cross-Language Perspective (121–155). Amsterdam & Philadelphia, PA: John Benjamins. Günthner, Susanne & Wolfgang Imo (Eds.) (2006). Konstruktionen in der Interaktion. Berlin: Mouton de Gruyter. Hawkins, John A. (1978). Definiteness and Indefiniteness: A Study in Reference and Grammaticality Prediction. London: Croom Helm. Himmelmann, Nikolaus P. (1996). Demonstratives in Narrative Discourse: A Taxonomy of Universal Uses. In B. Fox (Ed.), Studies in Anaphora (205–254). Amsterdam & Philadelphia, PA: John Benjamins. Hirvonen, Ilkka (1987). Konstruktionstyperna denne man och denne mannen i svenskan: En språkhistorisk undersökning. Helsingfors: Svenska litteratursällskapet i Finland. Imo, Wolfgang (2006). „Da hat des kleine glaub irgendwas angestellt“ – ein construct ohne construction? In S. Günthner & W. Imo (Eds.), Konstruktionen in der Interaktion (263–290). Berlin: Mouton de Gruyter. Imo, Wolfgang (2007). Construction Grammar und Gesprochene-Sprache-Forschung: Konstruktionen mit zehn matrixsatzfähigen Verben im gesprochenen Deutsch. Tübingen: Max Niemeyer.



Interactional construction grammar  Kangas, Kea, Jan-Ola Östman, Leila Mattfolk & Åsa Mickwitz (2003). A strawberry-shaped day. Paper presented at The 8th International Pragmatics Conference (IPrA), Toronto, 2003. Kay, Paul (1995). Construction Grammar. In J. Verschueren, J.-O. Östman & J. Blommaert (Eds.), Handbook of Pragmatics. Manual (171–177). Amsterdam & Philadelphia, PA: John Benjamins. Lambrecht, Knud (1994). Information structure and sentence form. Topic, focus and the mental representations of discourse referents. Cambridge: Cambridge University Press. Lambrecht, Knud (2004). On the interaction of information structure and formal structure in constructions: The case of French right-detached comme-N. In M. Fried & J.-O. Östman (Eds.), Construction Grammar in a Cross-Language Perspective (157–199). Amsterdam & Philadelphia, PA: John Benjamins. Langacker, Ronald W. (2008). Cognitive Grammar. A Basic Introduction. New York: Oxford University Press. Levinson, Stephen C. (1979). Activity types and language. Linguistics, 17–5/6 [219/220], 365–399. Lindström, Eva (2000). Some uses of demonstratives in spoken Swedish. In S.P. Botley & T. McEnery (Eds.), Corpus-based and Computational Approaches to Discourse Anaphora (107–128). Amsterdam & Philadelphia, PA: John Benjamins. Lindström, Jan, Anne-Marie Londen, Marianne Nordman, Marketta Sundman & Jan-Ola Östman (2002). Svenskan i Finland. Syntaktiska drag i ett jämförande perspektiv. Project plan. Helsingfors: Svenska litteratursällskapet i Finland. Linell, Per (1998). Approaching Dialogue: Talk, interaction and contexts in dialogical perspectives. Amsterdam & Philadelphia, PA: John Benjamins. Linell, Per (2004). On Some Principles of a Dialogical Grammar. In K. Aijmer (Ed.), Dialogue Analysis VIII: Understanding and Misunderstanding in Dialogue. Selected Papers from the 8th IADA Conference. Göteborg 2001 (7–23). Tübingen: Max Niemeyer. Linell, Per (2005). En dialogisk grammatik? In J. Anward & B. Nordberg (Eds.), Samtal och grammatik: Studier i svenskt samtalsspråk (231–328). Lund: Studentlitteratur. Linell, Per (2006). Can We Formalise the Linguistic Resources of Interactional Language? In M. Suominen, A. Arppe, A. Airola, O. Heinämäki, M. Miestamo, U. Määttä, J. Niemi, K.K. Pitkänen & K. Sinnemäki (Eds.), A Man of Measure: Festschrift in Honour of Fred Karlsson on his 60th Birthday. SKY Journal of Linguistics, 19, Special Supplement, 56–65. Lundström, Gudrun (1939). Studier i nyländsk syntax. Stockholm: Norstedt. Michaelis, Laura A. & Knud Lambrecht (1996). Toward a construction-based theory of language function: the case of nominal extraposition. Language, 72, 215–247. NFOS = Nyländska folksagor och -sägner, Vol. VI, 1896. Ordnade av A. Allardt & S. Perklén. Helsingfors: Tidnings- och tryckeriaktiebolagets tryckeri. Norén, Kerstin & Per Linell (2007). Meaning potentials and the interaction between lexis and contexts: An empirical substantiation. Pragmatics, 17, 387–416. Ochs Keenan, Elinor & Bambi B. Schieffelin (1976). Topic as a discourse notion: a study of topic in the conversations of children and adults. In C.N. Li (Ed.), Subject and topic (335– 384). New York: Academic Press. Östman, Jan-Ola (2008). Expedition Liljendal.€ In H. Palmén, C. Sandström & J.-O. Östman (Eds.), Dialekt i östra Nyland: Fältarbete i Liljendal med omnejd [Nordica Helsingiensia] (9–25). University of Helsinki: Department of Scandinavian Languages and Literature. Östman, Jan-Ola & Mirjam Fried (Eds.) (2005). Construction Grammars. Cognitive grounding and theoretical extensions. Amsterdam & Philadelphia, PA: John Benjamins.

 Camilla Wide Ono Tsuyoshi & Sandra A. Thompson (1995). What can conversation tell us about syntax? In P.W. Davis (Ed.), Alternative Linguistics. Descriptive and Theoretical Modes (213–271). Amsterdam & Philadelphia, PA: John Benjamins. Reinhammar, Vidar (1975). Pronomenstudier. Uppsala: Almqvist & Wiksell. SAG = Svenska Akademiens grammatik. Ulf Teleman, Staffan Hellberg & Erik Andersson, 1999. Stockholm: Svenska Akademien (Norstedts distr.). SAOB = Ordbok över svenska språket, utg. av Svenska Akademien. 1893–. Lund. (Available at http://g3.spraakdata.gu.se/saob/.) Schegloff, Emanuel A., Elinor Ochs & Sandra A. Thompson (1996). Introduction. In E. Ochs, E.A. Schegloff & S.A. Thompson (Eds.), Interaction and grammar (1–51). Cambridge, New York & Melbourne: Cambridge University Press. Steensig, Jakob (2001). Sprog i virkeligheden: Bidrag til en interaktionel lingvistik. Århus: Aarhus Universitetsforlag. Wide, Camilla (2002). Perfect in Dialogue. Form and functional potential of the vera búinn að + inf. construction in contemporary Icelandic [PIC Monographs, 3]. University of Helsinki: Department of Scandinavian Languages and Literature. Wide, Camilla (2008a). Bruket av demonstrativa pronomen och bestämd form i östnyländska samtal och intervjuer. In J. Lindström (Ed.), Språk och interaktion 1 [Nordica Helsingiensia 10] (107–149). University of Helsinki: Department of Scandinavian Languages and Literature. Wide, Camilla (2008b). En diskursfokuskonstruktion i östnyländska. In H. Palmén, C. Sandström & J.-O. Östman (Eds.), Dialekt i östra Nyland Fältarbete i Liljendal med omnejd [Nordica Helsingiensia 14] (53–109). University of Helsinki: Department of Scandinavian Languages and Literature. Wide, Camilla & Benjamin Lyngfelt (Eds.) (2009). Konstruktioner i finlandssvensk syntax: Skriftspråk, samtal och dialekter [Skrifter utgivna av Svenska litteratursällskapet i Finland, 716]. Helsingfors: Svenska litteratursällskapet i Finland.

Transcription conventions word word *word* ºwordº .word wor- [ ] = .hh (.) (0.9) (word)

expression discussed in the analysis emphasis laughter in speaker’s voice soft voice word uttered on inbreath speaker interrupts himself overlapping speech starts overlapping speech ends turns are latched audible inbreath short pause timed pause (seconds) speech which is difficult to decipher



Interactional construction grammar 

(---) [...] ((laughs)) prt

inaudible speech speech omitted transcriber’s comments discourse particle

part 3

Context and grammar

Contextual cues for particle placement Multiplicity, motivation, modeling Bert Cappelle

This chapter looks into a variety of contextual factors that play a role in a well-described case of grammatical variation: the particle placement alternation in transitive phrasal verbs (e.g., to pick up the book / to pick the book up). Most of the factors that linguists have argued to be of relevance to this alternation are phonological, morpho-syntactic, semantic or discourse-functional in nature (cf.€Gries 2003). I am not concerned here with re-investigating all known factors. Rather, what I intend to do is (i) to single out those factors, both familiar and less familiar ones, that can be construed as involving, one way or another, the context in which the phrasal verb occurs, (ii) to formulate the further motivation(s) behind these factors, and (iii) to briefly consider how these factors might be integrated into a constructionist model of the variation. The main emphasis lies on my first aim. The text is organized as follows. In Section€1, I introduce the ordering alternation at hand, as well as two important (and related) conditions on the morphosyntactic form which the direct object noun phrase (NP) should take for the alternation to be possible: the object should not be an unaccented pronoun nor should it be syntactically heavy – in sum, it should have moderate weight. Section€2 links these two related constraints to a first, often-mentioned contextual factor, namely the discourse-familiarity of the object NP, which has to do with whether that object refers to something that has already been mentioned or evoked in the preceding linguistic context (in which case that object will often be realized as an unaccented pronoun) or whether it introduces a new referent (in which case its identification may require lengthy and complex modifications of the noun). The subsequent sections deal with other, less well studied or less obviously context-related factors. In Section€3, I will deal with a factor that could easily be (and often has been) confused with discourse-familiarity, namely focus structure. This factor is concerned, simply put, with whether or not a clause element provides prominent information, where ‘prominent’ can only be evaluated, of course, in terms of contextual relevance and importance. Section€ 4 presents a

 Bert Cappelle

recently-discovered contextual factor: the presence of an instance of the transitive verb-particle construction in one ordering or the other in the immediately preceding linguistic context and its persistence effect on the word order of a subsequent occurrence of that construction. Section€5 looks into linguistic context in its widest sense: the impact of register and region. Section€6 is devoted to cues for particle placement coming from the context taken in a more extra-linguistic sense: here we look at how the choice for one or the other constructional alternative may be influenced by whether or not the referent of the direct object NP and the action it undergoes are ‘expected’, given certain frame-based knowledge. Sections 7 and 8, then, deal with factors that might not seem to be related to the notion of context in any straightforward way: whether or not the verb-particle construction is stative, and the degree to which the verb-particle combination is semantically compositional, respectively. I will ask (and answer) the question how viable it is to reformulate these factors as contextual ones. Section 9 provides some explanation of why the factors exert the influence that they do. Section€10, finally, reflects on how the impact of these contextual factors on the particle placement variation can be modeled in a constructionist framework. 1. Particle placement: The phenomenon 1.1

One alternation, two conditions

The particle of a transitive phrasal verb can occur on either side of the object noun phrase. The examples given in (1) illustrate this placement variation: (1) a. They messed up the show. (‘joined’ ordering) b. They messed the show up.€(‘split’ ordering) This variation is constrained by two conditions on the object NP, namely, (i) that it not be an unaccented pronoun and (ii) that it not be too heavy. If the object NP is an unaccented personal pronoun, the particle has to follow it; if it is decidedly heavy (more on which below), the particle has to precede it: (2) a. *They messed up it. b. They messed it up. c. They messed up everything we had all worked towards these last few months. d. *They messed everything we had all worked towards these last few months up.



Contextual cues for particle placement 

Whether a pronoun is accented or not is easy enough to ascertain in any given context, even when the mode of communication is written rather than spoken. In written discourse, different readers will normally reach a consensus as to whether the pronoun receives a “silent” accent. The pronoun only receives an accent if it is focused on, but this is a rather exceptional case for pronominal objects. Also, there is no gradience in ‘accentedness’ of an object pronoun. Either the pronominal object is accented (i.e. is focused on) or it is not (i.e. is not focused on). There are no in-between cases.1 In the rare case that a pronominal object receives an accent, it can follow the particle (e.g.€Bolinger 1971:€39–40; Fraser 1976:€17; Aarts 1989:€286), but it does not have to, as we will see in Section€3.2. In the much more common case that the pronominal object is unaccented, it obligatorily has to precede the particle. There simply is no other option. By contrast, the end-placement of heavy objects is not a categorical constraint but a tendency. That is, the requirement that particles precede heavy object NPs can only be expressed as a statistical generalization, not as an absolute rule, because there is only an increasing preference for end-placement of the object NP as its weight increases. Since the weight of the object NP (unlike the presence or absence of focal accent) is a matter of degree, we can only try to determine what can count as a sufficiently heavy NP for us to predict with a reasonable degree of confidence that the split ordering will be ruled out. Moreover, it is not immediately clear how we should define ‘weight’. Do we have to look at crude factors like number of words or syllables (i.e. length) or is weight to be defined relative to the internal constituency of the NP (i.e. complexity)? Or is there even another parameter that should be taken into account if we talk about the weight of a constituent? 1.2

Weight: A matter of length or a matter of complexity?

The earliest known formulation of the general principle of end-weight (Quirk et al. 1972) probably dates back to Otto Behaghel’s (1909/1910) “Gesetz der Wachsenden Glieder” (‘law of increasing elements’), of which he later gave the following concise formulation: Von zwei Gliedern von verschiedenem Umfang steht das umfangreichere nach. (Behaghel 1930, cited in Arnold et al. 2000:€29) (‘Of two constituents of different size, the larger one follows [the smaller one].’)

Interestingly, Dik (1997 [1978]: 404) mentions Behaghels law of increasing elements as the first statement of his own “principle of increasing complexity”, notwithstanding that Behaghel’s law clearly refers to size (i.e. length) and not to constituent-internal syntactic complexity. This goes to show that length and complexity are often treated as interchangeable parameters (cf.€Wasow 1997 for

 Bert Cappelle

disciussions of this point). There can be no misunderstanding, though, that length and complexity constitute different definitions of the concept ‘weight’. So, the question is: Which parameter defines weight best – length (i.e. number of, say, words) or complexity (i.e. internal syntactic intricacy of the NP caused by the presence of, e.g., a postmodifier)? Length apparently does play a role in constituent ordering. Charting the choice of particle-final placement against length of the object NP (measured in words) in a large set of corpus examples, Lohse et al. (2004:€243) observe a first marked drop in the choice of this ordering (from 40% to 18%) if the NP-length increases from two to three words, and a second marked drop (to a negligible 3%) if the NP-length reaches 5 words or over.2 Yet, as Lohse et al. further point out, these marked drops may in fact also correlate with increasing internal complexity of the object NP, since NPs of length three have more chance of containing a modifying adjective (besides a determiner) and since NPs of length five or over have more chance of containing postnominal modifiers like prepositional phrases or relative clauses. The idea that what actually determines particle placement is not so much the length of the object NP as its phrase-structural complexity has been suggested by various scholars, including Chomsky (1964 [1961]; 1975), Fraser (1966), and Ross (1986 [1967]).3 Complexity is usually taken to involve the presence of an embedded clause. Thus, while the object NP in (3a) is longer – both in number of words and in number of syllables – than the object NP in (3b), the latter, containing a short relative clause, is clearly less grammatical in the split ordering: (3) a. I called almost all of the men from Boston up. b. *I called the man you met up. (Chomsky 1964 [1961]: 127, n. 18) This contrastive pair of sentences seems very convincing. Surprisingly enough, though, corpus studies so far have not been able to confirm that it is indeed complexity and not sheer length that may prevent an object NP from occurring in the split construction. Gries’s (2003) book-length study is aimed at dealing with just this kind of issue: finding out which of the factors ever proposed in the literature genuinely influence particle placement, and which of them are ‘parasitical’, in the sense that their values usually co-occur with the values of these genuine factors without themselves having an independent impact. If I interpret Gries’s (2003:€110) results correctly, it looks as though complexity does not exert an influence on particle placement distinct from length. At least, his results appear to be rather equivocal. On the one hand, non-complex NPs (defined by him as either pronominal ones or as NPs consisting of only a head and possibly a determiner) and NPs with intermediate complexity (defined by him as NPs with an adjective, NPs with an of-PP or co-ordinated NPs) could both be shown to favor the split ordering



Contextual cues for particle placement 

independently of their length. On the other hand, complex NPs (defined as NPs with a finite or nonfinite postmodifying clause) apparently did not, independently of their length, prefer the joined ordering. I presume this non-finding is caused by the non-occurrence of complex object NPs that weren’t also considerably long, as in (3b), in his set of corpus examples, which contains fewer than 500 sentences. Similarly indeterminate results are reported by Wasow (2002:€32–40), whose experimental and corpus findings could neither corroborate nor reject the abovementioned claim (e.g.€ Chomsky (1964 [1961]) that it is complexity rather than length of the object NP that influences particle placement. 1.3

Two conditions for the price of one?

Given that pronouns are short and non-complex, we may capture a generalization if we merge the two conditions on the form of the object NP, mentioned in Section€1.1, into one single condition. Thus, (4) is the formulation of a more general constraint on particle placement:

(4) For the choice as to the placement of the particle to be in principle available, the object NP should have moderate weight.

By stating that the NP should have moderate weight, we at once exclude heavy NPs and unaccented pronouns from the set of object NPs that are allowed in the two constructions. However, object NPs consisting of a single, one-syllable, non-pronominal word can still enter into the alternation (e.g.€He sure knows how to pick {up girls / girls up}). This means that our question whether weight might involve more than just length and/or complexity (cf.€the end of Section€1.1) may not have been unwarranted. For an object NP to be light, it should not only have minimal length and complexity, but it should also be phonologically reduced. What distinguishes the set of unaccented pronominal objects from maximally light non-pronominal NPs is their lack of phonological strength. That is, their initial consonant can be dropped (in the case of him, her and them) and/or their vowel can be shortened (in the case of me and you) or even neutralized to a schwa (in the case of you, her, us and them), or they can be reduced to their final (syllabified) consonant (in the case of her and them).4 That phonological strength may indeed also play a role in particle placement seems to be confirmed by the fact that a pronominal object NP which receives an accent (thereby rendering impossible its phonological reduction – *pick ’em up) can occur to the right as well as to the left of the particle, as we have already pointed out above (compare He sure knows how to pick {*up ’em / ’em up} and He sure knows how to pick {up them / them up}).

 Bert Cappelle

1.4

Lexical restrictions

It should be borne in mind that object NPs of moderate weight cannot always occur in the two constructional variants because there are some verb-particle combinations which happen to be lexically constrained to have only one truly conventionalized manifestation, either the joined (V – Prt – NP) or the split (V – NP – Prt) ordering. For example, blow off ‘release’, drum up ‘obtain’, give off ‘emit’, hold out ‘foster (a feeling) in the face of hardship’, kick up ‘cause (trouble)’ and the full expression take up arms ‘become involved in a conflict’, among others, only seem to allow the joined ordering: (5)

a. b. c. d. e. f.

We all need to blow {off steam / *steam off} from time to time. It won’t be easy to drum {up support / *support up}. Do you like candles that give {off a smell / *a smell off}? Families of the missing continue to hold {out hope / *hope out}. If it happens again, I’m gonna kick {up a fuss / *a fuss up}. The peasants are threatening to take {up arms / *arms up}.

Conversely, the object NP always seems to occur between the verb and the particle in combinations like ask over ‘invite at one’s house’, push around ‘be bossy towards’, see through ‘bring to a happy conclusion’, tell apart ‘distinguish’, and in full expressions like push the boat out ‘celebrate regardless of expense’, laugh/work/… one’s ass/butt/socks off ‘laugh/work excessively’ and the popular humorous taunt {[Someone],} eat your heart out {[someone]}!: (6)

a. b. c. d. e. f.

It’s been ages since we invited {*over the neighbors / the neighbors over}. I don’t like the way he pushes {*around my mom / my mom around}. You can depend on us to see {*through the job / the job through}. She really pushed {*out the boat / the boat out} for her fortieth birthday. I’ve been working {*off my ass / my ass off} all day. Van Gogh, eat {*out your heart / your heart out}! (implies that the speaker or someone known to the speaker can do something better than the (absent) addressee, who is widely acclaimed for excelling in the relevant field, in this case that of painting)

Combinations like those illustrated in (5) and (6) merely represent the extremes on a continuum of word order preferences which lexicalized verb-particle combinations can have (cf.€Biber et al. 1999:€933; Gries and Stefanowitsch 2004:€110–113; Szmrecsanyi 2005:€130 ff.). Less extreme cases are, for example, find out ‘discover’ and pluck up ‘muster’, which usually pattern like the combinations in (5) and, towards the other extreme, combinations like answer back ‘reply or interrupt



Contextual cues for particle placement 

impudently when being scolded’ and get a word in edgewise ‘find an opportunity to say something’, which usually pattern like those in (6): (7)

a. b. c. d.

They will find {out the truth /??the truth out} someday. I must pluck {up my courage /??my courage up} and tell her I love her. In my days, we never answered {??back our parents / our parents back}. When he talked, nobody else could get {??in a word / a word in} edgewise.

Around the middle of the continuum are combinations that have a less clear preference for one ordering or the other, like beat up, hand over, shoot down, try out, etc. Note, finally, that when the object NP is an unaccented pronoun, it still has to precede the particle of a phrasal verb situated at the extreme represented by the combinations in (5): (8) a. The “angry youth” were allowed to blow off steam. Well they blew it off, alright, and stabbed a visiting Australian yeshiva student named Yankel Rosenbaum to death. (www) (cp.€*… blew off it) b. Still, many people find that if they want support for coping with a rare disease, they have to drum it up themselves. (www) (cp.€*… drum up it)

Apparently, the ‘grammatical rule’ requiring the particle to follow an unaccented pronoun always takes precedence over any ordering convention on the level of individual verb-particle combinations. Farrell (2005:€130) takes this as evidence that unaccented pronominal objects ought to be distinguished from other kinds of objects in that they attach to the right of verbs as clitics or as some kind of inflectional affixes (more or less as in French, where unaccented object pronouns attach to the left of the finite verb, e.g.€M’as-tu vu?, Me-have you seen?, i.e. ‘Have you seen me?’). We might therefore say that this cliticization/affixation mechanism, if that is what it is, applies across the board, staying blind to whichever fixed order of the particle vis-à-vis the object NP is otherwise observed with particular phrasal verbs.5 1.5

Summary of Section€1

To sum up so far, there is a general ordering variation in English between the nonverbal element (the ‘particle’) of transitive phrasal verbs and its direct object NP. In the ‘joined’ ordering, the verb and the particle are in adjacent positions, with the direct object NP occurring to the right of the particle (e.g.€stick out your tongue); in the ‘split’ ordering, the verb and the particle are separated from each other by the intervening direct object NP (e.g.€stick your tongue out).

 Bert Cappelle

If the direct object NP is extremely light, only the split ordering is possible; if it is very heavy, only the joined ordering is possible. A direct object is extremely light if it consists of a single syllable which is phonologically reducible, i.e. whose initial consonant can be dropped and/or whose vowel can be shortened and/or further neutralized. In practice, we can say that the direct object NP is extremely light if and only if it is an unaccented pronoun. A direct object is very heavy if it contains an embedded clause. In practice, such an object will also be quite long, and a length of five words or over has proven to be by itself a sufficiently accurate predictor for the choice of the joined ordering. If the direct object NP has moderate weight (i.e. if it neither is an unaccented pronoun nor has a length of more than four words), it can in principle occur before or after the particle. Individual verb-particle combinations, though, may still favor one ordering or the other. Still, even combinations with a very strong preference for the joined ordering cannot undo the ‘rule’ that unaccented pronouns must precede the particle. 2. Discourse-familiarity of the object NP 2.1

‘Old’ before ‘new’: Terminological specification

Discourse consists of both content which the speaker takes to be old or known to the hearer and content which he or she assumes to be new or unknown to the hearer. In constructing a sentence, the speaker ordinarily puts the informational content which is supposed to be old before that which is supposed to be new. ‘Old’ and ‘new’ can be interpreted in a variety of ways (cf., e.g., Prince 1981), for example as contextually given vs. contextually new, as ‘addressee-old’ vs. ‘addressee-new’, as ‘topic’ (what the sentence is about) versus ‘comment’ (what is said about the topic), as ‘theme’ versus ‘rheme’ (a distinction featuring prominently in the ‘Prague School’), or as ‘background’ / ‘presupposition’ versus ‘focus’ (cf.€Section€3 for the latter distinction). In this second section I am concerned with the first distinction mentioned, also known as the distinction between ‘discourse-old’ and ‘discoursenew’ information. Discourse-old information is information coded in parts of the sentence that refer (either verbatim, or, more often, via the use of anaphoric pronouns or other semantically related items – cf.€ Section€ 2.3 below) to entities, properties, propositions, etc. that have already been mentioned earlier in the discourse. Discourse-new information is conveyed by parts of the sentence (possibly all of it) that do not take up familiar elements from the preceding discourse.



Contextual cues for particle placement 

2.2

The relation between discourse-familiarity and weight

Discourse-familiarity usually correlates positively with weight, a fact that has not gone unnoticed by linguists. For example, Quirk et al. (1985:€1361) point out that “the new information often needs to be stated more fully than the given (that is, with a longer, ‘heavier’ structure)”, and Arnold et al. (2000:€34) give some further references to the generalization that “[i]tems which have been recently mentioned tend to be accessible to both speaker and hearer, and require less complex descriptions than items which are not”. Thus, because of their familiarity or lack thereof, respectively, discourse-old NPs can be referred to by means of short forms – most typically, of course, pronouns – while discourse-new NPs are often marked by the presence of modifications, appositions, etc. Now, given this correlation between weight and discourse-familiarity, Wasow (2002:€81) asks the obvious question: “[D]o weight and information status exert distinct influences on ordering, or is one just a side-effect of the other?” (cf.€also Arnold et al. 2000). That is: Do heavy constituents tend to be placed late in the sentence because they are usually new or do new constituents tend to be placed late in the sentence because they are usually heavy? Or, rather, is neither factor ‘parasitic’ on the other (cf.€again Section€1.2 for this imagery) and do they have independent influences on word order? Without going into this question in any depth, I confine myself to putting together two relevant facts known from the literature. First of all, the correlation between the length and/or complexity of clause constituents and the amount of new information they convey is not perfect. It is quite possible for long and complex constituents to convey discourse-old information and for short and simple ones to convey discourse-new information (Ward, Birner and Huddleston 2002:€1371). Secondly, weight and familiarity are independent word order paraÂ� meters. As a consequence, one parameter can be held constant to test the effect of the other parameter. This is the underlying method in Arnold et al.’s (2000) carefully conducted study (whose actual statistical method is more complicated than this text allows me to go into). Testing the individual effects of weight and familiarity on heavy NP shift and on the dative alternation (by means of corpus analyses and an elicitation experiment), Arnold et al. could confidently conclude that both factors significantly and independently contribute to end-placement. In other words, they revealed that the tendency of heavy constituents to occur at or towards the end of a clause is not epiphenomenal to (i.e. a side-effect of) their newness, and, vice versa, that the frequent end-placement of discourse-new constituents is not epiphenomenal to their weight.

 Bert Cappelle

2.3

Relevance of discourse-familiarity to particle placement

Gries’s (2003) study on particle placement supports Arnold et al.’s (2002) finding that discourse-familiarity is independent of weight. Thus, not only can the length (though less clearly the complexity) of the object NP be shown to be a certain, significant influence on its position vis-à-vis the particle (cf.€Section€1.2 above), but so can a variety of discourse-functional factors bearing on the discourse-familiarity of the object NP. More in particular, Gries proves the significance of such factors as the distance to the last mention of the object NP, the object NP’s ‘cohesiveness’ to the preceding discourse – that is, the degree to which it is evoked by the previous occurrence of coreferential items or of items that stand in a hyp(er) onymical (superordinate/subordinate) or in a meronymical (part/whole) relation with it – and the number of times the object NP has been mentioned in the preceding discourse (Gries 2003:€72–74; 89–92). By way of illustration, let us consider (9), the full context of which is cited and discussed by Gries (2003:€72–74). For lack of space, I only provide the actual sentence with the verb-particle combination.

(9) The lawyers took Valenzuela off to record his testimony. (BNC corpus, cited and discussed in Gries 2003:€72–74)

While the object noun in this example consists of no fewer than five syllables (Valen-zu-e-la) and hence should be placed after the particle on account of its considerable length, its mid-position is by far the most natural placement, since this proper noun refers to a discourse-familiar item, via its frequent anticipation in the form of pronouns (like he, his etc.) and via an alternative description (as a kidnapper) which cannot be mistaken to refer to the same person in the context concerned. Gries’s results also confirm Bock’s (1977), Chen’s (1986) and Peters’ (2001) earlier experiments on the role played by discourse-familiarity in the positioning of the direct object in transitive verb-particle combinations. In turn, Gries’s results have been corroborated by Szmrecsanyi’s (2005) findings. All these studies support, in part, what Erades (1961:€57), among others, already had intuited, namely that [t]he principle governing the place of the objects … is neither stress nor length nor rhythm, but something quite different: the news value which the idea denoted by the object has in the sentence. Objects denoting ideas that have news value, no matter whether they are nouns or pronouns, long or short, have end-position; those that have no such value come between verb and adverb. (Erades 1961:€57; cited in Bolinger 1971:€51)

Erades is not correct, however, in considering the news value of the object as the sole factor governing particle placement and, hence, in rejecting accent and length



Contextual cues for particle placement 

as possible co-determinants. Length does play a role, as we have seen, and so does accent, which is a symptom of focality, the factor to be discussed in Section€3. 2.4

Interim conclusion and transition

We have seen in Section€1 that unaccented pronouns are placed as close as possible to the verb, in other words, right after it and before the particle. This may not (just) be because unaccented pronouns are extremely light and therefore act as clitics but also because they convey information that is discourse-old or refer to elements that are otherwise familiar to the addressee in the speech situation (e.g.€me, you). Elements with such information status tend to occur early in the sentence, and the compulsory mid-positioning of unaccented pronouns in the verb-particle construction may be seen as a grammaticalization of this tendency. Conversely, long and heavily modified noun phrases are shifted as far as possible towards the end of the clause, in other words, after the particle, as we also stated in Section€1. Again, this is not (just) because such NPs have much weight but also because they typically refer to entities that are newly introduced into the discourse and because discourse-new constituents generally tend to be put towards the end. In the next section, we will discuss in considerable detail a factor that is often wrongly collapsed with discourse-familiarity, namely focus structure. Although discourse-new constituents usually do convey the most prominent and pertinent information in the clause (and in this sense contain ‘new’ information), it is nevertheless necessary to keep these two factors apart. 3. Focus structure As Erades’s (1961) quote given at the end of Section€2.3 insinuates, it is sometimes assumed that accent (i.e. stress, intonation) is a decisive factor in the placement of the particle and the object NP. This assumption had already been made by Poutsma (1928) and Kruisinga (1931/1932), but it is still voiced by present-day linguists. For example, Svenonius (1996:€52), commenting on native speakers’ preference for the joined construction with indefinite objects (e.g.€for I’ll send out flyers as opposed to I’ll send flyers out), states: “I suggest that intonation is actually what det[er]mines these preferences”. This view is repeated, more recently even, in Büring (2005:€25), where some further references are given (repeated here in his quote): “[…] object–particle order is preferred with unaccented objects, while particle–object order is preferred with accented objects (Chen, 1986; Dehé, [2000], [2001], 2002; Svenonius, 1996).”

 Bert Cappelle

Accented elements in a sentence signal ‘focal’ (or ‘focused’) information. This is the most prominent and pertinent information in a given context, as opposed to the ‘backgrounded’ or ‘presupposed’ information contained in what Ward, Birner and Huddleston (2002:€1370) refer to as the ‘focus-frame’. If speakers prefer to put an accented object after the particle, this is in accordance with a general principle by which focal elements are put at the end of sentences (the principle of end-focus – see, e.g., Quirk et al. 1985:€1356–1357). Now, a number of oversimplifications are being circulated when it comes to linking up particle placement with the role of sentence focus: i. Focal elements correspond to discourse-new elements: “It is natural to place focus pitch on an element which is informationally novel”. (Svenonius 1996:€52) ii. The nuclear accent of a sentence generally falls at the end: “It is also natural in English for there to be a rising pitch at the right edge of a simple sentence” (Svenonius 1996:€52). Hence, the natural place for a focal element, which is by definition marked by accentuation, is at the end of the sentence, so that the focal accent coincides with the natural sentence accent: “Accented items are usually sentence-finally, which corresponds to the natural order of English where end-focus is reserved for expressions denoting the focal referent of the utterance”. (Gries 2003:€39) iii. When uttering an active clause containing a transitive phrasal verb, the speaker has the choice to assign the focus to either the object NP or the particle. Depending on this simple choice, either the object NP or the particle is put at the end: “Whichever of the two constituents (the direct object or the particle) is to be emphasized, it will be positioned sentence-finally”. (Gries 2003:€52) In the following three sections, I will show that these at first sight straightforward rules overlook a number of complications. 3.1

Focal elements need not be discourse-new

Quirk et al.’s (1985:€ 1356–1357) principle of end-focus is defined as “the linear presentation from low to high information value”, where information value seems to be further formulated and illustrated in terms of discourse-familiarity: elements with low information value are those that are known from the preceding discourse (‘given’) and that can accordingly be pronominalized, while elements with high information value are unknown (‘new’) and require explicit expression with full lexical material (p.€1430–1431). The link between the notion of focus and that of discourse-familiarity is also established by Dehé (2002): The focused part of an utterance cannot be taken for granted at a corresponding time of utterance. It is not predictable and cannot be inferred/recovered from the



Contextual cues for particle placement 

preceding context or discourse situation. It is typically new information which has not been mentioned before. (Dehé 2002:€105)

But note that the use of typically in the last sentence is not an unwarranted hedge, since focal information is often but not necessarily discourse-new information. Consider the following short question-answer pair: (10) Did you prefer the movie or the book? — I preferred the book. The focus in the response sentence is the book. It is intonationally as well as informationally the most prominent element in the response. In a sense, it provides new information, since prior to the response, the first speaker did not yet know the specific value of the variable x in the focus-frame ‘I (= second speaker) preferred x’. Notwithstanding its status as focus and as ‘new information’, though, the NP the book in this second sentence is not discourse-new, since it has been mentioned in the preceding question. Thus, focus structure and discourse-familiarity are distinct concepts (Ward, Birner and Huddleston 2002:€1370). 3.2

Focal elements need not be sentence-final

The end of the sentence is generally acknowledged to be the prime focus position: “The focus typically appears at or towards the end of the clause” (Ward, Birner and Huddleston 2002:€1372). Granting this, we should not disregard the possibility of assigning any word or constituent of a clause the status of focus depending on the surrounding discourse, resulting in contrastive focus, a special case of marked focus (Quirk et al. 1985:€1365). This kind of focus being available for any element in any clause, it can also be applied, naturally, to the direct object NP in active clauses with a transitive verb-particle combination. Thus, it is perfectly possible for an object NP to occur in the split ordering and bear a marked focus accent. In other words, it need not be the case, contrary to what Diessel and Tomasello (2005:€91) write, that “[i]f the direct object carries stress accent the particle [precedes] it, even if the object is pronominal”, as in (11) Pick up HIM (not her).

(Diessel and Tomassello 2005:€91)

On the one hand, it is true that a pronominal object cannot occur sentence-finally (cf.€again example (2a) in Section€1.1) unless it is accented. Here is a similar, authentic, example: (12) Of all the dogs in the shelter, my Master picked out me!

(www)

 Bert Cappelle

In this example, the narrating dog (referred to by the pronominal object me) puts itself in contrast with all the other dogs in the shelter, and hence, me is contrastively focused. On the other hand, it is by no means obligatory to put a pronominal object, nor indeed any object NP (pronominal or not), at the end if it carries contrastive accent. This is illustrated by the following authentic examples (small capitals added): (13) a. Well, it was me who finally asked him out instead of the other way around. (www) b. …quitting cold turkey is not an option. The withdraw[a]l not only stresses the body out, but stresses the baby out as well. (www) The occurrence of a contrastively accented object in the split ordering is possible, as Dehé (2002) remarks, because “contrastive focus and the corresponding accent placement is not bound to a certain position. It can be assigned to every constituent in the sentence, and can therefore be realized at every structural position” (Dehé 2002:€172). Bolinger (1971:€51), too, discusses the possibility of mid-position of accented objects, giving examples like the following: (14) a. Let’s take our friénds over. b. They shot the whole pláce up.

(Bolinger 1971:€51)

It is important to note that Bolinger’s examples above could easily be uttered in circumstances in which the object, though accented, is not contrasted with another potential object in the context. This means that contrastive opposition is not a prerequisite for the occurrence of accented objects in the split ordering, a point to which we will return below, in Section€3.3 and, in more detail, in Section€6. The occurrence of an accented object in the split ordering may be thought to be in violation of the unmarked placement of the nuclear accent, which is sentence-final. Remember, though, our quote of Ward, Birner and Huddleston (2002:€1372) that “[t]he focus typically appears at or towards the end of the clause” (emphasis added). More specifically, the nuclear sentence accent characteristically falls on “the (stressed syllable of the) last open-class lexical item of the last clause element” (Quirk et al. 1985:€1357). So, given that particles are closed-class elements, it seems to me that examples like (14a-b) are in fact perfectly in keeping with this general prosodic principle. Note, for example, that sentence-final prepositions are also usually left unaccented (e.g.€What are you talking about?).6



Contextual cues for particle placement 

3.3

Other options than focus on either object or particle

In view of the fact that the particle placement alternation concerns the position of two elements relative to each other – the direct object and the particle – it is tempting to reduce the issue of the prosodic pattern of the verb phrase to focal accentuation on either of these two elements – see Van Dongen (1919:€331ff.) and Gries (2003:€24–25) for an early and recent example, respectively. Gries uses the following sets of question-answer pairs to illustrate the positional preferences (I underline the focused part in the answers): (15) a. What did he pick up? — He picked up the book. b. Where did he bring the book? — He brought the book back. (after Gries 2003:€25) Dehé (2002:€132) correctly observes that the domain of the focus can vary in size. That is, the focused part need not be restricted to the very word or smallest constituent containing the accented syllable. So, we could also have the following question-answer pairs. (16) a. What did he do? — He picked up the book. b. What happened? — He picked up the book.

(after Dehé 2002:€132)

Thus, in addition to the so-called ‘minimal focus’ on the object NP or on the particle, exemplified respectively in (15a) and (15b), we can have ‘non-minimal focus’ on the entire predicate and ‘maximal’ focus on the entire sentence, as in (16a) and (16b), respectively. Furthermore, Dehé considers the case in which the predicate minus the object NP is within the focus domain: (17) What did he do with the book? — He picked the book up. (after Dehé 2002:€132) According to Dehé, the focal scenario with a backgrounded (presupposed) object NP is the only one in which the split ordering is, and has to be, used.7,8 Dehé (2002) does not go into a case like (15b), where the focus is on the particle, but since such a sentence also has a backgrounded object NP, its obligatory use of the split ordering conforms to her claim. Dehé (2002) acknowledges, though, that the split ordering can also be used if the object NP is not backgrounded but (on the contrary) bears contrastive focus accent, as we remarked in Section€3.2. The joined ordering, still according to Dehé, is the default one, to be selected in all cases in which the object NP constitutes or is a part of the focus domain without being contrastively accented, as in (15a), (16a) and (16b).

 Bert Cappelle

This is not accurate, however. Consider again Bolinger’s (1971:€51) examples given in Section€3.2 above (in 14), repeated here in (18): (18) a. Let’s take our friénds over. b. They shot the whole pláce up.

(Bolinger 1971:€51)

Here, the object NP carries (presumably non-contrastive) accent but nonetheless occurs in the split ordering. I will come back to such examples in Section€6. The examples given so far in this section still do not exhaust all the focal possibilities. There are, in fact, four more cases, logically speaking. First, there can be minimal focus on the verb itself, namely when it is contrasted with another verb combining with the same particle: (19) a. Did he pick the book up? — No, it is more accurate to say that he heaved the book up.€It was a very heavy book, you know. b. He did not wash any of the dust off. He brushed it off… (www) In this case, the particle is preferably put in end-position (compare: Did he pick up the book? —?No, he truly heaved up the book). I suppose the reason for this is that a particle in the joined ordering is always accented to some degree. Accenting the verb and its immediately following particle produces a violation of the prosodic principle discussed by Gussenhoven (2004:€255–256) as the ‘NoClash’ constraint. To avoid two accents in adjacent positions, then, the speaker can either de-accent the particle in the joined ordering, which results in an unusual accent pattern for verb-particle combinations, or she can select the split ordering, which is often used for de-accented particles anyway (cf.€ again Bolinger’s examples above). In short, the split ordering allows the accent on the verb to stand out.9 Second, we may actually be dealing with two independent foci, one on the object NP and another on the particle. This possibility only arises with verbs that combine with more than one particle in a semantically similar way. (We therefore cannot use pick up: while we also have pick out, it is clear that the contribution of pick is not constant in these combinations.) Here is an example: (20) The next steps in the sequence must turn the red and amber lights off (www) before turning the green light on. Here again, the split ordering is preferred, but since the NoClash constraint is potentially violated, this preference must be for a different reason. A particle can only be contrasted with another particle if it makes an independent semantic contribution – independent of the verb, that is. This means, in effect, that it has some predicative power (over the object NP) of its own: in the example above, for example, the result of the actions involved is that the relevant lamps are off or on. Now,



Contextual cues for particle placement 

when both the object NP and the particle compete for end-position (which is the neutral position for focal elements), the particle will be preferred in this position, so that object and particle align themselves as in the grammatically very general subject-predicate construction.10,11 The final placement of the particle in (15b) may also be partly explained as inheriting the predicate-final ordering of clauses. Third, it may be the case that the verb and the object NP are two independent foci. Consider the following made-up example: (21) Moody Max was in a particularly destructive temper and decided that whatever made noise had to be turned off with violence. He smashed the TV off and shot the radio off. This situation is so rare that the preferred ordering cannot be deduced from corpus data. On the one hand, the joined ordering may be preferred because it keeps the focal elements, and hence their accented syllables, apart as far as possible. On the other hand, the particle in the verb-particle combination is always slightly accented in pre-object position, and its accent then potentially clashes both with the accent on the preceding verb and with the accent on the following noun. Keeping the particle de-accented in end-position resolves this clash. Elicitation experiments and judgment tasks would be needed to find out speakers’ preferences in this special focal scenario. Fourth, the final possibility is the trivial one in which neither the verb, nor the object NP nor again the particle falls within the focus domain: (22) How did he pick the book up? — He picked the book up with both hands. In this case, both orderings are in principle possible, but the ordering preferred is the one used in the question, as we will see in the next section (Section 4). Thus, if the question were How did he pick up the book?, then the answer would be more likely to repeat the joined ordering: He picked up the book with both hands. Although pleading against this ordering would be the discourse-old status of the object NP, this violation would be mitigated by the fact that the object-old NP would not appear in the canonical position for focal elements, which would be occupied by another element. 3.4

Summary of Section€3

In Table€1 below I summarize the different possibilities of assigning focal accentuation and the impact this may or may not have on the choice of the joined or the split construction.

 Bert Cappelle

Table 1.╇ The two orderings by focal element(s)

V V, Prt

V, Prt, NP V, NP

Prt

Prt, NP

Joined (V – Prt – NP)

Split (V – Prt – NP)

()  NoClash   OldObj-Left  Focus-Last  e.g. (16a–b) ()  NoClash

 e.g. (19a–b)

  OldObj-Left  Focus-Last  Subj-Pred   Subj-Pred

NP

 e.g. (15a)

other

  OldObj-Left

 e.g. (17)  e.g. (18a–b) ()  NoClash e.g. (21)  e.g. (15b)

  NoClash e.g. (20a–b)   Focus-Last e.g. (13a–b)  e.g. (22)

 = ‘the ordering on top is allowed under the focus structure on the left’ () = ‘the ordering on top is marginally allowed under the focus structure on the left’  = ‘the ordering on top is not allowed under the focus structure on the left’  = ‘the ordering on top violates the following principle under the focus structure on the left: …’ Focus-Last: The focus of a clause appears at the end of that clause. NoClash: Accented elements should not be in adjacent positions. Subj-Pred: When aligning a subject and its predicate, put the subject first. OldObj-Left: A discourse-old direct object NP is leftmost in the post-verbal field.

The table above also indicates how the different focus assignments may interact with some general prosodic and ordering principles to rule out one or the other particle placement construction. Note that selecting the split ordering generally yields least violations. It is important to remember from this section on focus structure that focal NPs can occur in mid-position. In Section€6, we will look in more detail at sentences



Contextual cues for particle placement 

in which this is the case. First, however, we will briefly report on a couple of studies into how the choice for the joined of the split construction is influenced by previous occurrences of one of these orderings (Section 4) and discuss the impact of factors like register and region (Section 5). 4. Previous occurrence of one ordering or the other It has been known for some time that speakers tend to reuse syntactic patterns that they have come across (either produced themselves or heard) in the preceding discourse. This tendency is known as ‘syntactic priming’ or ‘syntactic persistence’ – see Levelt and Kelter (1982), Weiner and Labov (1983), and Bock (1986) for some early studies. In this short section I confine myself to summing up the findings of two recent studies that discuss syntactic priming in relation with particle placement: Gries (2005) and Szmrecsanyi (2005). i. There is a general and strong tendency for speakers to stick to the same word order choice of the transitive verb-particle construction. That is, if they have encountered an instance of the joined construction in the preceding linguistic context, they will tend to repeat this construction when they subsequently utter another instance of the transitive verb-particle construction; likewise if they have first encountered an instance of the split construction, then this will tend to be the word order that they will use again for a next use of a transitive phrasal verb. ii. This persistence effect is somewhat stronger for a prior occurrence of an instance of the split ordering than for a prior occurrence of an instance of the joined ordering. iii. In written discourse, there is a stronger persistence effect between instances involving the same phrasal verb lemma (as in he’d fill all their bags up … he wouldn’t fill our bags up) than between instances involving different phrasal verb lemmas. iv. In dialogues, the second speaker often copies the word order along with the exact inflectional form and the lemma of the transitive phrasal verb used by the first speaker. v. For specific verb-particle combinations with a preference for one word order manifestation or the other (cf.€Section€1.4), there is only a marked persistence effect for the order they prefer anyhow. vi. The general priming effect decreases somewhat as the distance between the two instances increases. vii. Perhaps related to the previously-mentioned persistence factor, the longer the sentence in which the new instance occurs, the weaker the persistence effect from the previous instance on this new instance.

 Bert Cappelle

5. Lectal factors: Register and region This section looks into the wider context of the sentence in which a transitive verb-particle combination occurs: not just the preceding sentence(s) in the discourse but the entire setting in which the discourse is situated. Are we dealing with spoken or written discourse? Which regional variety of English does the speaker use? Which social background does the speaker have? Is the speaker a man or a woman? Such potential influences on linguistic variation can be grouped under what Geeraerts (2005) calls ‘lectal’ factors, a cover term for influences related to the speaker’s dialect, regiolect, sociolect, etc. Below I single out two such factors that are known to exert some influence on particle placement. 5.1

Written versus spoken discourse

As Gries (2003:€97; 2005:€381) and Cappelle (2006) have shown by means of research into corpora containing both spoken and written material, there is a significant preference for the split ordering in spoken discourse and a significant preference for the joined ordering in written discourse. Table€2 below presents the data for the British Component of the International Corpus of English (ICE-GB) given in Cappelle (2006): Table 2.╇ The two orderings by medium in ICE-GB

Joined (V – Prt – NP) Split (V – NP – Prt) Column totals

Spoken

Written

â•⁄ 731 (± 42%) â•⁄ 917 (± 58%) 1648 (100%)

543 (± 71%) 227 (± 29%) 770 (100%)

According to Bolinger (1971:€57, fn. 8), the split ordering is rather informal (“at the familiar end of the speech-level or register scale”). Hence, language users may feel more restrained to select this word order option in written discourse than they do in spoken discourse. Written language is, after all, much more of an ‘edited’ medium, allowing users to replace an option they might naturally have selected in spoken discourse by an option that is a little more formal. This inhibition against using an all too informal construction may also account, still according to Bolinger, for the fact that the split ordering occurred even more rarely in pre-twentieth century literature, which was relatively more censured than current literature. Other reasons why the split ordering is used less often in written language will be discussed in Section€9.



Contextual cues for particle placement 

5.2

Regional variation

Szmrecsanyi (2005:€ 133) observes that there is noticeable variation in ordering preferences among British English dialects. If the dialect of England’s Southeast is to be considered as the standard, then one finds the split ordering to be significantly disfavored in the Hebrides and, to a lesser extent, in the Midlands, while it is used more frequently in the North of England than it is elsewhere. This is an interesting finding, especially since no other researcher (that I know of) has ever considered the possibility that particle placement is subject to dialectal variation. It would be worthwhile to investigate whether there are also differences in the choice of ordering between speakers of more largely-defined varieties of English: British English, American English, Australian/New-Zealand English, and so on. Table 3.╇ The two orderings by corpus of regional variety of English (adapted from Lohse et al. 2004:€243) British

Joined (V – Prt – NP)

American

LOB corpus (written)

Brown corpus

Wall Street Journal corpus

Switchboard corpus

154 (± 65%)

324 (± 79%)

248 (± 86%)

521 (± 69%)

Written American 572 (± 82%)

Split (V – NP – Prt)

83 (± 35%)

Written + spoken American 1093 (± 76%) 86 (± 21%) 39 (±14%) 229 (± 31%) Written American 125 (± 18%)

Column totals

237 (100%)

Written + spoken American 354 (± 24%) 410 (100%) 86 (100%) 750 (100%) Written American 697 (100%) Written + spoken American 1447 (100%)

 Bert Cappelle

There is some indication that there might be. Taking a look at Lohse et al.’s (2004:€243) table of the distribution of split to joined orderings by corpus (adapted as Table€ 3), I notice that their written British English material, taken from the Lancaster-Oslo/Bergen corpus, contains a higher percentage of split orderings than their combined material from corpora of American English, more than half of which even consists of phone conversations (the Switchboard corpus). This difference between British English and American English in their dataset reaches statistical significance (χ2(1) = 11.81; p ≤ 0.001). 6. Expectedness 6.1

‘Context’: A fluid concept of linguistic and extra-linguistic senses

Up to now we have dealt with cues to particle placement provided by the linguistic context. In this section, I deal with cues to the ordering variation at hand that come from ‘context’ in a wider sense than we have used so far: not so much the sentence’s linguistic or deictic embedding, but the entire body of expectations the discourse participants bring to a discourse situation. These expectations have been shaped by the domestic, societal and cultural environment in which they have been raised or adopted. Thus, when I speak of contextual cues to particle placement in this section, I should be understood as referring to the way the speaker/writer and hearer/reader draw from their shared ‘knowledge of the world’, or, more precisely, the knowledge of this kind that the speaker/writer assumes to share with the hearer/reader. It should be clear, however, that no all too strict distinctions can be made between linguistic and extra-linguistic context. For example, to assess whether a direct object is discourse-old, it does not suffice to look for a verbatim occurrence of it in the preceding discourse. Rather, one must look for an item that is coreferential with the object, or that stands in some sort of taxonomic relation with it. Thus, one can only establish a coreferentiality relation in the discourse between, say, …Ennelis… and …the girl… if one knows that Ennelis is a girl’s name, which is an extralinguistic piece of knowledge. Similarly, some world knowledge is required to perceive the relation between discourse elements like …pheasant shoots… and …the birds…, and between …black coffee… and …the taste…: (23) a. “Mama, I am hungry,” 10-year-old Ennelis is saying as Rose gathers the girl up in her arms. (www) b. In the winter he used to have pheasant shoots there. HR’s been at many of them, driving the birds out. (www)



Contextual cues for particle placement 

c. I only like black coffee when I’ve ground it myself and I’ve chosen the beans, otherwise you really need milk to take the taste away. (www) 6.2

An example: Putting the kettle on

The following lengthy example will help to make clear what I mean by expectations based on the discourse-participants’ “domestic, societal and cultural environment”: (24) The kettle is the one thing we all have in common on these islands. Other than us lot, hardly anybody else in the world has them. According to the story I was told at school, it was watching a boiling kettle that gave Stevenson [sic] the idea of the steam engine. Without the steam engine, this country would not have been first off the starting block into the industrial revolution. … The kettle may have evolved over the years since Stevenson [sic] but every home in the land has one, from Buckingham Palace to the lowliest of doss houses. … Get invited into any home and the first thing they will say is “I’ll just put the kettle on.” (www) The boiling kettle is such a staple fixture in every British household that it needs no prior introduction in the discourse to be comfortable in mid-position. In the standard phrase I’ll (just) put the kettle on (or Shall I put the kettle on?), the kettle is typically discourse-new (in that it is probably not mentioned in the preceding discourse), but this lack of discourse-familiarity is amply compensated for by its high expectedness value, since its presence in a British home is fully taken for granted nation-wide, as the author of the example above rightly claims. In fact, it could be argued that discourse-familiarity (discussed in Section€2) is a special case of expectedness. After all, the mention of an NP (or of a superordinate, subordinate, or meronymical term) creates some expectation as to its reappearance. Put differently and perhaps more accurately, prior mention of an NP renders its referent mentally more accessible. Familiar items like a kettle in a British home are mentally accessible without having to be linguistically mentioned or evoked first, but there is a point in which mental accessibility is the factor that licenses the mid-positioning of discourse-familiar and ‘world’-familiar objects alike. The main reason why I have nevertheless discussed discourse-familiarity in a separate section is that it is a concept that is far easier to grasp and accordingly to quantify (even taking into account the complications mentioned above in Section€6.1) than the factor ‘expectedness on the basis of world knowledge’. Still, we can gain a better understanding of world knowledge if we think of it in terms of framesemantic knowledge, in Fillmore’s sense. To use our kettle example again, paying a visit to someone’s home in Britain can be conceived as a complex structure of related entities and events, among others the owner (or occupier) of the house offering

 Bert Cappelle

the visitor something to drink, usually a nice cuppa. Since the process of making tea involves the use of a boiling kettle, and perhaps because Britons perceive some impoliteness in forcing visitors straightaway to confirm that they would like to drink some tea (as would be the case if the host asked, “Would you like some tea?”), they offer to put the kettle on. In other words, the host’s suggestion to put the kettle on is a part of the ‘visiting someone’s home in Britain’ frame. 6.3

Bolinger (1971): Frame semantics avant la lettre

Without being explicitly couched in frame-semantic terms, some observations made by Bolinger (1971:€55–57) are clearly consonant with such a framework. He associates the split ordering with “the ordinariness of certain actions and the familiarity of certain objects”. That is, “[i]f the noun retains the accent but the particle is put after it, the sensation most often seems to be just that of what is familiar or usual under the circumstances”. Here is one of his examples: (25) It’s almost ten o’clock. Put your níghtie on, now, and run up to bed. (Bolinger 1971:€57) Here, a nightie, the action of putting it on, and the temporal location at 10 p.m. are apparently part of the bed-time routine of the hearer.12 A quite similar example is the following: (26) Do you want to squeeze some paste on? (father to young child, during a conversation overheard in the bathroom of a camping in the UK) Again, the speaker assumes the hearer’s familiarity with the tooth-brushing frame, in which a toothpaste and a toothbrush are fixed entities, and of squeezing some of the former on the latter is a fixed action. Moving beyond this anecdotal evidence, let us consider a simple corpus investigation. I entered the following two search patterns in the British National Corpus: (27) a. put down [possessive determiner] [noun] b. put [possessive determiner] [noun] down The predictions were that instances of the joined ordering, returned by the query in (27a), would contain fewer nouns whose referent is expected than instances of the split ordering, returned by the query in (27b). The results of the 50 most frequent results for these two queries are given in 0 and 0, respectively.13 By the way, the query with the joined ordering yielded (only) 182 matches, compared to 344 matches for the query with the split ordering.



Contextual cues for particle placement 

(28) 1. his glass (15 instances); 2. her cup (12); 3. his knife (8); 4. his cup (5); 5. her coffee (5); 6. her glass (4); 7. her knife (4); 8. his book (4); 9. his pen (4); 10. his paper (3); 11. my pen (3); 12. her book (3); 13. her pen (3); 14. her shopping (2); 15. her trowel (2); 16. her wine (2); 17. her coffee-cup (2); 18. her basket (2); 19. her cloth (2); 20. his briefcase (2); 21. his coffeecup; 22. his spade (2); 23. her fork (2); 24. his newspaper (2); 25. his notebook (1); 26. his notes (1); 27. his pint (1); 28. his pitchfork (1); 29. his pocket (1); 30. his scalpel (1); 31. his scone (1); 32. my scythe (1); 33. my shopping (1); 34. my suit-case (1); 35. their bibles (1); 36. their bowls (1); 37. their clues (1); 38. their deposits (1); 39. their guns (1); 40. their knives (1); 41. their names (1); 42. their pens (1); 43. their porridge (1); 44. their stakes (1); 45. your bag (1); 46. your case (1); 47. your drink (1); 48. your guns (1); 49. your handbags (1); 50. your knives (1) (29) 1. your foot (26 instances ); 2. your name (15); 3. his foot (15); 4. her head (14); 5. her foot (14); 6. his head (14); 7. my foot (11); 8. my name (11); 9. your head (10); 10. my head (9); 11. her cup (9); 12. their foot (8); 13. your hands (7); 14. your phone (6); 15. his glass (6); 16. her glass (5); 17. his cup (5); 18. your feet (5); 19. their names (5); 20. your hand (4); 21. your handset (4); 22. their name (3); 23. his hand (3); 24. her bag (3); 25. her name (3); 26. his name (3); 27. his money (2); 28. his mug (2); 29. his phone (2); 30. his paper (2); 31. his things (2); 32. its head (2); 33. her pen (2); 34. her basket (2); 35. her feet (2); 36. his goblet (2); 37. her hand (2); 38. my hand (2); 39. your age (2); 40. your bag (2); 41. your leg (2); 42. your pen (2); 43. your mallet (1); 44. your papers (1); 45. your plate (1); 46. your things (1); 47. your welly (1). 48. your heel (1); 49. your keys (1); 50. your glass (1) Even without considering the context of all the corpus hits, a pattern emerges.14 Each of the ten most frequently occurring nouns, apart from name, in the split pattern refers to a body part of whomever the possessive determiner refers to. Body parts, like a head, feet, hands, legs, etc. are (for all intents and purposes) inalienable possessions and therefore, they are expected entities. That somebody has a head, for example, is taken for granted. It is an entity that is an inherent part of the frame invoked by the entity to which the possessive determiner refers. Thus, as soon as a ‘person’ frame is evoked – this happens, for instance, as soon as the hearer makes a mental picture of the subject referent in She put her head down – these body parts are automatically also activated to some degree. The same goes for the noun name: people (ordinarily) have a name, so this entity is also co-activated along with the person referred to by the subject and/or possessive determiner. Thus, body parts and more abstract entities like somebody’s name or age are expected on the basis of the discourse participants’ familiarity with the ‘person’

 Bert Cappelle

frame (which in turn is inevitably evoked by the possessive pronoun and/or a coreferential NP in the preceding discourse). Interestingly, the first noun referring to an object not evoked by the ‘person’ frame (the noun cup) only comes in at the 11th position in the ranking for the split ordering. Conversely, there are no nouns referring to body parts among the fifty most frequent hits in the joined ordering and the first noun referring to an inalienable possession, names, only comes in at the 41st position in the ranking. These results go some way to supporting Bolinger’s (1971:€55) above-mentioned intuition. 6.4

Not just the expectedness of the object

We should emphasize here that Bolinger (1971) also considers the split ordering to be eminently suited to suggesting that the verbal action, and not just the direct object referent, is familiar or expected in a given situation. Sometimes, the latter more or less naturally implies the former. Thus, to use one of his examples, cattle entails the action of driving, which partly explains why the use of the split ordering is acceptable in the response below. (30) Where’s Joe? — He’s driving the cáttle in. He’ll be along in a minute. (Bolinger 1971:€56) Note also that the object NP here can, on the one hand, be discourse-new but may, on the other hand, not be completely unpredictable to the addressee (assuming that he or she knows Joe and knows he is a cattle-driver). Similar cases in which the split ordering can be used despite the object NP being discourse-new and falling within the focus domain are hanging the {washing / laundry} out, picking the kids up, dropping the kids off, etc. Not only are the verbal actions expressed in these phrases notionally closely associated with their respective objects, but the events in question may also take place regularly (e.g.€as part of a daily or weekly routine) or as a matter of course. The use of the split ordering, while by no means obligatory in such cases, may underscore this idea: (31) Molly finds that she doesn’t need the car with her all the time, as she no longer has to pick the kids up every day and now she is able to go to her computer class in the bus. (www) Ignoring the political correctness issues of the following example, we may invoke the notion of ordinariness to account for its shift from one ordering to the other for what is otherwise one and the same phrase (which I have underlined here for ease of identification; the italics for emphasis are in the original):



Contextual cues for particle placement 

(32) Something happened the other day that made me feel uneasy. Yet I shouldn’t have felt that way! My wife had left for work and I was hanging the washing out to dry. A neighbour from down the way was in his backyard doing the same. ‘Good day for drying’, he called. ‘Let’s hope the rain stays away.’ I had to think about what made me uneasy. Then it hit me. Two men hanging out the washing! When I was a kid that would never have happened. That was women’s work, after all! (www) The first occurrence of hang out + the washing is meant to sound as unsuspicious as possible, which may explain the use of the split ordering. On its second occurrence, hang out + the washing is used in the joined ordering, even though the direct object is discourse-old (and might therefore have been put in mid-position). In this second instance, the author wants to draw the attention to the extra-ordinariness of the situation. One likely reason why he uses the joined ordering here is that it permits him to put an accent on the particle, thus marking almost each word with a separate accent (Twó mén hánging óut the wáshing!) rather than giving hanging out + the washing one overarching accent (on washing). This way, form and (pragmatic) content match up nicely. Needless to say, the examples given in this section are merely suggestive. The precise role played by the frame-semantic expectedness of the verbal action (given the object) and/or the ordinariness of the entire event (as a Gestalt made up by the verbal action and the object) in particle placement is something that needs to be investigated in much more detail. 7 Stativity as a contextual factor? A little-mentioned semantic constraint on the particle placement alternation (but see Fraser 1976:€11; Huddleston 2002:€286, fn. 49; Cappelle 2005:€220) has to do with the aspectual nature of the verb: dynamic verbs allow a spatial particle to precede as well as follow the object NP, while stative verbs only allow a spatial particle to follow the object NP. Compare, for example: (33) a. He put {on his socks / his socks on}. b. He kept {?on his socks / his socks on}. (34) a. I pulled {up the blinds / the blinds up}. b. I left {*up the blinds / the blinds up}.

 Bert Cappelle

In what way is this constraint relevant to our discussion on the role of context on particle placement? To answer this question, let us consider some examples with prima facie stative verbs that do allow the particle to precede the object: (35) a. The nail holds up the picture. b. The bricks keep down the sheet. c. The load weighs down the car.

(McIntyre 2003)

Just because there is no overt motion involved in these examples, this does not mean that they describe stative events. On the contrary, on the basis of encyclopedic knowledge, especially the laws of mechanics, we know that holding up a picture requires a sustained impediment to the force of gravity. Similarly, keeping down a sheet which tends to be blown upwards requires the constant counteracting input of a downward-pushing force, by releasing the brick’s potential energy. And weighing down a car involves the sustained application of the load’s weight to work against the spring force of the car’s suspension system. So, while the above clauses depict steady situations, they also involve energy to maintain that steadiness. What is important to our concerns is that it is the context in which the verb is used which may reveal this ‘hidden’ dynamic aspect, and that the particle is part of that context. Thus, while keeping one’s socks on does not require energy – rather, energy is required to take them off – keeping them up may prove to be dynamic, in the way described above: (36) …a pair of those old fashioned garters that men wore to keep up their socks… (www) In short, the requirement that the verb not be stative for the two orderings to be allowed is context-dependent: the verb taken by itself does not tell us whether this requirement is met; we need to consider the context in which the verb is used. 8 Literalness as a contextual factor? As is well known, verb-particle combinations differ in the degree to which their component parts independently contribute to the meaning of the composite whole, some combinations being completely transparent (e.g.€kick the ball out), others being quite opaque (e.g.€chew out someone ‘reprimand someone for doing something wrong’). According to Cowie and Mackin (1993: ix), “[t]he fact that a particular combination of verb + particle is idiomatic need not affect its grammar”, by which they mean that both transparent and idiomatic combinations in principle allow the two orderings. Yet, it has been shown on the basis of corpus research that the degree of idiomaticity



Contextual cues for particle placement 

does have an impact on word order preferences, in that combinations with a semantically dependent particle significantly favor the joined ordering (e.g.€Gries 2003:€87; Lohse et al. 2004), a preference that is already manifest in early child language (Diessel and Tomasello 2005) and that has also been supported by means of psycholinguistic experiments (Dehé 2002:€97; Gonnerman and Hayes 2005). Gries (2003:€87) and Szmrecsanyi (2005:€132) also note a significant increase in the choice of the split ordering with literal particles, although Lohse et al. (2004) maintain that this increase does not, in their data set, amount to an actual preference for the split ordering with literal particles – there is just a higher proportion of split orderings. According to Gries (2003:€52), a characteristic of the split ordering is that “the particle is positioned in the canonical position for focal elements, i.e. clause-finally, so that the particle is processed more intensively than the direct object.” Hence, the split ordering “naturally underscores the spatial contribution the particle makes to the meaning of the utterance and would, therefore, be the natural choice for a speaker who intends to communicate a state of affairs where the spatial meaning is prominent” (Gries 2003:€52–53). In this account, the endplacement of literal particles is linked up with focus structure, which would entail that literalness is ultimately also a context-related factor. In fact, however, of all literal particles that appear in end-position, the ones that form the clause focus all by themselves probably make up a small minority. Recall that a particle may be in end-position when it is not the only element under focus, as in (37a-c), and even when it is not at all part of the focus, as in (38a-b). (37) a. What happened? — I threw my radio away. b. What did you do? — I threw my radio away. c. What did you do with that old radio? — I threw the damn thing away. (38) a. Did you give that old radio away? — No, I gave my computer away. b. Did you give that old radio away? — No, I threw the damn thing away. The reason why there is a higher proportion of split orderings with particles that can be independently processed than with other particles has less to do with the possibility of independent particles to have end-focus – which they can but do not often have – than with a higher processing cost of the split ordering with dependent particles, compared to the joined ordering with such particles. As Lohse et al. (2004:€246Â�–247) explain, if a particle depends on the verb for its interpretation, then processing benefits from putting that particle in close proximity to the verb. This way, its “lexical dependency domain” is minimized. This is actually also recognized by Gries (2003), who writes that it would be uneconomical to process the opaque meaning of a [transitive phrasal verb] but produce the parts that trigger this opaque meaning in possibly widely

 Bert Cappelle

disparate positions of the sentence. With literal phrasal verbs, by contrast, no preference for a construction is to be expected on grounds of semantic dependency because the low degree of interdependence does not require a particularly small distance between the component parts and, thus, licenses both word orders. (Gries 2003:€54)

As regards the question in the title of the present section, we have to conclude that the influence of literalness on particle placement cannot plausibly be formulated as a contextual, in the sense of focus-structural, factor. End-placed particles do not ordinarily convey the most prominent information in a given context. 9 Motivating the contextual factors of particle placement The above sections have made it clear that there is no one single context-related factor that governs particle placement in English. At the same time, Gries (2003) maintains that all factors that have ever been proposed (including those not involving context) can be subsumed under a single motivating principle, related to the processing cost of the object NP. His ‘processing hypothesis’ boils down to this: the harder the object NP is to process, the more it tends to follow the particle; the easier it is to process, the more likely it is to precede the particle. This hypothesis can be expressed even more succinctly in the words of Bever (1970): “Save the hardest for last”. In this section, I want to examine whether Gries’s processing hypothesis is viable: can it account for all of the factors we have dealt with? 9.1

Re discourse-familiarity

The ratio of split to joined orderings is higher for discourse-old object NPs than for discourse-new ones. This fact is straightforwardly related to processing issues. For discourse-old NPs, both the speaker and the hearer already have a conceptual representation with a high degree of mental activation (Gries 2003:€ 166). This makes these NP’s easily accessible and useful anchor points for information units that have not had any prior mention in the discourse. Proceeding from the familiar to the unfamiliar does not only facilitate comprehension. By postponing a discourse-new constituent, speakers also buy themselves some time to think about such matters as how they will refer to the newly introduced discourse entity; which potentially complex structure they may have to retrieve from the grammar in order to express all the information needed to help the hearer identify the referent; how the lexical concepts have to be integrated into this grammatical structure, and so on (Arnold et al. 2004:€33). Somewhat surprisingly, though, Gries (2003:€130, note 33) and Lohse et al. (2004:€258) point out that factors relating to processing



Contextual cues for particle placement 

facilitation exert a stronger influence in writing than in speech (cf.€also Section€9.4 below). Given that speed of planning and production plays a relatively small role in written discourse, it seems, then, that the postponement of discourse-new NPs predominantly targets comprehension rather than production facilitation. 9.2

Re focus structure

Gries (2003) seems to have conflicting ideas on whether focused items have a high or a low amount of processing cost. On the one hand, Gries argues that “focused elements require more processing effort” (p.€52) because the hearer’s brain has to allocate more attention to elements that are marked as important, thus increasing their processing cost. Insofar as focal elements appear clause-finally, there is support for his processing hypothesis. On the other hand, Gries agrees with claims in the literature that focal elements have high activation in the mind of the speaker (p.€172), which means that they are easily accessible and thus, from the perspective of the speaker at least, require low processing effort. This, he admits, is then hard to reconcile with the tendency to put focal material at the end. His solution is to argue that speakers do not only select the joined or the split construction on the basis of whether the particle or the object, respectively, happens to have the highest degree of mental activation – for example, if it is the object, it is inserted in the syntax first, resulting in the generation of a split construction – but also on the basis of the intended communicative intention. If an element is to be highlighted, the speaker will choose a constructional pattern that accords a prominent position to this element, which in English is usually at the end. Speakers thereby conform to the general Focus-Last principle, by which hearers look for the most informative element at the end of clauses, after they have been given all the background information required for its successful processing. We have seen that focal elements do not always occur in sentence-final position: any constituent in the clause, irrespective of its syntactic position, can contain the nuclear accent. We have also just noted, in Section€9.1, that written texts seem to attend to the needs of the reader more than spoken discourse does to the needs of the hearer, when it comes to the choice of word orderings that facilitate processing. We now have a possible explanation for why this may be so. In writing, the unavailability of prosodic cues to the contrary may force writers to stick to the general FocusLast principle. Conversely, in speech, temporal constraints on planning may force speakers to select the ordering that allows them to insert the most active item (the one under semantic focus) as early as possible, even if this results in the selection of a structure that deviates from the general Focus-Last principle. During the actual phonological production stage, they may then overlay the syntactic structure with a ‘correcting’ pitch pattern. It goes without saying that these are mere hypotheses.

 Bert Cappelle

9.3

Re previous occurrence of one ordering or the other

The occurrence of a construction raises its mental activation level, making it more likely to be selected again when the speaker is put before the choice of alternative orderings. It is not clear, however, how the higher accessibility of either the split ordering or the joined ordering as a whole ties in with Gries’s (2003) processing hypothesis, which is stated in terms of the processing cost of the direct object (relative to that of the particle) only. 9.4

Re register

The ratio of split to joined orderings is higher in speech than in writing. Gries (2003:€100–101) notes that this should not surprise us, since speech is characterized by a higher occurrence of short object NPs and of literal verb-particle combinations, which he shows to strongly favor the split ordering. However, the higher occurrence of split orderings in spoken discourse cannot just be written off as a simple side-effect of the well-established influence of lightness of the object and literalness of the phrasal verb. For one, although Lohse et al. (2004:€ 258) equally observe that the average length of direct objects is shorter in speech than in writing, they find “no significant difference in the distribution of dependent and independent particles”, so an explanation for the higher ratio of split orderings in speech should not be sought (in part) in the brute effect of a higher ratio of literal combinations in speech. For another, Lohse et al. (2004:€258) note that as the length of NPs increases, the ratio of split orderings decreases much more steeply in writing than in speech. So, if there are more instances of the joined ordering in writing than in speech, this is not only because there is a higher ratio of long objects in writing compared to speech, but also because, for a same number of long objects, the effect of these long objects on placement is more clearly noticeable in writing than in speech. Gries (2003:€130, note 33) also observes that in writing, the decline of split orderings is steeper for discourse-new NPs. These observations point to a finer attunement to the needs of the parser in written discourse (cf.€Section€9.1 above). Lohse et al. (2004:€258) furthermore find that in written discourse, the tendency to use the split ordering with literal particles is lower than in spoken discourse. (Again, it is not that writing is characterized by a lower occurrence of literal particles, as Gries (2003) reports. Instead, what they find is that, for any given number of literal particles, the split ordering is selected for them with a lower frequency in writing than in speech.) As we noted in Section€5.1, this may have to do with the connotative value attached to the split ordering, whose perceived familiarity may lead to its avoidance in written texts.



Contextual cues for particle placement 

9.5

Re region

Clearly, regional variation in particle placement cannot be construed as a manifestation of Gries’s (2003) processing hypothesis. It would be nonsensical to claim that speakers of British English, when using transitive phrasal verbs, tend to use object NPs with a lower degree of processing cost, which then accounts for their higher use of the split ordering compared to speakers of American English. One can only explain regional variation in terms of a higher entrenchment of the construction most frequently used in a particular regional variety of English and/or in terms of differences among regional varieties in the weights attached to certain factors for particle placement whose values may sometimes be in conflict. Consider, for example, the following question-answer pair: (39) If you could jump in a time machine and relive a certain moment of your life, which would it be? — I wouldn’t. I’d blow the damn thing up. (www) The object NP the damn thing refers back to a time machine mentioned in the question. Its status as a discourse-old entity plays a role in the use of the split ordering, which conforms to the principle by which discourse-familiar entities are given as early as possible to facilitate processing. On the other hand, the object is a three word long NP, and in general, objects of this length are more frequently put in the joined ordering (since long NPs are hard to process and may therefore be saved for last). So, there are at least two relevant processing-facilitating principles that lead to opposing word order choices. It may be plausible to assume that whenever two or more processing constraints are in conflict, speakers tend to select the constructional option which incurs the least serious violation(s), where seriousness is determined with reference to a constraint hierarchy, as in Optimality Theory (Prince and Smolensky 2004 [1993]), or with respect to weights attached to the constraints, allowing for higher predictability, as in probabilistic grammar.15 Suppose, now, for the sake of the argument, that there are only two constraints to be considered: OldObjLeft (i.e. discourse-old objects come first) and Heavy-Right (i.e. weighty objects come last).16 In the above example, putting a three-word-long object in midposition is felt to come at a much lower cost than putting a discourse-old object after the particle. This is, in any case, how by far most speakers of British English appear to rank the two constraints, based on a series of Google searches I did on 5€ September 2006. Using Google’s Language Tools function,17 I entered the strings “up the stupid thing”, “up the damn thing” and “up the bloody thing”, as well as “the stupid thing up”, “the damn thing up” and “the bloody thing up”, each time selecting ‘Great Britain’ as the country in which the search pages were to be located. I then performed the same searches, but with ‘United States’ selected as

 Bert Cappelle

Table 4.╇ The two orderings by country for some selected Google search strings Search pages in Great Britain

Search pages in the United States

Joined: ‘verb up the {stupid / damn / bloody} thing’

â•⁄ 14 (± 9%)

â•⁄ 298 (± 23%)

Split: ‘verb the {stupid / damn / bloody} thing up’

149 (± 91%)

â•⁄ 974 (± 77%)

Column totals

163 (100%)

1272 (100%)

country. After that went manually through all the hit pages to cull out noise examples. The results are given in Table€4. The difference in ratios of split to joined orderings for the two selected countries is significant (χ2(1) = 18.70; p ≤ 0.001). So, speakers of British English clearly let the OldObj-Left constraint prevail on most usage occasions.18 For speakers of American English, the situation is less clear. While most of them also appear to find that the selection of the split ordering is preferred in cases with discourse-old objects of word-length three, a sizeable proportion of them come up with the other ordering, which means that in American English, some less weight is attached to the constraint that discourse-old objects should be put early compared to British English and, by the same token, some more weight is attached to putting long objects late in the clause, while the actual ranking of these constraints with respect to each other is the same as in British English. There is probably some interaction between the severity of a constraint violation and the entrenchment of a construction. Thus, in Britain, where a violation of the OldObj-Left is seen as comparably more serious than in the US, we may expect the split ordering to be more entrenched than in the US. This deeper entrenchment may in turn have an impact on the weights attached to the ordering constraints. 9.6

Re frame-semantic expectedness

The ratio of split to joined orderings is higher for frame-semantically expected objects than for objects that do not form part of a particular frame evoked in the discourse. This seems to be quite naturally related to the assumption that all the elements belonging to an activated frame are themselves also activated to a higher degree than other elements. Since this higher activation of a frame-expected object (like a boiling kettle for making tea in the ‘visiting someone in Britain’ frame) also renders it less hard to process, its typical mid-position is predicted under Gries’s (2003) processing hypothesis.



Contextual cues for particle placement 

However, Gries’s (2003) processing hypothesis, which is stated in terms of the processing cost of the direct object only, makes no prediction as to the choice of word order for frame-semantically expected actions (as expressed by the verbparticle combination). On the basis of the general processing constraint to put familiar elements, including familiar actions, as early in the sentence as possible, we would expect a preference of the joined ordering, with verb and particle preceding the rest of the VP. But this is not what the examples given in Section€6.4 tentatively suggest. The more felicitous option for the split ordering in these cases may be made under the impulse of avoiding unneeded accents in the clause – recall our discussion of hanging the wáshing out vs. hánging óut the wáshing. If the particle is put in end-position, it need not receive an accent in addition to the accent on the object. This leads to what Bolinger (1971:€56), perhaps somewhat impressionistically, calls “the suffusion of the sentence as a whole with a kind of relaxed rhetorical effect”. 9.7

Re stativity

The split ordering is ordinarily selected for literal verb-particle combinations whose verb receives, in the context concerned, a stative interpretation: (40) a. It is hard to leave {out all this stuff / all this stuff out}. (suggests that an urge to put the stuff in, or even the projected tendency of the stuff itself to be included, has to be overcome by force of will) b. This morning, I just left {*down all the blinds / all the blinds down}. (no force needed) This constraint cannot be linked in any sensible way with Gries’s (2003) processing hypothesis. What motivates this factor, rather, is probably the fact that the split ordering underscores the family resemblance with a number of semantically related non-motional constructions, instantiated by, e.g., judge {*out the ball / the ball out} and picture {*off all the lights / all the lights off}. The possibility of putting the verb and the particle in adjacent positions is contingent on their semantic togetherness. Note in this respect that verbs and particles which resist being put in adjacent positions cannot be replaced by a simplex verb. Compare: (41)

a. b. c. d.

It is hard to {leave out / omit / exclude} all this stuff. This morning, I just left all the blinds down. (no simplex verb available) The umpire judged the ball out. (idem) Can you picture all the lights off? (idem)

In the end, therefore, the stativity constraint may be (negatively) motivated by one of the most basic word order principles, known as Behaghel’s (1932) “First Law”:

 Bert Cappelle

Das oberste Gesetz ist dieses, daß das geistig eng Zusammengehörige auch eng zusammengestellt wird. (Behaghel 1932:€4) ‘The supreme law is this: that which belongs closely together mentally is also placed closely together’.

9.8

Summary of Section€9

While this paper has essentially been aimed at getting a sharper picture of what kinds of contextual factors may influence the selection of one particle placement option rather than the other, it has now become clear that these factors can ultimately be motivated by the way in which they contribute to ease of processing (i.e. production/comprehension), though not exactly in the way intended by Gries (2003). More specifically, the processing ease of one or the other word order variant depends on such context-related parameters as: i. the early positioning in the variant of an element which has heightened mental accessibility as a result of its previous mention and, thus, which has the potential of serving as a ‘stepping stone’, especially for the addressee, to reach constructional elements with lower mental accessibility; ii. the variant’s conformance, in the absence of ‘corrective’ intonation, to the standard – and therefore processing-facilitating – Focus-Last principle, which directs the addressee’s attention to the position in the sentence where he or she expects to find the most important information, namely at or towards the end; iii. the variant’s degree of mental accessibility as a result of activation through previous use; iv. in writing: the degree to which the variant reflects the extra requirement posed by written discourse (as a medium that is more controlled and edited than spoken discourse), namely that the writer should further enhance comprehension, e.g.€by putting long objects towards the end and by not splitting up idiomatic verb-particle combinations; v. the variant’s degree of mental accessibility as a result of being the most entrenched ordering option in a regional variety, and/or whether or not it is the ordering option in a given usage situation with the least severe constraint violations as evaluated by variety-specific rankings; vi. the early positioning in the variant of a frame-semantically expected (hence accessible) element and its absence of uncalled-for accents (which are costly to produce, and, for the hearer, to give heed to). vii. the separation of elements in the variant that could wrongfully be interpreted as conceptual units if they were not separated.



Contextual cues for particle placement 

10 Towards modeling the contextual factors of particle placement We have seen how the multiple context-related factors that play a role in the particle placement alternation find their motivation in processing efficiency. The question, first of all, is whether we ought to give a place at all to these processing-facilitating factors in a grammar of English. Can’t they just be dismissed as a matter of performance/parole? After all, isn’t what matters to a grammarian just to know which word strings are grammatical rather than to know which ones are easy to process? This narrow view of what counts as grammar proper seems to underlie a statement like the following: “Given two equally grammatical outputs, a speaker is free to use other, non-grammatical, criteria to decide which of them to use” (Svenonius 1996:€54). However, constructionist frameworks, like Construction Grammar, do not a priori regard pragmatic, discourse-functional or even socio-linguistic considerations in the choice of a constructional variant as falling outside the realm of grammar. While the main achievement of Construction Grammar so far has been to link up lexicon and semantics more closely with syntax (which is then no longer considered to be an autonomous component of grammar), there is no denying that this framework is now also taking an active part in a general trend within contemporary linguistic approaches towards what Geeraerts (2006) calls the “recontextualization” of grammar, that is, towards reintegrating into grammar various forms of context that had been severed from it during the Chomskyan revolution: “the social context”, “the situational context of actual language use” and “the cognitive context of meaning and experience”. In this paper, we have considered a variety of elements which relate to these aspects of contexts and which provide information to the language user (in other words, which serve as contextual ‘cues’) as to which ordering option a speaker should select when using a transitive phrasal verb.19 These kinds of information are drawn upon by language users – they know what to do with these cues – and they should therefore be captured in a grammar that aspires at reflecting all the knowledge speakers have about the language they use. The next, and much less trivial, question is how we can incorporate these context-based cues in a constructionist model. I can only offer some very sketchy suggestions at this stage, pointing out some possible avenues for future endeavors. A fruitful way to start looking at grammatical variation from a constructional perspective is to look at the options as ‘allostructions’ (Cappelle 2006), that is, as formally distinct realizations of a more abstract construction that remains underspecified for one or more particular features – in our case, the exact placement of the particle vis-à-vis the direct object and the internal constituency of the VP (cf.€Farrell 2005). This is visualized in Figure 1 for transitive blow up.€It will be observed (i) that the semantics of the two allostructions exactly matches that of the more abstract construction, (ii) that the lexical leaves in the syntactic trees are

 Bert Cappelle

linked up with the lexically specified verb and particle in the otherwise more unspecified syntactic representation in the upper construction, and (iii) that the NP in each of the allostructions corresponds with the patient argument in the semantic representation. Boldface in the semantic representation means that the semantic roles involved must be realized syntactically. Of course, the agent role is syntactically expressed as a subject argument, in keeping with standard linking rules. We then have to ensure that the most appropriate ordering is chosen under certain contextual conditions. How are we to appeal to context-related processing constraints in a framework where one of the main tenets is that “[t]he totality of our knowledge of language is captured by a network of constructions: a ‘construct-i-con’” (Goldberg 2003:€219). That is, if a constructionist framework embraces “the idea that the network of constructions captures our knowledge of language in toto – in other words, it’s constructions all the way down” (Goldberg 2003:€223), does this mean that general processing-facilitating constraints should also be described as constructions? Constraints as constructions transitive blow up Sem CAUSE-BECOME < agt, pat, destroyed in an explosion > Syn [VP [V blow] [Prt up]]transitive

Joined blow up

Split blow up

Sem

Idem

Syn

VP

V

Sem Syn

NP

V

Prt

blow

up

V

Idem VP

NP

blow

Figure 1.╇ The two allostructions (joined and split) of transitive blow up

PrtP

Prt

up



Contextual cues for particle placement 

Some tentative proposals along these lines have been made by Stefanowitsch and Gries (2002) with respect to the Given-before-New principle. Instead of treating this as a universally valid principle – which it isn’t – with extra-grammatical status, they suggest a number of constructional treatments, one of which is to give ‘GivenX-New-Y’ (GXNY for short) the status of a general information-structure construction within the grammar. A consequence of such an approach is that the acceptability of an allostruction in a given usage context is not evaluated in terms of the number of constraints it does or does not violate but, rather, in terms of the number of general constructions it does or does not inherit. In essence, constraintviolation and construction-inheritance boil down to the same thing; they seem to me to be alternative ways of looking at a single phenomenon. Choosing between two constructional alternatives may then be described as an evaluation of the severity of each alternative’s failing to inherit one or more general constructions. This severity is partly determined by these constructions’ relative positions in a ranking, more or less in the way of Optimality Theory, but also partly by their absolute positions in that ranking (i.e. by their weights). As a highly simplified example, consider again the variety-related difference in distribution of ‘verb up the damn thing’ and ‘verb the damn thing up’. Recall that we hypothesized that American English and British English do not differ with regard to the relative ranking of the OldObj-Left constraint and the Heavy-Right constraint, but do so with regard to the degree to which the former constraint is felt to be more important than the latter (cf.€ Section€ 9.5). This means that we should represent the rank-ordered constraints (whether or not we construe them as constructions) as interval values and not just as ordinal values. Figure 2 shows, very schematically, the relative positioning of these two constraints in British English on a scale of importance. The further to the left a constraint is positioned, the less easily it can be violated. Of course, other constraints on particle placement should also be located somewhere on this scale, but I disregard them here for practical purposes. (Some of these other constraints in fact have to be located even further to the left than OldObj-Left.)20 Oldobj-left

Heavy-Right »

Very important

Not very important

Figure 2.╇ Two rank-ordered constraints on particle placement in British English

 Bert Cappelle

This figure abstracts away from differences that may exist among regional dialects within British English, as well as from possible differences among individual speakers. Also not represented in this figure (but see Figure 4 below) is the fact that OldObj-Left is a special instance of Given-X-new-Y and that HeavyRight may be further motivated by Hawkins’s (1994) ‘early immediate constituents principle’ (EIC), by which a heavy NP in end-position does not compromise quick identification of all the immediate constituents of a VP as it would if it were in mid-position (since then, the hearer would have to wait to add the particle to the list of immediate constituents until he or she had heard the NP in its entirety). Compare Figure 2 now with Figure 3, where the location of these two constraints on an importance hierarchy is shown for American English. Observe that the ranking order is maintained but that the difference in importance is smaller. Oldobj-left

Heavy-Right »

Very important

Not very important

Figure 3.╇ Two rank-ordered constraints on particle placement in American English

Such a partial constraint hierarchy can now be fused with the transitive blow up construction so as to enable evaluation of constructs like blow up the damn thing versus blow the damn thing up, where in each case the damn thing is a discourse-old object NP consisting of three words. A proposal on how this might be done is shown in Figure 4. On the right of this figure, we recognize a schematic rendering of the constellation more fully represented in Figure 1 above. While I have left out the semantic and syntactic particulars, I have added a piece of information, namely that, out of context, the joined ordering should be selected more readily than the split ordering, due to the semantic dependency of the particle and the verb. This relative ranking of the joined and the split ordering is therefore represented as an instance of Lohse et al.’s (2004) ‘lexical dependency domain (LDD)’ minimization principle.



Contextual cues for particle placement 

Given-X-new-Y

EIC

Minimize LDD

transitive blow up

Oldobj-left

joined blow up

Heavy-Right »

»

Very important

… [NP]i… { blow up [the damn thing]i

split blow up » Not very important

/

blow [the damn thing]iup }

Figure 4.╇ Simplified evaluation of the joined vs. split ordering for the transitive phrasal verb blow up, used with a 3-word discourse-old object in British English

On the same hierarchical scale we can also find again the OldObj-Left constraint and the Heavy-Right constraint, among quite a few others that have been left out for the sake of visual clarity. Their respective motivating principles are represented as well (cf.€the discussion below Figure 2). The distance between these two constraints iconically reflects the degree to which (I hypothesize) they are felt to differ in absolute, rather than just relative, importance by speakers of British English (cf.€Figure 2). Remember that this difference can be smaller or greater for speakers of other regional varieties. We now have four constraints/constructions placed on a single interval scale. This will suffice to illustrate how speakers may go about weighing up the two word order possibilities available with blow up when using this phrasal verb with a relatively long object referring back to a discourse-old entity. (The discourse-old status of the object NP is represented by means of a subscripted index, indicating its coreferentiality with an NP in the preceding context.) As far as the joined ordering is concerned, the speaker obeys the HeavyRight constraint and avoids the disjunction of the semantically dependent components of the phrasal verb. However, its merits are outweighed by the advantage of the split ordering. This ordering admittedly disintegrates the idiomatic unit of verb and particle but complies with the important constraint of putting

 Bert Cappelle

discourse-old objects as far as possible to the left. Accordingly, the split ordering will be opted for. In our representation, it deserves the pointing finger symbol, which is also used by practitioners of Optimality Theory to indicate the best candidate of any number of competing options. The joined ordering does not receive an asterisk for ungrammaticality, since it can be used, but its selection is predicted to be far less likely. In American English, the joined ordering is also the dispreferred option, but it reaches higher frequency than in British English. I have assumed, mainly for argumentative reasons, that this higher frequency is a result of the smaller difference in importance between OldObj-Left and Heavy-Right in American English. I should emphasize here that I do not feel strongly about this assumption and that it needs to be tested empirically by taking other factors into account. For example, it might just as well be the case that, compared to British English, more importance is attached in American English to the Minimize LDD constraint, leading to higher observance of the adjacency of the components of an idiomatic phrasal verb. Other issues that should be addressed in future research are the following: i. Are general principles like GXNY and EIC also ranked relative to each other, or is it just their construction-specific instantiations (in our case: OldObjLeft and Heavy-Right) that are rank-ordered in each specific construction? ii. Is what is shown in Figure 4 (minus the allostructs at the bottom) the result of a fusion of the transitive blow up construction with a more general, higherordered transitive V-Prt construction, or is it the actual representation of a construction that speakers can directly retrieve from their mental grammar? That is, does each lexically specified V-Prt construction contain only information that is unique to that very combination (such as the weights of each of its allostructions), or does it also contain information that is common to other lexically specific constructions (cf.€the “impoverished-entry” vs. “full-entry” debate; see Goldberg [1995:€74])? iii. Supposing that there is a more general constructional schema for transitive V-Prt combinations, how does that schema represent the relative importance of its two allostructions, given that this is different for individual V-Prt combinations? Does it set limits on how far apart these allostructions and on how heavy the weight of each of the two allostructions can be? Or does it contain default settings, which average out the strengths of all of the individual lexicalized V-Prt combinations? iv. Given that a constraint like Heavy-Right becomes more important as the object becomes longer and/or more complex, don’t we have to distinguish different heaviness subconstraints on the importance scale, each of them applying to an object of a certain length/complexity?



Contextual cues for particle placement 

However these issues may be resolved, though, the general evaluation mechanism as modeled here would remain largely unaltered. While I have outlined it in necessarily broad and suggestive strokes, the main virtues of the proposed model should be clear. It provides a unified treatment of constructions and constraints, and of various sorts of parameters involved in grammatical variation: lexical preferences, structural factors, as well as contextual cues for making a grammatical choice. Notes 1 This is not to say, of course, that there cannot be any acoustical variation in spoken discourse with respect to the height of pitch or amplitude a speaker wants to assign to an accented pronoun, but this is irrelevant to what is said in the text, viz. that it is generally clear whether or not a pronominal object receives an accent. So, accent (yes / no) is an abstraction of acoustic reality (since an accented pronoun can be accented to various degrees of strength), but not a distortion of it (since however strongly or lightly an accented pronoun is accented in a given context, it is clearly different from a non-accented pronoun in that context). That there is a clear difference between accented and unaccented object pronouns also appears from the fact that some object pronouns have two spellings, one of which can only be used if it is unaccented: you vs. ya/ye, her vs. ’er (e.g.€Fill ’er up!), them vs. ’em. Compare: (i) a. It was {you/her/them} who did it. b. *It was {ya/ye/’er/’em} who did it. 2 In his monograph on particle placement, Gries (2003:€ 83–85) obtained quite analogous results for different corpus data. Gries also considered length in terms of number of syllables, not just in terms of words. Both operationalizations of the factor ‘length’ proved relevant, though length measured in number of syllables appeared to be a slightly better predictor for particle placement (Gries 2003:€110). 3 See also Gries (2003:€14–15) for some complementary references and for brief discussion of the literature. 4 Except perhaps for some slight centralization, the pronoun it cannot be further reduced. However, this pronoun is in its present appearance already a reduction from the original form hit. An aspirated onset to the pronoun it can in fact still be heard in some isolated British and American dialects (cf.€AHD 2000: ‘Our living language’ note in the entry of it). 5 Besides, unaccented pronouns (me, you, him, her, …) are not shorter or simpler than many particles (up, in, out, off, …), so the light-before-heavy tendency (cf.€Behaghel’s law of increasing elements) cannot by itself account for the obligatory position of an unaccented pronoun on the left of the particle. Farrell’s (2005) cliticization/affixation analysis provides just such an additional explanation. 6 Admittedly, particles do get sentence-final accentuation in intransitive constructions (compare, e.g., Please come in! vs. *Please come in!), despite their status as closed-class items. 7 It is not clear whether the split ordering is chosen in (17) because this ordering allows a part of the focus domain (namely, the particle) to appear in the canonical position for focal elements (cf.€Gries 2003:€24–25, who refers to Halliday 1985:€185), or because this ordering simply allows

 Bert Cappelle a discourse-old constituent (namely, the object NP) to stay in non-final position (cf.€Section€2). I see no way of finding out whether the first reason is an independent aspect of the explanation, since I cannot think of an example in which the object NP is backgrounded (i.e. not a part of the focus domain) without also being discourse-old. (But note that, conversely, discourse-old constituents need not be backgrounded; cf.€Section€3.1.) Of course, the focality of the particle and the discourse-oldness of the object NP may be independent factors reinforcing each other in (17). 8 However, it is possible to use the joined ordering if the object NP is not actually backgrounded but nonetheless considered to be less important information than the action expressed by the verb-particle combination. In that case, the nuclear accent is placed on the particle (cf.€Bolinger 1971:€54). For example, consider the parenthetical clause in the sentence below:

(i) When you figure out which city you’re in, click on the Globulator icon (who thinks up these names?) and then click on the world map to select the correct city. (BNC)

9 Note that there are no adjacently accented syllables if the inflected verb ends in an unstressed syllable, as in He lifted up the book. 10 I would not go so far as to claim, though, that the object NP and the particle form a true ‘small clause’ here, as they do in (i) below:

(i) [SC The red and amber lights off] is a necessary starting situation for turning the green light on.

See Farrell (2005) for an extensive critique against the proposal, often made in the generative literature, that the object and the particle in a case like turn the green light on also form such a small clause. 11 We also find examples like the following, where the joined ordering is used:

(i) When he was ready to project a transparency, he turned off the lights before turning on the projector and ended up tripping over some electrical cords in the dark. (www)

It could be argued that in this example, the foci are not off vs. on (and the lights vs. the projector) but rather, the full verb-particle combinations turn on vs. turn off (and their respective objects). Some evidence that the verb is also focused on is provided by the impossibility to omit it in a conjunction reduction construction. Naturally, a speaker would never omit an element that is focused. Compare: (ii)

a. He turned off the lights and turned on the projector. b. *He turned off the lights and Ø on the projector.

By contrast, if only the particle and the object NP are focused, the verb itself can be omitted: (iii) a. He turned the lights off and turned the projector on. b. He turned the lights off and Ø the projector on. 12 This example also shows that the knowledge of a frame need not be shared by all members of a large linguistic community: the frame invoked in (25) – the ‘(older) child going to bed’ frame – is part of the shared world knowledge of a single family only. 13 Some of the determiner-noun sequences in (28) show only a part of the full object NP. For example, among the instances of her cup, we may have (and actually do have) an example like her cup of coffee. Similarly, a result like her shopping may mask the actual object NP, which is (in one of the two cases) her shopping bag, and his pocket is part of the NP his pocket Bible. However, this minor problem does not render our general findings suspicious in any way.



Contextual cues for particle placement  14 I am ignoring here an observation that is perhaps more conspicuous, namely that out of the 50 most frequently used NPs in the joined ordering, 15 contain two-syllable nouns and another 4 even 3-syllable nouns, while there are only 6 two-syllable nouns among the 50 most frequently objects in the split ordering. The correlation between the length of the object NP and ordering preferences was dealt with in Section€2 above. 15 I pass by the question as to what happens when option A violates lower-ranked but more constraints than option B. In Optimality Theory, A will still win out in such a case. In probabilistic grammar, a more sophisticated metric would indicate which option is the most suitable one, on the basis of all the factors and their associated weights. 16 So, we ignore here, among other things, the lexical dependency of the particle on the verb in the example under discussion. 17

http:// www.google.com/language_tools?hl=en

18 I allow myself to make the perhaps unwarranted assumption that, for all practical purposes, search pages from Great Britain reflect British English usage, and search pages from the US reflect American English usage. 19 We excluded literalness as a context-related factor in Section€9, but Geeraerts (2006) would probably consider this factor as belonging to “the cognitive context of meaning and experience”. 20 See Farrell (2005:€ 125–128) for the ranking of some broad syntactic word-order constraints. The most important constraint in his ranking is the one stipulating that heads have to be leftmost in the phrases they expand (so that the verb is the first element in the VP).

References Aarts, Bas (1989). Verb-preposition constructions and small clauses in English. Journal of Linguistics, 25 (2), 277–290. AHD (2000). The American Heritage Dictionary of the English Language, Fourth edition. Boston: Houghton Mifflin. Arnold, Jennifer E., Thomas Wasow, Anthony Losongco & Ryan Ginstrom (2000). Heaviness vs. newness: The effects of structural complexity and discourse status on constituent ordering. Language, 76 (1), 28Â�Â�–55. Behaghel, Otto (1909/1910). Beziehungen zwischen Umfang und Reihenfolge von Satzgliedern. Indogermanische Forschungen, 25, 110–142. Behaghel, Otto (1930). Von deutscher Wortstellung. Zeitschrift für Deutschkunde, 44, 81–89. Behaghel, Otto (1932). Die deutsche Syntax: Eine geschichtliche Darstellung, Volume 4: Wortstellung, Heidelberg: Winter. Bever, Thomas G. (1970). The cognitive basis for linguistic structures. In J. R. Hayes (ed.), Cognition and the Development of Language (279–362). New York: Wiley and Sons. Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad & Edward Finegan (1999). Longman Grammar of Spoken and Written English. Harlow: Longman. Bock, J. Kathryn (1977). The effect of pragmatic presupposition on syntactic structure in question answering. Journal of Verbal Learning and Verbal Behavior, 16, 723–734. Bock, J. Kathryn (1986). Syntactic persistence in language production. Cognitive Psychology, 18, 355–387.

 Bert Cappelle Bolinger, Dwight (1971). The Phrasal Verb in English. Cambridge, MA.: Harvard University Press. Büring, Daniel (2005). Semantics, intonation and information structure. Ms. University of California at Los Angeles. <semanticsarchive.net/Archive/GQ0YjgxM/buring.information. structure.v2005.pdf> Cappelle, Bert (2005). Particle patterns in English: A comprehensive coverage. Ph.D. dissertation. K.U.Leuven. Cappelle, Bert (2006). Particle placement and the case for ‘allostructions’. In: Doris Schönefeld (ed.), Constructions all over: Case studies and theoretical implications [special volume of Constructions], urn:nbn:de:0009–4-6839. < http://www.constructions-online.de/articles/ specvol1/683 > Chen, Ping (1986). Discourse and particle movement in English. Studies in Language, 10 (1), 79–95. Chomsky, Noam (1964 [1961]). On the notion “rule of grammar.” In J. A. Fodor & J. J. Katz (eds.), The Structure of Language: Readings in the Philosophy of Language (119–136). Englewood Cliffs, NJ: Prentice-Hall. Chomsky, Noam (1975). The Logical Structure of Linguistic Theory. New York & London: Plenum. Dehé, Nicole (2000). On particle verbs in English: More evidence from information structure. In N. M. Antrim, G. Goodall, M. Schulte-Nafeh, & V. Samiian (eds.), Proceedings of the 28th Western Conference on Linguistics 11 (WECOL 1999) (92–105). California State University at Fresno: Department of Linguistics. Dehé, Nicole (2001). Intonation patterns of particle verb constructions in English. NELS, 31, 183–197. Dehé, Nicole (2002). Particle Verbs in English: Syntax, Information Structure and Intonation. Amsterdam & Philadelphia: John Benjamins. Diessel, Holger & Michael Tomasello (2005). Particle placement in early child language: A multifactorial analysis. Corpus Linguistics and Linguistic Theory, 1 (1), 89–112. Dik, Simon C. (1997 [1978]). The Theory of Functional Grammar, Part 1: The Structure of the Clause, Second, revised edition, edited by Kees Hengeveld, Berlin & New York: Mouton de Gruyter. Erades, Peter A. (1961). Points of modern English syntax XL, English Studies, 42, 56–60. Farrell, Patrick (2005). English verb-particle constructions: Constituency and order. Language, 81 (1), 96–137. Fraser, Bruce (1966). Some remarks on the Verb-Particle Construction in English. In F. P. Dinneen (ed.), Problems in Semantics, History of Linguistics, Linguistics and English (45–61). Washington, DC: Georgetown University Press. Fraser, Bruce (1976). The Verb-Particle Combination in English. New York: Academic Press. Geeraerts, Dirk (2005). Lectal data and empirical variation in Cognitive Linguistics. In F. Ruiz de Mendoza Ibañez & S. Peña Cervel (ed.), Cognitive Linguistics. Internal Dynamics and Interdisciplinary Interactions (163–189). Berlin & New York: Mouton de Gruyter. Geeraerts, Dirk (2006). Towards a multifactorial grammar. Paper presented at the Second Biennial Conference on Cognitive Science, 9–13€June 2006, State University of St. Petersburg, Russia. Goldberg, Adele E. (1995). Constructions: A Construction Grammar Approach to Argument Structure. Chicago & London: The University of Chicago Press.



Contextual cues for particle placement  Goldberg, Adele E. (2003). Constructions: A new theoretical approach to language. Trends in Cognitive Science, 7 (5), 219–224. Gonnerman, Laura M. & Celina R. Hayes (2005). The professor chewed the students … out: Effects of dependency, length, and adjacency on word order preferences in sentences with verb particle constructions. In B. G. Bara, L. Barsalou, & M. Bucciarelli (eds.), Proceedings of the 27th Annual Conference of the Cognitive Science Society (785–790). Mahwah, NJ: Lawrence Erlbaum [Cognitive Science Journal Archive maintained by Rensselaer Polytechnic Institute: Cognitive Science Department]. < http://www.cogsci.rpi.edu/CSJarchive/Proceedings/2005/index.htm > Gries, Stefan Thomas (2003). Multifactorial Analysis in Corpus Linguistics: A Study of Particle Placement. New York & London: Continuum. Gries, Stefan Thomas (2005). Syntactic priming: a corpus-based approach. Journal of Psycholinguistic Research, 34 (4), 365–399. Gries, Stefan Thomas & Anatol Stefanowitsch (2004). Extending collostructional analysis: A corpus-based perspective on ‘alternations’. International Journal of Corpus Linguistics, 9 (1): 97–129. Gussenhoven, Carlos (2004). The Phonology of Tone and Intonation. Cambridge: Cambridge University Press. Hawkins, John A. (1994). A Performance Theory of Order and Constituency. Cambridge: Cambridge University Press. Huddleston, Rodney. (2002). The clause: complements. In R. Huddleston. and G. K. Pullum (eds.), The Cambridge Grammar of the English Language (213–321). Cambridge: Cambridge University Press. Kruisinga, Etsko (1931/1932). A Handbook of Present-day English, Part II: English Accidence and Syntax. Fifth edition. Groningen: P. Noordhoff. Levelt, Willem J. M., & Stephanie Kelter (1982). Surface form and memory in question answering. Cognitive Psychology, 14 (1), 78–106. Lohse, Barbara, John A. Hawkins & Thomas Wasow (2004). Domain minimization in English verb-particle constructions. Language, 30 (2), 238–261. McIntyre, Andrew (2003). The semantic and syntactic decomposition of get and have: An interaction between verb meaning and particle placement. Ms. Universität Leipzig. Peters, Julia (2001). Given vs. new information influencing the ordering of constituents in the verb-particle construction. In R. Brend, A. K. Melby and A. Lommel (eds.), LACUS Forum XXVII: Speaking and Comprehending (133–140). Fullerton, CA: LACUS. Poutsma, H. (1928). A Grammar of Late Modern English, Part I: The Sentence, Second edition. Groningen: P. Noordhoff. Prince, Ellen F. (1981). Toward a taxonomy of given-new information. In P. Cole (ed.), Radical Pragmatics (223–255). New York & London: Academic Press. Prince, Alan & Paul Smolensky (2004 [1993]). Optimality Theory: Constraint Interaction in Generative Grammar. Malden, MA, Oxford & Carlton, Victoria: Blackwell. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech & Jan Svartvik (1985). A Grammar of Contemporary English. Harlow: Longman. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech & Jan Svartvik (1985). A Comprehensive Grammar of the English Language. Harlow: Longman. Ross, John Robert (1986 [1967]). Infinite Syntax! (Language and Being 5). Norwood, N.J.: Ablex. [publication of: Constraints on Variables in Syntax. Ph.D. dissertation. Massachusetts Institute of Technology]

 Bert Cappelle Stefanowitsch, Anatol and Stefan Thomas Gries (2002). A construction-grammar approach to linear-precedence principles. Paper presented at the Second International Conference on Construction Grammar, 6–8€September 2002, University of Helsinki, Finland. Svenonius, Peter (1996). The optionality of particle shift. Working Papers in Scandinavian Syntax, 57, 47–75. Szmrecsanyi, Benedikt (2005). Language users as creatures of habit: A corpus-based analysis of persistence in spoken English. Corpus Linguistics and Linguistic Theory, 1 (1), 113–150. Van Dongen, W. A. Sr. (1919). HE PUT ON HIS HAT and HE PUT HIS HAT ON. Neophilologus, 4, 322–353. Ward, Gregory, Betty Birner & Rodney Huddleston (2002). Information packaging. In R. Huddleston and G. K. Pullum (eds.), The Cambridge Grammar of the English Language (1363– 1447). Cambridge: Cambridge University Press. Wasow, Thomas (1997). Remarks on grammatical weight. Language Variation and Change, 9 (1), 81–105. Wasow, Thomas (2002). Postverbal Behavior. Stanford: CSLI Publications. Weiner, E. Judith & William Labov (1983). Constraints on the agentless passive. Journal of Linguistics, 19, 29–58.

Conditionals and mental space set-up Evidence from German word order Ilona Vandergriff Recent studies of conditionals in English have cited putative evidence from German word order in support of the functional distinction between content and speech-act conditionals (Dancygier & Sweetser 2005; Declerck & Reed 2001) while counterexamples have largely been ignored. The present study takes a closer look at the syntactic marker in question, namely integrative and nonintegrative word order in wenn ‘if ’-initial conditionals. The findings raise doubts about the hypothesized correlation of the construction’s syntactic form and the cognitive domain in which it is interpreted. In this paper I explore how clausal integration interacts with mental space construction in conditional speech acts and show that integration does not mark content or predictive conditionality. Rather, clausal integration signals that the if-clause (P) is the sole space where the apodosis (Q) holds and that in all other spaces, ~ Q would take Q’s place. From the perspective of Construction Grammar (CG), the form-function mapping of syntactic integration and contrastivity observed in the German data provides a window on functional distinctions which may exist but are not formally marked in other languages.

1. Conditionals and mental spaces Conditionals set up a “mental space” (Fauconnier 1985), defined as “construct[s] distinct from linguistic structures but built up in any discourse according to guidelines provided by the linguistic expressions” (Fauconnier 1985:€16). These mental spaces can be set up in various domains, in time or place, for example.1 Typically, “if ” acts as a space builder in conditionals. All conditionals share the simple format of if P, Q, yet different types are associated with different meanings. For present purposes it will be important to distinguish between predictives, or content conditionals and non-predictives, or speech act conditionals.2 In (1), exemplifying a content conditional, the speaker is predicting whether or not a hacker by the name of Hiro will transfer some data.

 Ilona Vandergriff



(1) If Hiro reaches out and takes the hypercard, then the data it represents will be transferred from this guy’s system into Hiro’s computer. (Dancygier & Sweetser 2005:€33) Content conditionals show a causal or enablement dependency between two events or states of affairs. Here, if sets up a primary conditional space and an alternative space. Diagram 1 “represents two alternative mental space set-ups […] both interpreted as potential futures of the current base” (Dancygier & Sweetser 2005:€32). According to the authors, this alternative space gives rise to an if and only if implicature. Here if P, Q carries the implication if ~P, ~Q. Base/Present Hiro is offered the hypercard

If /Future neutral stance

Alt/Future

Hiro takes the card

Hiro doesn’t take the card

Ext/Future

Alt/Ext

data are transferred

data not transferred

Diagram 1.╇ Example (1) If Hiro reaches out and takes the hypercard, then the data it represents will be transferred from this guy’s system into Hiro’s computer



Conditionals and mental space set-up 

Predictives thus set up exactly two alternative scenarios. In Dancygier and Sweetser, alternativity is defined as the existence of exactly two incompatible mental spaces, generated in the same context from the same base space, which “cannot be combined into a single coherent space” (Dancygier & Sweetser 2005:€35). What makes prediction possible is that that these two alternative scenarios, P and ~P and their correlates Q, and ~Q, respectively, exhaust all possibilities. In these predictives neither the if-clause nor the main clause is asserted, rather it is their relationship. Dancygier and Sweetser explain: The if-clause cannot express an assertion because the conjunction sets up a hypothetical mental space, different from the Base space where assertions could be made. The main clause describes the predicted result in that same hypothetical space so it is not asserted either. What a predictive conditional asserts is the causal dependency and correlation between the events or states of affairs described by its clauses (2000:€122).

Conditional speech acts, by contrast, show a different relationship between P and Q. In (2) the purpose of the if-clause is to establish as shared ground the reason for producing Q. Dancygier and Sweetser do not show an alternative speech-act space. Rather, for ~P the speaker would produce no apodosis at all.

(2) If you need any help my name is Ann. (Dancygier & Sweetser 2005:€111)

Unlike predictive conditionals exemplified in (1), speech-act conditionals like (2) are described as one-scenario structures that do not allow a biconditional reading, as shown in (3).

(3) * If you don’t need help (then) my name is not Ann.

In contrast with predictives which neither assert P nor Q, speech-act conditionals do assert Q (Köpcke & Panther 1989; following König & van der Auwera 1988) with P specifying the speech-act space in which Q is relevant. In the conditional speech act the truth of Q is maintained regardless of the truth of P. According to Dancygier and Sweetser, the functional distinction between predictive content and non-predictive speech-act conditionals is reflected in their linguistic form in a number of ways. Predictive conditionals, for example, regularly occur with correlative then (as illustrated in (1)) whereas speech-act conditionals will typically not accept it (cf. (3)). Dancygier and Sweetser write that then points deictically to a particular mental space, defined temporally or circumstantially, and locates the event or state described in the apodosis in that mental space. By insisting on location within a particular mental space, the conditional use of then brings on a natural inference that the content of the apodosis does not hold in other alternative mental spaces (why else insist that it holds specifically in this one?) (2005:€144).

 Ilona Vandergriff Base/Present Customers may need to ask for help, using the staff member’s name

If/Speech-Act Space

Hearer needs help

Ext/Speech-Act Space

Speaker introduces herself

Diagram 2.╇ Example (2) If you need help my name is Ann

They argue that correlative then is ungrammatical in (3) because it uniquely identifies “P as the (sole) space wherein Q holds” (2005:€149). Since the saleswoman’s name is Ann regardless of whether P holds or not, correlative then is incompatible with the intended meaning. However, not all conditional speech acts are incompatible with correlative then. Dancygier and Sweetser find that some speech-act contexts readily accept then as illustrated in Example (4). (4) If you’re so smart, (then) why aren’t you rich? (Dancygier and Sweetser 2005:€149). In their discussion, they argue that (4) differs from (2) in three ways. First, it allows the speaker to emphasize the uniqueness of P as the only space in which Q is relevant. Second, the relation between the clauses in (4) implies a predictive or inferential interpretation to the effect that the speaker would normally expect a smart



Conditionals and mental space set-up 

person to get rich. And, third, (4) is “susceptible to contrasting interpretations” (2005:€149) in that it gives expression to the speaker’s incredulity that an intelligent person would be poor. Dancygier and Sweetser thus show that within the class of speech-act conditionals - all of which specify the relevance condition for producing Q - differences exist with respect to the precise nature of the link between P and Q. They argue that not all of the observed variables carry equal weight with respect to the use of correlative then. While many speech-act contexts allow inferential, predictive, or contrastive readings, the authors argue that it is the emphasis on P’s uniqueness rather than other factors that motivates the use of then. 2. Clausal integration in German Following initial wenn ‘if ’-clauses, conditional consequents or apodoses predominantly occur with integrative word order in German, i.e., the finite verb immediately follows the if-clause or protasis. This inversion is the unmarked ordering in all complex clauses where the subordinate clause precedes the main clause. The subordinate clause can thus be taken as a constituent of the larger clause. Integrative conditionals generally accept resumptive dann ‘then’ (Köpcke & Panther, 1989). In spoken German, if-clauses are sometimes not syntactically integrated. Instead, the main clause will have the verb position of an independent main clause. This ordering is considered standard in speech-act conditionals. Compare the following minimal pair in Table 1. Table 1.╇ Syntactic Integration, Resumption, and Non-integration

Speech-Act

Content

Pre-prefield

Prefield

Verbal Bracket

Middle Field

(5) Wenn du Lust hast, (dann)

machen

wir morgen einen Kindergottesdienst.

if you interest have then

make-FUT we tomorrow a Sunday-school

‘If you feel like it we’ll have Sunday school tomorrow.’ (Speaker is offering to organize Sunday school with the help of the hearer if s/he feels like it.) (6) Wenn du Lust hast, wir machen morgen einen Kindergottesdienst.3 if you interest have we make-FUT tomorrow a Sunday-school ‘If you feel like it, tomorrow we’ll have Sunday school.’ (Speaker is inviting the hearer to join him/her for Sunday school if s/he feels like it.)

 Ilona Vandergriff

The conditionals in Table 1 show that different syntactic orderings sometimes force different interpretations, a predictive interpretation in (5) and a relevance interpretation in (6). In (5) the hearer’s attitude will cause or enable Sunday school to take place. In (6), by contrast, Sunday school will take place regardless of how the hearer feels about it. The if-construction allows the speaker to present the information tentatively, in case the hearer’s interested. Predictive conditionals such as (5) with their real-world relationship between the clauses typically occur with integrative word order, whereas speech-act conditionals like sentence (6) signal their looser conceptual dependency via non-integrative word order (Gohl 2002; Günthner 1999; Auer 1996, 2000; Pasch 2003; König & van der Auwera 19884). If confirmed, these putative form-function correlations attested in German are relevant to research on conditionality beyond German as they appear to provide formal evidence of the functional distinction between predictive (i.e., content) and non-predictive (i.e., speech-act) conditionals (Dancygier 1998; Dancygier & Sweetser 2005; Declerck & Reed 2001). Drawing on results from Köpcke and Panther’s 1989 paper, Dancygier and Sweetser (2005:€139) conclude that “studies of German have provided even more extensive evidence [than English or Greek] in support of similar grammatically marked function distinctions.” For German, clausal integration and resumption (using e.g. dann ‘then’) are claimed to lend support to the functional distinction between content and speech-act conditionals. However, in the same study, Köpcke and Panther (1989) also discuss counterexamples which show that some speech-act conditionals occur with integration but neither Köpcke and Panther – nor Dancygier and Sweester - call into question the hypothesized correlation between speech-act conditionality and clausal nonintegration. Rather, Köpcke and Panther posit an overriding pragmatic parameter to explain why the putative form-function correlation does not hold in all cases.5 Table 2.╇ Köpcke and Panther’s Correlations between Word Order and Pragmatic Function (1989)6

Speech Act

Content

Clausal Integration

Clausal Non-integration

(7) Wenn er das erfährt, gibt es Ärger. (8) Wenn er das erfährt, das gibt Ärger! If he that finds-out gives it trouble. If he that finds-out that gives trouble. ‘If he finds out about it there’ll be trouble.’ ‘If he finds out about it, there will be trouble!’ (unmarked) (marked – signaling affect) (9) Wenn du meine Meinung hören willst, (10) Wenn du meine Meinung hören willst, die (dann) fallen die Aktien bald. Aktien fallen bald. If you my opinion hear want fall the If you my opinion hear want the stocks stocks soon fall soon ‘If you want my prediction, stock prices ‘If you want to hear my point of view, will drop soon.’ stock prices will drop soon.’ (marked – signaling speaker uncertainty) (unmarked)



Conditionals and mental space set-up 

While all speech-act conditionals allow non-integration, the authors present evidence which shows that integration and resumption are more acceptable in some speech-act conditionals than in others. Subjects were asked to rate German ifclauses with integration and resumption on an acceptability scale ranging from 1 (the sentence is acceptable without hesitation) to 5 (the sentence is not acceptable without hesitation.) The speech-act conditional Wenn Sie mich fragen, dann wird es bald schneien ‘If you ask me, then we’ll soon have snow’ received a rating of 1.73, whereas informants rated another resumptive speech-act conditional (Wenn du es noch nicht wusstest, dann ist Hans wieder im Lande. ‘If you didn’t know yet, then Hans is back in the country.’) relatively unacceptable at 4.68 (Köpcke & Panther 1989:€696). Based on their analysis, the authors claimed that clausal integration and resumption signals a decrease in assertability of the apodosis and hypothesized that speaker uncertainty about the truth of the apodosis can override the expected forms and trigger the variant orderings: … a relevance conditional may be grammaticalized as VS [verb-subject, i.e. inversion] if the speaker wishes to communicate that the content of the apodosis does not constitute an item of factual knowledge, but rather his/her opinion about the truth of the apodosis (1989:€685).

They continue “that the factor degree of certainty influences the word order in the apodosis. The higher the degree of the speaker’s uncertainty about the truth of the apodosis, the more likely s/he will resort to integrative word order in the consequent clause” (1989:€702). In effect, they assert that clausal integration correlates with both (unmarked) content conditionality and (marked) speaker uncertainty in speech-act conditionals. One formal marker is thus associated with two different pragmatic meanings. In sum, no one meaning has been linked to clausal integration in all types of conditionals. Köpcke and Panther’s (1989) findings that link clausal integration in speechact conditionals to lack of speaker certainty are based on a limited set of data. In all the examples they analyze, Q does indeed present the speaker’s opinion suggesting that the speaker may not be absolutely certain about the truth of the apodosis. However, Vandergriff (1995) raises questions about the validity of their findings. Her analysis based on a larger corpus shows that in some of the integrated speechact conditionals the speaker does not present an opinion but a fact. (11)

Wenn ich mal dabei bleibe, dann hat der Bundeskanzler gar in If I once with-it remain then has the Chancellor even in seiner Regierungserklärung am Freitag früh angeboten, his state-of-the-union-address on Friday early offered

 Ilona Vandergriff

darüber zu reden. about –it to talk. ‘If I pursue this for a moment, (then) the chancellor offered to talk about this in his state-of-the-union address on Friday morning.’ Vandergriff argues that there is no evidence of speaker uncertainty about the truth of the apodosis in (11). Her analysis is supported by Köpcke and Panther’s own test criterion for speaker uncertainty. They had argued that integrated and/or resumptive speech-act conditionals “may be followed by an additional sentence such as ‘aber vielleicht irre ich mich/habe ich unrecht, etc. (‘ but maybe I am mistaken/I am wrong’, etc.)” (1989:€703). Example (11) does not allow this reading. Instead, the speaker leaves no doubt about his certainty as he states the topic of the chancellor’s address. Vandergriff writes that the content of the apodosis or the speaker’s stance toward it are not relevant to the use of clausal integration. Instead, her analysis focuses on the interactional function of the syntactic marker. In (9) the speaker opts for clausal integration in expressing a discourse-dominating function. Building on Held (1992), she argues that “integrative word order reduces the proposition’s illocutionary force” (Vandergriff 1995:€ 79). This weakening reduces “the threat to the hearer’s negative face. The syntactic marker clausal integration, we can thus conclude, has primarily interactional function in the speech-act domain.” Both Köpcke and Panther’s (1989) and Vandergriff ’s (1995) studies build on König and van der Auwera’s (1988) work on the subject which is much broader in scope. Considering clause integration in German and Dutch conditionals, concessive conditionals and concessives, König and van der Auwera try to pinpoint what motivates clause integration and non-integration, respectively. Following Foley and Van Valin (1984), they suggest that the reasons for syntactic variation may be found on the semantic level: the more closely two clauses are semantically related, the more closely they are syntactically integrated. Conversely, “non-integrative word order occurs in complex sentences which assert or imply the independence and mutual irrelevance of two propositions or their coocurrence against a background of general incompatibility” (König & van der Auwera 1988:€128). In sum, previous research has shown that while clausal integration is primarily associated with predictive conditionality, it is compatible with speech-act conditionality under certain circumstances. Focusing on the speaker’s stance toward the content of the apodosis, Köpcke and Panther (1989) try to link clausal integration to a reduction of the validity of Q. Vandergriff ’s (1995) analysis, by contrast, claims that clausal integration reduces the illocutionary force of utterances with discourse-dominating functions such as topic change. In contrast to previous research that seeks to explain clausal integration for a limited set of conditional data, the present study will try to tease out what the



Conditionals and mental space set-up 

specific formal features contribute to the meaning of a conditional speech act in German. What precisely does clausal integration contribute in light of its conventional meaning “close semantic relation of the clauses?” After all, it would be economical to account for clausal integration, resumption and non-integration in conditional speech acts in the same way linguists account for clausal integration in general. To this end, I will firstly consider integration and secondly German resumptive dann as formal aspects of speech-act conditional sentences in German within the framework of Construction Grammar (Fillmore 1986, 1988, 1990a, 1990b; Fillmore, Kay & O’Connor 1988; Goldberg 1994; Shibatani & Thompson 1996). From this perspective, conditional sentences are examples of a construction, which Dancygier (1998:€ 5) defines as “a conventional pattern of linguistic structure which is paired with features of interpretation.” Constructions thus have predictable meanings, which are mapped onto the formal and semantic parameters or their combinations. 3. Clausal integration in speech-act conditionals It is a well established fact that all speech-act conditionals can occur and that the majority of them do occur with clausal non-integration. As outlined in the research overview, it is less well known that some speech-act conditionals do occur with clausal integration (König & van der Auwera 1988; Köpcke & Panther 1989). Therefore, the analyst needs to explain why syntactic integration – while not conventionally associated with speech-act domain conditionality – is compatible with it under certain circumstances. The following conditional, for example, can occur with both integrative and non-integrative word order. (12) Wenn du meine Meinung hören willst, die Aktien fallen bald. If you my opinion hear want the stocks go-down soon ‘If you want to hear my point of view the stocks will go down soon.’ (Köpcke & Panther 1989:€702) (13) Wenn du meine Meinung hören willst, fallen die Aktien bald. If you my opinion hear want go-down the stocks soon ‘If you want to hear my point of view the stocks will go down soon.’ (Köpcke & Panther 1989:€702)

 Ilona Vandergriff Base/Present Hearer may be interested in stock market predictions If/Future

Alt/Future

Hearer wants to hear speaker’s stock market prediction

Hearer does not want to hear speaker’s stock market prediction

Ext/Future

Alt/Ext

Stock prices will fall soon

Stock prices will not fall soon

Diagram 3.╇ Example (12) Wenn Du meine Meinung hören willst, die Aktien fallen bald. ‘If you want to hear my point of view the stocks will go down soon’

Example (12) is unambiguously interpreted in the speech act domain. The wenn clause purports to request the hearer’s permission as a condition for producing Q. The speaker’s prediction that stock prices will fall soon is relevant in the mental space described in P. Consider initially Diagram 3, which illustrates the most likely interpretation of sentence (12). In the alternative scenario, neither Q nor ~Q would be performed. For this reason, the alternative extension box is empty and the alternative scenario is shown as incomplete.7 If P, Q; if ~P, Ø, where Ø stands for an empty syntactic position. Without context, the integrated version (13) of the same conditional is pragmatically ambiguous. In a narrowly defined context in which the speaker’s opinion has regularly been followed by a fall in the stock market as though his prediction were a bad omen, this sentence can be construed to be a predictive conditional. Compare Diagram 4. In the absence of a suitable context, pragmatics would intervene here to block the predictive reading shown in Diagram 4. Instead, hearers will construct a more appropriate causal connection.



Conditionals and mental space set-up  Base/Present Hearer may be interested in stock market predictions If/Future

Alt/Future

Hearer wants to hear speaker’s stock market prediction

Hearer does not want to hear speaker’s stock market prediction

Ext/Future

Alt/Ext

Stock prices will fall soon

Stock prices will not fall soon

Diagram 4.╇ Example (13) Wenn Du meine Meinung hören willst, fallen die Aktien bald. ‘If you want to hear my point of view the stocks will go down soon’

Base/Present Hearer may be interested in stock market predictions

If/Speech-Act Space Hearer wants to Hearer wants to hear speaker’s hear person3’s stock market stock market prediction prediction

person4

person5

personn

Ext/Speech-Act Space Speaker makes prediction that Person3 makes stock prices prediction x will fall soon

prediction y

prediction z

predictionn

Diagram 5.╇ Example (13) Wenn Du meine Meinung hören willst, fallen die Aktien bald. ‘If you want to hear my point of view the stocks will go down soon’

 Ilona Vandergriff

Diagrams 3 and Diagrams 5 show conditional readings of (12) and (13) in the speech-act domain which differ in important ways. In the default reading shown in Diagram 3 Q holds in P, but the truth of Q (i.e., that the stocks will be falling) is maintained irrespective of the truth of P. Here, the speaker’s opinion Q is presented as relatively independent of P. In Diagram 5, by contrast, P is identified as the sole mental space where Q is relevant. In all other instances ( P2, P3, P4, Pn) contrasting apodoses (Q2, Q3, Q4, Qn) are inferred. I propose that the range of contrasting scenarios if P2, Q2, if P3, Q3, etc. can be captured in the sentential hedge aber man kann auch etwas anderes sagen ‘but one can also say something different’. The conditional speech act readily accepts the hedge. To test the hypothesized form-function mapping, I will check whether conditional speech acts which are not compatible with integration or resumption do or do not accept the hedge. Before I discuss these examples, it is important to clarify the concept of contrastivity, how it relates to alternativity (Dancygier & Sweetser 2005), and how it will help to explain formal variations in speech-act conditionals. Contrastivity is akin to Dancygier and Sweetser’s concept of alternativity defined as “two spaces [that] are incompatible” (2005:€35). Both concepts, contrastivity and alternativity, share the basic mental structure that Q will hold only in P, not in any other space. For ~P, something other than Q will be performed. The difference between alternativity and contrastivity is that alternative-based conditionals set up exactly two alternative, mutually exclusive scenarios. Because only two scenarios are set up, and because one of these is specified in toto, the speaker implicitly sets up a specific alternative mental scenario. Constrastivity, on the other hand, means that ~P, namely P2, P3, P4, Pn would trigger one of n-tuple ~Q’s (Q2, Q3, Q4, Qn). Because more than one contrast provides the background to the highlighted conditional speech act, no specific scenario is evoked in ~Q’s. As a result, the other contrasting options are not mutually exclusive and cannot be specified on the pragmatic scale. Contrastive speech-act scenarios not only provide a justification for performing Q – as all other speech-act conditionals do - but also signal that P is the sole space where Q holds. Compared to other conditional speech-acts such as Sentence 2 (Diagrams 2 and 3), which evoke two alternative scenarios - one in which Q is performed and one in which it is not - contrastive conditional speech acts restrict the relevance of Q by evoking contrasting options. This contrastivity and alternativity, respectively, signals that Q holds in P only and is thus functionally dependent on P. I will argue that this functional dependence is mapped onto the syntactic level, resulting in clausal integration or resumption. In consideration of contrastivity, we can now explain why many of the integrated speech act conditionals convey the speaker’s personal opinion. These conditionals present the speaker’s contribution as one in an array of possible



Conditionals and mental space set-up 

contributions among many in the speech-act space. Hence, Q appears constrained in its relevance, albeit not in its validity. What needs to be shown is that the hypothesized contrastivity can account for clausal integration even in conditional speech acts, where Köpcke and Panther’s pragmatic parameter failed, namely, in cases where the speaker appears quite certain about the truth of the apodosis. The sentential hedge “aber vielleicht irre ich mich/habe ich unrecht” [but maybe I am mistaken/I am wrong’] (1989:€703) is too narrow here where the apodosis is presented as a fact. Going back to (11), cited again here as (14), for example, the speaker wishes to project the image of a competent professional who is certain about the truth of the apodosis. (14)

Wenn ich mal dabei bleibe, dann hat der Bundeskanzler gar in if I once with-it remain then has the Chancellor even in seiner Regierungserklärung am Freitag früh angeboten, his state-of-the-union-address on Friday early offered darüber zu reden. about-it to talk. ‘If I pursue this for a moment, (then) in his state of the union address on Friday morning the chancellor offered to talk about this.’

Wenn.. noch von meinem Fach her was gestattet if still from my field from something permitted ist, zur juristischen Form der--der Kaufverhandlungen, so is to-the legal form of-the purchase-negotiations then stellt also sich jedwede Handlung dieser represents therefore itself each action of-this Art als ein Vertrag dar kind as a contract verb ‘If I may add something from my professional background to the legal process of the purchase talks, (then) every process of this type is a legal contract…’ (Vandergriff, 1995:€269) (15)

In Example (14), the speaker appears to contrast his topic maintenance (dabei bleibe ‘stick with it’) with topic change, which would trigger other discourse contributions. If he moved on to a different topic he might mention the Foreign Minister’s comments on a similar topic, for example, or ask a political opponent a question. In (15) the speaker explores the present discourse topic further by contributing information from his area of expertise rather than moving on to a new topic. As hypothesized, both (14) and (15) accept the sentential hedge. In order to provide support for the hypothesized form-function correlation between clausal integration and contrastivity, it must be shown that conditional

 Ilona Vandergriff

speech acts that do not accept the hedge also do not allow clausal integration or resumption. (16) Wenn du es noch nicht wusstest, die Aktien fallen bald. if you it still not knew the stocks go-down soon (17) ? /*Wenn du es noch nicht wusstest, (dann) fallen die Aktien bald. if you it still not knew then go-down the stocks soon ‘If you didn’t know the stocks will go down soon.’ With the protasis wenn du es noch nicht wusstest ‘ if you didn’t know’ the speaker sets up an alternative-based mental space, one in which addressee does not yet know as opposed to one where he does already know. For ~P wenn du es schon wusstest ‘if you already knew’ the speaker would produce no apodosis. Unlike the apodosis P in (12) and (13) which evokes contrasting P1, P2, P3, Pn such as the hearer wanting to hear John’s or Susan’s opinion on the stock market P in (16) and (17) evokes exactly one ~P. Not surprisingly, the sentential hedge is not accepted. (* If you didn’t know (I am telling you that) the stocks are falling but one can also say something else.) Example (17) demonstrates that speech-act conditionals that set up exactly one alternative in ~P and are not open to a range of contrasting ~P’s do not accept integration. This grammaticality judgment is supported by Köpcke and Panther’s 1989 experimental results, reported on in the research overview. In summarizing the results, Table 3 shows that speech-act conditionals can be classified into those that evoked contrasting scenarios and are therefore compatible with integration and resumption and those which do not. Predictive conditionality is conventionally associated with integrative word order. Integrative conditionals, in turn, regularly allow the insertion of correlative dann or so ‘then’ (Köpcke & Panther 1989). The conditions which constrain clausal integration in conditional speech acts should also hold for resumption (marked Table 3.╇ Clausal Integration and Relation of P and Q in Conditional Speech Acts Domain

Clausal Integration

Relation of P and Q

Speech-Act

Integration Resumption

Contrastive Scenarios If P, (then) Q is said; If P2, (then) Q2 is said; If Pn, (then) Qn is said No Contrastive Scenario If P, Q is said; If ~P, Ø is said

Non-integration

P is the sole space where Q holds

P is one space where Q holds



Conditionals and mental space set-up 

by dann or so ‘then’). Since German dann, in turn, is “semantically similar” to English then and is “used with a function somewhat parallel to that of then in conditional apodoses” (Dancygier & Sweetser 2005:€171), it will be interesting to see if the findings for English then also apply to German dann/so. Before I look at some examples, it will be useful to describe the function of German dann and English then. English then, much like its equivalent German counterparts, seems to function as a temporal and deictic pointer, indicating a mental space in which the event or state described in Q will hold. In fact, it appears that German dann is even more similar to English then than Dancygier and Sweetser (2005) claimed. In stating only that German dann is used in predictive but not in speech-act conditionals, they stop short of describing the full range of functions dann interacts with. German dann is actually compatible with speech act conditionality, just like its English counterpart, and subject to some of the same restrictions.8 Both dann and then occur only in speech-act contexts that restrict Q’s relevance to the particular mental space described in P. Consider Austin’s famous example, which – according to Sweetser and Dancygier (2005) - some speakers deem acceptable with correlative then. (18) If you’re hungry (then) there are biscuits on the sideboard. I agree with Dancygier and Sweetser who argue that correlative then is acceptable in certain contexts, for example, if the addressee has just indicated that s/he is hungry. We can generalize that all speech-act conditionals with correlative then identify P as the sole space where Q is relevant and that any alternative to P such as P2, P3, Pn would correlate with Q2, Q3, Qn,, respectively. P (i.e., if you are hungry) appears as one of many possibilities, for instance, in opposition to if you’re thirsty or if the baby is hungry, each matched to its relevant extension in the alternative speech act space. For example, if you’re thirsty might cause the speaker to go on to say (then) there is a beer in the fridge whereas if the baby is hungry (rather than you) (then) there is some applesauce. In German, speech-act conditionals involving contrastivity are also compatible with resumptive dann. (19) Wenn du Hunger hast, (dann) stehen da Kekse auf dem Buffet. if you hunger have (then) stand there biscuits on the sideboard. ‘If you are hungry (then) there are biscuits on the sideboard.’ Based on Dancygier and Sweetser’s (2005) findings for English conditional speech acts, I have argued that German dann much like English then emphasizes that Q’s relevance is restricted to the state or event described in P. As a deictic marker the correlate picks out one option P among possible options (P2, P3, P4, Pn) along with its extension (Q2, Q3, Q4, Qn).

 Ilona Vandergriff

By setting up an alternative space or multiple contrasting spaces, the mental space described in P becomes the sole space where Q will hold. Contrastivity and alternativity in conditionals, then, necessarily involve the semantic-pragmatic dependence of the apodosis on the protasis. In speech-act conditionals where no contrastivity is implied as in (2) or (16), Q will hold in spaces other than P, reducing Q’s dependence on P. The relative assertability of Q and its functional dependence on P are formally marked by clausal integration, resumption, and non-integration, respectively. With clausal integration marking a decrease in assertability and greater semantic-pragmatic dependence, integration and resumption are thus iconic markers in German conditionals, including speech-act conditionals. They signal that the protasis P is not only one space but rather the only space in which the utterance is relevant. I have shown above how a speaker can link an utterance Q to a relevance condition and how clausal integration or resumption mark the content of the apodosis as one option among an array of scenarios. We are now in a position to see why clausal integration has been linked to politeness functions (Vandergriff 1995). For example, in the context of discourse-dominating functions such as topic change, the speaker’s utterance may appear as a potentially face-threatening act. Restricting its relevance to a sole space minimizes the threat and provides the addressee with options on the discourse record whereas non-integration would present Q with full illocutionary force. By signaling to the hearer that Q is only one of many alternative options, the speaker reduces Q’s independence and assertability. However, it is not always more polite to reduce the independence of Q. Dancygier and Sweetser point out that in some contexts, for example, when making an offer as in If you’re hungry, then there are biscuits on the sideboard “it seems rude to insist on the specific circumstances under which the biscuits are offered, by adding then” (2005:€149). Clausal integration (and resumption) therefore is not a politeness marker but may serve a politeness function in speech-act conditionality in contexts where insisting on exclusiveness is more polite. While it seems at first glance that discourse-dominating functions that threaten the hearer’s negative face (Brown & Levinson 1989) by requesting permission to speak or manipulate the topic progression will tend to show clausal integration while offers do not, further research on conditionals in extended discourse will be necessary to explore the link between alternativity/contrastivity, clausal integration/resumption, and politeness functions.



Conditionals and mental space set-up 

4. Conclusions In conclusion, the findings provide strong evidence that the syntactic differentiation of clausal integration and non-integration, respectively, marks neither the distinction between predictive and non-predictive conditionals, nor between content and speech-act conditionals. Instead, clausal integration signals the semanticpragmatic dependence of the apodosis on the protasis, emphasizing that P is the sole space where Q holds. The concept of contrastivity in conditionals (paralleling Dancygier and Sweetser’s concept of “alternativity”) as outlined here helps to bridge some of the previous research results. Whereas Köpcke and Panther’s (1989) analysis tried to link clausal integration in speech-act conditionals with speaker stance, the present study has shown that it signals contrastivity, with P as the only space where Q is relevant. Speaker stance may also be involved but only as a pragmatic epiphenomenon. In other words, clausal integration signals that the opinion is only one among many and not that the opinion is limited in its validity as Köpcke and Panther hypothesized. Whereas Köpcke and Panther’s account posited an overriding parameter to account for clausal integration in speech-act conditionals, Construction Grammar requires a clear form-function mapping. Each form or combination of forms is conventionally associated with a certain meaning or function. The proposed account which views clausal integration as an iconic marker aligns well with Construction Grammar and is consistent with studies that link syntactic dependence to semantic/pragmatic dependence. My findings thus complement and further illuminate König and van der Auwera’s result, who linked non-integration to “separate assertability” (1988:€126). Clausal integration and resumption with correlative dann constitute aspects of constructional meaning and hence play an important role in communication as they facilitate the hearer’s search for possible interpretations. In Sanders et al. (1993:€94) understanding a discourse “means constructing a coherent representation of that discourse…” (cited in Rouchouta 1998:€ 21). From this perspective, both markers along with other aspects of constructional meaning guide the hearer in his/her interpretation. The form-function correlation between contrastivity and alternativity on the one hand and clausal integration (as well as correlative dann and so ‘then’) in German on the other, emphasize the dependence of the apodosis on the protasis in German conditional speech acts much like correlative then does in English conditional speech acts. Further research will be necessary to determine the extent of the functional overlap between the correlative markers in German and English.

 Ilona Vandergriff

Notes 1. Temporal when signals the speaker’s positive stance toward the contents of P, whereas if signals non-positive stance. Dancygier and Sweetser point out that “if is not try neutral in stance, but is indeterminate between a range of stances including almost everything except complete positive stance P or ~P” (2005:€53). German wenn, by contrast, is ambiguous between temporal and conditional meanings, i.e., the speaker stance is not formally marked. In most cases, speaker stance can be disambiguated from context, in some cases it remains unclear. Only in combination with subjunctive mood wenn-clauses are unambiguously conditional (Auer 2000:€7). 2. Examples and glosses for 1) – 2) from Sweetser (1990). In addition to the content conditionals and speech-act conditionals discussed, Dancygier and Sweetser distinguish two other types of conditionals, namely epistemic such as If John went to that party, (then) he was trying to infuriate Miriam (If I know that John went to the party, then I conclude that he went to infuriate Miriam.) and metalinguistic conditionals of the type Ok, I’ll have a tomahto, if that’s how you pronounce it (If my pronunciation is acceptable I’ll take the tomato.)(All examples and glosses for 1) – 3) from Sweetser, 1990; examples 1) – 4) cited in Dancygier 1998.) 3.

Example has been adapted/standardized from Günthner (1999:€210).

4. König and van der Auwera (1988) describe the correlation between form and function as tendential rather than absolute. They make reference to the “general rule which requires or, at least favors, non-integrative word order in Dutch and German, whenever an initial adverbial clause relates to the speech act performed in uttering the following clause, rather than to the proposition expressed by it.” (1988:€110) 5. Neither Declerck and Reed (2001) nor Dancygier and Sweetser (2005) discuss the German counterexamples to the hypothesized link between content conditionality and integration, and between speech-act conditionality and non-integration. 6. Examples and glosses from Köpcke and Panther (1989). 7. While my analysis parallels Dancygier and Sweetser, my diagram differs from theirs for similar speech-act conditionals. Their diagram for speech-act conditionals (see Diagram 2) shows the alternative set-up as non-existent rather than incomplete. The modified diagram allows the analyst to highlight that even in an alternative space no alternative to Q would be performed. 8. While English then and German dann/so seem semantically and functionally similar, further study involving authentic data will be necessary to determine the area of overlap. Generally speaking, clausal integration and German dann seem possible in contexts whose English parallels would be compatible with then. However, the reverse does not seem to be the case, as resumption in German seems to enjoy broader distribution than in English.

References Auer, Peter (1996). The pre-front field in spoken German and its relevance as a grammaticalization position. Pragmatics, 6(3), 295–322. Auer, Peter (2000). Pre- and post-positioning of wenn-clauses in spoken and written German. In E. Couper- Kuhlen & B. Kortmann (Eds.), Cause, condition, concession, contrast (173–204). Berlin & New York: de Gruyter.



Conditionals and mental space set-up  Brown, Penelope & Stephen C. Levinson (1987). Politeness. Some universals in language usage. Cambridge: Cambridge University Press. Dancygier, Barbara (1998). Conditionals and prediction. Cambridge: Cambridge University Press. Dancygier, Barbara & Eve Sweetser (1996). Conditionals, distancing, and alternative spaces. In A. Goldberg (Ed.), Conceptual structure, discourse and language (83–98). Stanford, CA: CSLI Publications. Dancygier, Barbara & Eve Sweetser (2000). Constructions with if, since and because. In E. Couper-Kuhlen & B. Kortmann (Eds.), Cause, condition, concession, contrast (111–142). Berlin & New York: de Gruyter. Dancygier, Barbara & Eve Sweetser (2005). Mental spaces in grammar. Cambridge: Cambridge University Press. Declerck, Renaat & Susan Reed (2001). Conditionals. A Comprehensive Empirical Analysis. Berlin & New York: Mouton de Gruyter. Fauconnier, Gilles (1985). Mental spaces: Aspects of meaning construction in natural language. Cambridge, Mass.: MIT Press. Reprinted 1994, Cambridge: Cambridge University Press. Fillmore, Charles (1986). Varieties in conditional sentences. ESCOL 3 (Eastern States Conference on Linguistics), 163–82. Fillmore, Charles (1988). The mechanisms of ‘Construction grammar.’ In S. Axmaker, A. Jaisser, & H. Singmaster (Eds.), Proceedings of the 14th Annual Meeting of the Berkeley Linguistics Society, (35–55). Berkeley, CA: Berkeley Linguistics Society. Fillmore, Charles (1990a). Epistemic stance and grammatical form in English conditional sentences. In M. Ziolkowski, M. Noske, & K. Deaton (Eds.), Papers from the 26th Regional Meeting of the Chicago Linguistic Society, (137–62). Chicago: Chicago Linguistic Society. Fillmore, Charles (1990b). The contribution of linguistics to language understanding. In Proceedings of the 1st Symposium on Cognition, Language, and Culture, A. Bocaz (Ed.), 109–28. Universidad the Chile. Fillmore, Charles, Paul Kay, & Mary C. O’Connor (1988). Regularity and idiomaticity in grammatical constructions. Language 63(3): 501–38. Gohl, Christine (2002). Zwischen Kausalität und Konditionalität: Begründende wenn-Konstruktionen. Deutsche Sprache 30, 193–219. Goldberg, Adele (1994). Constructions: a Construction Grammar approach to argument structure. Chicago: Chicago University Press. Günthner, Susanne (1999). Wenn-Sätze im Vor-Vorfeld. Ihre Formen und Funktionen in der gesprochenen Sprache. Deutsche Sprache 27(3), 209–235. König, Ekkehard & Johan van der Auwera (1998). Clause integration in German and Dutch conditionals, concessive conditionals and concessives. In J. Haiman & S. Thompson (Eds.), Clause combining in grammar and discourse (101–129). Amsterdam, Philadelphia: Benjamins. Köpcke, Klaus-Michael & Uwe-Klaus Panther (1989). On correlations between word order and pragmatic function of conditional sentences in German. Journal of Pragmatics 13, 685–711. Pasch, Renate et al.€(2003). Handbuch der deutschen Konnektoren. Berlin, New York: de Gruyter. Selting, Margaret (1993). Voranstellungen vor den Satz. Zeitschrift für germanistische Linguistik 21, 291–319. Shibatani, Masayoshi & Sandra A. Thompson (1996). Grammatical constructions: Their form and meaning. Oxford: Oxford University Press. Sweetser, Eve (1990). From etymology to pragmatics. Cambridge: Cambridge University Press. Vandergriff, Ilona (1995). A study of German conditionals: Domain, verb position, and pragmatic function. (Unpublished dissertation, University of California at Berkeley).

Semantics and pragmatics in construction grammar The case of modal verbs* Ronny Boogaart Leiden University

1. Introduction At least since the work of Traugott (1989) and Sweetser (1990:€49–75), modal verbs have been considered a schoolbook example of polysemy. However, for any given modal verb, it is notoriously problematic to determine both the number and the kind of modal ‘meanings’ to be distinguished as well as the exact relationships among these (see e.g. Nuyts; 2005 and Narrog; 2005 for overviews and discussion). In this paper, I will approach the problem from the perspective of construction grammar (in the general sense outlined by Croft & Cruse; 2004:€225–257). I will argue that constructions function as constraints on polysemy: in actual language use, modal verbs occur exclusively as part of modal constructions and such constructions as a whole are, in principle, monosemous rather than polysemous. In (1)–(3), the problem is illustrated for the Dutch verb kunnen (‘can’). (Examples taken from the novel Hersenschimmen, by J. Bernlef, Querido, 1987). Following the threefold distinction made by Nuyts (2005), this verb allows for dynamic (1), deontic (2) and epistemic (3) readings.1 (1) Gedachten kun je niet lezen. Thought can you not read ‘One can’t read thoughts’ (2) Van mij kunt u rustig naar huis gaan. From me can you quietly to house go ‘As fas as I’m concerned, you can go home’ (3) Kan dat ze hem ergens gevonden hebben. Could that they him somewhere found have ‘It could be that they found him somewhere’

 Ronny Boogaart

To account for such flexibility, most linguists have taken either a monosemous or a polysemous view. On the monosemy view, more commonly referred to as the vagueness approach2, verbs such as kunnen (‘can’) and moeten (‘must’) have one general, encompassing meaning that is compatible with their various interpretations. For these particular verbs, as well as for their English counterparts can and must, this characterization must be something like ‘possibility’ and ‘necessity’, respectively (Kratzer 1977, 1981). On the polysemy account, the meaning of a modal verb is represented as a network of different senses – some of them more peripheral than others – that are related to one another in systematic ways, for instance by means of metaphorical extension (Sweetser 1990:€56ff). Obviously, the categories of dynamic, deontic and epistemic modality illustrated in (1)–(3) make good candidates for the different senses associated with the modal verbs kunnen and moeten – or their equivalents in other languages, for that matter. The discussion on monosemy and polysemy is a classic one. In fact, some people may feel that it is outdated, especially since the rise of cognitive and constructionist approaches to language. Even though it is well known that both the monosemy approach and the polysemy approach to meaning have their own specific advantages and problems associated with them (see Section 2), it is clear that, in cognitive linguistics at least, the dispute has been decided in favour of polysemy. And it has been a landslide victory at that.3 In fact, polysemy is sometimes regarded as a defining feature of the cognitive approach to language. This is most evident when Langacker labels the search for minimal lexical semantics, i.e. monosemy, as one of the ‘ghosts from our theoretical past’ (2005:€151), suggesting it may be compared, for instance, to the generative grammarians’ quest for the largest generalizations possible.4 In this paper, however, I will argue that precisely within a constructionist approach to language, monosemy is neither a ‘ghost’, nor something ‘from the past’. In fact, the taxonomic construction hierarchies that have been proposed in construction grammar may well be labeled monosemy networks. This paradoxical term is introduced in this paper to capture the claim that a constructionist approach to language combines the advantages of the polysemy and the monosemy approach while at the same time overcoming the problems attached to either approach individually. I will start out in Section 2 by presenting in more detail the monosemy and the polysemy approach as applied to the semantics of modal verbs. In Section 3, I will apply the principal claims of construction grammar to constructions that include a modal verb as one of their constituting parts. In Section 4, it is concluded that, from a constructionist perspective, examples such (1)–(3), do not illustrate one form with three different meanings, but rather three different constructions, each with their own meaning.



Semantics and pragmatics in construction grammar 

2. Monosemy versus polysemy in the domain of modality Examples such as those given in (1)–(3) raise the question of what exactly is ‘the meaning’ of a modal verb such as kunnen (‘can’) and how does this meaning relate to the different usages exemplified in (1)–(3)? There are, at least, three possible answers to this question and in Section 3, I will add a fourth one, from the perspective of construction grammar. In this section, I will present the advantages and problems of the first three possible answers that may be characterized as homonymy, monosemy and polysemy. While these matters will be familiar to most readers and excellent discussions of the issue can be found elsewhere in the literature (e.g. Kirsner 1993; Geeraerts 1993; Tuggy 1993; Croft 1998; Sandra 1998; Fortuin 2000; and, with special reference to modal verbs, Coates 1983; Kronning 1996; Janssen 2001 and Papafragou 2000), some remarks are necessary in order to compare the constructionist solution of Section 3 to the answers that have been around in the literature to date. Moreover, I will apply the issue specifically to modal verbs and, even more specifically, to the verb kunnen, the Dutch equivalent of English can. 2.1

Homonymy

The first possible answer to the question ‘What is the meaning of the verb kunnen?’ is that there is no such thing as the meaning of this verb, but rather that the verb simply has (at least) three different meanings, as represented in Figure 1. For a linguist, this answer, which basically treats kunnen as a homonym, is obviously not very satisfactory. The main reason is that Figure 1 does not show: a. what exactly the three meanings have in common, nor b. how exactly the three meanings are different from one another. FORM 1 kunnen

MEANING 1 dynamic

Figure 1.╇ Homonymy

MEANING 2 deontic

MEANING 3 epistemic

 Ronny Boogaart

The first aspect is especially problematic in view of the fact that other modal verbs, including modal verbs in other languages, show the same kinds of usage types, whereas the representation of Figure 1 makes it seem like more or less of a coincidence that these three particular meanings are expressed by one and the same form. (In this respect, the situation for kunnen is clearly different from the ambiguity of a word like bank, the standard example of genuine homonymy.) The second aspect relates to the fact that, in addition to some common semantic core, there are also clear differences between the three readings. Research on language acquisition and grammaticalization, for instance, has made clear that, both ontogenetically and diachronically, these three meanings occur in a definite order and that one may be derived from the other – either via metaphorical extension (Sweetser 1990) or by conventionalization of implicatures (Traugott 1989) – but not vice versa. A final problem for the homonymy ‘solution’, and the one that causes most frustration for practicing linguists, is that it is often extremely difficult to classify a specific use of a modal verb as belonging in one of the three categories. Take, for instance, the use of zou kunnen (‘could’) in (4) (taken from the Corpus of Spoken Dutch) (Brantjes 2005). (4) Ik wou uh ik dacht dat jij aan de lange muur zou kunnen begin[nen]. lit. I wanted uh I thought you could start at the long wall. Most researchers working on modality would classify the use of kunnen in (4) as an instance of dynamic modality, expressing ‘participant-internal ability’ or ‘capacity’, much like in (1). However, it is clear from the context that the utterance in (4) is used as a directive, i.e. it suggests – rather strongly, in fact – that the speaker wants the addressee to perform the action (this is clear, for instance, from the fact that the speaker first begins his sentence with ik wou (‘I wanted’) and then corrects himself to the more polite form ik dacht (‘I thought’)). From this perspective, then, the utterance belongs in the domain of deontic modality, like that in (2). Clearly, we are dealing here with an indirect speech act in which a claim about the ability of the addressee to perform a certain action (dynamic modality) has to be understood as a rather strong incentive for the addressee to actually start performing this action (deontic modality). The issue of indirect speech acts constitutes, of course, one of the core issues of pragmatic theory and it is well beyond the limits of this paper to treat it in any great depth. It suffices to note, for now, that Figure 1 does not offer any insight into the issue of indirect speech acts. For many reasons, then, the ‘homonomy’ solution to the problem illustrated by (1)–(3) is not really a solution and in recent years it has not really been advocated as such for modal verbs at all. In the literature, it mainly functions as an ideal strawman when arguing in favour of the other two answers to this question. In fact,



Semantics and pragmatics in construction grammar 

both approaches can use it as such since the monosemous approach basically tackles the first problem ((a) above), whereas the polysemous solution quite adequately handles the second one ((b) above). Neither one, however, is particularly suited, or even interested, in accounting for uses such as the one demonstrated in (4), which would require a separate pragmatic ‘module’ on either account. 2.2

Monosemy

Of the monosemy and the polysemy approach, the first one differs most radically from the strawman in Figure 1. On this account, the answer to our question of how to account for the different readings of kunnen in (1)–(3) is that, in actual fact, these readings are not all that different. It is argued that the verb has only one meaning, which is constituted by the conceptual content that is shared by all of its uses. Drawing on existing accounts of English can (most notably those of Kratzer, 1991; Papafragou 2000 and, in particular, Perkins, 1983), we might hypothesize for Dutch kunnen that its unitary, all-encompassing semantics is ‘x enables y’, as represented in Figure 2. In this representation, y is the propositional content of the entire clause, without the modal information as expressed by the verb. x is a pragmatic variable, to be resolved by the context, the nature of which will determine the kind of modality involved. In the case of dynamic modality, x is supposed to be constituted either by participant-internal properties (abilities, capacities), as in (1), or by participantexternal circumstances that enable an agent to perform a certain action. If x is an authority or an ethical norm, we are automatically dealing with deontic modality; in that case, x may, more specifically, also be the speaker himself, as is the

FORM 1 kunnen

MEANING 1 X enables (is compatible with) Y

Figure 2.╇ Monosemy (vagueness)

 Ronny Boogaart

case in (2), which gives rise to a directive-deontic reading. The category of epistemic modality, as in (3), is a bit more complex but it may nonetheless be captured by ‘x enables y’ as well, in which x is the body of knowledge or evidence available to the speaker at the time of speaking and y is not so much the state of affairs itself as it is the conclusion that this state of affairs is holding. Pro monosemy The monosemous characterization of the meaning of kunnen in Figure 2, rather convincingly, gives insight into the common semantic denominator of its different uses, like those given in (1)–(3); in fact, it is the only thing that it does. Thus, unlike in Figure 1, the fact that these three different ‘meanings’ are expressed by one and the same form is not merely a coincidence. This, indeed, constitutes the principal merit of the vagueness approach to meaning, that has been advocated most consistently throughout the years by linguists of the so-called Columbia School (Contini-Morava et al. 2004 and references cited there). With respect to modal verbs, the monosemous approach is also the prevalent one in formal linguistics, ranging from Kratzer’s (1977) work to the relevance-theoretic approach of Papafragou (2000). The monosemous perspective has one other important advantage, namely that it allows for a clear distinction between the contribution of semantics and pragmatics, as represented in Figure 2'.

FORM 1 kunnen

Semantics MEANING 1 X enables Y Pragmatics INTERPRETATION 1 (dynamic)

INTERPRETATION 2 (deontic)

INTERPRETATION 3 (epistemic)

Figure 2'.╇ Vagueness (semantics versus pragmatics, meaning versus interpretation)



Semantics and pragmatics in construction grammar 

The only contribution of the semantics of the verb kunnen to the interpretation of the utterance consists of imposing the rather global and schematic constraint of Figure 2. All the rest, including the choice between the different kinds of modality, is determined by features of the context of use, which – at the very least – will have to provide a value for x (corresponding to Kratzer’s conversational background and Papafragou’s modal restrictor). Contra monosemy While the monosemous approach thus solves the first problem of the homonomy approach of Figure 1, it does not provide a solution for the second problem, that of accounting for the differences and precise relationships between the different readings of kunnen illustrated in (1)–(3), for instance, from a diachronic or a crosslinguistic perspective. The latter two perspectives will be taken up in the next subsection, on polysemy, but also from a synchronic, language-specific perspective, there are clear differences between the three readings, starting with their relative frequency. This is evidenced in Table 1, which lists the frequency of the dynamic, deontic and epistemic readings of kunnen in the novel Hersenschimmen.5 Table 1.╇ Frequency of the different readings of kunnen in the novel Hersenschimmen Modal type Dynamic Deontic Epistemic Other (blends, undecided) Total

n

%

227 â•⁄ 16 â•⁄â•⁄ 9 â•⁄ 14 266

85,3 6 3,4 5,3 100

The differences in frequency between the readings allowed for by kunnen, in particular between the dynamic reading and the other two, are quite significant, but since the monosemous analysis focuses exclusively on what the different uses have in common, such differences are not noticed, let alone explained. Furthermore, there are at least two other problems to the monosemy approach. The first one is obvious: since the semantic characterization will have to cover all the uses of a given form, it can only be highly abstract. So much so in fact that one may well wonder whether language users have, or indeed need to have, access to this abstract level of analysis at all. And, crucially, as abstract as the semantics of kunnen proposed in Figure 2 might be, it still feels like stretching the notion of ‘enablement (x,y)’ when applying it to actual cases, such as those given in (1)–(4). The final objection to the monosemous approach to be mentioned here can be characterized as the rule-list fallacy (Langacker 1987:€29) or the generality fallacy

 Ronny Boogaart

(Croft 1998:€156 ff.), which boil down to saying that if it is possible for a linguist to come up with a general semantic characterization for a given form, this does not mean that language users do this as well (the generality fallacy), nor that, in addition, individual types of uses of the form are not stored independently (the rule-list fallacy). This is precisely why, as mentioned in the introduction, Langacker (2005:€151) has called the exclusive strive for abstract generalizations and rules ‘a ghost from the past’ and this, of course, seems totally in accordance with the basic insights that motivated the development of construction grammar – see Section 3. 2.3

Polysemy

In Section 2.1, I mentioned two problems with the idea of homonymy as applied to the verb kunnen: (a) it does not show what the different uses have in common, and (b) it does not show how exactly the various uses differ from one another. While the monosemous analysis discussed in the previous subsection clearly solves the first problem (but not the second one), the polysemous analysis is aimed precisely at accounting for the second problem (but not for the first one). The idea that modal verbs are polysemous is a well-established one in cognitive linguistics (Mortelmans 2007:€869–970). In fact, the very existence of examples such as (1)–(3) may be taken as a sufficient condition for a polysemous analysis: one and the same form allows for quite different readings that, nonetheless, are felt to be connected in some way. The latter is what distinguishes the polysemous analysis from simply assuming homonomy (or ambiguity). Thus, rather than saying that the verb kunnen has three different meanings, it is now assumed that ‘the meaning’ of this verb can be represented as an entire network of different ‘senses’. If we restrict ourselves to the dynamic, deontic and epistemic sense of kunnen, the resulting network might look like the one in Figure 3. Figure 3 is meant to capture two ideas that are central to the polysemous approach to modal verbs. First, the common semantic denominator of the different senses associated with kunnen – constituted by something like ‘x enables y’ (Figure 2) – is not represented here at all. This is not to say that such an overall schematic characterization is principally incompatible with the polysemous approach, but rather that polysemists are either not too interested in it, or deny outright its relevance from a contemporary, usage-based approach to language.



Semantics and pragmatics in construction grammar 

FORM kunnen

MEANING dynamic

deontic

epistemic

Figure 3.╇ A polysemy network for kunnen

Second, the three different senses of kunnen in Figure 3 are not on an equal footing: the bold line around the dynamic sense illustrates the idea that this sense is somehow the prototypical one, i.e. the most ‘central’, ‘concrete’, ‘literal’ or ‘core’ sense (and, at least in the case of kunnen, also the most frequent one, as was obvious from Table 1), whereas the other senses are more peripheral and, in fact, can be shown to be derived from the prototype. The latter is usually argued to be true from the synchronic perspective of the adult language speaker, but it is all the more clear from the perspective of grammaticalization and language acquisition, where the dynamic sense can be shown to be prior to the deontic and the epistemic sense. How exactly the three senses are related is not given by the simplified polysemy network of Figure 3, but it is one of the goals of the polysemous approach to describe precisely these types of relations. (Mortelmans 2007:€875–880 offers a recent overview of different solutions to the latter problem.)

 Ronny Boogaart

Pro polysemy The last issue provides an important argument in favour of a polysemous analysis, namely the argument from grammaticalization. Since the polysemy network of Figure 3 is ‘directed’, as is witnessed by the arrows going from the core to the peripheral, to the even more peripheral sense, it allows for elegant diachronic generalizations on the extent to which a given form has ‘grammaticalized’. For instance, whereas both Dutch mogen and English may seem to have lost their original dynamic use, the Dutch form has hardly grammaticalized beyond the deontic sense, whereas the English form is used both deontically and (even predominantly) epistemically these days. All this is not to say that a polysemy network is necessarily identical to the ‘grammaticalization chains’ that are used in diachronic research, but still, it is hard to see how this kind of language change could have taken place if forms, at any synchronic point of their diachronic development, were not to some extent polysemous. After all, grammaticalization is a gradual development in which older uses are not lost immediately after a new kind of use starts to develop. It is a further advantage of the polysemous approach, as compared to the monosemous approach of Figure 2, that it is better suited for the comparison of different modal constructions within one language as well as for cross-linguistic comparison. The semantic characterization of a modal verb in the monosemous analysis can only be highly abstract, which makes comparative analysis, both within one language and across languages, extremely difficult. At this level of abstraction or schematicity, different forms may be hard to distinguish from one another, whereas their actual use may be quite different. For instance, the general semantic characterization of both Dutch kunnen and English can has to be something like ‘x enables y’, but this does not garantuee at all that they are always used in similar contexts. On the polysemous account, it can more easily be shown to what extent the use of ‘similar’ forms overlaps and to what extent they diverge from one another. Thus, whereas Dutch kunnen clearly has all three senses represented in Figure 3, English can does have a dynamic and a deontic sense, but the form is not really used to express epistemic possibility (for which English exclusively uses could, which is hardly felt to be the past tense form of can anymore, or may). Contra polysemy There are, however, also quite a few problems attached to the polysemy approach to the semantics of modal auxiliaries. I will discuss three of those, the last one of which, in particular, will show the need for a constructionist approach to the semantics of modals, which will be presented in more detail in Section 3. The first problem, obviously, is that we have now lost the idea that there is a common semantic core to the various uses of the modal or, at least, the polysemous approach is not principally – and, in some cases, even principally



Semantics and pragmatics in construction grammar 

not – occupied with trying to formulate it. In addition, two problems that we noted above for the monosemous approach are (almost) equally applicable to the polysemous approach, namely the rule-list fallacy and, interestingly enough, the ‘vagueness’ of the semantic characterizations used. As for the first, the fallacy, of course, also works the other way round: if it is possible for a linguist to come up with a network of different but related senses for a given form, this does not exclude the possibility that, in addition, there is a schematic semantic characterization available that is compatible with all the different senses of that form. Moreover, one can now also commit the polysemy fallacy (Sandra 1998), which consist of assuming, without further empirical support, that the network of senses as envisaged by a clever linguist, will also be part of the linguistic competence of language users. Besides, even clever linguists differ substantially when it comes to the design of their modal network, both with respect to the number of senses distinguished and with respect to the relationships assumed to exist between them. As an example of the first, Goossens (1983) may be mentioned, since he distinguishes no less than twelve different senses for kunnen. As an example of the latter, Van Ostaeyen & Nuyts (2004) have argued that the epistemic sense of kunnen does not derive from the deontic sense, as Sweetser would have it and which was represented in Figure 3, but rather that both the deontic and the epistemic sense are based on the dynamic sense, as they appear in Figure 3'. Now, none of these analyses is necessarily wrong, but the point is that there are still no clear criteria when it comes to determining either the number or the kind of senses for any given form, or the exact nature of the relationships between them. The principal problem of polysemy networks, then, is that they always seem to be to some extent arbitrary; in other words, the senses are given by the analyst more than by the language. The final drawback of the polysemous analysis, namely the ‘vagueness’ of the semantic characterizations, may need a bit more elaboration since, as I have already claimed, it is an advantage of the polysemous approach over the monosemous one that the senses in the polysemy network of Figure 3 are decidedly less vague compared to the one unitary meaning of Figure 2. However, ‘vagueness’ is a matter of degree, of course, and the senses proposed within a polysemy approach to semantics are still rather ‘vaguely’ defined when compared to yet another approach, namely the constructionist approach to be presented in Section 3. Of course, vagueness or schematicity in itself is not a problem – in fact, for monosemists vagueness is a claim rather than a problem – and some degree of abstraction from actual usage events will be an indispensable part of any grammatical theory,

 Ronny Boogaart

FORM kunnen

MEANING dynamic

deontic

epistemic

Figure 3'.╇ An alternative polysemy network for kunnen (Nuyts)

including construction grammar. However, since the traditional notion of monosemy is rejected by polysemists precisely on account of its high level of abstraction, it might be worthwhile to point out in somewhat more detail the problems that arise within polysemous analyses of modals precisely because of their own level of abstraction. At the same time, the discussion will show the need for a ‘constructionist’ approach to language and thus pave the way for the following section. To illustrate this crucial point, I will first have another look at the Dutch examples (1) and (4) and then offer some additional data in support of a constructionist approach from the study of cross-linguistic variation. If we have to choose between the senses of Figure 3 then the example in (1), repeated here for convenience, comes out as a case of ‘dynamic modality’, and thus, more specifically, as describing an ability or capacity of the subject referent. (1) Gedachten kun je niet lezen. Thought can you not read ‘You can’t read thoughts’



Semantics and pragmatics in construction grammar 

It is, however, quite different from the standard example of ‘dynamic modality’ given in the literature, which would be something like He can play the piano. On closer scrutiny, it does not really make sense to say that (1) is about a property of the subject referent or that it is ‘agent oriented’ at all. The sentence can, for instance, not really be paraphrased with to be able to (or the Dutch equivalent thereof). Rather, such sentences with a reduced form of the second person pronoun, receiving a generic reading, are used in Dutch to present a property of the objectNP rather than the subject-NP. (If the full form of the pronoun is used, the sentence is about the ability of the addressee). Very often, as is the case in (1), the object-NP will be preposed. This passive-like (agent backgrounding) function of (1) is, thus, more adequately paraphrased as ‘thoughts are not readable’ or ‘thoughts cannot be read’. This might seem like a minor problem and a minor divergence of the ‘prototypical’ dynamic sense of kunnen, but it already shows that the meaning of language-specific patterns can be more complex than, and actually even different from, what is captured by the highly schematic sense of ‘dynamic modality’. Turning to the category of deontic modality, example (4) likewise shows the inadequacy of the polysemy approach when applied to actual instances. In fact, it does so in two respects. (4) Ik wou uh ik dacht dat jij aan de lange muur zou kunnen begin[nen]. lit. I wanted uh I thought you could start at the long wall. First, as already noted in Section 2.1, these ‘deontic’ uses of kunnen are really somewhere in between the dynamic and the deontic sense. From a polysemous perspective, we might assign kunnen in (4) to the category of dynamic modality and then leave the ‘directive-deontic’ impact of such utterances to pragmatics. However, separating semantics and pragmatics in this way seems artificial and it obscures the simple fact that knowing Dutch includes knowing that one can use the specific construction instantiated in (4) to politely suggest to the addressee to perform a certain action. Thus, indirect speech acts, at least to the extent that they are conventionalized, may well be considered constructions (cf. Stefanowitsch 2003), i.e. conventionalized pairings of a certain form – which in the case of (4) will include at least zou and a specific intonation pattern – and a specific meaning (cf. Levinson’s 2000 utterance types). Second, even if there were a way to assign the utterance in (4) to the deontic sense of kunnen in the polysemy network, such a label would not tell us much. The semantic pole of ‘indirect speech acts constructions’ will have to include quite subtle pragmatic information on the type of situations in which it is appropriate to use them, far beyond what is, schematically and vaguely, captured by the notion of ‘deontic modality’ in the polysemy network.

 Ronny Boogaart

A final illustration of the vagueness of polysemy networks is taken from crosslinguistic comparison. It was argued above that the polysemy approach is quite better suited for cross-linguistic comparison than is the monosemous approach. Thus, ‘similar’ forms across languages, such as Dutch kunnen and mogen and their English near-equivalents can and may, can be compared as to which senses of the polysemy network they are used to express. Such an undertaking is practically impossible at the level of abstraction at which the monosemous approach formulates its language-specific generalizations. However, at the same time, cross-linguistic comparison brings out quite a serious problem for the polysemous approach. To illustrate this, let us, once again, look at Dutch kunnen as it is used in the novel Hersenschimmen, and compare it to the official English and German book translations, entitled Out of Mind and Hirngespinste, respectively. Table 2 shows the verbs used most often in English and German as translations of Dutch kunnen (recall that the Dutch original contained 266 occurrences of kunnen, see Table 1). I will illustrate the relevance of Table 2 for the present discussion on the basis of the top left corner of the table which shows the frequency of English can and could in Out of Mind. It can be noted that the English translation uses the verb can (including could) (303 instances) quite a bit more than the Dutch original uses kunnen (266 instances). The difference is, in fact, much bigger than it might appear at first sight, since of the 303 instances of English can/could, only about two thirds (201 instances) are based on an occurence of kunnen in the original Dutch text, leaving 102 instances of can/could that are not based on kunnen. Now, whereas the polysemy network for English can/could and Dutch kunnen will undoubtedly look highly similar, these observations suggest that these networks are not very good at describing, let alone explaining, the language-specific distribution of these forms, which is, apparently, quite different in either language. Table 2.╇ English and German renderings of Dutch kunnen in the official book translations of Hersenschimmen English Total can/could in Out of mind: Based on Dutch kunnen: Not based on Dutch Kunnen: Total may/might in Out of mind: Based on Dutch. Kunnen: Not based on Dutch Kunnen: Total be able to in Out of mind: Based on Dutch kunnen: Not based on Dutch kunnen:

German 303 201 102 â•⁄ 42 â•⁄ 10 â•⁄ 13 â•⁄ 17 â•⁄ 17 â•⁄â•⁄ 0

Total können in Hirngespinste: Based on Dutch kunnen: Not based on Dutch kunnen Total dürfen in Hirngespinste: Based on Dutch kunnen: Not based on Dutch kunnen: Total sollen in Hirngespinste: Based on Dutch kunnen: Not based on Dutch kunnen:

259 213 â•⁄ 46 â•⁄ 49 â•⁄â•⁄ 2 â•⁄ 47 â•⁄ 53 â•⁄â•⁄ 5 â•⁄ 48



Semantics and pragmatics in construction grammar 

Thus, there are many contexts in which English uses a form of can whereas Dutch does not (and vice versa, in fact). Dutch may, for instance, use another modal verb in a particular context, as is the case in ‘epistemic’ questions of the kind exemplified in (5). (5) a. Zou Vera al naar bed zijn? Would Vera already to bed be b. Could Vera be in bed already? Thus, if one is wondering whether or not a state of affairs is holding, one could formulate a question – either to oneself or to some addressee – of the form given in (5), of which there are 11 instances in the novel. In all cases, English uses the verb could, whereas Dutch has opted for the past tense of zullen (though the present tense would also be possible in such cases). For English, one might say that the use of could in (5b) is in accordance with our earlier observation that can does not really allow for epistemic uses; thus, (5b) may be regarded as the interrogative version of (6b). In Dutch, however, there is no affirmative version for (5a) using a form of the verb zullen at all. The sentence in (6a) is not ungrammatical, but it does not allow for the epistemic reading that Vera is possibly in bed already, like English (6 b)does.6 (6) a. Vera zou al naar bed zijn. Vera would already to bed be b. Vera could be in bed already. In Dutch, then, both the polysemy network for kunnen and the one for zullen will have to include a sense of ‘epistemic possibility’. However, such a move still would not tell us that Dutch uses zullen rather than kunnen in the highly specific ‘epistemic interrogative construction’ exemplified in (5a) and there is of course no way to predict this on the basis of the kind of information represented in polysemy networks or existing classifications of modality. As a final illustration of the same point, the high number of instances for can in the English translation of this novel can be attributed to a large exent to one particular group of verbs that are typically combined with can in English, but that do not need any modal verb at all in Dutch. This group is constituted by, on the one hand, the ‘sensory verbs’ to see, to hear, to feel and to smell and, on the other hand, the ‘mental verbs’ to understand and to remember. Thus, whereas in English these verbs are typically used as a complement to can, as in I can see it and I can’t understand, in Dutch they are mostly used as main verbs without any modal qualification in the form of an auxiliary, as in Ik zie het (‘I see it’) and Ik begrijp het niet (‘I do not understand’). Taken together, this group of verbs accounts for 50 instances of the English ‘surplus’ of can in the translation of Hersenschimmen. This

 Ronny Boogaart

particular use of can in English constitutes a clear example of ‘dynamic modality’, more specifically of participant-internal ability, as does the use of kunnen to the extent that it is used with these verbs in Dutch at all. Children who acquire English and Dutch will at some point (quite early in fact for this particular sense) have arrived at the generalization that can/kunnen is used to express ‘ability’, but clearly this general meaning does not, as such, tell them when exactly to use it – or, for that matter, not to use it. Thus, for English learners, more or less fixed combinations such as can see and can feel will have to be learnt independently and presumably stored as such. This conclusion could, of course, also have been arrived at on the basis of language-specific research of English, but cross-linguistic analysis has the potential of bringing to light the crucial role of language-specific conventions: English typically uses can with verbs like see or feel, but, as is shown by the case Dutch, this is merely a convention of English that does not automatically follow from the fact that can expresses the general sense of ‘dynamic possibility’.7 In this section, I have pointed out the shortcomings of the polysemous approach to modal verbs, especially when it comes to describing the meaning and use of language-specific constructions that contain a modal verb as one – but only one – of their constituting parts. At the same time, I have, therefore, illustrated the need for a constructionist approach to modal verbs. In Section 3, I will illustrate in more detail what such an approach could look like. First, I will sum up the findings of Section 2. 2.4

Summing up

In this section, I have compared the monosemous and the polysemous approach to the semantics of modal verbs. While the monosemous approach is very good at explaining why all the various uses of one and the same verb are expressed by one particular form, the polysemous approach, instead, focuses on these various uses themselves, as well as on the precise relationships amongst them. In addition, each approach has some more specific advantage over the other. The monosemy approach allows for a clear distinction between the contribution of semantics and that of pragmatics, as was represented in Figure 2’; in polysemy accounts, it is less clear to what extent the distinction between epistemic, deontic and dynamic modality is semantic or pragmatic in nature. If specific senses are represented independently, then the role of pragmatics is presumably more limited than it is on the ‘minimalist’ approach to semantics of monosemists, but it is hard to see how much more limited it actually is, or where precisely pragmatics comes in. The polysemous approach, on the other hand, has a crucial advantage over the monosemous one in that the notions of dynamic, deontic and epistemic modality (or perhaps some subtly different classification of senses) are necessary to allow for



Semantics and pragmatics in construction grammar 

a comparison of different modal verbs, or non-verbal modal expressions for that matter, both within one language and accross languages. In addition, these senses seem to play an explanatory role when considering consecutive stages in language acquisition, or in the diachronic development of modal expressions. Finally, the monosemous and the polysemous approach to meaning turned out to share a crucial problem. The level of abstraction or schematicity they are operating at does not really allow for a detailed description of the way in which modal expressions are actually used in a specific language. This will be most obvious for the monosemy approach, in which the ‘vague’ or minimal semantics of linguistic elements is actually being claimed, but it is hardly less of a problem for the polysemous approach, as anyone can witness who has ever tried to classify actual instances of modal expressions according to the modal classifications that have been proposed in the literature. In the following section, I will argue that a constructionist approach to the semantics and pragmatics of modal verbs combines the advantages of the monosemous and the polysemous approach without, however, inheriting their shortcomings. 3. From modal verbs to modal constructions 3.1

Introduction

Construction grammar presents an exception to the idea that functional and cognitive linguists have always been more interested in the meaning and use of language, and thus in semantics and pragmatics, than in aspects of its form, i.e. syntax. In fact, construction grammar has been characterized as ‘the cognitive linguistic approach to syntax’ (Croft & Cruse 2004:€225). (Which is not to say that all branches of construction grammar are necessarily cognitive or functional in nature.) The latter description suggests that construction grammar is not concerned with the semantic and pragmatic issues that we have been discussing, as applied to modal verbs, in the preceding sections. Now, while this is true to some extent – and for some branches of construction grammar more so than for others – the very starting point of construction grammar precisely consists of the claim that the two, i.e. form and meaning, cannot be separated. In fact, I will show that, precisely because of its focus on issues of form, the framework of construction grammar provides a new perspective on the classic issue of monosemy versus polysemy that is the central concern of the present paper. The label of construction grammar is really an umbrella-term covering various schools of linguistics, arguably including Langacker’s Cognitive Grammar, but in line with the rather theoretical, and basically programmatic, outlook of the present

 Ronny Boogaart

contribution, I will not be formulating my findings in terms that are specific to any of these schools. Rather, I will start off from the set of assumptions that Croft & Cruse (2004:€265) have identified as ‘the three essential principles of construction gammar’ and that are thus, by definition, shared by the different ‘constructionist’ schools. These have been summarized in (7). (7) a. Constructions are symbolic units b. All grammatical knowledge can be represented as a construction. c. Constructions are linked to one another in a taxonomic hierarchy. The principle in (7a), i.e. the symbolic nature of constructions, means that constructions are conventionalized pairings of form and meaning. Until now, in Figures 1–3, I have been visualizing both forms and meanings as circles; constructions I will represent as squares, as in Figure 4, containing both a form (phonologial pole) and a meaning (semantic pole). The symbolic nature of words has long been accepted in linguistics, but construction grammar extends the symbolic claim to all kinds of grammatical knowledge, as formulated above in principle (7b), thus ranging from morphemes to words via idiomatic phrases to the most abstract grammatical patterns. This was captured above when I pointed out that even when construction grammar is ‘doing syntax’ it is still, as a matter of principle, occupied with semantic issues as well. The third and final principle in (7c) captures the idea that a language does not consist of simply a list of constructions. This follows naturally, in fact, from the second principle: if morphemes, words, multi-word expressions and abstract grammatical patterns are all constructions, then it will be clear that the more specific (lower level) constructions contain the more abstract (higher level) constructions as constituting parts. Such relationships are represented in taxonomic hierarchies, or construction networks, in which constructions inherit formal and semantic aspects of more abstract constructions higher up in the taxonomy, while at the same time exhibiting more specific features that make them into independent constructions of their own.

Form

Meaning

Figure 4.╇ A construction



Semantics and pragmatics in construction grammar 

In the following subsection, I will apply these basic principles of constructionist approaches to the problem, discussed at length in the previous section, of accounting for the many different readings allowed for by modal verbs. 3.2

Towards a network of modal constructions

A crucial difference between construction networks and the polysemy networks that we have been discussing in Section 2.2 is that, following principle (7a) and Figure 4, a network of constructions is a network of form-meaning pairings, whereas a polysemy network is a network of senses associated with one particular form. Construction taxonomies are concerned with forms (and their accompanying meanings), whereas polysemy networks are concerned with meanings. The two are not necessarily incompatible because they are operating at different levels of linguistic analysis: since lexical items, just like grammatical rules, are constructions (principle 7b), we could simply say that there is a kunnen-construction in Dutch that, obviously, has kunnen as its phonological pole and some version of the polysemy network for modal verbs, like the one in Figure 3, as its semantic pole. Even though, in my view, a monosemous analysis of constructions, in which constructions are pairings of one form and one meaning, is more in accordance with the sign-based spirit of construction grammar (principle 7a and Figure 4) and, specifically, with the notion of inheritance (see below), it is true that the existence of polysemy is not a priori excluded by the ‘essential principles’ cited in (7). In fact, Goldberg (1995:€ 31–32) claims that ‘constructional polysemy’ is only to be expected given that polysemy is the norm for words, and constructions are not principally different from words (cf. principle 7b). At this point, then, it might feel like we could start the debate on monosemy vs. polysemy as applied to kunnen all over again and that construction grammar as such has no new insights to offer into this matter. This, however, is far from true. While treating lexical items as constructions does not help us much in accounting for the different readings of modals, lexical items are, of course, not the only kinds of constructions that are relevant for the present discussion.8 In fact, it is one of the merits of constructionist approaches that they have shifted the attention away from abstract patterns and meanings to relatively specific and concrete ‘low-level’ constructions that can be observed in actual language use. From the latter perspective, there is, of course, no such thing as the lexical item kunnen being used in isolation. Rather, the verb kunnen functions as just one element among others in an almost indefinite number of lower level constructions. All these ‘modal constructions’ can be said to inherit the form and function of ‘the kunnen-construction’, but the meaning of the latter is, basically, an abstraction of all the different constructions that kunnen is compatible with and it may well be questioned – as it

 Ronny Boogaart

is by construction grammarians – if constructions that are so high up in the taxonomy have any role to play other than in the most interesting linguistic papers. (No irony intended.) Therefore, the issue of ‘the meaning’ of the verb kunnen by itself loses much of its urgency since it will be the primary task of construction grammarians to determine the form and meaning of the lower level modal constructions in which kunnen is only one of the constituting parts. The meaning of these constructions, like those illustrated in (1), (4) and (5), will be so much more specific, just like their forms are, that neither ‘vagueness’ nor ‘polysemy’ necessarily enters the picture. In fact, this is why construction networks might be labeled, paradoxically, monosemy networks rather than polysemy networks. Note that, in my view, monosemy does, thus, not necessarily imply vagueness. I hope to make this point more clear in the remainder of this section. What might such a construction network for kunnen look like, and what position in the network – if any – is occupied by the senses of dynamic, deontic and epistemic possibility that were shown, in the previous section, to be indispensable for the purpose of cross-linguistic comparison and the study of diachronic developments? A tentative answer to this question is visualised in Figure 5. As an attractive – and yet untenable (see below) – hypothesis, we may assume that rather than constituting senses in a polysemy network, the notions of dynamic, deontic and epistemic modality constitute the semantic pole of three independent modal constructions, that are linked by the fact that all three of them inherit the form and meaning of the more abstract kunnen construction.9 In addition, of course, the lower level constructions will have many more different ‘parent constructions’, including both lexical and grammatical constructions, but these are not represented in Figure 5. If we take the notion of inheritance to mean complete inheritance, thus to capture the idea that constructions necessarily inherit all formal and semantic specifications of their parent constructions, then we cannot but assume a monosemous analysis for the top node in Figure 5. Otherwise, the three lower-level constructions would inherit the entire polysemy network from the kunnen construction, which is incompatible with the gist of our hypothesis.10 Thus, Figure 5 seems to be an ideal way of combining the monosemous and the polysemous approach to the semantics of kunnen, thereby escaping the rule-list fallacy.



Semantics and pragmatics in construction grammar 

Kunnen

X enables Y

Form ?

Form ?

Form ?

Dynamic

Deontic

Epistemic

Figure 5.╇ Towards a construction network for kunnen

However, since constructions, other than senses in a polysemy network, are pairings of form and meaning, our hypothesis requires us to formulate the distinctive formal characteristics of each construction. Thus, we need to determine what exactly, in addition to kunnen, is contained in the formal component of the three constructions in Figure 5. This, in fact, is a definite advantage of construction networks over polysemy networks, since it non-arbitrarily restricts the number of constructions to be distinguished: at least in theory, constructions are ‘given’ by the language rather than just by the analyst (cf. the criticism of polysemy networks in Section 2.3). Thus, different conventionalized usages of a given form are stored independently in the taxonomy only to the extent that they can be shown to correspond to formal features. Thereby, the sign-based nature of the network is preserved: all constructions are pairings of a particular form and a particular meaning. Even though Figure 5 does a good job illustrating the advantages of the constructionist approach to modals compared to monosemous and polysemous analyses, it cannot be totally right. More specifically, it may very well be questioned if the categories of dynamic, deontic and epistemic modality could actually figure at all in a language-specific network of modal constructions. In one of the few explicitly constructionist papers on modal verbs, Wärnsby (2002) has actually tested the hypothesis that I formulated above and that was visualised in Figure 5. More specifically, she

 Ronny Boogaart

used corpus-linguistic data to see if epistemic instances differ systematically from non-epistemic ones in terms of the linguistic environment they appear in. Actually, this was done before in non-constructionist terms, for instance by Coates (1983) and Heine (1995). There clearly do seem to be such differences, most notably when it comes to the nature of the complement (as stative versus eventive, cf. Diewald 2006 for German, and Barbiers 1995 for Dutch). In addition, it is clear that, in example (2), the adjunct van mij (‘from me’) triggers a deontic reading of permission for kunnen. And, in a similar vein, the impersonal complementation construction in (3) can only be epistemic. In fact, this type of complementation construction seems to constitute the only pattern in which modal verbs in Dutch unambiguously receive an epistemic reading (Nuyts 2001:€189–192); this reading follows quite naturally from the general semantics of ‘the complementation construction’ as a kind of epistemic stance marker proposed by Verhagen (2005:€94–113).11 (2) Van mij kunt u rustig naar huis gaan. From me can you quietly to house go ‘As fas as I’m concerned, you can go home’ (3) Kan dat ze hem ergens gevonden hebben. could that they him somewhere found have ‘It could be that they found him somewhere’ Nonetheless, Wärnsby concludes her paper by claiming that it is ‘doubtful’ whether the constructionist approach to modality is ‘fruitful’ (2002:€8), mainly since one will be faced with a ‘bewildering web of constructions’ (2002:€7). The basic flaw in Wärnsby’s paper, as well as in Figure 5, is the expectation that the supposedly universal conceptual notions of dynamic, deontic and epistemic modality as such constitute the semantic pole of any language-specific construction. This, in fact, should already be clear from the specific modal constructions that we have been discussing so far and that were illustrated by means of (1), (4) and (5) below (cf. also the findings on constructions containing sensory and mental verbs discussed in Section 2.3). (1) Gedachten kun je niet lezen. Thought can you not read ‘You can’t read thoughts’ (4) Ik wou uh ik dacht dat jij aan de lange muur zou kunnen begin[nen]. ‘I wanted uh I thought you could start at the long wall.’ (5) a. Zou Vera al naar bed zijn? Would Vera already to bed be b. Could Vera be in bed already?



Semantics and pragmatics in construction grammar 

These examples illustrate a couple of the shortcomings of the representation in Figure 5. First, the examples (1), (4), and (5) exemplify cases of dynamic, deontic and epistemic modality, respectively. However, each of them is, of course, an instantiation of just one of the many dynamic, deontic and epistemic constructions that Dutch has. Thus, if we would label all of the modal constructions of Dutch in these general terms, then we are faced with an, indeed, bewildering number of synonymous constructions, which should be at least as dissatisfying to any linguist as the situation of homonymy discussed in 2.1. Second, and related to the first point, the actual semantics and pragmatics of the constructions in (1), (4) and (5) is, of course, much more specific than can be captured by any of these highly schematic labels. In fact, it is even to be expected that some language-specific constructions cannot really be classified as belonging to one of the three categories at all: they could be ‘in between’, as I argued for (4), or outside of these categories alltogether, which, for instance, seems to be the case for the use of moeten (‘must’), mogen (‘may’) and willen (‘want’) in conditional and concessive constructions (Boogaart 2007, cf. the ‘post modal’ uses in Van der Auwera & Plungian 1998). Thus, if we want to include the notions of dynamic, deontic and epistemic modality in a language-specific construction network, then we are basically faced with the same problems as noted in 2.2 for the polysemous approach. Therefore, when investigating the particular synchronic network of modal constructions in a specific language, it is not useful to employ the notions of dynamic, deontic and epistemic modality as a necessary part of the semantic component of all constructions that have a modal verb as one of their constituting parts. The fact that linguists when trying to classify actual utterances as representing one of these types of modality are often at a loss, or disagree with one another, should be telling us something, other than that they are not clever enough. Still, as was observed in Section 2.3, the senses of dynamic, deontic and epistemic modality (or another modal classification of ones choice) are necessary, or – at the very least – quite handy, tools when analyzing cross-linguistic differences, language acquisition and grammaticalization. Therefore, these labels should be regarded as pertaining to possibly universal conceptual content that probably all languages will have means to express in some way. Thus, they are part of a ‘semantic map’ (Haspelmath 2003), or a ‘conceptual space’ (Croft 2001), on which language-specific construction networks may be projected – as has been, somewhat simplistically, presented in Figure 6. (A more detailed semantic map for modality is proposed by Van der Auwera & Plungian 1998.)

 Ronny Boogaart

Kunnen

X enables Y

1

4 Dynamic

2

3 Deontic

5

Epistemic

Figure 6.╇ Construction network projected onto ‘semantic map’

The numbers (1)–(5) in Figure 6 refer to the specific constructions instantiated in the ‘constructs’ (1)–(5), discussed at various points in the preceding sections. The exact form and meaning of these particular constructions are more complex than can be treated in the present paper, but that is precisely the point to be made here: the notions of dynamic, deontic and epistemic modality do not suffice when it comes to characterizing the semantic pole of modal constructions in a given language, and sometimes they are not applicable at all.12 4. Conclusion At the outset of this paper, examples (1), (2) and (3) were provided to show that one ‘form’ can have multiple ‘meanings’. However, from the constructionist perspective on modal verbs developed in the course of this paper, we should now conclude that examples (1), (2) and (3) differ not only in meaning but also in form. It is true that all three utterances contain an instance of the verb kunnen, but that is only one of their constituting parts. It is not the verb kunnen as such that is dynamic in (1), deontic in (2), and epistemic in (3). Rather, the three examples are instantiations of particular modal constructions, that, as a whole, receive a particular kind of modal reading. The examples in (1)–(3), then, do not really illustrate



Semantics and pragmatics in construction grammar 

one form with different meanings, but different forms, each with their own meaning and, thus, monosemy rather than polysemy. Even though the notion of ‘form’ may be – literally – more complex than was envisaged in the traditional monosemous account of modal verbs, modal constructions may still be considered monosemous pairings of form and meaning. The semantic component of such complex constructions does not have to contain the kind of ‘abstract’ or ‘vague’ characterizations that have apparently given monosemy its bad name (Section 2.2). In fact, the semantic component of lower level constructions will even be more concrete and specific than the ‘senses’ distinguished in the polysemous approach (Section 2.3). It may include, for instance, subtle pragmatic information about the use of a given construction, as is most clear for the kind of indirect speech act construction exemplified in (4). However, when going up in the taxonomy to higher level constructions, the semantic pole, unavoidably, gets more abstract or schematic. On the continuum from lexicon to grammar, modal verbs may, in many respects, be considered grammatical words. Being auxiliaries, they can be combined with any other verb to form a complex VP-construction. In other words, modal verbs have a great many ‘daughter constructions’ and their semantics, by definition, will thus be fairly abstract. In Figure 5 and Figure 6, it is assumed that there is such a thing as a ‘kunnen construction’, with the kind of vague semantics proposed in Figure 2 (cf. Janssen 2001 for a similar proposal for Dutch moeten ‘must’). In doing so, I may have opened the door for ‘a ghost from the past’ (Section 1) or I may have committed the generality fallacy (Section 2.2). However, I am not claiming that such a construction is a psychologically real entity. In fact, I agree with Kay (2004) that it cannot be determined on the basis of linguistic data whether or not language users actually make this kind of generalization (cf. Sandra 1998). Still, “the relevant obligation of the grammarian (…) is to abstract from the data of the language, all the generalizations (…) that a speaker-hearer might extract” (Kay 2004: nt.5; cf. Fortuin & Boogaart 2009). One issue that will have to constitute a primary concern in the future development of constructionist approaches such as the one proposed in this paper concerns the contribution of pragmatics. Clearly, construction grammar allows for a great deal of pragmatics to be incorporated in the semantic component of constructions, but I am talking here specifically of pragmatics in a different sense, namely to refer to general inferencing on the basis of linguistic and, particularly, non-linguistic information. In the traditional monosemous approach, all aspects of the interpretation of a modal except its ‘minimalist’ semantics was ‘left to pragmatics’, so to speak (see Figure 2'). Because of its focus on form-meaning pairings, and a wider conception of ‘form’, constructionist approaches depend significantly less on such a general appeal to pragmatics. However, not everything in construction

 Ronny Boogaart

grammar can obviously be ‘left to syntax’. In some instances, the intended reading of the modal will not be explicitly indicated at all and it is only linguistic information across clause boundaries or non-linguistic information which can provide clues as to its interpretation. It might be questioned, as it has been by Janssen (2001), if it is possible to make a principled distinction between, for instance, deontic readings of the modal that are triggered by clause-mate particles and those that are triggered by non-linguistic features of the context of use, such as the hierarchical relationship between speaker and addressee. Notes *

Email: [email protected]. Thanks are due to Theo Janssen, Esther Pascual and Olga Fischer as well as to the series editors and an anonymous reviewer for comments on an earlier draft. 1. Many other different classifications of modal notions have been proposed in the literature. In Section 3, I will argue that from a constructionist perspective on modals, the classification of modal concepts loses much of its relevance. 2. I prefer to use the term monosemy rather than vagueness since, as I will argue below, constructions may be monosemous without being vague at all. See Noordegraaf (2005) on the history of the term monosemy. 3. A noteworthy exception is constituted by researchers working within the Columbia School of linguistics, see Contini-Morava et al.€ (2004) for a recent collection of papers within this framework, including a comparison with Cognitive Grammar (Langacker 2004). 4. In the same paper, Langacker takes a rather extreme monosemous view himself when it comes to defining the ‘meaning’ of grammatical categories. Since modal verbs are, in many respects, grammatical words (see Section 4), this might open the door for a monosemous analysis of modals even on Langacker’s account. Indeed, in his other work, he attributes a highly schematic semantics to the meaning of the English modals (e.g. Langacker 1990:€27). 5. This type of corpus is obviously too limited in size and genre (written literary fiction) to be able to draw conclusions about the frequency of the readings in general, but it may be noted that the results are quite similar to those arrived at in the large-scale research reported on by Nuyts (2001). The distribution of the various modal readings differs significantly depending on genre: whereas the epistemic reading, for instance, is relatively rare in general language corpora, it is quite frequent in highly specific genres such as weather reports (Brantjes 2005). 6. The affirmative Dutch sentence in (6)a could get either an evidential, hear-say reading (‘Someone told me that Vera is in bed already’) or a hypothetical/counterfactual reading (‘If Vera were still alive, she would be in bed already.’). 7. There are subtle differences of a similar kind even between different varieties of Dutch. Thus, northern standard Dutch as spoken in the Netherlands typically uses kunnen with the verb vinden (‘to find’), whereas the southern variety of Dutch spoken in Belgium significantly more often uses the verb vinden without kunnen. However, with other verbs, such as geloven (‘to believe’) the situation is the reverse (Diepeveen et al. 2006:€33–39).



Semantics and pragmatics in construction grammar  8. One might argue that modal verbs are grammatical rather than lexical items, but given the basic principles of construction grammar, in particular principle 7b, this should not make a real difference for the way they are treated within construction grammar (but see Section 4.). 9. The top node of the construction network in Figure5 is constituted by the lexical item kunnen (‘can’) and its abstract unitary meaning. Representing lexical items as higher level constructions is not standard practice and it may seem counterintuitive, but it is just taking seriously the theoretical claim of construction grammar that words are themselves constructions as well. Since lexical items, per definition, have a great many ‘daughter constructions’ their position in the network must be such that their form and meaning can be inherited by all of these daughters. 10. Therefore, monosemy goes hand in hand with complete inheritance (Kay & Fillmore 1999) rather than with default inheritance (Goldberg 1995). One might argue, in fact, that complete inheritance is the constructionist way of representing monosemy. 11. A similar situation is described for Old English by Fischer, who argues that this restriction on the epistemic reading of modals provides evidence against the received view that the development of the epistemic reading constitutes an instance of scope increase. Rather, “the modal becomes epistemic only via a more elaborate construction type” (2007:€268). 12. Needless to say, in drawing Figure6 I have not even begun to draw the network for the kunnen-constructions of Dutch – for one thing, two levels hardly make up a hierarchy or a taxonomy. The methodological problems involved in determining what exactly constitutes a construction are manifold; excellent discussions of some of them are offered, for instance, by Wray (2002) and Hoffmann (2004). Let me just say at this point that contrastive analysis of two or more languages, such as reported on in Section 2.3, is an interesting way to bring to light collocations and idioms, i.e. constructions, that may go unnoticed in language-specific research.

References Barbiers, Lambertus C. J. (1995). The syntax of interpretation. The Hague: Holland Academic Graphics. Boogaart, Ronny (2007). Conditionele constructies met moest(en) en mocht(en) in BelgischNederlands en Nederlands-Nederlands. www.neerlandistiek.nl, 07.05. Brantjes, Jenneke (2005). Modaliteit en constructiegrammatica. Doctoraalscriptie, Vrije Universiteit Amserdam. Coates, Jennifer (1983). The semantics of modal auxiliaries. London: Croom Helm. Contini-Morava, Ellen, Robert S. Kirsner, & Betsy Rodríguez-Bachiller (Eds.) (2004). Cognitive and communicative approaches to linguistic analysis. Amsterdam: John Benjamins. Croft, William (1998). Linguistic evidence and mental representations. Cognitive Linguistics, 9, 151–173. Croft, William (2001). Radical construction grammar. Oxford: Oxford University Press. Croft, William & D. A. Cruse (2004). Cognitive Linguistics. Cambridge: Cambridge University Press. Diepeveen, Janneke, Ronny Boogaart, Jenneke Brantjes, Pieter Byloo, Theo Janssen, & Jan Nuyts (2006). Modale uitdrukkingen in Belgisch-Nederlands en Nederlands-Nederlands: corpusonderzoek en enquête. Amsterdam/Münster: Stichting Neerlandistiek VU/Nodus Publikationen.

 Ronny Boogaart Diewald, Gabriele (2006). Context types in grammaticalization as constructions. Constructions. Special Volume 1: Constructions all over – case studies and theoretical implications, Ed. Doris Schönefeld. http://www.constructions-online.de/articles/specvol1/ Fischer, Olga (2007). Morpho-syntactic Change. Functional and Formal Perpectives. Oxford: Oxford University Press. Fortuin, Egbert (2000). Polysemy or monosemy: Interpretation of the imperative and the dative-infinitive construction in Russian. Amsterdam: Institute for Logic, Language and Computation. Fortuin, Egbert & Ronny Boogaart (2009). Imperative as conditional: from constructional to compositional semantics. Cognitive Linguistics, 20, 641–673. Geeraerts, Dirk (1993). Vagueness’s puzzles, polysemy’s vagaries. Cognitive Linguistics, 4, 223–273. Goldberg, Adele (1995). Constructions: a construction grammar approach to argument structure. Chicago: University of Chicago Press. Goossens, Louis (1983). Can and kunnen: Dutch and English potential compared. In F. Daems, L. Goossens (Eds.), Een spyeghel voor G. Jo Steenbergen, huldealbum aangeboden bij zijn emeritaat (147–158). Leuven: Acco. Haspelmath, Martin (2003). The geometry of grammatical meaning: semantic maps and crosslinguistic comparison. In Michael Tomasello (Ed.), The new psychology of language: cognitive and functional approaches to language structure (211–242). Mahwah, New Jersey & London: Lawrence Erlbaum. Heine, Bernd (1995). Agent-oriented vs Epistemic modality: some observations on German modals. In J. Bybee & S. Fleischman (Eds.), Modality in grammar and discourse (17–53). Amsterdam-Philadelphia: John Benjamins. Hoffmann, Sebastian (2004). Are low-frequency complex prepositions grammaticalized? On the limits of corpus data – and the importance of intuition. In Hans Lindquist & Christian Mair (Eds.), Corpus approaches to grammaticalization in English (171–210). Amsterdam/ Philadelphia: John Benjamins. Janssen, Theo A. J. M. (2001). The semantics of Dutch moeten ‘must, should, have to’ from a typological and a relevance-theoretical perspective. In P. Bogaards et al.€ (Eds.), Quitte ou double sens (109–133). Amsterdam/Atlanta: Rodopi. Kay, Paul & Charles J. Fillmore (1999). Grammatical constructions and linguistic generalizations: the What’s X doing Y? construction. Language, 75, 1–33. Kay, Paul (2004). Pragmatic aspects of grammatical constructions. In G. Ward & L. Horn (Eds.), Handbook of pragmatics (675–700). Oxford: Blackwell. Kirsner, Robert (1993). From meaning to message in two theories: Cognitive and Saussurean views of the modern dutch demonstratives. In R. Geiger and B. Rudzka-Ostyn (Eds.), Conceptualizations and mental processing in language (80–114). Berlin and New York: Mouton de Gruyter. Kratzer, Angelika (1977). What must and can must and can mean. Linguistics and Philosophy, 1, 337–355. Kratzer, Angelika (1981). The notional category of modality. In H.J. Eikmeyer & H. Rieser (Eds.), Worlds, words, and contexts. (38–74). Berlin: De Gruyter. Kratzer, Angelika (1991). Modality. In A. von Stechow & D. Wunderlich (Eds.), Semantics: an international handbook of contemporary research (639–650). Berlin: De Gruyter. Kronning, Hans (1996). Modalité, cognition et polysémie: sémantique du verbe modal devoir. Acta Universitatis Upsaliensis (Studia Romanica Upsaliensia 54), Almqvist & Wiksell International. Langacker, Ronald (1987). Foundations of Cognitive Grammar. Stanford: Standford University Press. Langacker, Ronald (1990). Subjectification. Cognitive Linguistics, 1, 5–38.



Semantics and pragmatics in construction grammar  Langacker, Ronald (2004). Form, meaning, and behavior: the Cognitive Grammar analysis of double subject constructions. In E. Contini-Morava et al (Eds.), 21–60. Langacker, Ronald (2005). Construction grammars: Cognitive, Radical, and less so. In Ruiz de Mendoza, Francisco J. and Sandra Peña Cervel (Eds.), Cognitive Linguistics: Internal Dynamics and Interdisciplinary Interaction (101–159). Berlin/New York: Mouton de Gruyter. Levinson, Stephen (2000). Presumptive meanings: the theory of generalized conversational implicature. Cambridge: MIT Press. Mortelmans, Tanja (2007). Modality in Cognitive Linguistics. In Geeraerts, D. and H. Cuyckens (Eds.), The Oxford Handbook of Cognitive Linguistics (869–890). Oxford: Oxford University Press. Narrog, Heiko (2005). On defining modality again. Language Sciences, 27, 165–192. Noordegraaf, Jan (2005). Over monosemie: Bally, Humboldt, Hemsterhuis. In J. Noordegraaf (Ed.), Een kwestie van tijd: vakhistorische studies (134–148). Münster: Nodus Publikationen. Nuyts, Jan (2001). Epistemic modality, language and conceptualization: a cognitive-pragmatic perspective. Amsterdam: John Benjamins. Nuyts, Jan (2005). The modal confusion: on terminology and the concepts behind it. In A. Klinge & H. H. Müller (Eds.), Modality: studies in form and function (5–38). London: Equinox. Papafragou, Anna (2000). Modality: Issues in the semantics-pragmatics interface. Amsterdam: Elsevier. Perkins, Michael R. (1983). Modal expressions in English. London: Frances Pinter. Sandra, Dominiek (1998). What linguists can and can’t tell you about the human mind: a reply to Croft. Cognitive Linguistics, 9, 4, 361–478. Stefanowitsch, Anatol (2003). A construction-based approach to indirect speech acts. In Klaus-Uwe Panther & Linda Thornburg (Eds), Metonymy and pragmatic inferencing (105–126). Amsterdam and Philadelphia: Benjamins. Sweetser, Eve E. (1990). From etymology to pragmatics. Cambridge: Cambridge University Press. Traugott, Elizabeth C. (1989). On the rise of epistemic meanings in English: an example of subjectification in semantic change. Language, 57, 33–65. Tuggy, David (1993). Ambiguity, polysemy and vagueness. Cognitive Linguistics, 4, 273–291. Van der Auwera, Johan & Vladimir A. Plungian (1998). Modality’s semantic map. Linguistic Typology, 2, 79–124. Van Ostaeyen, Gert & Jan Nuyts (2004). De diachronie van kunnen. Antwerp papers in Linguistics 109. Verhagen, Arie (2005). Constructions of intersubjectivity. Oxford: Oxford University Press. Wärnsby, Anna (2002). Modal constructions? The Department of English in Lund: Working Papers in Linguistics 2. (http://www.englund.lu.se/images/stories/pdf-files/workingspapers/vol02) Wray, Alison (2002). Formulaic language and the lexicon. Cambrdige: Cambridge University Press.

Index of constructions A Apodosis 193-209 B Bare Parenthesis Construction 58 C Clausal integration 193, 197-210 Concessive repair 102-103, 109 Content conditional 11, 193-194, 210 Context focussing parce que Construction 8, 45 Correlative then 195-197, 207, 209

F Focus structure 145, 155-163, 173, 175 I Insubordinate parce que 47

R Reactive construction 100, 103, 107 Responsive construction 99100, 106

J ‘Joined’ ordering 146-189

T Topicalization 127

M Modality 84-87, 90, 214-219, 224-225, 227-228, 232-236

X X-och-x construction, Swedish 99-109

P Participle / participial 9, 58, 65, 70-71, 76-77, 82-84, 86-87, 91 Predictive conditional 193-209

Subject index A Abstractions from concrete utterances 95 Activation status 132 Activity 113, 131-139 Affordances 7, 35-38, 40, 41, 99 Affordances of physical settings 35-37, 40 Allostruction 181-183, 186, 190 Alternativity 195, 204, 208-209 Apodosis 193-209 B Background 1, 5, 7-8, 20, 33-34, 44-48, 58, 67-68, 80, 84, 105, 121, 132-134, 152, 156, 159, 164, 175, 188, 200, 204-205, 219, 225 Bare Parenthesis Construction 58 C Can 12, 213-215, 239 Clausal integration 193, 197-210 Cognitive process 40, 97 Coherence 6, 55-56, 131-138 Common ground 5, 8, 10, 33, 37, 43-46, 53-54, 59, 122-123 Complexity 39, 40, 147-149, 153154, 186, 189 Compositional / non-compositional 2, 43, 66-67, 74, 76, 82, 146 Compositionality 68, 83, 107 Concessive repair 102-103, 109 Consciousness 108, 132 Constitutive role of context 23 Constraints as construction 182 Construct 23, 43-47, 106, 114, 122, 127-128, 134, 138, 182, 193, 202 Construction 1-14, 27-42, 43-47, 55, 57-60, 63, 65, 67-68, 83-94, 97-110, 111-140, 145-148, 150, 155, 161-164, 174-192, 193, 198, 201, 209-211, 213-215, 220, 222-241

Construction Grammar 1-6, 8-12, 38, 43-44, 67-68, 97, 106108, 111, 113, 136, 181, 193, 201, 209, 213-215, 220, 224, 229-231, 237, 239 Constructional scheme 138 Content conditional 11, 193-194, 210 Context focussing parce que Construction 8, 45 Context(s) 1-14, 17-42, 43-45, 49, 51-53, 57, 59, 60, 63-70, 72, 75-76, 79-93, 97-99, 102-103, 105-109, 111-139, 146-147, 152, 154, 156, 158, 163-164, 166, 169, 171-174, 179-189, 195-197, 202, 207-208, 210, 216-219, 222, 227, 238, 240 Contrast 56, 71, 92, 109, 112-113, 123, 127, 129, 131, 135, 148, 157160, 193, 197, 204-211 Contrastive focus 157-159 Contrastivity 193, 204-209 Conventionalization 65-66, 82, 88-90, 216 Conversation 4-5, 27, 42, 43, 61, 97, 100-106, 109-110, 111-115, 120-129, 132, 134, 137, 139-140, 166, 168, 219, 241 Conversation Analysis (CA) 4-5, 100, 111, 113, 121 Co-occurring constructions 104 Corpus data 3, 27, 37-38, 161, 187, 240 Correlative then 195-197, 207, 209 Co-selection 104 Cross-linguistic 111, 222, 224, 226, 228, 232, 235 Cypriot Greek 7, 27, 29, 32-33, 36-38, 40, 42 D Diachrony 3, 65-66, 68, 72, 88, 216, 219, 222, 229, 232

Dialogical theory 97 Discourse dependency 50 Discourse pattern 50, 55-57, 64, 84, 86 Discourse topic 133, 205 Discourse-familiarity 145, 152157, 167, 174 E Emphasis 21, 140, 145, 158, 170, 197 Enabling condition 104 Encyclopaedic knowledge 6, 11 Evidence 11, 13, 26, 29, 33, 38-42, 45, 87, 90-92, 121, 151, 168, 188, 190, 193, 198-200, 209, 218-219, 239 Expectedness 166-171, 178-179 External syntax (of constructions) 75, 84, 88, 106-107 Extra-linguistic context 6-7, 18, 19, 22, 33, 166 F Focus structure 145, 155-163, 173, 175 Formal idiom 101 Formalisation 107 Form-function mapping 193, 204, 209 Frame semantics 6, 9, 12, 23, 39, 67, 90, 168 Frequency 2, 29, 32, 40, 90, 176, 186, 219, 226, 238 Function 1-5, 8-9, 14, 25, 29-32, 40, 45-49, 52-59, 63-68, 71-72, 74, 76, 82-83, 86-90, 93, 99, 101, 105, 110, 111-115, 117, 120, 127-140, 145, 154, 177, 181, 190, 193, 195, 198, 200, 204-211, 213, 216, 225, 229, 231, 240-241 Functional potential 99, 101, 105, 110, 131, 135-136, 140

246 Subject index G Gesetz der wachsenden Glieder 147 Gibsonian affordances 7 Grammaticalization 1, 10, 13, 61, 63, 67, 90-94, 93, 108, 138, 155, 210, 216, 221-222, 235, 240 H Homonymy 215-216, 220, 235 I Iconic / iconicity 11, 47-48, 53, 55, 59, 185, 208-209 Illocutionary force 7, 9, 18, 2627, 32-33, 35, 37, 40, 49-50, 53, 200, 208 Incrementality of interpretation 33, 35 Indirect speech act 12, 18, 216, 225, 237 Generality fallacy 219-220, 235 Polysemy fallacy 223 Rule-list fallacy 219-220, 223, 232 Inflectional form 84, 163 Inflectional paradigm 9, 66, 70, 87 Inheritance network 45, 57-58, Insubordinate parce que 47 Interactional linguistics (IL) 10, 99, 109, 111, 113, 136, 138 Inversion 197, 199 J ‘Joined’ ordering 146-189 K Knowledge of the world 19, 166 Kunnen 12, 213-228, 231-234, 236-239 L Language variation and change 65 Lectal factors 164-166 Length 23, 145, 147-149, 152-155, 167, 176-178, 186, 187, 189, 191, 231 Lexical restrictions 150-151 Lexico-grammatical change 63 Lexico-grammatical network 88 Literalness 172-174, 176, 189

M Meaning 1-2, 6-12, 17-18, 21-26, 34, 37, 39, 41, 43, 52-53, 57-59, 63-93, 97-101, 105-110, 114, 129-131, 139, 172-173, 181, 189, 191, 193, 196, 199, 201, 209-211, 213-241 Meaning potential 10, 99, 105, 108, 109, 139 Mental space 11, 193-211 Method 105-107, 110 Minimal context 7, 27, 29-36, 40, 42 Minimalism 107, 108 Mixed-category morphology 67 Modal verb 213-216, 222, 227228, 235 Deontic 213-214, 216-225, 228, 232-238 Dynamic 5, 21, 39, 63, 97-98, 107, 115, 171-172, 213-225, 228, 232-236 Epistemic 47, 59, 213-215, 218-224, 227-228, 232-236, 238-239 Modality 84-87, 90, 214-219, 224-225, 227-228, 232-236 Modeling 145, 181 Moeten 214, 235, 237 Monosemy 12, 214-215, 217-219, 224, 228-229, 231-232, 237-239 Motivation 27, 82, 84, 145, 181 Must 214, 235, 237 O One clause at a time 98, 110 Other-responsive 101 P Participle / participial 9, 58, 65, 70-71, 76-77, 82-84, 86-87, 91 Particle placement 3, 10-11, 145-192 Perspective 6, 11, 36, 39, 64, 80, 82, 87-88, 114, 117, 120, 133, 135, 175, 181, 192, 201, 209, 213-216, 218-219, 221, 225, 229, 231, 236, 238 Politeness 32, 39, 41-42, 168, 208, 211 Polysemy / polysemous structure / emerging polysemy 3, 66, 87, 89, 213-215, 217, 219229, 231-233, 237

Polysemy network 89, 221-227, 231-233 versus construction network 232-233, 235-236, 239 Potential 10, 25, 35-36, 65-66, 68, 71-78, 84, 86-99, 101, 105, 108-110, 114, 125, 131, 135-136, 139-140, 158, 164, 172, 180, 228, 240 Pragmatic consequences 6-7, 21-22, 26, 37 Pragmatic role 132 Predictive conditional 193-209 Prefaces 30, 32 Processing hypothesis 174-179 Pronoun(s) 10, 36, 71, 73, 112, 116-130, 135, 137, 145-155, 170, 187, 210, 225 Protasis 197, 206, 208-209 Prototype 68-69, 79, 81, 221 Q Qualitative 19-20, 121 Quantitative 60-61, 105, 119, 121 R Reactive construction 100, 103, 107 Recontextualization of grammar 181 Reference 7, 18, 26, 37, 41, 49, 56, 58, 66, 71, 76, 123, 131-138 Region 112, 146, 163-165, 177-178, 180, 184-185, 211 Register 2, 40, 91, 107, 146, 163166, 176 Relevance 7-8, 39, 44, 91, 145, 154, 197-200, 204-210, 218, 220, 226, 240 Requests 27, 29-32 Responsive construction 99100, 106 Resumption 197-201, 204, 206, 208-210 S Semantic map 235 -236 Schematic idiom 101 Self-responsive 101-104 Semantic network / semantic space 68-69, 82, 89 Sequential context 97, 99 Sequential position 131-132, 136

Subject index 247 Shared knowledge 5, 120, 129-130 Social action and interaction 97 Speech act 2, 7, 11-12, 18, 28-33, 41-42, 52, 108, 114, 193-210, 216, 225, 237, 241 Speech-act conditionals 11, 193, 195, 197-210 Split ordering 146-148, 151-152, 157-179, 184-189 Spoken language 3, 42, 45, 61, 90, 92, 97, 111-112, 115, 119-120, 137, 138 Stativity 171-172, 179-180 Syntactic integration 58, 193, 197, 201

Syntactic persistence 163, 189 Syntactic priming 11, 163, 191 T Token frequency 29, 32 Topic 2, 127, 132-136, 139, 152 Topicalization 127 Truth conditions 8, 17-18 Type frequency 29 U Unaccented pronoun 145-146, 149, 151-152, 155, 187 Usage events 98-99, 105, 108, 223

V Vagueness 214, 217-218, 223, 226, 232, 238 Valence 68, 70-71, 73-75, 77, 79, 84, 86, 130 W Weight 145, 147-155, 172, 177-178, 183, 186, 189, 197 Written language 8, 164 X X-och-x construction, Swedish 99-109

In the series Constructional Approaches to Language the following titles have been published thus far or are scheduled for publication: 9 8 7 6 5 4 3 2 1

Bergs, Alexander and Gabriele Diewald (eds.): Contexts and Constructions. 2009. v, 247 pp. Barðdal, Jóhanna: Productivity. Evidence from Case and Argument Structure in Icelandic. 2008. xiii, 209 pp. Hilpert, Martin: Germanic Future Constructions. A usage-based approach to language change. 2008. ix, 205 pp. Iwata, Seizi: Locative Alternation. A lexical-constructional approach. 2008. xiv, 239 pp. Leino, Jaakko (ed.): Constructional Reorganization. 2008. vi, 155 pp. Fried, Mirjam and Hans C. Boas (eds.): Grammatical Constructions. Back to the roots. 2005. viii, 246 pp. Östman, Jan-Ola and Mirjam Fried (eds.): Construction Grammars. Cognitive grounding and theoretical extensions. 2005. viii, 325 pp. Fried, Mirjam and Jan-Ola Östman (eds.): Construction Grammar in a Cross-Language Perspective. 2004. vi, 209 pp. Kuno, Susumu and Ken-ichi Takami: Functional Constraints in Grammar. On the unergative– unaccusative distinction. 2004. ix, 242 pp.