A Typology of Purpose Clauses (Typological Studies in Language)

A Typology of Purpose Clauses

Typological Studies in Language (TSL) A companion series to the journal Studies in Language. Volumes in this series are functionally and typologically oriented, covering specific topics in language by collecting together data from a wide variety of languages and language typologies.

Editor Spike Gildea

University of Oregon

Editorial Board Balthasar Bickel

John Haiman

Andrew Pawley

Bernard Comrie

Martin Haspelmath

Doris L. Payne

Denis Creissels

Bernd Heine

Franz Plank

William Croft

Paul J. Hopper

Anna Siewierska

Nicholas Evans

Andrej A. Kibrik

Dan I. Slobin

Carol Genetti

Marianne Mithun

Sandra A. Thompson

Leipzig

Leipzig / Santa Barbara Lyon

Albuquerque Canberra

Santa Barbara

St Paul

Leipzig Köln

Pittsburgh Moscow

Santa Barbara

Volume 88 A Typology of Purpose Clauses by Karsten Schmidtke-Bode

Canberra Oregon

Konstanz

Lancaster Berkeley

Santa Barbara

A Typology of Purpose Clauses

Karsten Schmidtke-Bode Friedrich Schiller University, Jena

John Benjamins Publishing Company Amsterdam / 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 Schmidtke-Bode, Karsten. A typology of purpose clauses / Karsten Schmidtke-Bode. p. cm. (Typological Studies in Language, issn 0167-7373 ; v. 88) Includes bibliographical references and index. 1. Grammar, Comparative and general--Clauses. 2. Grammar, Comparative and general-Subordinate constructions. I. Title. P297.S35    2009 415--dc22 isbn 978 90 272 0669 5 (hb; alk. paper) isbn 978 90 272 8898 1 (eb)

2009028110

© 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 Acknowledgements Abbreviations and notational conventions 1. Aims and scope of the book

vii ix 1

2. Theoretical and methodological foundations 5 2.1 Typological generalizations and distributions  5 2.2 A usage-based approach to explaining typological generalizations  9 2.2.1 Functional-typological accounts of language structure  9 2.2.2 The usage-based nature of language universals  14 2.2.3 Interim summary  16 2.3 A functional definition of ‘purpose clause’  17 2.4 Language sampling  20 2.5 Previous work  26 3. The grammar of purpose: Documentation and explanation 3.1 Preliminaries  29 3.1.1 On the identification of purpose clause constructions  29 3.1.2 On the quantitative analysis of constructions  33 3.2 Coding of the situation: The form of the verb  34 3.2.1 Finiteness and deranking in purpose clauses  35 3.2.2 Tense, aspect and mood in purpose clauses  42 3.3 Coding of participants: Argument structure  50 3.3.1 The notional subjects of purpose clauses  51 3.3.2 Further constraints on argument realization in p ­ urpose clauses  62 3.3.3 Argument structure and the ecology of constructional systems  69 3.4 Constructional gestalt features and the coding of purpose  71 3.4.1 Structure and distribution of primary gestalt features  72 3.4.2 Secondary gestalt features and the semantics of purpose  88 3.4.3 ‘No marking’: Purposive inferences and ambiguities  103

29

vi

A Typology of Purpose Clauses

3.5 Positioning patterns of purpose clauses  110 3.5.1 Data and analysis  110 3.5.2 Explanation and discussion  116 3.6 Semantic and pragmatic idiosyncrasies in purpose clauses  129 3.6.1 Avertive (‘lest’) constructions  129 3.6.2 Further issues  144  urpose clauses in the syntactic and ­conceptual space 4. P 147 of complex sentences 4.1 The synchronic status of purpose clauses  147 4.1.1 The conceptual and syntactic space of complex sentences  147 4.1.2 Coordination and subordination: (A)symmetry in purpose clauses  150 4.1.3 Purpose clauses and other adverbial clauses  151 4.1.4 Purpose clauses and complement clauses  157 4.1.5 Purpose clauses and relative clauses  165 4.2 The diachronic development of purpose clauses  170 4.2.1 Grammaticalization and the dynamic nature of grammar  170 4.2.2 Constructional intraference: From purpose to complement (and relative) clauses  172 4.2.3 The purpose-reason-result cline(s)  176 4.2.4 From ‘motion-cum-purpose’ to TAM constructions  178 4.2.5 From avertive (‘lest’) clauses to apprehensional epistemics  185 4.2.6 The origins of purpose clauses  186 4.2.7 Summary: The developmental trajectories of purpose clauses 196 Conclusion and outlook

199

References

205

Index of authors Index of languages Index of subjects

221 224 226

Acknowledgements In my younger and more vulnerable years my father gave me some advice that I’ve been turning over in my mind ever since: “Whenever you feel like criticizing any one,” he told me, “just remember that all the people in this world haven’t had the advantages that you’ve had.”  (F. Scott Fitzgerald, The Great Gatsby)

The advantages that I have had during all stages of this publication are enormous, indeed. Therefore, I  would like to seize the opportunity to thank a number of people who I am greatly indebted to, as they have made invaluable contributions on my way to completing this book. First and foremost, I wish to express my gratitude to Johanna Nichols and Holger Diessel, who are undoubtedly responsible for arousing my interest in linguistic typology and who have strongly encouraged me to conduct the present study. Throughout the entire research process, I have benefited from their ideas, professional enthusiasm and great support. On my way to finalizing the manuscript, I have received feedback and valuable suggestions from numerous colleagues and friends. I would like to thank in particular Daniel Wiechmann, John Thomson and Katja Hetterle for insightful discussions about quantitative analysis, astute nativer-speaker advice and patient proof-reading of draft versions, respectively. I am also very grateful to a number of language specialists, notably Keren Rice, Igor Nedjalkov and John Roberts, who provided most insightful comments on the grammars of Slave, Evenki and Amele, respectively. Needless to say, none of them is to blame for any errors in the interpretation of the data. I alone am responsible for coding, interpreting and discussing the material used in this book. Special thanks also go to an anonymous reviewer for extremely helpful comments and suggestions for improvement, as well as to the late Mickey Noonan for his editorial guidance and Kees Vaes from John Benjamins Publ. Co. for impeccable assistance and communication during the publishing process. I have also been in the very fortunate and privileged position to be funded most generously by the German National Academic Foundation (Studienstiftung des deutschen Volkes), which opened up an array of research possibilities that would otherwise have been beyond my reach, including my stay at the Linguistics Department of the University of California at Berkeley. But most importantly, I owe the greatest debt of gratitude to my wife Tina and my family for their constant emotional and intellectual support. It is with deepest affection that I dedicate this book to them and to the memory of my grandfather, who would have loved to see it published.

Abbreviations and notational conventions For each example sentence from a language other than English, an idiomatic translation is provided alongside a literal (interlinear ‘morpheme-by-morpheme’) one. The interlinear abbreviations given below were adopted by the Framework for Descriptive Grammars project (1991) and they largely correspond to the Leipzig Glossing Rules, which have now become standard practice in linguistic typology. Abbreviations for interlinear morpheme translation 1, 2, 3 I, II, III a abl abs abstr acc adv aff agt all an ant antip aor appl art asp aux ben cau caus clf cmpm co

1st, 2nd, 3rd person morphological classes transitive agent ablative (‘from’) absolutive abstract accusative adverbial (case) affirmative agent(ive) allative (‘to’) animate anterior antipassive aorist applicative article aspectual auxiliary benefactive causal causative classifier compound marker coordination

com comp compl conn cont conv cop cor dat def dem denoml des det dim dir dist ds dtr du dur dyn erg evid

comitative complementizer completive connector, connective continuous converb copula coreference dative definite demonstrative denominalizer desiderative determiner diminutive directional, directive distal, distance different subject (switch-reference) detransitivizer dual durative dynamic ergative evidential

x

A Typology of Purpose Clauses

ex excl exp f ff fin foc freq frust fut g gen gl gnr hod hyp imm imp impf inan inch incl incompl ind indf inf instr int intr io irr iter lest loc log m mdl med

exclusive exclamation expected (aspect) femine fulfillment finite focus frequentative frustrative future ditransitive recipient genitive goal generic hodiernal (past, future) hypothetical immediate (past, future) imperative imperfective inanimate inchoative inclusive incompletive indicative indefinite infinitive instrumental intentional (mood, aspect) intransitive indirect object irrealis iterative avertive marker, apprehensional epistemic locative logophoric masculine modal medial

modifier mod mot motion masdar (gerund) msd neuter n n- non- (e.g. nsg non singular, npst nonpast) negative, negation neg non-finite nf nominative nom nominalizer noml nonlocut disjunct indexation near (future, past) nr object obj oblique obl optative opt transitive patient p passive pass plural pl plpf pluperfect (past perfect) punctual pnct polite pol possessive poss potential pot preposition prep preverb prev perfect prf perfective prfv prog progressive proprietive prop pros prospective (aspect) proximal, proximate prox present prs prvt privative (‘without’) past pst ptc participle particle ptcl purpose, purposive purp quot quotative reduplication rdp ref referential



refl rel rem res rl s sbj seq sg sim spec src srel sress ss

Abbreviations and notational conventions

reflexive relative clause marker remote (past, future) resultative realis intransitive subject subject sequential, consecutive singular simultaneous specific, specifier source superelative superessive same subject (switchreference)

sub subj sugg sup temp th tns ton top tr trns ven vis vrbl

subordinate, subordinator subjunctive suggestive superlative temporal thematic vowel tense tonal marker (affix) topic transitive transitivizer ventive visual (evidential) verbalizer

Symbols In both original language and interlinear morpheme translation x y x–y x+y x=y

word boundary between x and y morpheme boundary between x and y x and y form a compound or a derivative stem x and y are joined by clisis

In original language only ø null expression of meaning [x] x is a syntactic constituent (purpose clauses are generally bracketed this way) / end of intonation unit In interlinear morpheme translation only (x) x is not overtly marked in the original (i.e. null expression of meaning) x.y x and y are grammatical (sub-)categories of one original language morpheme Notational conventions In example sentences, structures to be highlighted are printed in boldface. L ­ anguage-particular lexemes or grammatical markers are generally given in italics.

xi

xii A Typology of Purpose Clauses

The meaning of linguistic elements is usually given in ‘inverted commas’, and whenever reference is made to conceptual domains (e.g. the concept of purpose), they are capitalized. Alternatives are given in {winged brackets}. Ungrammaticality is indicated by an *asterisk, and semantically odd sentences carry a ?question mark.

chapter 1

Aims and scope of the book The present study seeks to provide the first comprehensive typology of purpose clause constructions across the world’s languages. Purpose clauses are part of complex sentence constructions which encode that one verbal situation, that of the matrix clause, is performed with the intention of bringing about another situation, that of the purpose clause, as in the following examples from English:

(1) Maria went to the bakery [in order to get some croissants].



(2) Brendan put the bike into the garage [so that it would not get wet in the rain].



(3) I brought a book [for Aaron to read on the plane].

Complex sentence constructions have figured prominently in the last decades of typological research. While some clause types have continuously received particular attention, such as relative clauses (e.g. Comrie 1998), complement clauses (e.g. Horie 2000) or conditional constructions (e.g. Khrakovskij 2005), purpose clauses are considerably less well-studied. On the one hand, an in-depth (monographic) study of the phenomenon is available for only one language, i.e. English (Jones 1991). Cross-linguistic research, on the other hand, has focused on particular, relatively isolated features of purpose clauses (e.g. Haspelmath 1989, ­Kazenin 1994, Cristofaro 2003 and 2005, Verstraete 2008). However, no attempt has yet been made to integrate all aspects into a comprehensive and empirically well-founded cross-linguistic survey. This is precisely the aim of the present study. Based on a geographically and genetically diverse sample of 80 languages, I will provide, firstly, a synchronic documentation of the grammatical properties of purpose clauses. The central goal is to uncover the unity and diversity of the linguistic means by which purposive relations are encoded in human languages. Explanations for universal tendencies of morphosyntactic coding will be sought in the conceptual characteristics of purposive situations, in the communicative functions of purpose clauses, and in the cognitive-psychological mechanisms involved in language use. Therefore, the explanatory apparatus to be applied is embedded in the more general functional and particularly usage-based approach to language structure. Secondly, purpose clauses will be investigated in relation to other types of complex sentences. By virtue of modifying the proposition in the matrix clause,

2

A Typology of Purpose Clauses

purpose clauses are a specific semantic type of adverbial clause combining. It will be shown, however, that purposive relations deviate from typical adverbial relations in a number of important ways, and at the same time need to be characterized by features commonly associated with other types of complex sentences, notably complementation structures. This constellation often results in peculiar coding properties of purposive constructions that are not shared by (most) other adverbial clauses in a given language. Moreover, at the synchronic level, one frequently gets the impression that purpose clauses are under a certain classificatory tension in the variation space that complex sentences unfold across languages. They appear to be ‘expelled’ by adverbial clauses as a set and show considerable overlap instead with complement clauses, (some) relative clauses and the domain of deontic modality. This static impression is, however, a result of principled pathways of diachronic extension, which are rooted in cognitive and communicative pressures on the use of purpose clauses in actual discourse. For this reason, the typological survey of purpose clauses presented here will also take stock of what is known about the historical pathways along which purpose constructions develop. The goals of my investigation are reflected in the structure of the book. In Chapter 2, I will outline the theoretical and methodological foundations for the cross-linguistic study of purpose clauses. This will include some important theoretical premises of functional-typological work, a notional (i.e. structure-independent) definition of purpose clauses ensuring cross-linguistic comparability, and a presentation and justification of the language sample that forms the database of the current study. An important methodological aspect in regard to which this book deviates from other typological works is the tight integration of language-particular and cross-linguistic work. Specifically, we will exploit the rich data on English purposive constructions supplied by previous research in formal, discourse-pragmatic and historical linguistics, which often provides useful parameters and stimuli for a systematic cross-linguistic investigation. Conversely, a usage-based approach to typological distributions requires us to relate cross-linguistic data about grammars to actual usage events, particularly exhaustive frequency data. At several stages of the investigation, we will thus draw on corpora of English to supplement (or motivate) typological generalizations. In an incidental way, then, the present study thus also aims to make a contribution to a recent research programme in English linguistics (cf. Kortmann 2007), which seeks to assess the relative typicality of grammatical phenomena of English against the variation space of human languages. Chapter 3 constitutes the first (and largest) of two empirical parts of the book. It will document and analyse the ways in which purposive relations are encoded in the world’s languages. Since it is common for one language to have several strategies for the expression of purpose, it was essential to compile a database of all dis-



Chapter 1.  Aims and scope of the book

tinct purposive constructions found across the 80 languages sampled. This procedure does not only ensure that less common coding patterns are considered as well; it also allows for precise quantitative analyses since constructions define much narrower, fine-grained typological variables and categories than languages. Working with these variables, in turn, paves the way for the application of significance tests whenever feasible. The chief grammatical properties to be analysed in this way are the form of the verb, argument structure configurations (including co-reference patterns), clause-linking and purposive morphology, the positioning patterns of purpose clauses and, finally, some selected semantic and pragmatic idiosyncrasies such as the encoding of negative purpose in so-called avertive (‘lest’) constructions. Significant distributions and correlations will be formulated as probabilistic implications, each of which is tested for geographical independence so that areal patterns can be distinguished from genuinely universal ­tendencies. Chapter 4 will be concerned with the place of purpose clauses in the conceptual and syntactic space of complex sentence constructions. This requires a shift from the construction-specific analysis to a ‘whole-language’ perspective since we wish to probe into the synchronic connection of purpose clauses with other types of complex sentences. By systematically relating purposive constructions to all traditional classes of clause combining (adverbial clauses, complementation, relativization, coordination), a special status of purpose clauses in the typology of complex sentences will be empirically uncovered. Finally, we will explore the historical trajectories that give rise to this peculiar synchronic status. The concluding Chapter 5 contains a brief summary of the most important results of this study and points to some promising directions for future research. For environmental reasons, the publication will dispense with an elaborate appendix. As an alternative, the reader is invited to visit the companion website of this book: . It provides high-resolution (and occasionally coloured) images of all original graphs contained in this book, as well as illustrative excerpts from the databases that provide an impression of coding and analytical practices. On the whole, I hope that the integration of typological and language-­specific analyses, the combination of synchronic and diachronic perspectives, and a synthesis of documentation and explanation will provide a comprehensive and coherent picture of purpose clauses that unveils universal preferences in coding purposive relations, but at the same time does justice to the fascinating breadth of variation that the grammars of human languages display.

3

chapter 2

Theoretical and methodological foundations This chapter will delineate some important theoretical and methodological concepts that are essential to a typological study of the kind envisaged here. It is divided into five sections. In §2.1, I will broadly sketch the fundamental premises of crosslinguistic research, focussing on the characteristics of typological generalizations and distributions. §2.2 will situate the present study in the so-called functional-­ typological and more generally the usage-based approach to language, and is meant to expound in considerable detail the explanatory apparatus relevant to my investigation. In §2.3, it will be shown that a serious commitment to the basic principles of language typology requires us to define the domain of enquiry in structure-independent terms. Therefore, a functional, universally applicable definition of complex sentences and purpose clauses will be provided. In §2.4, the issue of language sampling will be addressed. Since the present study aims to provide a survey of the cross-linguistic range of variability of purpose clauses, it is of vital importance that the primary database be a diversified and representative sample of the world’s languages. I will thus outline and justify the policies which guided the sampling of languages for this analysis. Finally, in §2.5, I will briefly take stock of previous research on purpose clauses that will provide a valuable ­basis for the present study.

2.1  Typological generalizations and distributions The present study is situated in the framework of modern linguistic typology as pioneered by Greenberg (1963). In this respect, it is concerned with the crosslinguistic comparison and classification of linguistic structures, with the aim of uncovering both the diversity and systematic constraints on structural variation in human languages. The latter can be conceived of as empirical generalizations drawn inductively from a large body of cross-linguistic data, and usually go by the name of language universals. Accordingly, a principal goal of this book is to provide a comprehensive description of the linguistic means by which purposive relationships are encoded across the world’s languages, and, by way of comparing purpose clause constructions in a representative sample of languages, to reveal significantly recurrent trends of grammatical coding in this domain. Importantly, the universals thus obtained differ tremendously from the homonymous notion in formal, notably generative, linguistics. In those approaches,

6

A Typology of Purpose Clauses

linguistic universals are commonly seen as genetically endowed principles of linguistic organization, collectively known as Universal Grammar (henceforth UG; cf. Chomsky 2000 for a classic exposition). By virtue of being innate and speciesspecific, the ingredients of UG are necessarily ‘absolute universals’ in Bickel’s (to appear) sense: They form an indispensible part of a descriptive metalanguage that is necessary to capture the grammatical organization of human languages. According to Newmeyer (2007: 135), one may succinctly formulate the (mainstream) generative position as follows: “There exists a set of formal categories FC1, FC2, FC3, … FCn provided by universal grammar from which languages may choose, without it necessarily being the case that each be manifested in each language.” In other words, universals in formal linguistic theory intend to capture the deep structure of human languages (e.g. Anderson 1976 on the ‘deep’ universality of the category ‘subject’) and derive from an innate predisposition for language, i.e. from what has been called ‘representational innateness’ (Elman et al. 1996: 25) or even a ‘language organ’ (Anderson and Lightfoot 2002). This position is not only extremely controversial from a psychological and ontogenetic view (e.g. Dąbrowska 2004), but has also proven difficult for the very practice of cross-linguistic research. Apart from the fact that no authoritative list of UG components is currently available (cf. Tomasello’s [2004] critique), the most serious challenge is that hardly any grammatical category seems to be a necessary prerequisite for the description of every language, or can even be equated across languages (cf. Croft 2001, Dryer 1997, Haspelmath 2007). For this reason, linguistic typology in the Greenbergian tradition has concerned itself with the observable surface structures of human languages. Universals are but extrapolations from significant statistical trends in samples of languages (or structures, for that matter). Rather than falling out deductively from a preconceived formal architecture of grammar, they are empirical results of inductive research. To be sure, typologists, too, need a metalanguage to define their domain of investigation (more on this below); but they would generally be reluctant to consider the categories and constructs of this metalanguage to be of any biological reality for individual speakers (cf. Haspelmath 2008a on this point). In keeping with this practice, the present study relies on a universally applicable metalanguage for defining purpose clauses (cf. §2.3 below) and formulates alleged universals of coding strategies in purposive constructions as probabilistic (and hence exception-ridden) generalizations. More often than not, such statements will take the form of conditional implications: For instance, we may find that if a purpose clause construction allows for its direct object to be controlled by a matrix clause argument, it will also license its subject to be controlled this way. Whenever such implications are formulated, it should be borne in mind that they have a statistical (rather than a strictly logical) structure (cf. Cysouw 2003),



Chapter 2.  Theoretical and methodological foundations

and should thus always taken to be interpreted as ‘If p, then F(q) > F(not-q)’ (cf. Bickel 2006a), with F being shorthand for the observed frequencies of q and not-q in the sample. Note that such statistical implications are not necessarily less powerful constructs as compared to the absolute principles of UG. On the contrary, they “give us some additional insight into the nature of grammar: If a property A is implied by a property B, then it is quite likely that the two are causally related” (Haspelmath 1997: 8). This insight paves the way for developing explanatory models for grammatical structure which are anchored, not in a universal formal architecture of grammar, but in (functionally) motivated relationships between structures in a given domain (so-called ‘intra-domain implications’ in Haspelmath 2008b), as we will see shortly. To sum up thus far, the picture obtained by Greenbergian typological research represents, as it were, the cross-linguistic counterpart to the ‘orderly heterogeneity’ of sociolinguistic variation within a single language (Weinreich et al. 1968). The basic idea is that grammatical coding is subject to vast, but structured (nonrandom) variation across the world’s languages, and typologists, just like most sociolinguists, take seriously the possibility that patterns of variation in the data can be due to extralinguistic factors rather than to innately specified grammatical knowledge. The comparison with sociolinguistics (and other ‘behavioural’ sciences more generally) may also be justified in light of more recent developments in linguistic typology. As Bickel (2007) points out, modern typology has become a full-grown distribution-oriented discipline. On this view, the cross-linguistic distribution of a structural variable can thus be understood from several revealing perspectives.1 Firstly, a typological distribution is a geographical object, with non-accidental skewings in space. In the form of macro-areas (Dryer 1989), Sprachbund, enclave and spread zone effects (Nichols 1992, Bickel and Nichols 2006), or large-scale geographical mappings of linguistic variables (e.g. in The World Atlas of Language Structures [Haspelmath et al. 2005]), areal patterns have gained new prominence in typological research. In the analysis of purpose clauses, therefore, we will not only seek to control for areal confounds in the sampling process, but also, once the sample is established, consider the distribution of particular coding strategies and construction types in geographical space. Secondly, a typological distribution is a historical object, which may show significant skewings in time. While the present study does not address the historical stability or ‘accidentality’ (Daniel 2007: 71) in the distribution of purpose clauses, thus effectively assuming a stationary distribution, it is cognizant of the fact that universal trends in grammatical coding are in essence diachronically transmitted, 1.  The following conceptual outline of typological distributions is based on Bickel (2006a).

7

8

A Typology of Purpose Clauses

i.e. brought about by pressures on diachronic change (cf. also Bickel 2008a). We will thus devote an entire chapter to the historical pathways of purpose clauses and how they impinge on the synchronic distribution of related construction types, such as complement clauses. Finally, a typological distribution is a statistical object. Just as language-internal (i.e. sociolinguistic) variation can be submitted to precise quantitative analysis and statistical evaluation (e.g. Séguy’s [1973] groundbreaking work on measuring dialectal differences), so too do typological data lend themselves well to rigorous empirical treatment. This solid quantitative underpinning of cross-linguistic research is certainly one of the most vital developments in recent linguistic typology (cf. Cysouw 2005 for an overview) and, at the time of writing this book, still in its infancy. Therefore, while the present study could not fully exploit the array of methods that has only most recently been made available to the typologist (e.g. logistic regression analyses for universal pressures [Bickel 2008a] or multidimensional scaling [Croft and Poole 2008]), it nevertheless incorporates a good deal of quantitative thinking about the typological distribution of purpose clauses. To begin with, the present database of purpose clauses does not only take languages as data points, but shifts the locus of analysis to individual purpose clause constructions (cf. also §3.1.1 below). Each distinct construction is characterized by a particular constellation of structural elements and, for the purpose of cross-linguistic comparison, can be analysed with respect to a multitude of theoretically and descriptively relevant variables (e.g. TAM properties, argument reduction, clause order, etc.). Depending on the quality and quantity of the information available, variables were generally developed on a fine-grained level and operationalized in such a way that their distribution could be submitted to statistical evaluation whenever feasible. It is in this way that it becomes possible to quantify particular coding preferences in the grammar of purpose. Additionally, as will be discussed in more detail below, the present work deviates from much typological research in including rich quantitative data from language-specific analyses of purpose clauses. In particular, it often taps electronic corpora of English to reveal significant patterns of usage of purposive constructions and relates them to the conventionalized grammatical properties found in the typological study of purpose clauses. This procedure is ultimately grounded in the theoretical backdrop to this study, in which universal properties and constraints of morphosyntactic coding are seen as conventionalizations of usage preferences and their ultimate explanation hence as rooted in pressures on language use. The following sections will elaborate on this framework in a more detailed and systematic way.



Chapter 2.  Theoretical and methodological foundations

2.2  A usage-based approach to explaining typological generalizations 2.2.1  Functional-typological accounts of language structure Most typological research does not stop at the descriptive and analytical level of cross-linguistic comparison. The ultimate goal of many comparative studies is to provide a principled explanatory account of the distributional skewings revealed by cross-linguistic generalizations. Whereas generative researchers working on cross-linguistic data have often accounted for variation by recourse to UG-based ‘parameters’, i.e. “mental switches” (Chomsky 2000: 8) that categorically set the value of a linguistic variable to a particular phenotype,2 most ‘Greenbergian’ typologists have preferred a rather different explanatory framework. In their view, similarities in grammatical coding are motivated by universally operative factors relating to language use and to the language user. As Bickel (2007: 240) puts it, “linguistic structures tend to be systematically interrelated among themselves and with other anthropological patterns”, so that we find “close correlations between universal preferences in structure with universal preferences in cognition and communication.” Linguistic diversity, on the other hand, is the result of weighting the multitude of cognitive and communicative pressures in different ways, with each language-specific construction reflecting a conventionalized solution to this weighting problem. The appeal to extralinguistic explanatory variables and their competition in actual usage is the essence of what has come to be known as the ‘functional-typological approach’ or functionalism more generally (cf. Givón 1990 or Van Valin 2001 for overviews). A popular trend within the functionalist literature, which will also be adopted in this book, is its evolutionary, Darwinian orientation: Language structures are often seen as adaptations to complex environmental pressures on the way language is used in communicative interaction, and on the way it is accessed and processed in speech production and comprehension. Along these lines, Deacon (1997) has likened the emergence of cross-linguistic structural similarities to the convergent evolution of analogous forms in unrelated biological species: Just as “the dorsal fins of sharks, ichthyosaurs, and dolphins are independent convergent adaptations of aquatic species” (Deacon 1997: 116), namely to the problem of moving smoothly through water, so do statistical universals of language structure reflect similar, but independently arisen solutions to communicative and cognitive pressures. The relative frequencies of different alignment systems, for example, 2.  While the idea of parameters is still highly relevant to generative accounts of cross-linguistic variation, it has to be acknowledged explicitly that there has been a substantial reconceptualization of parametric explanations within the generative movement. The reader may be referred, for instance, to Newmeyer (2005) or the papers in Biberauer (2008) for recent views.

9

10

A Typology of Purpose Clauses

are jointly motivated by the communicative necessity to discriminate the semantic roles of a and p in transitive clauses, and the economical choice to identify the single participant of intransitive clauses (s) with either a or p. On this account, both nominative-accusative and ergative-absolutive systems are equally well-motivated, but indiscriminate and tripartite systems would be expected to be comparatively rare, which is confirmed by empirical data (e.g. Comrie 2005).3 Similarly, the joint forces of conceptual, discourse-pragmatic and processing pressures can account for the extremely high frequency of languages that place the subject before the direct object, rather than the other way around (cf. Tomlin 1986). Conceptually, language users appear to prefer a linear ordering that adheres to the ‘action-chain’ model (Langacker 1999: 30ff.). On this view, subjects typically encode semantic roles that are ‘energy sources’, whereas prototypical direct objects represent ‘energy sinks’, and the natural flow of energy proceeds from source to sink. In addition, however, it has been shown that subjects typically also encode discourse-old or topical information (e.g. Li and Thompson 1976: 484) and hence would be preferred in clause-initial position, or at least before the focal- or comment-encoding direct object. Finally, Hawkins (2004) argues that object NPs are asymmetrically dependent on their corresponding subject NPs with respect to assigning several syntactic and semantic properties on-line, and hence ideally follow subjects for ease of processing. These relatively simple and well-known examples demonstrate the complex array of potential factors that cause universal structural skewings, and it is the task of the following sections to attempt a more systematic overview of those forces that are particularly relevant to the study of purpose clauses (or complex syntax, for that matter). Let us begin with a factor that is invoked by linguists of many theoretical persuasions, i.e. the idea that structural properties of language are constrained by our cognitive apparatus. As we saw earlier, however, typologists do not generally subscribe to the view that those constraints reside in innate domain-specific principles of grammar (what we called ‘representational innateness’), but in cognitive mechanisms of a domain-general sort. This is what Elman et al. (1996) dub ‘architectural innateness’, which basically refers to the anatomy and physiology of the human (mind and) brain. Perhaps the best-known architectural constraint arises from the linearity of speech: When language is produced, hearers are necessar3.  Comrie’s overview definitely testifies to the extreme rarity of tripartite systems (4∕190 languages = 2.1%). The relatively high frequency of neutral (i.e. indiscriminate) systems (98∕190 = 51.6%) should be taken with a grain of salt, though, since Comrie here only considers overt case marking as a means to establish alignment systems: Especially the neutral class comprises cases in which rigid word order patterns disambiguate syntactic functions. Therefore, at least some languages in this class would be considered nominative or ergative by criteria other than case, reducing the total number of indiscriminate systems.



Chapter 2.  Theoretical and methodological foundations

ily confronted with a linear sequence of linguistic elements to which they need to assign the appropriate structural and conceptual properties that ultimately map onto the communicative message intended by the speaker. On-line pressures of language processing, therefore, have been studied intensively in psycholinguistic research and have also been brought to bear on typological distributions, most cogently on phenomena of constituent order, as in Hawkins (1994, 2004). Accordingly, they will constitute an important explanatory variable in the ordering constraints on purpose clauses. In this domain, they will be shown to join forces with another processing principle: It has variously been argued that on-line processing is facilitated if the experience to be conveyed is similar to its linguistic expression (cf. Clark 1971 for a classic psycholinguistic study). This refers, of course, to the well-known concept of iconicity (Haiman 1980, Givón 1985). If applied to phenomena of constituent order (e.g. affix ordering in Bybee 1985, or clause order in Diessel 2008), the relevant notion is that of diagrammatic iconicity, i.e. a match between the experiential and the verbalized order of the concepts involved. We shall thus examine more closely to what extent diagrammatic iconicity has an impact on the ordering distribution (and other properties) of purpose clauses. Apart from on-line processing considerations, some more overarching architectural constraints arguably affect linguistic structure. Specifically, the relatively young but immensely fast-growing field of cognitive linguistics (cf. Evans and Green 2006 for a comprehensive overview) has developed the idea that language is systematically grounded in human cognition, to the extent that linguistic knowledge is essentially seen as akin to other kinds of knowledge. Language, on this view, is basically auditory or visual information that needs to be perceived, attended to, categorized into meaningful elements, stored in and retrieved from memory, etc. In other words, it is amenable to general cognitive processes also involved in non-linguistic tasks. Croft and Cruse (2004: 45ff.) have devised an extensive list of general cognitive processes recognized in psychology and of their application to language production and comprehension. Let us briefly highlight the notions that will prove relevant to the analysis of purpose clauses. One of the most basic perceptual processes is that of attention, the allocation of cognitive resources to particular aspects of a perceptual event, so that those aspects become the focus of consciousness (cf. Chafe 1994). Importantly, even within this focus, we are selective and ‘detect’ (in Tomlin and Villa’s [1994] terminology) specific parts of information as more relevant to a given task than others. This allows us to highlight or profile different facets of a percept. In language use, this ability is often exploited in such a way that the verbalization of one facet effectively evokes another facet of the same domain, or even comes to stand for the entire domain. Thus we can talk about faces in a crowd even though, of course, we are referring to people of whom faces are just a part. This construal operation

11

12

A Typology of Purpose Clauses

is known as conceptual metonymy in cognitive linguistics (cf. Leite 1994 for its p ­ ervasiveness in discourse). Metonymic construals rely on the hearer’s inference capacity, and it will be argued that such contextually- or metonymically-driven inferences can be crucial in the evolution and diachronic development of grammatical formatives, precisely also purpose markers. A second important set of cognitive operations relates to the fact that the perception of incoming stimuli or information is always guided by prior experience and hence involves what Croft and Cruse (2004: 54) call judgement and comparison. This encompasses all processes of categorization (cf. Murphy 2002), as they rely on judgements of similarity to previous encounters. Category formation by similarity judgements is particularly conspicuous in language when new meanings are accommodated under an already existing form, leading to polysemy in the lexicon or grammar, and when existing grammatical constructions are analogically extended to new contexts (such as the gradual diffusion of the English Progressive to new verb classes, cf. Potter 1969). We will encounter both phenomena in the historical development of purposive constructions. A specific type of analogical extension that has spawned much research in cognitive linguistics is the phenomenon of conceptual metaphor (cf. Lakoff and Johnson 1980 for a classic reference). In this construal operation, a conceptual category or domain is apprehended in terms of another, usually more concrete or physically grounded domain. Thus we can talk of being in love as if this emotional state were a container, i.e. a bounded entity in physical space. Again, metaphors have a key role to play in the extension of grammatical constructions to new, notably more abstract, lexical contexts and thus in the growing productivity of a construction. Therefore, diachronic phenomena such as ‘rule generalization’ or ‘semantic bleaching’ (e.g. Aitchison 2001) usually involve a certain amount of metaphorical conceptualization, and it is in this context that we will encounter and elaborate on the notion of metaphor later on. In addition to those fundamental cognitive constraints and operations, it is the instrumental embedding of language in social interaction that shapes grammatical structure in profound ways. The central idea is that across cultures (and hence languages), human beings “share a great deal of prelinguistic and extralinguistic experience” (Sweetser 1990: 7), which gives rise to certain basic “communicative functions that can be fulfilled by all languages” (Haspelmath et al. 2001: v; cf. also Lee 1988: 211–12 on this point). For example, members of all speech communities undoubtedly experience situations of motion or rest in space (go, stand, sleep), manipulation of objects (catch, eat, carry) or physical transfer (give), which typically involve one, two and three participants, respectively. In other words, there seems to be an experientially motivated and presumably universal inventory (sometimes called a ‘conceptual space’ in typology, cf. Croft 2001: 92–8)



Chapter 2.  Theoretical and methodological foundations

of event-participant constellations. Argument structure constructions in individual languages (e.g. the Absolutive case construction in Epena Pedee) then map onto this universal space in a language-specific way. Figure 1 depicts the universal space of event-participant distinctions, introducing the participant labels that will be used throughout this book. It also exemplifies how language-specific case constructions stretch across this space and thus create their own ‘semantic maps’ (cf. Haspelmath 2003, Croft 2003: ch. 5 for a general synopsis of the semantic map model in typology). Figure 1 illustrates an important point of typological research: Form is highly language-specific, i.e. the choice of bound morphological marking on NPs in this particular way is a conventionalized pattern of the Epena Pedee speech community. Language universals, on the other hand, are generalizations pertaining to how such language-specific forms map onto functions, i.e. they capture which semantic maps recur in similar ways and significantly often across the world’s languages. The value of the semantic map model as a typological tool, then, is that it “allows us to relate functional categories identifiable across languages […] to their syntactic expression within particular languages” (Croft 2001: 97). When we deal with the semantics of purpose markers and constructions, we will thus frequently appeal to such maps. Communicative aspects of language constrain grammar also in more ­specific ways. Some syntactic constructions, for instance, are adaptations to particular

intransitive event

transitive event

ditransitive event A

A S

T P G

S = Participant of a single-participant event A = most agent-like participant of a twoparticipant event P = most patient-like participant of a twoparticipant event T = theme (or patient) of a three-participant event G = goal of a three-participant event

Epena Pedee (ChocÓ: Colombia; Harms 1994: 10–11) wã-hí. Mi akhṍre-Ø wa-hi. my father- go- ‘My father left.’ pa Mi wẽẽ rá-pa khoráa-Ø kha-pá-ri. my woman- basket- weave-- ‘My wife weaves baskets.’

S A

P

Figure 1.  The conceptual space of participant roles and a semantic map for Epena Pedee core cases

13

14

A Typology of Purpose Clauses

ways of information packaging. Passive or cleaving constructions thus typically regulate the flow of information by foregrounding participants that would not normally be given special emphasis in simple active declarative clauses. Similarly, quite a few information-packaging constructions realign the order of constituents in a clause (e.g. inversion, dislocation, extraposition in Huddleston and Pullum’s [2002: 1366] taxonomy) to adjust topical and focal information in a proposition. Along these lines, we will inspect the distribution of purpose clauses in discourse and relate it to their structural properties across languages. Finally, the very nature of human behaviour in communicative interaction leaves its mark on the structure of language. Some well-known functional-typological principles are, in fact, anchored in the classic Gricean maxims of conversational practice. According to the ‘maxim of quantity’, for instance, people tend to make their contributions to the conversation as informative as required, but, crucially, no more informative than necessary (cf. Grice 1975: 45). On this account, communicative behaviour is rational and aims at efficiency (‘least effort’) whenever possible. A major domain of application of this economical behaviour is that of frequently occurring constructions. From a psychological point of view, frequent exposure and practice lead to routinization, entrenchment and predictability (Bybee and Thompson 1997, Haspelmath 2008c). It is, therefore, not surprising that frequent linguistic items are prone to economical coding across the world’s languages. The inversely proportional relation of frequency and overt coding has long been known as Zipf ’s Law (cf. Zipf 1935: 29). It is embodied in Du Bois’ famous slogan that “grammars code best what speakers do most” (Du Bois 1985: 363) and by now supported by a wealth of empirical data, particularly also typological phenomena (cf. Haspelmath 2008d). Since economy is such a pervasive principle of linguistic organization, it will surface repeatedly in the typology of purpose clauses, from the coding properties of individual constructions to efficiency in the processing of complex sentences. Incidentally, frequency effects also have an important theoretical dimension: As they reflect our experience with language, they are central to an approach that anchors typological distributions in language performance and incremental change across usage events. It is this approach that constitutes the final theoretical building block of this book, and thus deserves to be fleshed out more explicitly.

2.2.2  The usage-based nature of language universals What has always been implicit in the functionalist enterprise is the idea that selection pressures operate on individual utterances and the linguistic units embedded in those utterances. In conversational practice, this means that some variants will inevitably be selected more often than others, until they become conventional-



Chapter 2.  Theoretical and methodological foundations

ized in the speech community (cf. Croft 2000 for a detailed exposition, including its sociolinguistic dimension). In other words, structures and constraints emerge, as it were, from language use (Hopper 1987), so that the grammar of a language is but a systematic inventory of ‘frozen’ preference patterns. As Hawkins (2004: 3) puts it, “grammars have conventionalized syntactic structures in proportion to their degree of preference in performance, as evidenced by patterns of selection in corpora and by ease of processing in psycholinguistic experiments.” This is what Hawkins calls the ‘Performance-Grammar-Correspondence Hypothesis’ (henceforth PGCH), which is at the heart of an overarching theoretical approach to language that Langacker (1987) first termed the ‘usage-based model’. Over the last decade, this approach has become an important paradigm for the study of linguistic representations and ontogenetic development,4 but in its emergentist, PGCH orientation, its scope extends well beyond the individual and ultimately becomes highly relevant to typological distributions. This has two immediate consequences for the present study, a thematic and a methodological one. Thematically, a commitment to the usage-based approach inevitably leads us to consider the historical dimension in the synchronic typology of purpose clauses. We noted earlier that typological distributions are historically grown objects, emerging as the cumulative effect of several speech communities choosing well-motivated variants at the expense of alternative ones in language use (cf. Croft 2000, Good 2008, Keller 1994 for particular mechanisms of diachronic convergence). Variants themselves are generated because form-function mappings are constantly negotiated in actual discourse. Occasional or peripheral meanings can be reanalysed as the central one(s) associated with a particular form; forms themselves are often modified for economical or expressive purposes; and languages may ‘acquire’ innovative or alternative form-function mappings in discursive interaction: “When human beings use symbols to communicate with one another, stringing them together into sequences, patterns of use emerge and become consolidated into grammatical constructions.” (Tomasello 2003: 5) The best-known phenomenon in this context is grammaticalization. It can roughly be defined as the diachronic process in which independent lexical items (including demonstratives, cf. Diessel 1999) assume the characteristics of grammatical morphemes, or in which grammatical items and constructions adopt new grammatical functions. The important insight of grammaticalization theory for 4.  On this psychological plane of the usage-based approach, our linguistic competence is exemplar-based, abstracted from the sum total of individual linguistic experiences during our lifetime (Bybee 2006). It is thus built up from and refined by a vast body of input data (Tomasello 2003), and never identical to any other human being’s competence, i.e. the linguistic categories and abstractions we operate with may be slightly or even substantially diverse across different speakers of the same language (Dąbrowska 2008).

15

16

A Typology of Purpose Clauses

typologists is that across languages, there are strikingly similar pathways of change in similar linguistic environments. For example, case markers, tense and aspect constructions, markers of subordination and many more grammatical formatives develop along the following ‘cline of grammaticality’ (cf. Hopper and Traugott 2003: 7):

(4) content item > grammatical word > clitic > inflectional affix

Moreover, this pathway is usually unidirectional and involves a restricted set of typical semantic, morphosyntactic and even phonetic changes recurring in language after language. These processes will be treated in more detail in §4.2, when we ask where purposive markers and constructions come from, and whether they typically develop into certain classes of other markers, i.e. give rise to further grammaticalization. We will see that this historical dimension also provides a valuable clue to understanding the synchronic status of purpose clauses in the system of complex sentences, and their specific relationship to other types of subordinate clauses. The methodological consequence of adopting a usage-based approach derives from the fact that motivations for universal trends in structure are to be sought at the level of individual usage events. Without doubt, those can be studied most fruitfully in rich usage data of specific languages, notably in large electronic collections of authentic data. English is currently unique in being documented in several outstanding corpora, such as the tagged (i.e. word-class annotated) British National Corpus (BNC) or the syntactically parsed International Corpus of English (ICE). This enables us to systematically complement typological material with performance data from English, just as required by the PGCH, and to determine more reliably whether certain grammatical constraints are, in fact, grounded in language use. Needless to say, not every variable in the typology of purpose clauses is amenable to this methodological programme, but whenever sensible and suggestive, the present study will make use of corpus material from English. With some features, we will fortunately be able to fall back on previous studies (e.g. from quantitative discourse analysis), whereas others require a completely novel analysis in their own right. The relevant corpora and methods will be introduced more thoroughly in the respective chapters.

2.2.3  Interim summary At the end of this section, we are now in a position to succinctly characterize the theoretical and methodological framework in which this study is located. In the words of Bickel (2007: 248), [m]odern typology is a discipline that develops variables for capturing crosslinguistic similarities and differences of structures both within and across languages (qualitative



Chapter 2.  Theoretical and methodological foundations

typology), explores clusters and skewings in the distribution of these variables (quantitative typology), and proposes theories that explain the clusters and skewings (theoretical typology).

In this book, the theoretical approach to explaining typological clusters and skewings will be functional in nature, emphasizing the interplay of cognitive and communicative pressures on structural properties of language. In addition, it will be fundamentally usage-based in claiming that typological distributions and generalizations are ultimately anchored in language performance and diachronic shifts or conventionalizations of performance preferences. Given that modern typology is about capturing similarities and differences in language-specific structures, an elementary question is which structures are taken into account. We will deal with this issue in the following section.

2.3  A functional definition of ‘purpose clause’ One of the greatest challenges for any typological study stems from the fact that “[t]he fundamental prerequisite for cross-linguistic comparison is cross-linguistic comparability, that is the ability to identify the same grammatical phenomenon across languages” (Croft 2003: 13). Crucially, because of the arbitrariness of language-specific forms (cf. de Saussure 1922), it is notoriously difficult to define the domain of investigation in purely structural terms. Grammatical systems simply differ too radically from one another to devise a list of necessary and sufficient criteria that embraces the variation space of a particular construction. After all, we are not only interested in capturing the typological prototypes of purpose clauses, but also rarer, more ‘exotic’ construction types that may exhibit grammatical properties outside the realm of a highly restrictive formal definition.5 Additionally, even though some grammatical categories may seem strikingly similar across languages (‘past tense’, ‘passive’, ‘subject’), they are hardly ever identical because the very criteria for establishing those categories in the first place are fundamentally language-specific (cf. Croft 2001 for an elaborate argumentation). ­Consequently, 5.  At first glance, this may seem like an outright objection to the recent movement of autotypology (Bickel and Nichols 2002), in which typological structures and variables are defined on structural rather than conceptual grounds. Crucially, however, the start structure, i.e. the very phenomenon to be investigated, is often also broadly defined in semantic-conceptual terms (“semantic equivalences” in Bickel 2006b) in this approach (without any commitments to universal semantic concepts, though). Once the overall domain is thus delimited, variables for describing structural variation in the domain can, of course, be entirely morphosyntactic and derived bottom-up from language-specific descriptions. This procedure is, in fact, not too dissimilar from the present study. The following outline of how to define purpose clauses cross-linguistically is thus not incompatible with the autotypological agenda.

17

18

A Typology of Purpose Clauses

cross-linguistic comparison cannot be based on structural criteria with very specific distributions in individual languages. These problems have, of course, long been noticed by typologists. Greenberg (1966: 74) already suggested that “in identifying […] phenomena in languages of differing structure, one is basically employing semantic criteria” (cf. also Kibrik 1985). Since the scope of typological research has widened considerably ever since, now often also comprising variables pertaining to social interaction (e.g. Brown and Levinson 1996 on politeness strategies) and discourse-pragmatic organization (e.g. Bickel 2003 on referential density), it is more correct to speak of functional (rather than strictly semantic) definitions. Moreover, many typological definitions are not purely functional, but contain some formal notions (e.g. Diessel’s [1999] definition of demonstratives, or Pustet’s [1993] identification of copulas across languages). For these reasons, a more comprehensive notion of a cross-linguistic tertium comparationis has emerged in the typological literature. This is what Haspelmath (2008a) calls a ‘comparative concept’, i.e. a linguistic construct specifically created for the purpose of typological research. Crucially, comparative concepts abstract away from language-specific formal categories, and also from language-specific semantic categories, i.e. the particular meaning(s) associated with an individual linguistic unit. Instead, they rely, firstly, on universal conceptual-semantic concepts, “meanings which presumably any human being can conceive of ” (Haspelmath 2008a: 13). As I have tried to demonstrate above, those meanings reside in basic communicative functions, and purposeful, i.e. intentional, action is precisely such a universal functional concept. Searle (2006: 2) argues, in fact, that perception, consciousness and intentionality constitute “biologically primitive […] prelinguistic capacities” that are central to our understanding of human existence.6 It is simply a vital characteristic of human beings that we act, not only instinctively, but largely consciously and intentionally. We direct our actions to achieve certain goals that match our intentions. We can thus define the purpose of a particular action as a “reason formulated in terms of [the] intended outcome” of that action (Jackson 1995: 57). This conceptual-semantic structure of purpose is illustrated in Figure 2. Figure 2 also captures some more specific conceptual properties of purposes. For 6.  It must be emphasized that intentional action is not meant to evoke the highly culture-specific idea of a ‘purposeful’ conception of life, a societal goal- or ambition-orientation that involves, for instance, extensive planning and conceptualizations of time as a valuable, limited resource (cf. Lakoff and Johnson 1980: 8). What it refers to instead is purposes on a much smaller scale, such as those daily ones driven by our instinct of self-preservation (e.g. getting something to eat, or some rest, shelter, etc.). It is those purposes that every human being can arguably conceive of, and that grammaticalized constructions such as purpose clauses typically encode across ­languages.



Chapter 2.  Theoretical and methodological foundations

INTENTION

ACTION

(desired) RESULT

Figure 2.  The conceptual structure of purposes (A similar conceptual structure in terms of semantic primitives is given by Wierzbicka (1998: 183), who notes as the primitive ingredients of purpose ‘want, because and think’.)

one thing, the successive arrows symbolize that purposes are intrinsically futureoriented; that is, intentions give rise to actions which in turn may yield the desired outcome. Importantly, though, there is no necessity for the desired result state to come about: not every intention is successfully realized by action. This will prove to be a crucial characteristic in the coding of purpose clauses. In sum, the central conceptual ingredients of purpose are intentionality, target-directedness, future orientation, and a hypothetical result state. Comparative concepts rely, secondly, on universal formal concepts. Crucially, these are not categories required by any particular restrictive theoretical framework (e.g. X-Bar Theory, LFG, HPSG, etc.), but formal concepts that are themselves well-grounded in communicative functions of language shared across speech communities. One such formal concept, which is highly relevant to our topic, is that of the clause. The clause is often referred to as the pivotal unit of syntactic representation since it can be conceived of as the linguistic instantiation of a proposition, a chunk of conceptual knowledge formed whenever we construe a particular situation in the world and the participants engaged in this stateof-affairs (cf. Payne 1997: 71).7 It has been argued from several perspectives that clauses, by virtue of encoding propositional acts (Searle 1969: 29), should be regarded as the “locus of grammar and interaction” (Thompson and Couper-Kuhlen 2005: 481) and, therefore, taken to be a universal concept in syntax. Because of their central communicative function, clauses even figure as universal formal concepts in syntactic theories that are generally very sceptical of formal notions beyond language-specific categories (e.g. Croft 2001, Van Valin and LaPolla 1997). Going one step further, languages also typically provide their speakers with grammatical constructions that reflect a combination of two or more situations or propositions. In these cases, clauses are linked in specific ways, forming what is conventionally called a complex sentence. In what follows, I will thus assume that 7.  Situation is supposed to be a technical term similar to Langacker’s (1987: ch.7) process, a temporal and relational conceptual unit commonly expressed by verbs. It is a cover term subsuming all aspectual construal classes traditionally referred to as states, activities, achievements and accomplishments (cf. Comrie 1976, Croft to appear [a]).

19

20

A Typology of Purpose Clauses

a complex sentence expresses a specific contingent relationship between (at least) two situations in (at least) two clauses (cf. also Langacker 1991: 417ff., Longacre 1985: 235). In our case, the specific relationship between the two situations is purposive. One of the two clauses, which we will call the matrix clause, encodes the proposition of the ‘action’ part of our conceptual model of purposes (cf. Figure 2 above), while the purpose clause proper encodes the proposition pertaining to the ‘desired result’ state. We have now arrived at the definition of purpose clauses that was already given at the very beginning of this book: Purpose clauses are part of complex sentences which encode that one verbal situation, that of the matrix clause, is performed with the intention of bringing about another situation, that of the purpose clause. Notice that this definition qualifies as being external or functional since it does not include any structural notion that is not itself defined functionally. This definition will guide our identification of purpose clauses in the world’s languages. In fact, given the previous argumentation that intentionality is central to human behaviour, it is not at all surprising that most languages have at least one conventionalized complex construction specifically dedicated to the expression of purpose. These constructions, which are easily spotted in reference grammars, will be considered the prototypical candidates for purpose clauses and hence constitute the body of the database used in this study. In addition, however, there are quite a few constructions that are polysemous or vague with respect to the notion of purpose, and provided that the reference grammar is informative enough about those structures, they will also be taken into account. Notice also that our functional definition of purpose clauses does not restrict us to subordinate clauses in the traditional sense. On the contrary, particularly interesting (albeit marginal) construction types will be found in the domain of coordination, such as strong purposive implicatures of ‘and’-coordination gradually developing into more specialized purpose constructions. Our universally applicable metalanguage for identifying purpose clauses prevents us from glossing over such remarkable phenomena. In the context of defining purpose clauses, it should finally be mentioned that the alternative term ‘final clause’ is also used by many, especially more traditional, grammarians. In order to avoid potential confusion with ‘final’ as a variable for describing linearization and constituent order (e.g. ‘verb-final language’), I will not use this term as a synonym of purpose clauses.

2.4  Language sampling In addition to catering for the cross-linguistic comparability of purpose clauses, we now need to ensure that the languages chosen for the comparison are (a) rep-



Chapter 2.  Theoretical and methodological foundations

resentative of the world’s languages, so as to capture the linguistic diversity of the phenomenon in question, and (b) independent of one another, so as to make both tenable and revealing generalizations. Both criteria are, of course, related and can be met by carefully controlled sampling. In the typological literature, there have been several original and influential proposals for adequate sampling (e.g. Bell 1978, Dryer 1989, Perkins 1989, Rijkhoff et al. 1993, Maslova 2000). Nichols (2004) has recently drawn together the major insights of these approaches in a practically oriented guide to sampling, which is also informed by her own research experience. Nichols classifies the range of samples used in typological studies according to a number of criteria, and in the following I will present my sample with respect to these criteria. The first and probably most important criterion is the chief classificatory principle used to establish the sample. This book is similar to many typological studies in using a genealogically based sample. A  representative subset of the world’s languages is sensitive to the fact that historically closely related languages tend to be similar in their typological behaviour since they have inherited many typologically relevant properties from a common ancestor language. Therefore, genealogical independence is a major criterion for sample design. It is determined on the basis of a well-established and reliable genetic classification of the world’s languages. The best-known classifications are Ruhlen (1991) and the Ethnologue database (latest version Gordon 2005). However, it is widely recognized that Ruhlen follows a ‘lumping’ approach, grouping together vast numbers of languages into a single family. This method is based on Greenberg’s idea of multilateral comparison (cf. Greenberg 2005 for an overview), by which he arrived at extremely large groupings, such as Amerindian (for basically all of Northern, Middle and South America except for three other families) and IndoPacific (for all non-Austronesian languages of Oceania). Greenberg’s and hence Ruhlen’s classification is not uncontroversial, and a number of typologists prefer ‘splitting’ approaches until a genetic relationship between two languages can be reconstructed by means of the comparative method. Nichols uses the term ‘stock’ for the highest level of genetic classification, i.e. the highest node in a genealogical tree “reached by the comparative method and permitting reconstruction” (Nichols 2004: 5, cf. also Nichols 1992: 25 and Bell 1978: 148). Consequently, the number of stocks will be much larger than the number of ‘families’ in the macro-groupings by Greenberg and Ruhlen: Nichols (2004) estimates that presently there are 200–300 reconstructed stocks in the world, and this classification will be adopted here. A genealogically based sample then ideally takes one language per stock, unless a stock is either particularly “old, geographically widespread and not contiguous in distribution, structurally diverse or genetically diverse, i.e. having more than two branches at the root level” (Nichols 2004: 6). For ex-

21

22

A Typology of Purpose Clauses

ample, it would be perfectly legitimate to consider more than one Indo-European language because these fulfil all of the above criteria simultaneously, whereas some small and relatively young stocks do not. For both theoretical and practical reasons (cf. below), it is not possible to consider all stocks in this study; I rather decided in advance on compiling a medium-sized sample of 80 languages (topdown sampling). Another criterion for characterizing the sample is its stratification. It is wellknown that stocks are not distributed evenly across the globe, so a genetically based sample should attempt some areal stratification. The idea is that the number of stocks in geographically independent areas is reflected proportionally in the sample. Dryer (1989, 1992) suggested distinguishing six macro-areas, and ­Nichols (2004) takes up this idea, with some modifications. Her areal stratification is shown in Table 1. Based on the distribution of stocks in these areas, and considering the number of available descriptions, Nichols makes the following suggestions for top-down sampling (Table 2). The sample chosen for the present study follows these recommendations. Calculating the number of languages that should enter an 80-language sample while reflecting the exact proportions of stocks to area, one arrives at the following picture (Table 3). The last column already points to some severe sampling difficulties. Two areas, Oceania and South America, are understudied in the sense that many of their native languages have not yet been documented in a reference grammar. Many North American languages, by contrast, could not be taken into account because we lack detailed information on the topic in question, i.e. complex sentences in general and purpose clauses in particular. Crucially, however, the accuracy of the results is improved if only well-described languages are considered. The availTable 1.  Number of linguistic stocks, by area (Nichols 2004: 10) Macro-areas

Approx. number of stocks

Old World

Africa Eurasia Total

27 23–24 ca. 50

Pacific

Austronesian New Guinea Australia Total

1 60 15 ca. 76

New World

North America Mesoamerica South America Total

ca. 50 12 ca. 50–80 ca. 100–30

World total

ca. 240–70



Chapter 2.  Theoretical and methodological foundations

Table 2.  Recommended numbers of stocks for various sample designs (adapted from Nichols 2004: 10) Area

Africa Eurasia Oceania & NG Australia North America Mesoamerica South America Total

Maximum number of stocks

Welldescribed stocks

Lgs. in a large sample

Lgs. in a medium sample

Prportional (based on the percentage of stocks that each area contains)

28 24 83 30 50 12 80

20 20 30 25 40 6 20

20 20 20 20 20 5 15

10 10 10 10 10 3 7

9 8 27 10 16 4 26

5 4 13 5 8 2 13

307

161

120

60

100

50

Note: Oceania & NG = all Pacific islands plus New Guinea.

ability of comprehensive descriptions thus comes in as a so-called ‘convenience criterion’ for the sampling procedure employed here. In contrast to the Americas, many stocks of the Old World, particularly those in Eurasia and Africa, are known to be very old, geographically wide-spread and typologically diverse. Furthermore, we are fortunate to have excellent descriptions for many languages of these stocks. Therefore, there is some good reason to adjust the original proportions, to the effect that both Africa and Eurasia get a much stronger representation in the sample.8 This will be taken into account when formulating and testing empirical generalizations later on. Wherever multiple lanTable 3.  Ideal representation of stocks in a top-down 80-language sample Area

Africa Eurasia Oceania and New Guinea Australia North America Mesoamerica South America Total

Stocks per area (in %)

Ideal number of languages in the sample

9 8 27 10 16 4 26

7 6 22 8 13 3 21

100

80

Problems

under-documentation insufficient descriptions under-documentation

8.  Notice that if we had adopted Ruhlen’s or Greenberg’s genetic classification, we would have selected much fewer Amerindian languages in the first place. The sample employed here may thus in effect not be too different from traditional typological samples.

23

24

A Typology of Purpose Clauses

Table 4.  Representation of stocks in the present sample Area

No. of languages

Africa Eurasia Oceania and New Guinea Australia North America Mesoamerica South America

14 19 12 10  9  3 13

guages were selected from one stock, I tried to maximize areal dispersion. In this way, structural similarities shared between two languages in close geographical vicinity (so-called Sprachbund effects) can be reduced in advance and hence will not distort the results of the investigation. Finally, as the present study intends to represent the diversity of means by which languages encode purposive relations, multiple languages may be chosen from a stock if they differ in significant or remarkable ways on particular structural variables of purpose clauses. Some Pama-Nyungan languages, for instance, vary with respect to the organization of grammatical relations in purpose clauses, and contribute different types of purposive constructions that would go unnoticed by strictly independent (‘probability’) sampling.9 Therefore, to the extent that the present sample is now stratified both genetically and geographically, and that the degree of internal diversity of large stocks is reflected by picking out more than one language, it may well be called a variety sample. In fact, the sampling procedure roughly conforms to Bakker’s (to appear) characterization of variety sampling, even though its scope of 80 languages is rather modest in comparison to many well-known variety samples (e.g. most of the surveys in The World Atlas of Language Structures, cf. Haspelmath et al. 2005). ­ able  4 presents the macro-areal distribution of the sample languages, while T ­Table 5 provides the sample in full. Drawing on the interactive reference tool of the The World Atlas of Language Structures (Haspelmath et al. 2005), we can depict the sample graphically (see ­Figure 3). It should be noted that although the map contains English as a data point, this is only because English will figure prominently as a ‘control group’ throughout this book. Specifically, given the wealth of available information on English purpose 9.  When this study was close to completion, Bickel (2008b) proposed an innovative, computationally sound procedure for genealogically-based sampling. Specifically, it takes statistically significant structural variation within stocks into account and would, therefore, have been a viable option for the present sample.



Chapter 2.  Theoretical and methodological foundations

Table 5.  The sample Africa

Stock

Family/genus

No. of lgs. Language(s)

Afro-Asiatic

Chadic Cushitic Gur Benue-Congo Atlantic/Cangin Kadugli Nilotic Songhay Central Sudanic Central Khoisan

2 2 1 3 1 1 1 1 1 1

Hdi, Hausa Somali, Wolaytta Supyire Babungo, Kana, Nkore-Kiga Noon Krongo Lango Koyra Chiini Ma’di Nama (Khoekhoe)

Greek Indic Iranian

1 1 1 2 1 1

Modern Greek Punjabi Persian Basque, Korean Hungarian Abkhaz

1 1 1 1 3 1 1 1 1

Lezgian Georgian Turkish Evenki Tibetan, Meithei, Qiang Mandarin Chinese Kolyma Yukaghir Semelai Hmong Njua (Miao)

1

Kannada

4

Martuthunira, Yidiɲ, Ngiyambaa, Duuŋidjawu Kayardild Wardaman Nunggubuyu Wambaya Tiwi Ungarinjin

Niger-Congo Nilo-Saharan

Khoisan Eurasia

Indo-European

Isolates Uralic Northwest Caucasian Nakh-Daghestanian Kartvelian Turkic Tungusic Tibeto-Burman Sino-Tibetan Sinitic Yukaghir Austroasiatic Mon-Khmer Hmong-Mien (Miao-Yao) Dravidian Australia

Pama-Nyungan Tangkic Gunwinyguan West Barkly Tiwian Wororan

Oceania and Austronesian New Guinea

Yangmanic Nunggubuyu

1 1 1 1 1 1

Central Malayo- 1 Polynesian Eastern Malayo- 2 Polynesian Western Malayo- 2 Polynesian Trans-New Guinea Madang 2 Engan 1

Tetun Maori, Boumaa Fijian Tukang Besi, Acehnese Amele, Kobon Kewa

25

26

A Typology of Purpose Clauses

Table 5.  (cont.) Stock

N. America

Mesoam. S. America

Family/genus

No. of lgs. Language(s)

Solomons-East Central Solomons Papuan Lower Sepik-Ramu Lower Sepik West Papuan Border Eskimo-Aleut Inuit Athapaskan Siouan Kiowa-Tanoan Algic

1

Lavukaleve

1 1 1 1 1 1 1 1

Uto-Aztecan

2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2

Yimas Abun Imonda West Greenlandic Slave Lakhota Kiowa Nishnaabemwin (Eastern Ojibwe) Tümpisa Shoshone, Ute Yaqui Jamul Tiipay Tzutujil Lealao Chinantec Chalcatongo Mixtec Paumarí Huallaga Quechua Hixkaryana Wari’ Barasano Sanuma Ndyuka Yagua Pirahā Epena Pedee Mapuche (Mapudungun) Warao, Trumai

Hokan Mayan Oto-Manguean

Numic Cáhitic Yuman

Arauan Quechuan Carib(an) Chapacura-Wanham Tucanoan Yanomam Creole Peba-Yaguan Mura Chocó Araucanian Isolates

Chinantecan Mixtecan

clauses, we will either take English as our point of departure for specific analyses (thereby setting language-particular findings in a cross-linguistic perspective), or draw on English corpus data in order to substantiate those universal trends in purpose clauses whose explanations make reference to language use (e.g. economical coding). In other words, English is not part of the typological sample proper, and will hence not enter the statistical computations performed on this sample.

2.5  Previous work The present study draws on valuable previous research on purpose clause constructions. To start with the language-specific literature, every comprehensive grammar of the better-known and extensively documented languages contains



Chapter 2.  Theoretical and methodological foundations

Figure 3.  The sample

some section on purpose clauses, sometimes with excellent discussions (see, for example Huddle­ston and Pullum 2002: 725–31 for English) or even quantitative data (e.g. van Klinken 1999 on Tetun). In addition, there are a number of papers that deal with (a specific aspect of) purpose clauses in a single language; these will be introduced and discussed at the relevant places in this study. However, the only full-length study of purpose clauses that I am aware of is Jones’ generatively oriented treatise on English purpose clauses (1991). No comparable monographic study seems currently available for any other language. Cross-linguistic, i.e. comparative research on purpose clauses has been fairly limited in scope since it has usually focused on particular aspects that are compared across languages. Cristofaro (2005) examines characteristics of the verb form in purpose clauses, and in her study on subordination systems (Cristofaro 2003), overt participant coding is also taken into consideration. Kazenin (1994) compares the argument marking in coordinate, purpose and relative clauses with regard to patterns of syntactic ergativity. While I was conducting this study, I became aware that Verstraete had just completed an investigation into purpose and reason clauses; thus Verstraete (2008) makes insightful observations on modal marking in purpose clauses, and points to the special status of purpose clauses in subordination systems from the perspective of Systemic Functional Grammar. Finally, Haspelmath (1989) carried out a well-known study in grammaticalization that exploits cross-linguistic data in order to suggest a common historical pathway of morphosyntactic change from purpose clauses to infinitival complements. These major studies, and a number of others in which purpose clauses are discussed yet are not the primary topic of investigation, will be brought to

27

28

A Typology of Purpose Clauses

bear on the present study. The major contribution of this book consists in integrating the diverse set of studies on purpose clauses with a large-scale empirical analysis into a coherent account that characterizes universal as well as language-­ particular aspects of purpose clauses and thereby enriches our understanding of the ­phenomenon.

chapter 3

The grammar of purpose Documentation and explanation

3.1  Preliminaries This chapter will map out the space of morphosyntactic variation in the encoding of purposive relations across the world’s languages. Its first major goal is to provide a comprehensive documentation of both coding and behavioural properties of purpose clauses. To this end, each section of the present chapter will be devoted to one major aspect of the morphosyntax of purpose clauses. We will start by investigating the basic syntactic structure of purpose clauses, as manifested by the characteristics of verb form and argument structure (§3.2 and §3.3 respectively). Afterwards, a detailed description will be given of the markers that lend a clause its purposive semantics (§3.4). These comprise typical purposive markers such as specific conjunctions or affixes, but also additional features in the clause that contribute to evoking a purposive interpretation. Two further sections will then deal with properties concerning purpose clauses as a whole, i.e. their position vis-à-vis the associated matrix clause (§3.5) and selected semantic and pragmatic idiosyncrasies of purposive constructions (§3.6). In the entire discussion, considerably more attention will be paid to those aspects that have not been subject to detailed previous investigations. The second goal of this chapter consists in seeking explanations for cross-linguistically recurring features of purpose clauses, drawing, where appropriate, on the rich apparatus of cognitive-functional explanations introduced in Chapter 2. These will not be dealt with in a separate paragraph, but instead be invoked directly in the discussion of each morphosyntactic feature. In this way, I attempt a synthesis of documentation and explanation for each ­section. Before we embark on the actual analysis, however, two preliminary comments are necessary. In particular, we need to stake out purposive constructions as our primary object of investigation, and introduce some statistical concepts for the quantitative analysis of constructions.

3.1.1  On the identification of purpose clause constructions As specified by our definition, purpose clauses are complex morphosyntactic con-

30

A Typology of Purpose Clauses

structions for the expression of purposive relations between two states-of-affairs in the world. Crucially, languages may develop more than one such construction in their inventory of complex sentences and each language in our sample may, consequently, contribute more than one purpose clause for analysis. But how can we identify constructions in the first place? Following a recent movement in syntactic theory, a construction is taken to be a complex linguistic sign in the Saussurean sense, a specific constellation of formal features (significans) that is associated or paired directly with a particular semantic or discourse-pragmatic function (significatum). This characterization might be reminiscent of concrete morphemes such as tree, un- or -ed, and in most versions of Construction Grammar (cf. Goldberg 2003 for an overview), these are, in fact, just the smallest and most substantive (i.e. lexically filled) constructions in a language. Syntactic constructions, such as the Present Perfect or the Passive in English, differ from words not in kind but in complexity and schematicity. That is, they are also symbolic units, but their formal pole usually comprises several syntactic elements, many of which are not defined as specific lexical items but as variable slots (e.g. an infinitival verb form, a subject category, a prepositional phrase, etc.). According to Croft (2001: 52), the identification of a construction (for both native speakers in communicative interaction and for the linguistic analyst) is essentially a cognitive problem of categorization (cf. also §2.2.1 above). What makes up a particular construction and differentiates it sufficiently from other constructions in a language is its unique constellation of ‘gestalt features’ (i.e. formal properties) that ultimately maps onto a specific communicative function. In order to express purposive semantics as defined here, for instance, speakers of English can opt for a finite subordinate clause, introduced by the conjunction so that, which follows the associated main clause. Crucially, though, the clause usually also contains a modal auxiliary, so that the construction is sufficiently different from a result clause, a semantically closely related concept that lacks the purposive characterisformal properties of the English finite purpose clause construction We went to the concert early So so that we would get good seats. lexically modal main specific auxiliary verb slot conjunction postposed position in relation to matrix clause

CONSTRUCTION syntactic properties morphological properties phonological properties

semantic properties pragmatic properties discourse-functional properties

FORM symbolic correspondence (link) (CONVENTIONAL) MEANING

Figure 4.  The anatomy of a construction and formal (gestalt) features of English finite p ­ urpose clauses



Chapter 3.  The grammar of purpose

tic of ­hypotheticality. Figure 4 illustrates the theoretical notion of a construction (adapted from Croft 2001: 18) and characterizes an example clause as an instance of the English Finite Purpose Clause construction. It should be obvious that constructions, the gestalt features they contain and the syntactic categories they define can only be language-specific. In Ute, for example, purpose clauses are structurally very different from English, with one construction being characterized by a particular combination of suffixes, i.e. a subordinating or complementation suffix, a future tense suffix with inherent hypothetical orientation, and concatenative possessive marking on the subordinate subject:

(5) Ute (Uto-Aztecan: Colorado; Southern Ute Tribe 1980: 258) Mama-ci 'uwáy maĝá-vaa-ku-'u. woman-poss she.poss feed-fut-sub-him ‘(The man entered the house,) so that the woman would feed him.’

The task of language typology, then, is to sample and compare such language-specific constructions by means of a set of variables applicable to all of them (e.g. position, marker of purpose, argument reduction relative to a main clause construction, etc.). In keeping with this agenda, I scanned the 80 sample languages for the distinct purposive constructions they contain. This procedure yielded a subcorpus of N = 218 constructions, each coming with its own language-specific set of gestalt features.1 We can thus carry out analyses on languages as such, but also on constructions as individual data points. In this introductory section, we will briefly ask two basic questions concerning the overall distribution of those data points. First, we are interested in how many purpose clause constructions are contributed on average by a language, and in whether or not the major geographical macro-areas are comparable with respect to the number of constructions they host. Table 6 provides a succinct quantitative overview of these issues. All macro-areas make respectable and quantitatively comparable contributions to the database (Mesoamerica is exceptional since it is, of course, a much smaller macro-area with fewer data points in the first place). The table also shows that on average, each language has at least two distinct constructions encoding a purposive relation. As will become clear throughout the analysis, separate purpose constructions in a single language are usually associated with different semantic or discourse-pragmatic properties. A second initial question concerns the number of gestalt features that typically characterize purpose clauses across the world’s languages. As was touched 1.  An overview of those constructions and an illustrative excerpt of the analytical data frame can be found on the companion website.

31

32

A Typology of Purpose Clauses

Table 6.  Overview of purposive constructions per language and macro-area Macro-area

Number of languages Number of purpose clause Average no. of PCs in the sample constructions (PCs) per language

Africa Eurasia Australia Oceania North America Mesoamerica South America

14 19 10 12  9  3 13

37 49 30 37 28 4 33

2.6 2.6 3.0 3.1 3.1 1.3 2.5

Total

80

218

2.7

upon above, this deceptively trivial question can be of vital importance from the ­theoretical perspective of Construction Grammarians. On this account, overt coding properties are perceptual stimuli that can serve as ‘cues’, in psycholinguistic terminology, to the categorization of a clause as an instance of a purposive construction. If we analyse all 218 constructions in our sample for the number of overt gestalt features they contain, the following picture emerges (Figure 5).2 In almost half of all purposive constructions (106∕218 = 48.6%), there is only one identifying cue, notably a subordinating morpheme denoting purposive semantics. In §3.4.1, we will deal with the remarkably broad typological variation of this particular element. More often than not, however, the gestalt of purpose constructions is commonly also defined by a joint effort of two (38.1%) or less commonly 120 100 80 60 40 20 0 No. No.ofofcxns. lgs.

one

one or two

two

one, two or three

two or three

three

four

106

13

83

1

5

9

1

Figure 5.  Number of gestalt features in purpose clauses (N = 218) 2.  Since we are primarily interested in overt coding cues here, we disregard, for instance, the position of the purpose clause or implicit arguments (which are rather behavioural properties).



Chapter 3.  The grammar of purpose

three or more markers (4.6%). Apparently, there is some variation with respect to the optional occurrence of features (cf. columns 2, 4 and 5). Yet, as §3.4.2 will uncover, the additional gestalt features occurring in purpose clauses are amazingly uniform across languages, as they fall into a very limited number of distinct types. As a result of this first overview, we can contend that all languages sampled for the present study have at least one, more commonly two or three, syntactic constructions that are conventionally used to encode purposive relations. There is no gross areal bias in the quantitative distribution of purposive constructions, which makes it possible to compare the geographical macro-areas with respect to the parameters in question.

3.1.2  On the quantitative analysis of constructions The organization of the present database into two subcorpora makes it possible to address research questions both at the level of whole languages (‘How many languages have a grammaticalized ‘lest’-construction?’) and at the more fine-grained construction-specific level (‘Which types of markers are used in a finite purpose clause?’). At both levels, I operated with categorically defined parameters, so that potential correlations between individual variables as well as distributional skewings in the data — be they areal or structural — can be detected. Whenever feasible, alleged associations between variables are submitted to statistical analysis. As has been pointed out in the literature, many classic statistical tests are not felicitously applicable to typological samples (cf. Janssen et al. 2006, Perkins 2001: 427, Cysouw 2005: 566f.).3 Following the most recent proposal by Janssen et al. (2006), I will apply non-parametric significance tests throughout the analysis. In particular, the tests employed do not place any tight constraints on the underlying distribution of the data. The Fisher Exact test, for example, is preferred to Pearson’s Chi-squared test for 2×2 tables since it is not contingent on a specific probability distribution and can be applied to expected frequencies smaller than 5. Tables exceeding the 2×2-outline are submitted to randomized Chi-squared testing. In this 3.  Specifically, the inferential logic of parametric tests is violated by the peculiar conditions of the typological sampling procedure: With small families and language isolates, sampling is (nearly) exhaustive, so that in these cases, the sample and the population are identical. As a consequence, random sampling becomes impossible, and the principal aim of parametric tests, i.e. to assess whether a skewing in the sample reflects a skewing in the population, cannot be achieved because “if the sample is identical to the population”, any skewing is “significant by definition and no testing is necessary or required” (Janssen et al. 2006: 422). In other words, the (near-) exhaustive sampling carried out for small families and language isolates influences the probability value of the skewing in the overall sample, to the effect that the skewing will more likely be judged to be significant.

33

34

A Typology of Purpose Clauses

procedure, the observed data are compared to randomly-generated alternative t­ ables with the same margin totals. More technically, the χ²-value of the observed table is evaluated against those of 100,000 alternative tables constructed in a socalled Monte-Carlo simulation (cf. Robert and Casella 2005).4 If only very few tables in this simulation can match or outrank the observed χ²-value, it is highly unlikely that our present distribution has arisen by mere chance. In other words, if our observed table scores amongst the five percent highest χ²-values of all tables that entered the simulation, we can conclude that there is a statistically significant association between the variables under investigation (cf. Janssen et al. 2006 for a full exposition of the method). Whenever randomization tests are carried out in the present analysis, I will report the χ²-value of the table in question, the number of alternative tables that were generated for the statistical evaluation (B = 100,000), and an associated probability p indicating the likelihood of obtaining our data in the chance-based simulation. In addition, an appropriate effect size measure, such as φ (for 2x2 contingency tables) or Cramer’s V (for larger tables), will be provided for all statistical tests. In contrast to ordinary p-values, effect size measures do not vary with the sample size and hence offer a more reliable test for the association of two typological variables (cf. also Cohen 1994, Gries 2006). As a final technical remark, it is worth mentioning that non-parametric permutation statistics such as the Fisher Exact test and randomized Chi-squared testing are not designed to make statistical inferences beyond the observable cases (cf. Janssen et al. 2006: 430). However, a logical inference to the entire population from which the present sample was drawn is perfectly possible since the sample itself was designed to be representative of the population underlying it (cf. §2.4 for justification). That is, on the grounds of carefully controlled sampling, we can argue that the statistically significant skewings and patterns we observe in the data are likely to carry over to the population of languages the sample is supposed to ­represent.

3.2  Coding of the situation: The form of the verb This section will explore the form of the verb in purpose clauses, the parameter that has been investigated most extensively in previous research, notably by Cristofaro (2003, 2005). The variables of interest chiefly refer to what has traditionally been called ‘finiteness’, i.e. overtly coded information that restricts the interpretation of the verb form in terms of time reference, modality and person or number specifications. In the typological literature, it has been pointed out repeatedly that finite4.  The simulations and the significance tests were computed in the open-source statistical software package R (R Development Core Team 2007).



Chapter 3.  The grammar of purpose

ness is a complex scalar or gradient phenomenon rather than a binary distinction (cf. Givón 1991: 853) and is, therefore, not readily applicable to cross-linguistic research (Cristofaro 2007, Koptjevskaja-Tamm 1994: 1247). Some constructions practically resist being characterized in terms of finiteness (e.g. most serial verbs), while many other verb forms are marked for some of the above distinctions only, but not for others, the specific distributions being extremely diverse and difficult to capture by a binary distinction between ‘finite’ and ‘non-finite’ predicates. Therefore, a recent proposal has been to distinguish between semantic finiteness, the condition for an independent interpretation of a clause, and (various degrees of) morphological finiteness, i.e. overt ‘finite’ marking of a verb form (Maas 2004). A similar, by now well-known solution is to replace finiteness by the notion of ‘deranking’ (Stassen 1985). This notion basically rests on a dichotomous distinction between so-called ‘balanced’ verb forms, which are potentially capable of appearing as the verbs of independent declarative clauses, and so-called ‘deranked’ verb forms, which are defective to some degree in that they cannot occur as main verbs in canonical declarative clauses. Deranking can be reconciled with finiteness if the latter is interpreted from a discourse-functional vantage point. Givón (1991: 853) argues that finiteness is “the systematic grammatical means used to express the degree of integration of a clause into its immediate clausal environment.” By means of deranking a verb, the clause in which the verb appears is essentially deprived of its status as an independent proposition (cf. also Givón 1980: 338) and hence needs to be associated with another proposition for its full interpretation. From this perspective, finiteness as described in reference grammars, i.e. the languagespecific set of criteria for coding dependent events, is but a heuristic device for the assessment of the balanced or deranked status of a clause. In the following, we will thus ask whether purpose clauses are characterized cross-linguistically by a rather high or low degree of integration, viz. whether they are prone to deranked or balanced expression (§3.2.1). We will then, in §3.2.2, home in on the best-known variables of traditional finiteness. Specifically, the focus will be on uncovering and explaining the particular phenotypes of tense-aspect-mood (henceforth TAM) marking in purpose clauses.

3.2.1  Finiteness and deranking in purpose clauses When it comes to describing the complex sentence system of a particular language, most field workers or grammarians do actually operate with the notion of finiteness. Although the individual criteria are language-specific and hence differ necessarily from one description to another, they mostly capture our functional interpretation of finiteness and actually correspond to the notion of deranking. In Korean, for instance, finiteness is not defined by traditional features of

35

36

A Typology of Purpose Clauses

TAM or person marking, but by a set of sentence-type and speech-act suffixes that o ­ ccur obligatorily on verbs of independent clauses (cf. Sohn 1994: 53). Subordinate clauses, by virtue of lacking those suffixes, are described as non-finite by Sohn, which precisely amounts to a deranked status of subordinate verb forms. All of the 80 languages in the sample were analysed from this perspective, i.e. the finiteness decisions of the respective authors were evaluated in functional, deranking terms. Before we quantify the cross-linguistic preferences in finiteness, some of the recurrent typological patterns on this parameter shall be illustrated briefly. To begin with, a typical candidate for a fully finite or balanced purpose clause is found in Wari’. It is marked for its own independent time reference and shows overt participant coding:

(6) Wari’ (Chapacura-Wanham: Brazil; Everett and Kern 1997: 97) Mi' xi'-on pain carawa ca cao' wa give 1pl:incl.rl:fut-3sg.m prep:3n animal infl:n.rl:pst eat inf [ma' ta ma'am ca]. [that.prox:hearer infl:rl:fut full 3sg.m ‘We will give him something to eat, so that he will be full.’

The opposite situation arises when there is no independent specification of temporal relations, the time reference of the purpose clause being dependent on that of the matrix clause. This phenomenon, which Comrie (1985: 56) labels ‘relative tense’, is a criterion for deranking in Warao:

(7) Warao (isolate: Venezuela; Romero-Figueroa 1997: 21) [Waba-naka-mioroi] kokotuka ine nahoro-te. [die-neg-purp everything 1sg eat-npst ‘I eat everything in order that I do not die.’

In this example, the purposive event is located temporally in the same ‘time sphere’ (Declerck 1991: 16) as the matrix event, i.e. non-past. This does, of course, not mean that the events are conceived of as simultaneous. The non-past sphere is coarse enough to allow for a successive interpretation of the situations. At any rate, the crucial point here is that this purpose clause construction is different from causal and concessive clauses in Warao, which, just like main verbs, carry their own temporal markers and hence signal independent time reference (‘absolute tense’). Consequently, purpose clauses like the one in (7) must be regarded as non-finite or deranked. The absence of TAM marking is also characteristic (though not a necessary feature) of subordinate verb forms that are inherently deranked because they cannot normally occur in independent main clauses. Specifically, this set of forms comprises the well-known infinitives (cf. [8]) and nominalizations (cf. [9]), but



Chapter 3.  The grammar of purpose

also converbs and participles (which we will illustrate later on in the study).

(8) Hungarian (Uralic: Hungary; Kenesei et al. 1998: 56) Anna elküldte Péter-t [a könyvet olvas-ni]. Anna sent Peter-acc the book-acc read-inf ‘Anna sent Peter to read the book.’



(9) Basque (isolate: Spain, France; Hualde and Ortiz de Urbina 2003: 739) [Zuek hemen sartzeko], lehenago guk irten egin beharko dugu. [you here enter.noml before we.erg leave do must.fut aux ‘In order for you to get in, first we will have to get out.’

In this category, we also find verb forms which are made dependent by the co-occurrence of a special mood marker that cannot normally be used on independent main verbs. Persian purpose clauses are a case in point since their verbs are usually prefixed with the subjunctive inflection, which is reserved for subordinate clauses and hence a marker of deranked status (Mahootian 1997: 247). Notice, finally, that the absence of TAM marking on purposive verbs in some languages, such as Kana, may look deceptively non-finite, resembling ‘bare infinitives’ in others: (10) Kana (Benue-Congo: Nigeria; Ikoro 1996: 279) Dá ló [(lō)k ō-bṵ̀ ̰ὲ lé ŋwíí]. lick salt [purp you-father good child ‘Take enough salt so that you can father a good-looking baby.’ However, we should consider, first, that the main verb does not display any TAM marking either and, secondly, that there is a distinct non-finite (infinitival) construction in Kana used for the expression of purpose: (11) Kana (Ikoro 1996: 280) [lóó à-kp̰à̰ mɛ̅]. N pīè study medicine purp inf-save me ‘Study medicine so as to save me.’ Thus we have reason to classify the purpose clause in (10) as fully balanced. Similarly unmarked verb forms can also be found in other languages (e.g. balanced purpose clauses in Mandarin Chinese). Apart from TAM distinctions, languages may also indicate finiteness by means of verbal agreement (indexation) markers. By implication, a lack of such markers is taken to instantiate a non-finite or deranked construction. Kiowa, for example, has an incorporated purpose construction which lacks pronominal indexation

37

38

A Typology of Purpose Clauses

prefixes. The main verb, by contrast, carries precisely such a prefix: (12) Kiowa (Kiowa-Tanoan: Kansas, Oklahoma, Texas; Watkins 1984: 229) Gyà-kí·+kɔ·tɔ+tòt. 1sg.agt:3sg.obj-meat+buy+send.prf ‘I sent him to buy meat.’ The last and trickiest category is that of serial verbs, in which two verb forms are welded into a complex verbal expression, as in Acehnese: (13) Acehnese (Austronesian: Indonesia; Durie 1985: 196) Ku = jak = seumayang kee dilee u = krueng. 1 go pray 1 now to = river ‘I am going to the river to pray.’ Typical serial verbs are used to conceptualize the two situations involved as being part of one complex event (Payne 1997: 307), and it is the entire complex that can receive TAM or person marking. For this reason, we will isolate serial verbs in the following counts of deranking and code them as n.a. ‘not applicable’. We can now proceed to analysing the typological distribution of deranking in purpose clauses. At the construction-specific level, we can assign each construction the status of balanced, deranked or n.a., the results of which are displayed in Table 7. Table 7.  Verb forms in purpose clauses (N = 218) Verb status

Absolute frequency

Relative frequency (in %)

Balanced Deranked n.a.

76 137 5

34.9 62.8 2.3

Total

218

100.0

As we can see, at the construction-specific level, deranked purpose clauses outweigh balanced purpose clauses considerably. This difference by itself is statistically significant under an exact binomial test (p < .001), with a moderate effect size (g = 0.14).5 Many languages have more than one purposive construction, though. At a higher level, then, we can classify languages according to whether they generally prefer balanced or deranked purpose clauses, i.e. which of the clause types is instantiated 5.  g is an effect size measure for exact binomial tests, which basically captures the difference between an observed proportion π1 (here: the share of deranking languages [137] in the population of deranking and balancing languages [137+76 = 213]) and the proportion under the null hypothesis (π0 = 0.5). According to Cohen (1969: 142), g ≥ 0.15 reflects medium effect sizes, while g ≥ 0.25 reflects large effect sizes. The effect observed here is thus to be interpreted as significant though only moderate in size.



Chapter 3.  The grammar of purpose

Table 8.  Verb forms in purpose clauses (N = 80) Verb status preferences

Absolute frequency

Relative frequency (in %)

Only balanced Only deranked Mixed balanced and derankeda

15 36 29

18.8 45.0 36.2

Total

80

100.0

a

mixed = the language has both balanced and deranked purposive constructions, but not necessarily the same amount of each

by a larger number of constructions. This yields the results shown in Table 8. We can see that languages with ‘only deranked’ purpose clauses are again much more frequent than those which have ‘only balanced’ purpose clauses (p < .01, g = 0.21). That is, the construction-specific findings are basically replicated at the language level. These results can be strengthened by comparison to previous research on a much larger typological sample. Cristofaro (2005) specifically concentrates on deranking in purpose clauses in a sample of 170 languages, i.e. more than twice as many data points as surveyed here. She finds that in 102 of these 170 languages (60.0%), purpose clauses are invariably deranked, as opposed to only 38 (22.4%) languages which have only balanced purpose clauses. This distribution is, of course, highly significant under an exact binomial test and also achieves a stronger effect size (p < .001, g = 0.23). It is documented in WALS and displayed here in Figure 6. The map already hints at the importance of geography as a factor influencing structural distributions. In our database, too, we find that the six macro-­areas

1. Balanced [38] 2. Balanced/deranked [30] 3. Deranked [102]

Figure 6.  Form of the verb in purpose clauses (adapted from Cristofaro 2005)

39

40

A Typology of Purpose Clauses

Table 9.  Geographical distribution of the finiteness status of purpose clauses Macro-area

Purposive constructions

Languages

Balanced

Balanced

Deranked n.a.

Deranked Mixed

Africa Eurasia Oceania and NG Australia North America Mesoamerica South America

16 11 11 17 14 2 5

21 35 19 18 14 2 28

0 3 2 0 0 0 0

4 1 2 2 2 2 2

6 10 3 3 4 1 9

4 8 5 7 3 0 2

Total

76

137

5

15

36

29

we distinguished are not equally skewed for deranking in purpose clauses. As ­Table 9 and the ensuing diagram (Figure 7) show, some areas have a very strong proclivity for deranked verb forms (notably Eurasia and South America), while the skewing in other areas is much less pronounced (notably Africa, Australia and North America). The areal distribution of constructions is significant under a randomized Chi-squared analysis (χ2 = 15.16, df = 6, B = 100,000, p < .05, Cramer’s V = 0.267, discounting all ‘n.a.’ cases). At the level of languages (Fig. 7), the significant areal differences between balancing and deranking are relativized by the fact that most macro-areas host languages with a pronounced tendency towards mixed purpose clauses. This often 10 9 8 7 6 5 4 3 2 1 0 Africa

Eurasia

balanced

Australia

deranked

Oceania and NG mixed

North America

Mesoamerica

South America

Figure 7.  Number of languages with balanced, deranked or mixed purpose clauses, per ­macro‑area



Chapter 3.  The grammar of purpose

happens for functional reasons: On the one hand, different purpose clause constructions may fill discrete ecological niches, as it were, each being recruited for a specific constellation of participants and events. In Noon (Atlantic: Senegal), for instance, deranked (infinitival) purpose clauses are reserved for contexts of directed motion and must share their subject argument with the matrix clause. Balanced purpose clauses are more flexible in those regards and typically cover non-motion and different-subject contexts. We will return to this ecological issue in §3.3 below. On the other hand, the difference between balanced and deranked purpose clauses may also serve specific semantic and pragmatic purposes. In Amele (Trans-New Guinea/Madang: Papua New Guinea), for example, the difference between finite and non-finite purpose constructions lies in the different ‘strength’ of the respective purpose. Speakers of Amele can basically choose between a finite purpose clause overtly marked with a future tense suffix, and an infinitival purpose clause. However, as Roberts (1987: 58) puts it, “[w]hen the verb is in the future tense the notion of ‘purpose’ is more strongly expressed than with […] the infinitive”. Cross-linguistically, such situations appear to arise when the non-finite construction is highly polysemous, functioning simultaneously, for instance, as a strategy for sentential complementation. This is precisely the case in Amele, where the infinitive construction is also used for some kinds of complements, albeit in a slightly different form. In sum, the common co-existence of balancing and deranking of purpose clauses in one and the same language is an adaptive result to specific communicative needs. Taking all distributions into account, we can conclude at this point that (i) deranking is the globally preferred coding strategy for purposive verbs, to a statistically highly significant degree, and that (ii) exceptions to this trend pattern in a geographically meaningful way, with especially Africa and certain parts of the Pacific Rim (North America, East Asia, Oceania) hosting a significant number of languages with only balanced purpose clauses. On a global plane, it appears, therefore, that purpose clauses are susceptible to losing their status as a balanced or independent clause. Since we started out from Givón’s functional idea that deranking reflects the degree of integration of a clause into its superordinate construction, purpose clauses would have to be seen as prone to syntactic integration. Some reference grammars provide explicit evidence for this hypothesis. For infinitival purpose clauses in Lakhota (Siouan: USA), for example, Buechel (1939: 298) observes that “the principal verb and its infinitive are so closely bound together that they practically form a compound verb.” The same applies to the incorporated purpose construction in Kiowa, which we saw above, and to so-called ‘purposive VPs’ in Kewa, an Engan language of the

41

42

A Typology of Purpose Clauses

Trans-New Guinea stock: (14) Kewa (Engan: Papua New Guinea; Franklin 1971: 97) [Ádo-la] pá-lua. [see-purp go-1sg.fut ‘I will go to see it.’ In this tightly packed construction, no material can be inserted in between the main and the purposive verb. Such deranked and thus integrated purpose clauses are even more revealing if they are seen against the systems of adverbial clauses of which they are part. From this perspective, it turns out that in no less than 16 languages in our sample (16∕80 = 20.0%), purpose clauses constitute (or include) the only non-finite adverbial constructions, while all other adverbial clauses are invariably finite. We find this pattern, for instance, in Tiwi, Noon, Tzutujil, Imonda, Tetun, Wolaytta and Ndyuka. The reverse situation, i.e. that purpose clauses can have finite expression, while all other typical adverbial clauses are non-finite, is attested only once in the sample: Hewitt’s (1987: 39) description of Abkhaz (Northwest Caucasian: Georgia) suggests that a recently grammaticalized finite purpose clause is, in fact, the only adverbial clause in the language that can contain a balanced verb form. We will discuss this construction in more detail in Chapter 4. For now, we note it as a crosslinguistically exceptional pattern that yields a classic implicational statement: (15) Adverbial Deranking Hierarchy (Cristofaro 2003: 168) Purpose > Before, After, When > Reality condition, Reason That is, if any of the adverbial relations on this scale is coded by a deranked verb form, then all relations to the left are also likely to be coded by deranked verb forms. In our implicational formula (p → F(q) > F(not-q)) introduced above, F(not-q) for purpose clauses empirically amounts to only one language in our sample, as we have just observed. From a functional point of view, then, purpose clauses are more integrated with the matrix clause than all other adverbial clauses. In the following section, we will investigate more closely how the specifics of TAM configurations in purpose clauses contribute to this phenomenon, or may even be decisive for explaining it.

3.2.2  Tense, aspect and mood in purpose clauses The fact that the majority of purpose clauses come as deranked constructions and are hence often deprived of tense-aspect marking has a fairly straightforward explanation. It is part of the conceptual structure of purposive situations (cf. §2.3



Chapter 3.  The grammar of purpose

above) that purposes are intrinsically future-oriented. In linguistic terms, purpose clauses inherently have what Noonan (1985: 92) calls ‘determined time reference’ in relation to the matrix clause situation. Consequently, there is no strict communicative need to overtly specify the temporal location of the purposive situation. We find here a classic case of economical behaviour rooted in the predictability of information in discourse. Speakers can afford to omit overt temporal information, thus being able to make an economical choice for a shorter, less overtly marked non-finite purpose clause construction (cf. also Jespersen 1924 or Haiman 1983 for the general idea). This motivates the cases in which tense-aspect marking is absent from a purposive construction. As Givón notes, “the more predictable — i.e. continuous, coherent, non-switching — a clausal feature is vis-à-vis its immediate inter-clausal context, the more likely it is to be left unmarked — i.e. less finite” (Givón 1991: 876, emphasis in original). Not every purpose clause dispenses with temporal marking, though. If tenseaspect specifications are overt, they tend to emphasize the temporal posteriority of the purposive situation with respect to the main clause situation. This can even be the case in deranked constructions such as nominalizations. In Ute, for example, a nominalized purpose clause is still marked for future tense: (16) Ute (Uto-Aztecan: Colorado; Southern Ute Tribe 1980: 268) Ta’wá-ci ‘u kaní-naaʁa yųgá-pųgá [tųká-vaa-ci]. man-sbj he house-in enter-rem [eat-fut-noml ‘The man entered the house in order to eat.’ More generally, the predominant TA markers in purpose clauses are from the future (or at least non-past) temporal domain, as evidenced by the frequency counts set out in Table 10. Tense systems with a past versus non-past distinction conflate present and future time reference in a single grammatical marker, so that the occurrence of a purpose clause with ‘non-past’ marking is perfectly harmonious with its inherent future-orientation. Imperfective aspect, similarly, has an indirect link with the future time sphere. As both Comrie (1985: 72) and Chung and TimberTable 10.  Frequent tense-aspect markers in purpose clauses (N = 218) Tense-aspect marker a

Future tense Non-past or present tense Imperfective aspect Prospective, immediate or directed aspect a 

Absolute frequency 24  8  8  3

Notice that the category ‘future tense’ in this frequency count comprises all forms that contain a future marker, including participial constructions that are inherently future-oriented or purposive markers that simultaneously function as future tense formatives (cf. the Abun preposition su-bere, with bere = lexeme ‘future’, ‘later’).

43

44

A Typology of Purpose Clauses

lake (1985: 206) point out, there is a cross-linguistically frequent association between the past time sphere and perfective aspect on the one hand, and between non-past time and imperfective aspect on the other. These ‘natural links’, as Comrie calls them, are due to the logical implication that on-going situations have not been completed yet and hence cannot be presented in their entirety, thus being incompatible with perfective aspect marking; conversely, events in the absolute past are necessarily completed and thus lend themselves well to a perfective presentation. In this light, imperfective aspect can be seen as a means of evoking present or future time reference. In fact, tenseless languages such as Mandarin Chinese rely on these strong implicatures of imperfective aspect for the expression of future meaning. In other words, imperfective aspect marking in purpose clauses underlines or re-emphasizes the inherent futurity of purposive situations. A very similar logic applies to prospective and immediate aspect markers. As illustrative examples of aspect marking, consider the following purpose clauses from Evenki, containing an imperfective marker, and from Korean, highlighting prospective aspect: (17) Evenki (Tungusic: Siberia; Nedjalkov 1997: 52) Girki-vi tala ung-che-tyn [haval-d’a-da:-n]. friend-poss.refl there send-pst-3pl [work-impf-conv-3sg.poss ‘They sent their friend there to work (so that he worked).’ (18) Korean (isolate: Korean; Sohn 1994: 73) Na-nun [ton-ul pe-l mokcek-ulo] kongpwuhay-yo. 1sg.top [money-acc earn-pros purpose-instr study-pol ‘I am studying in order to earn money (‘for the purpose of…’). Finally, there are language-specific idiosyncrasies of tense-aspect marking in purpose clauses that cannot be subsumed under general coding strategies falling out from the conceptual structure of purpose. An interesting phenomenon of this sort is found in Georgian (Kartvelian: Georgia). In this language, the specific choice of tenses in purpose clauses with a past-tense matrix verb depends on the conceptualization of the purpose and its achievement: if the purpose is construed as having been realized in the phase between devising the original plan and the moment of speech, the dependent verb is marked by the Pluperfect Indicative. If, on the other hand, the fulfilment of the purpose is still open at the moment of speech, the Aorist Subjunctive inflection will be used (Hewitt 1987: 23–4). In addition to tense differences in a narrow sense, we can also observe a systematic alternation in mood here, i.e. between indicative versus subjunctive. More generally, overt mood marking in purpose clauses does certainly not come as a great surprise. Purposive situations are inherently modal in a two-fold



Chapter 3.  The grammar of purpose

Table 11.  Frequent modality markers in purpose clauses (N = 218) Modality marker

Absolute frequency

Hypothetical, irrealis or potential mood Desiderative or optative mood Intentional mood Subjunctive

23 15  3 18

way. On the one hand, they are necessarily hypothetical because the outcome of a purposeful action is yet to be achieved, i.e. non-realized at the moment of speech. As Hewitt (1987: 40) aptly puts it, “[a]s the accomplishment of any intention may be foiled as events unfold, it is clearly appropriate if a language should choose to have recourse to a non-factual mood for the representation of purpose.” On the other hand, Palmer (1986: 174) points out that purposes by their very nature contain a desiderative element, since they refer to someone’s intention to realize a certain goal or to make a certain situation obtain in the future. Both of these semantic shades of modality are attested in purpose clauses. Of the 218 purpose constructions in our sample, 67 are clearly marked for some kind of modality. The most frequent semantic types of overt mood manifestations are collected in Table 11.6 Subjunctive forms are often semantically complex in that they may simultaneously convey a notion of potentiality and desire, so these markers, too, can frequently be interpreted as emphasizing the intrinsic modal orientation of purposive situations. Hypothetical markers recurrently turn out to be ‘minimal features’ in distinguishing purpose from result clauses. English finite purpose clauses, for example, usually contain a putative modal verb so as to be distinguished from result clauses (sometimes called consecutive clauses): (19) I put on a jacket so that I wouldn’t be freezing in the Great Hall. [purpose clause] (20) I put on a jacket so that I wasn’t freezing in the Great Hall. [result clause] Whereas the proposition of result clauses is logically entailed, the purposive situation is certainly desired, but has no entailment that it was actually brought about. In other languages, the hypothetical nature of purpose clauses may not be coded by a modal auxiliary, but by specific potential or irrealis mood inflections. Example (21) ­illustrates potential mood marking in Chalcatongo Mixtec, whereas (22) records irrealis mood inflection in Mapudungun: 6.  Notice that the modal markers listed here also comprise those cases in which a purposive morpheme itself also functions as a modal marker in the language, for instance as an irrealis mood morpheme.

45

46

A Typology of Purpose Clauses

(21) Chalcatongo Mixtec (Oto-Manguean: Mexico; Macaulay 1996: 157) Ni-kíí = ø [xa = kéé = ø staà]. compl-come = 3 [purp = eat.pot = 3 tortilla ‘She came to eat.’ (22) Mapudungun [Mapuche] (Araucanian: Chile; Smeets 1989: 268) Ina-ye-ngüma-y-iñ [ta-ñi ñawfü-a-m]. follow-obl-cry-ind-1nsg-pl [the-3.poss get.consolation-irr-noml ‘We cried (with her) so that she would find comfort.’ Instead of an overt irrealis marker, languages may also choose to mark purpose clauses by the absence of a realis marker which is obligatory in other contexts. This happens in Acehnese (Austronesian: Indonesia). In (23), the copula na appears as a marker of actual states and events; hence the clause is interpreted as a habitual present or past event. In (24), by contrast, this realis marker is absent and the situation is interpreted purposively, i.e. indicative of an intention or a future event: (23) Acehnese (Austronesian: Indonesia; Durie 1985: 248) Lôn na = lôn = jak u = keude. I be = 1 = go to = town ‘I go to town habitually.’ / ‘I have gone to town.’ (24) Acehnese (Durie 1985: 248) Lôn lôn = jak u = keude. I 1 = go to = town ‘I am going (to go) to town.’ Another way of saying that the result state of a purposive relation is hypothetical is to claim that it is not presupposed. In contrast to several other adverbial clauses (e.g. when-temporal clauses, cf. Keenan 1971), purpose clauses do not presuppose their propositional content. A standard test for presupposition is negation. Compare the following sentences: ? When Sue finished school, she immediately got an offer for a job, but in fact (25)   she never finished school.

(26) We went to the concert early so that we would get good seats, but in fact we didn’t get good seats (because it had been too crowded already). This standard test suggests that the content of purpose clauses, unlike that of temporal when-clauses, is not presupposed. The specifics of modal marking are sensitive to this characteristic. If a speaker presupposes the content of an adverbial clause to be true, she cannot felicitously question or deny it (cf. [25]), to the e­ ffect that modal attitudes cannot normally be marked in presupposed adverbial



Chapter 3.  The grammar of purpose

clauses.7 In purpose clauses, on the other hand, hypotheticality and non-presupposition are inherent semantic features, and hence their occurrence with modal verbs is perfectly possible. Crucially, however, this type of modal assessment is “not the speaker’s but that of the agent of the main clause”, as Verstraete claims (2008: 772). According to his view, purpose clauses have the special property of describing the content of the mental state of the agent of the associated matrix clause, i.e. they spell out the desire or intention for whose realization the main clause action is carried out. In the grammar of purpose, this property is often reflected in optative, desiderative or intentional mood markers: (27) Lezgian (Nakh-Dagestanian: Azerbaijan, Russia; Haspelmath 1993: 393) [K’wal.i–z ččara kas hax t-awu-raj lahana], [house-dat strange person go.in neg-do-opt purp ada rak agal-na. he.erg door close-aor ‘He locked the door so that no stranger could come into the room.’ (28) (Boumaa) Fijian (Austronesian: Fiji; Dixon 1988: 287) Au ’ana va’a-.levu [me+u bula.bula vina’a]. 1sg eat greatly [des+1sg health good ‘I eat a lot in order to be healthy.’ (29) Lealao Chinantec (Oto-Manguean: Mexico; Rupp 1989: 50) duʔM ʔaL-ʔiL-ié:Li. KaL-h´ęʔHi pst-move.aside.trns.inan.3 purp neg-int-ruin.iii3 ‘He moved it aside so that it would not be ruined.’ In other words, the propositional content of a purpose clause is bound, as it were, to the agent of the matrix clause and not within the immediate control of the speaker.8 It is for this reason that modal judgements in purpose clauses can only be associated with the main clause agent. As argued by Verstraete, the peculiar ­behaviour of purpose clauses in this respect is similar to that of reason (as 7.  Note, for instance, that phrases like when I might, when he may etc. are notoriously absent from adverbial clauses, as documented by their zero-occurrence in corpora like the ICE-GB. A simple text search or a more elaborate fuzzy tree fragment search for adverbial clauses introduced by when and containing modal auxiliary yield telling results. Specifically, the token frequencies are: may = 0, might = 3 (two of which are actually conditional in meaning [when = ̂ if] and one appears in the collocation may well), must = 0, should = 0, ought to = 1. This goes to show that attitudinal marking by modal verbs is extremely infrequent in presupposed when-clauses. 8.  For a similar argument see also Hargreaves (1991), who investigates intentional action in Kathmandu Newari and invokes essentially the same notions as Verstraete in order to account for the peculiarities of the Newari inflectional system.

47

48

A Typology of Purpose Clauses

­ pposed to causal) clauses, constructions of intended endpoint (‘until X would o o ­ ccur’) and some complement clauses. Agent-binding, as he calls it, explains why these constructions have modal organization even though they are subordinate clauses: “[O]ne would not expect marking of attitudes in constructions that do not themselves encode speech acts” (Verstraete 2008: 760),9 but since attitudinal marking is associated with a matrix clause participant rather than the person making the speech act, modals are a licensed and in fact often an essential part of those constructions. In sum, the result state of purposive situations is intended, desired and hypothetical from the perspective of the matrix clause agent. It thus spells out a mental state relation, in Verstraete’s terms, which cannot be presupposed by the speaker. This specific constellation motivates (i) why certain types of modal marking arise naturally from the conceptual structure of purpose clauses, (ii) why purpose clauses differ from other adverbial clauses as to whether they license attitudinal marking to begin with, and (iii) why purpose clauses may exhibit a certain degree of overlap with other agent-binding constructions, notably complement clauses of mental state predicates. This intermediate position of purpose clauses between adverbial modification and clausal complementation will be a recurrent theme in this book. Before we leave the domain of modality, we should also note that there is an interesting structural skewing in our data on mood in purpose clauses. Even though we saw above that deranked verb forms can show overt mood indicators (cf. [22] again), it is far more common for mood marking to occur on balanced verb forms, i.e. those that were classified as balanced on mood-independent grounds (e.g. by the presence of tense-aspect marking or non-reduction of person indexation). Let us call these latter constructions ‘finite’ for the following analysis and divide our database into finite, non-finite and n.a. constructions (similar to Table  7 above, but discounting mood here). It then turns out that overt mood marking is greatly skewed towards finite verb forms: Of the 67 mood markers in our constructional sample, 50 occur in finite constructions. As the spine plot10 in Figure 8 illustrates, this amounts to a significant proportional difference between finite and non-finite purposive verb forms in that finite constructions are overtly marked for mood in 50.0% of all cases, while this only applies to 13.3% of all non-finite constructions. 9.  Just like all other subordinate clauses, typical purposive constructions do not have illocutionary force, i.e. they do not normally constitute independent speech acts. 10.  Spine plots are essentially scaled bar charts, i.e. the heights of the bars display the conditional relative (or proportional) frequencies of overt and absent mood marking, respectively (hence the range between 0.0 and 1.0 on the right-hand side of the diagram). The widths of the bars correspond to the relative frequencies of finite, non-finite and n.a. forms in our sample. The general logic behind spine plots is thus to plot, for two categorical variables x and y, P(y|x) against P(x) (cf. also Hummel 1996). This design will be used frequently in the following analyses.



1.0

Chapter 3.  The grammar of purpose

0.4

0.6

0.8

mood marking

n.a.

non-finite

finite

0.0

0.2

no mood marking

Figure 8.  Spine plot for the association of finiteness and overt mood marking.

The distribution is highly significant under a Fisher exact test (p < .001), with a solid effect size (χ2 = 33.82, df = 2, φ = 0.394). It appears, therefore, that mood marking is more likely to be made overt (to a statistically significant degree) if purpose clauses contain a finite (i.e. tensed or indexed) verb form. This pattern is quite common: Basque, Supyire and Abkhaz, to name but three typical examples, each have multiple purposive constructions. While their non-finite purpose clauses are completely unmarked with respect to TAM, the finite constructions have to carry an overt mood morpheme. As (25)(26) above have demonstrated, this is often necessary for reasons of disambiguation, especially to distinguish purpose from result clauses. In conclusion, this section has investigated typical tense-aspect-mood constellations in purpose clauses. We can now summarize how these patterns help us understand the prevalence for the deranking and hence the tight clausal integration of purposive constructions. First, purpose clauses have predetermined time reference, so that TA information can in principle be left out economically, resulting in deranked verb forms. Second, purpose clauses have an intrinsic modal orientation, for which reason modality marking could either be left out economically (along with TA marking) or receive overt coding by special mood forms (subjunctives, optative or hypothetical mood, some of which are exclusively dependent moods). Both cases often result in deranked verb forms. Third, the phenomenon of agent-binding (Verstraete 2008), i.e. the fact that typical purpose clauses represent a mental state associated with the agent of the main clause, makes ­purposive

49

50

A Typology of Purpose Clauses

­constructions conceptually integrated into the event frame of the superordinate clause. As Cristofaro (2003: 254) puts it, “If a participant of one state of affairs (SoA) has a will or an interest in the realization of another SoA, then the two SoAs are conceptually closer than if there is no element of will or interest.” It is for those reasons that purpose clauses are often deprived of their independent syntactic status. Importantly, this unique constellation of semantic features and functional pressures cannot (or only partly) be found in other adverbial relations, which is reflected in a cross-linguistic implicational hierarchy (Cristofaro 2003). In essence, this scale predicts that if the most commonly distinguished adverbial relations (e.g. temporal, conditional, causal relations) are expressed by deranked verb forms in a language, purpose clauses will also be deranked (cf. [15] above). This hypothesis finds significant empirical support in the present sample. In the following section, we will explore whether similar tendencies and motivations can be discerned for participant coding in purpose clauses.

3.3  Coding of participants: Argument structure This section is concerned with the syntactic arguments that surface in purpose clauses and their associated matrix clauses, coding the participants involved in purposive relations. In contrast to the verbal categories discussed in the preceding section, the argument structure in purpose clauses is not logically predetermined by the conceptual structure of purposes, although we have already seen that typical purposive situations are bound to their matrix event via agent-binding. The discussion will focus on the instantiation of the purposive ‘subject’ first (§3.3.1), before other arguments and more specific constraints are examined in §3.3.2. At the end of the section, in §3.3.3, we will eventually integrate argument-structural and verb-related properties of purpose clauses. Throughout the discussion, we will make use of the participant labels introduced in Chapter 2, i.e. chiefly S, A, P, and G. Those labels basically reflect the fundamental roles that are defined by one-participant, two-participant and threeparticipant situation types, respectively. Since languages differ widely in how they cluster these participant roles for syntactic purposes (e.g. in individual case marking or indexation constructions), I adhered to the more fine-grained participant labels in coding the data. In this way, the grammatical patterns and constraints of purposive constructions can be captured most accurately for quantitative analysis. In discussing typological generalizations, however, it will often prove useful to refer to higher-level entities, i.e. particular clusters of participant roles. Specifically, the notion of ‘subject’ will figure prominently in the analysis. It is mandatory to keep in mind that this term does not intend to convey that there is a cross-



Chapter 3.  The grammar of purpose

linguistically valid formal category of subject that can be identified by the same distributional properties and criteria across languages. On the contrary, it is defined functionally here as the single participant of an intransitive and the agentive participant of a canonical (i.e. non-passivized) transitive clause, thus referring to a notional cluster of the {SA} participant roles. Put differently, the term ‘subject’ captures the typically instigating participant of a situation, no matter whether S and A behave syntactically the same way (e.g. by nominative-accusative coding) or are categorized into different sets (e.g. by ergative-absolutive coding) in individual languages.

3.3.1  The notional subjects of purpose clauses One of the most prominent variables in the typology of complex sentence constructions has been the overt instantiation of arguments (e.g. in Keenan and Comrie’s [1977] seminal work relativization, or more recently Haspelmath’s [2005] examination of the subjects of desiderative complements). In keeping with this line of research, I analysed all of the 218 purposive constructions in our sample with respect to whether they need to express their arguments overtly or rather leave them implicit. In this section, we will concentrate on the notional subject of purpose clauses. In coding the data, a straightforward three-way distinction was made between explicit subjects, implicit subjects and ‘both types’, i.e. individual constructions which allow for both the overt and the covert expression of their subject. Note that ‘implicit’ does not only refer to the absence of full noun phrases or their pronominal counterparts, but also to the lack of indexical markers such as agreement morphemes where we would usually expect them (as in [12] above from Kiowa). I then distinguished further between regularities that were explicitly stated as grammatical rules (e.g. constructions that delete their subject obligatorily) and those that suggest themselves from inspecting the data (we shall call these ‘attested’ patterns). Two constructions could not be analysed along this parameter (a purposive NP in Qiang and a ‘reverted’ purpose construction in Lakhota which expresses the purpose in the matrix clause). In addition, two constructions were briefly described but not illustrated in the reference materials so that no reliable statement can be made about their argument structure. Taking all of these cases into account, the results of the analysis are provided in Table 12. Taking an aggregate view on the data first, it turns out that there is no dramatic difference between implicit and explicit subjects. However, if we just consider the conventionalized preferences, i.e. the subject patterns which form obligatory rules, the difference becomes more pronounced and also achieves marginal statistical significance under an exact binomial test (46 versus 69 cases, p = .04, g = 0.1). Again it is Cristofaro’s (2003) study which can help to set this finding into a more

51

52

A Typology of Purpose Clauses

Table 12.  Subject expression in purpose clauses (N = 218) Purposive subject is left…

Absolute frequency

Aggregate relative frequencies (in %)

Explicit (attested pattern) Explicit (obligatory rule)

47 46

42.7

Implicit (attested pattern) Implicit (default)a Implicit (obligatory rule)

27 4 69

45.9

Both possible N/A No illustration

21 2 2

9.6 0.9 0.9

218

100.0

Total a 

Implicit (default) refers to a construction which can have overt subjects, but does so only very rarely; in other words, the implicit pattern is the dominant one in language use.

revealing perspective. According to her, “there is a tendency for languages not to express arguments in purpose relations only” (Cristofaro 2003: 171). That is, while in other adverbial clauses of a particular language, arguments always need to be overt, they are left implicit in purpose clauses. This leads Cristofaro to establish the following implicational hierarchy: (30) Adverbial Argument Hierarchy (Cristofaro 2003: 173) Purpose > Before, After, When, Reality condition, Reason This hierarchy is strikingly reminiscent of the ‘adverbial deranking hierarchy’ we saw in the previous section; however, it is even more telling here since purpose clauses outrank all other adverbial relations with regard to the lack of overtly expressed arguments. How can we account for this universal tendency? As Cristofaro (2003: 77) points out, the absence of arguments in dependent clauses is commonly due to one of the following factors: (i) the situations coded by the main and the dependent clause respectively share a certain number of common participants;11 (ii) “the information pertaining to the missing argument is easily recoverable or irrelevant in the discourse context” (ibid.); (iii) arguments that logically belong to the dependent clause are associated with the main clause, as in raising constructions, although they do not encode a participant common to both situations. In order to shed light on this issue, I examined our sample with respect to possible co-reference, i.e. participant-sharing patterns. The key issue, then, is which participants of the matrix clause the notional subject of the purpose 11.  Argument omission under co-referentiality conditions is also known as ‘equi-deletion’ (cf. Noonan 1985: 65ff. for the standard application of this term to sentential complements). Another term for the sharing of a certain number of semantic elements between two propositions is ‘interlacing’ (Lehmann 1988: 204).



Chapter 3.  The grammar of purpose

clause is co-referential with. The coding of the data proceeded from the previous participant labels, but it is, of course, very common for co-reference relations to be licensed for clusters of participants, such as {SA}, {SAP}, {PG}, etc. Most of the coreference constraints, therefore, are formulated in terms of such clusters. In addition, two more categories are needed to capture all of the patterns in the data. On the one hand, there are constructions in which the purposive subject can, but need not take up a participant from the matrix clause. Its reference is thus relatively unconstrained. Such constructions will be referred to as ‘mixed’. On the other hand, there are constructions for which the available data only testify to non-coreferentiality, i.e. a novel or entirely different subject participant in the purpose clause. It cannot, of course, be excluded that argument sharing is allowed in those constructions, but in the absence of evidence to that effect, we have to assume disjunct reference and code them as ‘different’. The results of the analysis are displayed in Table 13. The careful reader might wonder why the grand total of all constructions reduces to 209 in this analysis; this is due to the fact that we carry over the ‘NA’ and ‘no illustration’ cases from the previous analysis (N = 4) and add novel ones for which co-reference relations are not illustrated, and some rare constructions which split the purposive subject to allocate different co-reference relations to S and A (we will treat those exceptional cases in more detail below) (N = 5). All of these cases were discarded for the present analysis. The most important information in this table can be summarized as follows. When the purposive subject is obligatorily implicit (N = 69), then it is invariably controlled by a matrix clause participant. This is the matrix subject in the overwhelming number of cases (53∕69 = 76.8%), yielding constructions that are often explicitly called ‘same-subject purpose clauses’ in reference grammars. Infinitival purpose clauses in Turkish provide a typical example (see (31)). Table 13.  Co-reference relations of purposive subjects (N = 218)

SAPG

P

PG

Oblq

Mixed

Different

Total

Total

SAP

Explicit (attested) Explicit (obligatory) Implicit (attested) Implicit (default) Implicit (obligatory) Both possible

Purposive subject co-referential with matrix…

SA

Overtness of purposive subject

14 9 23 3 53 3

3 3 0 1 9 1

3 1 0 0 6 1

2 1 0 0 1 0

0 0 1 0 0 0

1 0 0 0 0 0

14 22 3 0 0 13

10 7 0 0 0 1

47 43 27 4 69 19

105

17

11

4

1

1

52

18

209

53

54

A Typology of Purpose Clauses

(31) Turkish (Turkic: Turkey; Kornfilt 1997: 73) Hasan kitab-ı [[ø san-a ver-mek] için] al-dı. Hasan book-acc [ you-dat give-inf for buy-pst ‘Hasan bought the book in order to give (it) to you.’ Less frequently, the co-reference relation defines {SAP} or even {SAPG} as controllers of obligatorily implicit subjects. In Yimas, for instance, the purposive subject “can never be overtly expressed and is always coreferential with a human core argument of the matrix verb” (Foley 1991: 402). In the following example, it is the matrix P argument that controls the subject of the purpose clause: (32) Yimas (Lower Sepik-Ramu: Papua New Guinea; Foley 1991: 402) ŋaykum m-ŋa-wampak-c-um [ø kumpwi woman.ii.pl nr:dist-1sg.p-throw-prfv-3pl [ child.i.pl taŋkway-cakal-k]. vis-grasp-irr ‘It was the women who sent me to look after the children.’ Implicit argument rules can also carve up the co-reference space in more finegrained ways. The Australian aboriginal language Martuthunira, for example, has grammaticalized three distinct suffixes indicating particular co-reference relations in purpose clauses. Specifically, one inflection is used for co-reference with matrix {SA} participants (‘same-subject inflection’ as in [33a]), another for the cases in which the purposive S is co-referential with the matrix {PG} (as in [33b]), and yet another when the purposive subject is either co-referential with a matrix oblique participant or else introduces a new referent (‘different-subject inflection’ as in [33c]). (33) Martuthunira (Pama-Nyungan: Western Australia; Dench 1995: 252) a. Kayarra kanarri-lha [nganaju nhawu-lu]. two come-pst [1sg.acc [see-purp.ss ‘Two people came to see me.’ b. Ngayu kartungu parla-marta purra-minyji [punga-waa-rru]. [fall-purp.s = p-now 1sg.nom 2sg.acc stone.prop hit-fut ‘I’ll hit you with a stone, so that you fall down.’ c. Nhulaa murla kangku-ø ngurnu-mulyarra kanyara-mulyarra near.you meat take-imp that.obl-all man-all [kampa-lwala wuruma-l.yarra nganarna-a]. [cook-purp.ds do.for-sim 1pl.ex-acc ‘Take that meat to that man so he can cook it for us.’



Chapter 3.  The grammar of purpose

As can be read off Table 13 as well, explicit subjects are distributed very differently. Especially when they are obligatorily explicit (N = 43), they tend towards noncoreferentiality (either requiring or at least licensing reference to a non-matrix participant, 7+22∕43 = 67.4%). A typical explicit-subject context is displayed in (34) from Hixkaryana: (34) Hixkaryana (Carib: Brazil; Derbyshire 1979: 30) Kosohtxemehe, [rahohsira atunano yehtxoho me]. 1sg.treat.myself [not.catch:noml.me fever its.be:noml denoml ‘I treat myself with medicine so that the fever won’t get me.’ In order to statistically evaluate the distribution of overt subject marking in a meaningful way, it is useful to reduce Table 13 to slightly larger categories. Firstly, it is sufficient to distinguish between implicit, explicit and ‘both’ subject types (i.e. to conflate attested and obligatory cases). Secondly, we shall categorize the matrix participant clusters into ‘subjects’, ‘objects’ (including {P} and {PG}), ‘subjects and objects’ (comprising {SAP} and {SAPG}), ‘obliques’, and the already familiar ‘mixed’ and ‘different’ groups. The resulting distribution can be submitted to a randomized Chi-squared analysis, which reveals a highly significant skewing in the data (χ2 = 96.90, df = 10, p < .001, B = 100,000, Cramer’s V = 0.481). The individual contributions to this skewing can be visualized in a so-called association plot (see Figure 9, next page). The display in Figure 9 essentially follows the outline of our original contingency table, i.e. it charts how each of the three purposive subject categories on the left-hand side is associated with the different types of matrix participants along the horizontal axis. Crucially, however, the display does not reflect the actual observed frequencies in each cell, but the difference between the observed and expected frequencies in each cell, i.e. what is called the residual in statistics.12 Boxes raising from the baseline indicate that the observed frequency is higher than expected (positive residuals), falling boxes indicate a sign in the opposite direction (negative residuals). While the size of the boxes is proportional to the residuals and the number of data points (thus also reflecting the original frequency distribution at a glance), the shading of the boxes indicates whether the respective cell 12.  Specifically, association plots capture the Pearson residuals, which are calculated on the basis of the formula ri,j−µˆ i,j ri,j =  √µˆ i,j . r signifies the observed frequency of the feature combination i,j and µˆ is the expected frequency of this combination under the null hypothesis. In the Chi-squared analysis, µˆ is estimated from the row and column totals of the contingency table.

55

A Typology of Purpose Clauses

Both possible

Different

Mixed

Objects Oblq

Subjects

Subjects and objects

Pearson residuals: 4.00

Explicit

2.00

0.00

–2.00 Implicit

56

–4.00 –4.39 p-value = 2.2695e-16

Figure 9.  Association plot for co-reference relations of purposive subjects. Association plots, implemented as assoc in the R package vcd, allow for straightforward visualization of residual analysis (cf. Meyer et al. 2003, 2006). In particular, they indicate how the difference between observed and expected frequencies in each cell (i.e. the respective residual) contributes to the overall χ2-value of a contingency table. The shadings reflect which residual values are individually significant. According to Meyer et al. (2006: 27), residuals are “approximately standard normal”, which implies that residuals between 0 < |x| < 2 are not individually significant (light grey shading), residuals between 2 < |x| < 4 are significant at α = .05 (medium grey shading), and residuals from |x| > 4 are significant at α = .0001(dark grey shading).

makes an individually significant contribution to the overall skewing of the table (cf. the note to Figure 9). We can thus see that the statistically most decisive associations (i.e. the darkest shadings) are found in two cells: (i) To an overwhelming degree, implicit purposive subjects are referentially bound to the subject of the matrix clause, giving rise to the large number of ‘same-subject’ purpose constructions. (ii) Whereas explicit and ‘both-types’ subjects are each positively associated with mixed co-reference, implicit subjects strongly avoid this reference pattern. Not surprisingly, implicit subjects also eschew entirely different referents to a significant degree. Having established now that purposive constructions are considerably skewed in terms of subject choice, we can proceed to a yet higher level and briefly rephrase our findings for the database of languages (N = 80). Here it turns out that all languages in the sample have at least one purposive construction whose subject is (or can be) co-referential with a matrix participant. In the common wording of implicational universals: If a language has a purpose construction whose notional subject introduces a referent not found in the matrix clause, then it also has a purpose clause whose subject argument is co-referential with a participant



Chapter 3.  The grammar of purpose

of the matrix clause. This statement also entails that those languages which do not generally have purposive constructions with implicit subjects do always license co-referentiality even though subjects are explicit. This set of languages includes, for example, Nama (Khoekhoe), Somali, Chalcatongo Mixtec, Lealao Chinantec, Nishnaabemwin, Slave, Abun, Kobon, Mapuche, Trumai and Wari’. Incidentally, this small list of languages also shows that it is predominantly languages from Africa and the Pacific Rim that have only explicit-subject purpose clauses, mirroring almost exactly the distribution of balanced verb forms. But since co-reference in general and subject co-reference in particular have been shown to be significant trends across the entire sample, we are dealing with a genuine universal that needs to be motivated. Table 12 above provides unequivocal data in support of a usage-based explanation. Recall that constructions with obligatorily implicit subjects never show disjunct reference (cf. the values for ‘mixed’ = 0, and ‘different’ = 0 in Table 13). In other words, the referent of the purposive subject is always predictable quite narrowly from the set of matrix clause participants, and in most cases corresponds to the matrix {SA}. As was argued in Chapter 2 and in §3.2 above, highly predictable patterns in discourse tend to get shortened (by the general logic of the Gricean maxim of quantity). Unless they impede successful communication, such economical patterns frequently conventionalize into categorical grammatical constraints, giving rise to synchronic labels such as ‘same-subject purpose clause’ (cf. Haspelmath 1999 for the diachronic rise of the same phenomenon in desiderative complementation). Obviously, this frequency-based explanation would be even more convincing if it were demonstrated that languages with only explicit-subject purpose clauses also exhibit a strong usage asymmetry in favour of subject sharing, but unfortunately I do not have access to exhaustive corpus data from any such language. A helpful piece of information, however, can be found in the description of Lavukaleve (Solomons-East Papuan). In this language, purpose clauses are like other subordinate clauses in usually requiring overt subject and direct object indexation prefixes. Terrill (2003: 439) states, however, that in actual language use, “subject prefixes are rare with purposive verbs” because “the subject of the purposive verb is almost always the same as the subject of the main verb.” Therefore, while the purposive construction normally inherits the property of other subordinate clauses to require subject agreement, it can dispense with it in actual discourse due to the predictable co-reference relation. It is safe to conclude, then, that Zipfian frequency effects lead to the universal skewing in purposive subject ­realization. But what, we may ask, is the ultimate reason for the overwhelming participant-sharing preference? Let us start by recapitulating that one crucial semantic component of purposive relations is that a participant is interested in the realization of the purposive situation. This is what Givón (1991: 528ff.) calls ‘preference’,

57

58

A Typology of Purpose Clauses

i.e. the will or desire of a participant that a certain situation take place. In the parlance of Verstraete’s (2008) theory, there exists a mental-state relation between the agent of the matrix clause and the content of the purpose clause. Crucially, however, the typological investigation has shown that the agent of the matrix event is frequently also the agent of the purpose event, quite often even in situations where the purposive subject is made explicit. There thus appears to be a yet more specific version of ‘agent-binding’, i.e. a profound communicative preference that binds purposive subjects to matrix agents. Some idiosyncratic grammatical phenomena in purpose clauses can, in fact, be interpreted as direct reflections of this specific notion of agent-binding. It is well worth considering these peculiar cases before we go on motivating the concept of agent-binding. The first phenomenon comes from the language-specific literature on purpose clauses. Aissen (1984) discusses a so-called ‘motion-cum-purpose’ construction in Tzotzil, a Mayan language of the Chiapas region in Mexico. The matrix clause in this construction is headed by any of 12 intransitive motion verbs, which govern a purpose clause in subjunctive mood: (35) Tzotzil (Mayan: Mexico; Aissen 1984: 559) Ch-ba chitom] li Xun = e. [s-man-ø incompl-go [a3-buy-subj pig the Juan = def ‘Juan will go buy pigs.’ While the main verb ba ‘go’ is like other lexical verbs in being inflected for aspect, it is idiosyncratic in not cross-referencing its subject (i.e. S-argument) at the end of the sentence (here: Xune). Instead, the subordinate predicate is marked with an ergative cross-referencing prefix (‘A3’), substituting for the absolutive prefix we would normally expect on the matrix verb. The construction is still easily understood by hearers since matrix and purpose clause are interpreted to share their subject argument. In other words, the matrix subject is ‘controlled’ by the subordinate subject in this case. Importantly, however, this peculiar agreement rule does not apply to subjects per se, but typically to agents.13 This becomes evident in constructions that re-align the relationship between syntactic functions and thematic roles, such as the passive. An example of a typical Tzotzil Passive construction is given in (36). In this sentence, the indirect object of the corresponding ac13.  Aissen (1994: 676) points out that “the mover […] generally coincides with the individual that would be identified as agent by a theory of thematic roles”, but that there are also (presumably rather rare) counterexamples, as discussed by Haviland (1993). Here we will thus exemplify the prototypical case, which shows agent-sensitive properties. Ultimately, it may turn out, however, that the agreement rule is amenable to pragmatic control (as Haviland [1993: 40] advocates), a phenomenon we will encounter in other languages below. More generally, I am grateful to Volker Gast for illuminating comments on this and other examples in the Tzotzil data.



Chapter 3.  The grammar of purpose

tive clause (‘me’) advances to direct object (as indicated by the special g>p suffix) and is then promoted to subject status by means of the Passive inflection. As a socalled ‘derived-S’ argument (Van Valin and LaPolla 1997: 268), it now takes the absolutive suffix appropriate to this role (‘B1’). The active-clause subject, by contrast, appears in an oblique NP but still carries its A-role indexation prefix (‘a3’): (36) Tzotzil (Aissen 1984: 562) L-i-7ak'-b-at jmoton y-u7un kamikotak. compl-b1-give-g>p-pass my.present a3-by my.friend.pl ‘I was given my present by my friends.’ When this construction is used for expressing the purpose of a motion verb, we would expect the subject of the motion verb to be controlled by the subject of the subordinate clause (in keeping with [35]). Interestingly, this is not what happens. Instead, the matrix subject is controlled by the agent of the purpose clause, regardless of its syntactic status: (37) Tzotzil (Aissen 1984: 562) 7ech' [7ak'-b-at-ik-on jmoton y-u7un kamikotak]. pass give-g>p-pass-subj-b1 my.present a3-by my.friend.pl ‘My friends passed by to give me my present.’ There is thus some evidence that typical ‘controllers’ of indexation in Tzotzil auxiliaries are agents (or instigating participants) in the purposive event. We will return to this construction in the context of syntactic change in Chapter 4. Another phenomenon in support of agent-binding is the systematic restriction of an otherwise unconstrained grammatical construction to {SA}-contexts. In Kannada (Dravidian: India), for example, both the Infinitive and the Gerund constructions can be used for same-subject purpose clauses, with their internal subject being equi-deleted. Interestingly, however, the Gerund does not normally behave this way. In other constructions, it usually contrasts with the Infinitive precisely in terms of overt subject expression: “The [G]erund is an equi-structure only in purpose clauses, whereas the [I]nfinitive is always one” (Sridhar 1990: 74). That is to say, the overt argument potential of the Gerund reduces in purpose clauses to the {SA}-co-reference relation with the matrix clause; the subject of the Gerund is no longer free choice and overtly expressed, but invariably bound to the agent of the main clause. Finally, and most radically perhaps, some languages only have a straightforward grammatical construction for same-subject, i.e. agent-binding, purpose clauses. Other co-reference relations and entirely different subjects simply cannot be expressed by an ordinary purpose clause and hence need a more roundabout paraphrase. Such a ‘constructional gap’ is attested,

59

60

A Typology of Purpose Clauses

Table 14.  Subject choice in Tetun purpose clauses (Van Klinken 1999: 316)

Number of examples Different subject in purpose clause Pause before purpose clause Need of purpose to happen?

hodi construction

atu construction

bat(u) construction

ne’ebé construction

100 0%

62 6%

260 53%

119 72%

36% 6% yes no (unless irrealis)

30% no

53% no

for instance, in Tümpisa Shoshone (Uto-Aztecan: California; Dayley 1989: 355). Such constructional idiosyncrasies, in combination with the previous quantitative analysis, make us confident that the specific version of agent-binding acts as a communicative principle guiding the grammaticalization of argument constraints on purpose clauses. In the literature, several ways of accounting for the same-agent preference have been proposed. One strand of research argues that the agent-bound subject participant of the purpose clause typically benefits from the matrix action. According to Haspelmath (1989: 304), the reason why we talk about same-subject situations more often is precisely that “people are egoistic and act for their own purposes”. A complementary (not contradictory) line of argumentation is concerned with the degree of involvement of the subject in both situations. As Cristofaro (2003: 133) puts it, “A participant’s desires or interests are more likely to be referred to the occurrence of states-of-affairs involving that entity rather than states-of-affairs where that entity has no role.” Therefore, irrespective of the degree to which egocentrism is involved, it is clear that the likelihood of a purpose being achieved is higher if the person having a purpose is directly involved in the realization of that purpose, i.e. if s/he can control the action. In fact, some languages have grammaticalized several purpose clause constructions, each of which is associated with a different degree of control on the part of the subject and, consequently, with a different probability of successful realization. This is nicely illustrated by a comparison of four purposive constructions in the Fehan dialect of Tetun (Austronesian: West Timor); see Table 14. The striking correlation between same-subject choice and the need for the purpose to happen in the hodi construction is explained by Van Klinken (1999: 316) as follows: “One would expect that an actor has greater control over what he or she will do than over what someone else will do, with the result that having the same subject in the two clauses implies a greater degree of control (in general) than if the two clauses have different subjects.” A very similar distinction is made in Hdi. This language has two distinct purposive subordinators ngá and dá, which are in complementary distribution depending on whether the subject exercises control



Chapter 3.  The grammar of purpose

over the realization of the purpose. In the dá construction, for example, control is always implied: (38) Hdi (Chadic: Cameroon, Nigeria; Frajzyngier and Shay 2002: 491) Lá-m-là [dá n`ghá-tà mà tùghwázàk xàdí yá ɓá]. go-in-go [purp look-ref prep hibiscus here dem please ‘Go inside and look at [eat] what’s in this hibiscus here.’ Crucially, “the preposition ngá cannot replace the preposition dá, nor can the preposition dá replace the preposition ngá” (Frajzyngier and Shay 2002: 491) in any context. In sum, the intentions expressed in purpose clauses are most likely to be realized if the action is carried out by the person with whom the intention originated; in this way, s/he can exert a comparatively high degree of control over the realization. Conversely, the implicit subject of purpose clauses is generally assumed to be that participant who is associated with a volitional, agentive and hence controlling semantic role (i.e. A and most S arguments of the matrix) or the one who is most likely to be available for such a role in the discourse-pragmatic context. As Bach (1982: 37) points out, such ‘availability states’ are crucial in identifying the implicit subject when there is more than one possibility, as in the following English sentence: (39) I bought you War and Peace [ ø to read to the children]. Two readings are clearly possible here, one in which the main clause agent ends up doing the reading, and one in which the main clause recipient (the g argument) reads to the children. The latter is certainly the more natural one, and this, Bach argues, is due to the fact that the recipient has War and Peace available to him and is hence much more able to control the realization of the reading process than the main clause agent, who simply bought the book but then handed it over to the recipient. In other words, the implicit subject of a purpose clause is commonly taken to be that entity in the discourse environment which is in a position to exert control and bring about the intended situation (cf. also Farkas 1988 for a more formal account of this phenomenon). This typically coincides with the {SA} argument of the matrix clause, but as we have just seen, it can also be a pragmatically more salient matrix argument, motivating the second most frequent coreference relations for implicit subjects across languages ({SAP} and {SAPG}, cf. Table 13 again). In fact, purpose clauses do not seem to be any different from other implicitsubject constructions in this respect. In particular, they pattern in much the same way as converbal clauses with covert subjects. Based on extensive corpus data from English, Kortmann (1995: 227) proposed a ranking of ‘controller accessibility’ for

61

62

A Typology of Purpose Clauses

the converbal subject. On this hierarchy, too, matrix {SA} arguments outrank {PG} arguments with regard to their availability as subject co-referents, which in turn outrank non-matrix participants. While Kortmann’s notion of control captures essentially the syntactic term, however, the control we invoked for purpose clauses is conceptual in nature, having to do with actual control of a participant over the realization of the intended event. In sum, the prototypical and cross-linguistically most significant scenario in purpose clauses is for a participant to act for his or her own purpose and in such a way that the control over the implementation of the intended event is maximized.

3.3.2  Further constraints on argument realization in purpose clauses After having considered the purposive subject at length, we should also take a look at the realization of non-subject arguments in purpose clauses and concern ourselves with some more specific complications that are not easily captured by the previous analyses. To begin with, some constructions can also leave a non-subject argument implicit. Across the sample, we find a total of 34 constructions with this possibility, and the argument roles concerned form a continuum ranking from {P} over {P, Instrument} to {Instrument, Oblique}, with {P} arguments taking the lion’s share of the data (28∕34 = 82.4%). Those latter constructions are often called ‘object-gap purpose clauses’ (henceforth OPCs), and a suitable illustration is provided by (40) from Mandarin Chinese: (40) Mandarin (Chinese: China; Li and Thompson 1981: 619) Tā gĕi-le wŏ yi wăn tāng [ø hē ø]. 3sg give-prfv I one bowl soup   drink   ‘S/He gave me a bowl of soup to drink.’ OPCs are a well-researched domain in English linguistics, most notably in the generative literature. Jones (1991) offers the most detailed treatment of OPCs. He observes that these are characterized by two properties: (i) The implicit object is ‘obligatorily controlled’ (a term from generative theories of syntax) by an argument of the matrix clause, as in the following examples from English and Ma’di: (41) They brought John along [ø to talk to ø]. (42) Ma’di (Nilo-Saharan; Central Sudanic: Sudan, Uganda; Blackings and Fabb 2003: 203) Búkù ɲí ɲí (là-lέ. book 2sg ben (ton)-read-purp ‘Here is a book for you to read.’



Chapter 3.  The grammar of purpose

(ii) The subject of OPCs, by contrast, need not be controlled by an overt argument, but can be inferred pragmatically, in the way suggested above. Thus in the sentence (43) I brought a Shiraz [ø to enjoy ø with dinner]. there is no logical implication that the purposive subject is strictly co-referential with the matrix subject; instead, it could also be the case that the matrix subject is a teetotaller and brought over the wine for his hosts to enjoy with dinner. Interestingly, if there is an overt subject, it tends to have a somewhat different syntactic status than ordinary purposive subjects: (44) I brought a Shiraz [for John to enjoy ø with dinner]. Jones suggests that the bracket notation in (44) is misleading since the alleged ­subject for John could appear on its own, without the purpose adjunct, and the sentence would not lose its primary communicative value: (45) I brought the Shiraz for John. It seems, then, that the notional subject of an OPC can be construed as a goal by itself, and hence the purpose clause is omissible without jeopardizing the grammaticality of the sentence.14 In formal terms, this would be called ‘non-obligatory control’. A good example from a non-European language is provided in (48): (48) Trumai (isolate: Brazil; Guirardello 1999: 402) Kumaru-k Atawaka dakchї ka t’ak-es Kumaru-erg Atawaka cook caus manioc.bread-dat [iyi ø ma-n ø ahak]. [iyi   eat-3abs   purp ‘Kumaru made Atakawa make manioc breads to eat.’ Guirardello states that (48) is ambiguous with regard to subject control since the absolutive suffix in the purpose clause can potentially index either the ergative or the absolutive argument in the matrix clause. This language-particular agree14.  Notice that in other syntactic constructions with a similar ‘for X’ structure, this is not the case: (46) a. He called for Ed to be sacked. b.   *He called for Ed. (47) a. I arranged for her to go by bus. b.   *I arranged for her. In these cases, the subordinate subject is truly internal to the subordinate clause and hence cannot appear alone (at least not in the sense conveyed by examples [44] and [45], cf. also Huddleston and Pullum 2002: 1203).

63

64

A Typology of Purpose Clauses

Table 15.  Co-reference relations of implicit non-subject arguments in purpose clauses Implicit participant of purpose clause

Implicit participant co-referential with matrix… SA

P

P & Oblq

Oblq

Mixed

Different

Total

P P & Instr Instr & Oblq

2 0 0

22  5  1

1 0 0

1 0 0

1 0 0

1 0 0

28  5  1

Total

2

28

1

1

1

1

34

ment rule thus reflects the conventionalization of non-obligatory subject control in OPCs. A broader typological study of OPCs is, unfortunately, made impossible by the lack of sufficient data. Based on the subsample of 34 languages for which at least one example of a non-subject-gap purpose clause is attested, we can nevertheless try to set Jones’ research on English into a cross-linguistic perspective. In particular, the co-reference relations of implicit non-subject arguments can be e­ stablished from the available data and displayed in the following way (Table 15). As can be seen in Table 15, all three types of non-subject-gap purpose clauses are heavily skewed towards co-reference with the matrix P argument, and they are also almost equally skewed in this direction (χ2 = 1.56, df = 10, B = 100,000, p = .999). In other words, the typologically unmarked case is a structure in which the matrix P controls the implicit purposive P. Exceptions to this trend are very rare. For one thing, an implicit subordinate P argument is hardly ever bound to a higher-ranking matrix argument. Having a matrix-{SA} controller, as in the following example from Kayardild, is thus a rather unusual case: (49) Kayardild (Tangkic: Queensland/Australia; Evans 1995: 510) Nyingka warra-j, [dathin-inja dangka-ntha raa-nyarra-nth ø]. 2sg.nom go-imp [that-comp.obl man-comp.obl spear-lest-comp.obl   ‘You go away lest that man spear (you)!’ A possible reason for the rarity of such constructions may lie in pressures on online processing. For the processing of relative clause constructions, it has been argued that particular argument configurations in main and subordinate clause are comparatively harder to parse than others. MacWhinney and Pléh (1988: 106) maintain that this applies specifically to patterns that involve a shift in perspective. Starting out from the subject of the main clause, the conceptual ‘ground’, as it were, humans seem to disprefer the attachment of a subordinate proposition that introduces a different perspective, i.e. one that takes a new participant as the ground, and then shifts back to the original perspective. On this account, rela-



Chapter 3.  The grammar of purpose

tive clauses like The dog [that the cat bit] chased the monkey are SP patterns, first taking the dog as a reference point (S), then focussing on the cat as a force acting on the dog (thereby construing the latter as a patient P), and finally shifting back to the dog as an active participant in the main clause. Experimental research has demonstrated that such SP patterns are much harder to process than PS, PP and SS perspectives, where fewer or no shifts are involved. If this line of argumentation is on the right track, then we may be allowed to characterize (49) precisely as an example of SP perspectivization: It starts out from the matrix S, only to shift to the subordinate {SA}, which is then presented as acting on a P argument that refers back to the originally perspectivized participant. Needless to say, this analogy to relative clause processing remains entirely speculative in the absence of any psycholinguistic data. A second rarity in OPCs is what is actually excluded by Jones’ idea of obligatory control, i.e. that the implicit P argument in the purpose clause is not controlled by a matrix participant, but has an entirely different subject. We have reason to believe that although infrequent, this pattern is not impossible, as illustrated by Barasano: (50) Barasano (Tucanoan: Colombia; Jones and Jones 1991: 155) [ø ĩa-rã] yʉa. eha-bʉ [ see-an.pl arrive-pst.3 1pl ‘We arrived in order to visit (you).’ Before such patterns are treated as highly exceptional cases that deviate specifically from certain processing preferences in purpose clauses, it should be borne in mind that languages differ tremendously with regard to what Bickel (2003) calls ‘referential density’ on a more general plane. Some languages simply tolerate the sparsity or even absence of overt arguments over considerable stretches of discourse, and so rare patterns in particular constructions need not necessarily fall out from universal pressures on clause structure. Clearly, future research on purpose clauses might wish to accumulate more pertinent data on this issue. As a final question on OPCs, we may ask for the correlation between subject and non-subject gaps in purpose clauses. More precisely, the question is whether the existence of an object-gap in a purposive construction also implies that the corresponding subject is also omitted (as in [41], [43] and [48]). This issue can easily be dealt with if we analyse the 34 relevant constructions for their coding strategy with subjects. This procedure reveals that 21 of those constructions (61.8%) leave their subjects implicit (either obligatorily or as the only attested pattern); 6 constructions (17.6%) leave a choice (i.e. both overt and covert subjects are attested); and 7 constructions (20.6%) need to have an overt subject while having an implicit object. This latter pattern is found in nominalized (as opposed to ­infinitival)

65

66

A Typology of Purpose Clauses

­ urpose clauses in Lango and Turkish, in different-subject purpose clauses in Lezp gian, or finite purpose clauses in Slave. The difference between implicit and explicit subjects is significant under an exact binomial test (p = .01, g = 0.25), so that there is, indeed, an asymmetry such that object gaps often imply subject gaps in purpose clauses. Again, this trend would need to be substantiated (or refuted) by research on a much richer database. Apart from object gaps, there are some other idiosyncratic constraints on argument instantiation in purpose clauses that are worth mentioning. The first is a relatively well-known deviation from the very notion of ‘subject’ as defined so far. We tacitly assumed in the previous analyses that purposive subjects can be characterized as either explicit or equi-deleted. While this assumption worked well for most of the constructions in our sample, there are also languages whose control constructions do not operate on {SA} participants as a set but on each of them differently. In these cases, it is common for equi-deletion to apply to {SP} rather than to {A}, resulting in what Dixon (1994) called ‘syntactic ergativity’. This phenomenon is found in one of the sample languages, namely in Yidiɲ. This language has a so-called ‘coreferential purpose construction’ which defines an ergative pattern of argument deletion. More precisely, the construction needs to feature at least one participant shared between matrix and purpose clause and as a rule, this participant must be in either S or P function in both clauses. Acceptable patterns of argument sharing, then, are SS, SP, PS and PP (although the notion of S sometimes also refers to {SA} in the matrix clause, but never in the purpose clause). If the participant occupies the S or P role in the matrix clause and the A role in the purpose clause, the Antipassive construction has to be used: (51) Yidiɲ (Pama-Nyungan: Queensland/Australia; Dixon 1977: 348) Gana biwi duga [miɲa:gu gunda:j-iŋa:lna]. ptcl stick:knife.abs get.imp [meat.purp cut.antip.com.purp ‘Pick up the Biwi knife to cut the meat with it!’ In the Antipassive construction in (51), the A argument of the purpose clause is transformed into an absolutive (derived-S) argument and hence acquires the behavioural properties of such S arguments. In this case, it can function as the coreferential argument in a purpose clause. Antipassive constructions are thus very similar to the syntactic (‘roundabout’) paraphrases for different-subject contexts in languages that only have a productive same-subject, i.e. {SA}-sharing purpose clause. Interestingly, while {SA}-A and P-A sharing patterns are not sanctioned in Yidiɲ, this does not mean that A cannot contract any co-reference relation with the matrix participants. As long as there is a syntactic ‘pivot’ that adheres to the {SP} co-reference constraint, other participant-sharing relations can be licensed by the construction. As (52) shows, the pivotal element shared between matrix



Chapter 3.  The grammar of purpose

and purpose clause has to be in the S or P function in the purpose clause. Once this basic requirement is met, additional participant sharing operates more freely. Thus in (52), the oblique participant in the matrix clause is co-referential with the deleted A participant of the purpose clause: (52) Yidiɲ (Dixon 1977: 281) ŋayu wiwi:ɟɲu yiɲɟu:nda waguɟanda [ŋaɲaɲ ɟumba:lna]. 1sg.sa give-:j-i-pst this-dat man-dat [1sg.p swive-purp ‘I gave myself to this man, for him to swive ( = make love to) me.’ Another interesting quirk resulting from ergative organization is found in Abkhaz. In this language, one option for coding purposive relations is to choose a masdar. In general, this form patterns ergatively since it can only take indexation prefixes for S and P roles in the clause; the A participant, by contrast, cannot be expressed by verbal agreement but would have to occur in an oblique phrase (cf. Hewitt 1989: 112). In purpose clauses, crucially, the subject of the masdar clause is co-referential with the matrix subject and hence often left implicit, as in the following example: (53) Abkhaz (Northwest Caucasian: Georgia; Hewitt 1987: 37) [Ary a-xàc’a yə-ba-ra-a–z] s-aa-r-štə-yt’. [this art-man his-see-msd-it-purp me-hither-they-send-aor ‘They sent me here to see this man.’ According to the text material available on Abkhaz, the subjects of such transitive purpose clauses (which we would expect as oblique phrases) are always equideleted. Unfortunately, the data on intransitive subjects in masdar constructions are vanishingly rare (cf. their absence in Hewitt’s [2005] text collection); but if, by analogy, they were also economically deleted from the masdar clause,15 we would get an interesting situation because the operation of {SA} equi-deletion would lead to behavioural accusativity in a construction whose agreement marking generally operates on an ergative-absolutive basis.16 In the absence of more pertinent data, however, our argumentation remains tentative. 15.  There is evidence of intransitive S deletion in same-subject complement clauses with masdars (e.g. Hewitt 2005: 259), so it would not come as a surprise if the same pattern were found in same-subject purpose clauses. 16.  This phenomenon, if applicable to Abkhaz, shows that languages may define grammatical relations on a construction-specific basis, so that coding constructions like case marking or indexation can pattern differently from behavioural constructions like conjunction reduction, control or gapping. Kazenin (1994) presented an implicational hierarchy for syntactic ergativity, but this and other typological distributions of argument clustering are currently evaluated against a rich and very fine-grained cross-linguistic database (cf. Witzlack-Makarevich 2007).

67

68

A Typology of Purpose Clauses

A more reliably established split of a very different kind is found in Jamul Tiipay, a Hokan language spoken in the border region between California and Mexico. The language has two major constructions for the expression of purposive relationships. The finite purpose clause can appear with any subject, a­ lthough ­typically there is some co-reference relation with the matrix clause: (54) Jamul Tiipay (Hokan: Baja California; Miller 2001: 260) Tom peyii paa [nya’wach varaax uuniiw-x-ich]. Tom here arrive [we.sbj card.pl compete-irr-purp ‘Tom came here so that we (he and I) could play cards.’ There is an interesting constraint on the matrix clause, though, in that the matrix subject has to be a first or third person participant. If it contains a second-person subject, a completely different construction has to be used, the Purposive Auxiliary construction: (55) Jamul Tiipay (Miller 2001: 274) [Varaax m-uuniiw-x–m–i-ch] me-yiw-me-yu? [cards 2-compete-irr-2-say-ss 2-come-2-be ‘Did you come here to play cards?’ In this construction, the auxiliary -i, historically derived from the lexical verb -‘i ‘say’, encodes purposive meaning. It obligatorily inflects for a second-person subject, whose referent is identical to that of the matrix subject. Such a person split is unique among the purposive constructions investigated in the present study, and it can be demonstrated that this idiosyncratic fact of Jamul Tiipay can be explained in diachronic terms. Therefore, we will return to this example in §4.2. A final observation concerns, not the deletion of arguments in purpose clauses, but their appearance in a different form. Koptjevskaja-Tamm (1993) has shown that it is common for the notional subjects of nominalizations to appear in the possessive form. The essence of any nominalization is to construe a verbal situation as a nominal entity, as a result of which it can inherit some distributional properties of prototypical nominal expressions, i.e. nouns. Previous arguments of the verb then often occupy the ‘argument positions’ of a noun, such as that of possessor. It is thus not surprising to find some nominalized and, to a lesser extent, infinitival purpose clauses with possessive arguments. Such constructions are encountered, for instance, in Abkhaz, Evenki, Ute and Huallaga Quechua. A single example will suffice to illustrate this pattern: (56) Huallaga Quechua (Quechuan: Peru; Weber 1989: 205) Chaya-mu-ra-n [wara-nnin Pillku-ta aywa-na-n-paq]. arrive-afar-pst-3 [tomorrow-sup Pillku-obj go-noml-3poss-purp ‘He arrived to go to Pillku the following day.’



Chapter 3.  The grammar of purpose

This example concludes our survey of the argument-structural properties of purpose clauses. In the final paragraph of this section, we will summarize and systematize our findings.

3.3.3  Argument structure and the ecology of constructional systems The previous paragraphs have examined typical participant constellations in purposive events and their reflections in the argument-structural properties of purpose clauses. Perhaps the single most important concept is that of agent-binding, i.e. the existence of a mental-state relation between the content of the purpose clause and the agent of the main clause, and a more specific version of this concept which binds the {SA} argument of the purpose clause to a controlling participant in the matrix clause. These concepts have received ample qualitative and quantitative support from the constructions in our sample. It was also shown that the most predictable co-reference relation, i.e. the one holding between matrix and purposive subject, is prone to economical coding and often results in implicit subordinate subjects. Such argument reductions lead to ‘interlacing’ (Lehmann 1988), i.e. a high degree of syntactic integration of the purpose clause into the matrix clause. This pattern is, of course, ultimately due to agent-binding: “Semantic integration is higher if the agent of the main SoA can control the realization of the dependent SoA, and if the dependent SoA originates from an act of deliberate manipulation on the part of the agent of the main SoA” (Cristofaro 2003: 118). Since we argued earlier that clausal deranking is an indicator of semantic integration, it is now even more transparent that argument reduction due to conceptual integration (i.e. agent-binding) is an important contributor to the deranked nature of many purpose clause constructions. In fact, the absence of balancing features (such as TAM marking) on the verb and implicit person marking often go hand in hand, yielding a tightly-packed, economical construction for the expression of the default purposive situation, a strongly agent-bound, {SA}-co-referential event. It is those constructions that are most productively grammaticalized across the world’s languages, and in individual languages, they can be complemented by another construction that codes the more unusual, unexpected cases. Those latter constructions are often far less constrained: they do not require same-subjects (often even excluding them), they can (or prefer to) take overt TAM marking, they are not bound to particular matrix predicates (such as motion verbs), etc. Lezgian (Nakh-Daghestanian: Azerbaijan, Russia) can be cited as a prime example of this division of labour:

69

70

A Typology of Purpose Clauses

(57) Lezgian infinitival purpose clause with obligatorily controlled subject (Haspelmath 1993: 392) Axpa čun ğweči wac.u–n qerex.da-l [jal jağ-iz] acuq’-na. then we.abs small river-gen bank-sress [rest hit-inf sit.down-aor ‘Then we sat down at the bank of a small river to rest.’ (58) Lezgian finite purpose clause with different, overt subject and mood marking (Haspelmath 1993: 393) [Wun masa-bur.u-laj usal ta-xu-j luhuz], [you.abs other-noml.pl-srel poor neg-be-opt purp patal č’ugwa-zwa. za wiri zehmet-ar wun I.erg all work-pl you.abs for pull-impf ‘I’m doing all the work for you so that you may not be poorer than others.’ In other words, the constructions parcel out their work in such a way that each of them fills a particular niche, in ecological terms: Tighter morphological, syntactic and lexical co-occurrence constraints on one construction are often compensated for by the other, and there is usually little functional overlap between the constructions, i.e. they are well-adapted to their specific function in the constructional inventory of the language. We are now also in the position to summarize how verb and argument encoding interact to bring about the special status of purpose clauses on adverbial deranking and argument hierarchies. Coding patterns in both domains are sensitive to the convergence of agent-binding (or semantic integration of the two events), predetermined time reference, and a doubly modal orientation (desiderative-­ Table 16.  Cross-linguistic coding of adverbial relations (adapted from Cristofaro 2003: 178) Hierarchical pattern

Morphosyntactic phenomenon Functional factors

Purpose, Before, After, When > Others

lack of T distinctions

Purpose, Before, After > Others

lack of A distinctions

Purpose > Others

lack of person agreement distinctions, lack of arguments

agent-binding: sharing of participants between main and dependent state of affairs (not predetermined), semantic integration, preference

lack of TAM distinctions

semantic integration; preference; the dependent state of affairs is unrealized

case marking/adpositions, possessor coding

semantic integration

predetermination



Chapter 3.  The grammar of purpose

intentional and hypothetical). Among the typical adverbial relations, this constellation of conceptual features and their economical exploitation in language use is unique to purposive events. For this reason, they claim the upper end of the deranking and argument hierarchies. In Cristofaro (2003), this is summarized in a succinct format (see Table 16). With verb encoding and argument structure, we have addressed the most important parameters for the internal syntactic organization of purpose clauses. This structural scaffolding, as it were, needs to be fleshed out with the semantics of purpose, which will play a key role in the next section.

3.4  Constructional gestalt features and the coding of purpose In this section, we will return to our Construction-Grammar perspective on purpose clauses and concern ourselves with the ‘gestalt features’ that conspire to classify a particular clause as an instance of a purposive construction. In psycholinguistic terminology, we are interested in the ‘cues’ (Bates and MacWhinney 1989) that, on the one hand, guide the listener’s categorization processes in parsing a complex sentence construction; on the other hand, those cues are also relevant to the long-term organization of knowledge about grammatical constructions (Croft 2001: 52). In complex sentence constructions such as purpose clauses, an important cue to the overall structure to be parsed lies in clause-linking morphology, precisely because such marking signals that a second proposition needs to be attached to an already established proposition (for more on the notion of ‘attachment’, cf. Hawkins [to appear] and §3.5 below). In many cases, clause-linking markers also incorporate information about the semantic relationship that needs to be established between the two propositions, but this information can, of course, also be expressed solely by other cues in the clause, be distributed over multiple cues, or simply be left unexpressed. Not surprisingly, then, this opens up a considerable space of variation in the construction of purpose clauses across languages. The task of the present subchapter is to dissect this variation space in a systematic way. We will start by homing in on what we may call ‘primary’ gestalt features (§3.4.1). Most purpose clauses contain at least one salient marker that acts as a clause-linking device or as an element that allows for the on-line construction of a complex sentence ‘node’ (Hawkins 2004: 18), i.e. a cue to the biclausal structure. This element can, but need not, simultaneously convey the notion of purpose. This may need some clarification. If we recall the infinitival purpose clause from Lezgian that we just saw in (57), we notice that it practically only contains a verb stem suffixed by the Infinitive morpheme (and overt arguments, if necessary). In the absence of a more overt signal of clause-linking or a more specific semantic relation, the Infinitive

71

72

A Typology of Purpose Clauses

suffix thus functions as a primary gestalt feature that constructs the complex sentence node. In other languages, like Punjabi, the Infinitive inflection on the verb is only a ‘secondary’ gestalt feature since the clause boundary is usually marked by a postposition or conjunction that constructs the complex sentence node and establishes the interpropositional relation of purpose. In this case, then, the infinitival morpheme functions more locally to simply derank the verb in the purpose clause. Conversely, we sometimes find that an infinitive or (more ­frequently) a nominalization marker does the primary construction-and-attachment job, but other gestalt features in the clause help to evoke the purposive semantics of the construction. In Mapudungun, for instance, nominalizing morphology is complemented by irrealis mood inflection in purpose clauses. This and other secondary gestalt features will be the topic of §3.4.2. Finally, we will have to ask ourselves whether the gestalt features surfacing in purpose clauses are jointly sufficient to interpret the construction as purposive (§3.4.3). This question necessarily cuts across the categories in §3.4.1 and §3.4.2 since it may well be that a construction has both a clause-linking morpheme (say, a nominalizer) and a secondary gestalt feature (e.g. a future tense affix), but still cannot be reliably distinguished from other adverbial clauses (it being, for instance, a general sequential clause or a quotative construction). Also, the last section takes a closer look at constructions that come without any overt marking, such as serial verbs or juxtaposed clauses.

3.4.1  Structure and distribution of primary gestalt features We will first examine the primary construction-and-attachment features that characterize purpose clauses across languages (as defined by the criteria above). Our constructional sample (N = 218) yields a total of 225 primary marking strategies since in some cases, a construction can be marked by any of two or three alternative markers while the overall syntactic shape of the construction remains the same. Just like English non-finite purpose clauses can be marked by to, in order to or as to (the difference being associated with register- or style-specific variation rather than with the semantics of the construction), Abkhazian non-finite purpose clauses, for example, can be marked by either -r+c, -ra+ze or ra+ne. For this reason, the number of primary marking strategies slightly outnumbers the totality of constructions. I then grouped the total of 225 markers into semantico-syntactic categories by determining both their morphosyntactic status (e.g. conjunction, affix, adposition, clitic etc.) and their semantic content (e.g. adverbial meaning, nominalizer, TAM marker etc.). This leads to the groupings and quantifications illustrated by Figure 10. From a quantitative point of view, several observations can be made here. The trend that purpose clauses are marked most often by an adverbial conjunction is in



Chapter 3.  The grammar of purpose

14

no marking 1

suprasegmental marking

2

clitic

7

particle or discourse connector 5

TAM marker

4

auxiliary

13

preposition

26

postposition 1

case prefix

14

case suffix 2

participle

5

infinitive

15

nominalizer

55

adverbial verb suffix

61

adverbial conjunction 0

5

10

15

20

25

30

35

40

45

50

55

60

65

Figure 10.  Primary gestalt features in purpose clauses (N = 225)

keeping with Dryer’s (2005a) finding that most of the world’s languages prefer free conjunctions, either clause-initially or clause-finally, to mark adverbial clauses more generally. Notice that the class of so-called ‘particles’ and ‘discourse connectors’ (the terms are adopted from the respective reference grammars) could also be argued to fall into the category of adverbial conjunctions in a more traditional sense, thus making the preference for conjunctions even stronger. What the present count also replicates is Dryer’s observation of a very strong asymmetry among adverbial affixes: just as in Dryer’s corpus, all instances of such morphemes are suffixes here, i.e. there is no adverbial prefix in my database. This suffixing preference carries over to other types of primary markers: if languages employ a case affix for expressing purpose, it will be a suffix rather than a prefix (14: 1); the same holds for nominalizing morphology, where the suffixing-prefixing ratio amounts to 15: 0. Among adpositions, finally, postpositions also outnumber prepositions by 2: 1 (although this may have to do with the large amount of OV languages in the sample, cf. §3.5 below). The strong suffixing preference is wellknown to be found much more widely than just in purpose clauses, and various explanations for this phenomenon have been put forward (see, for instance, Bybee et al. 1994 for a diachronic account and Cutler et al. 1985 for a psycholinguistic explanation). Returning to adpositions for a moment, which account for 17.4% of all primary markers, it has to be noted that it can be difficult to decide whether a particular marker is an adposition or a conjunction. As it turns out in more than a handful of languages, the division is not clear-cut, with many adpositions also

73

74

A Typology of Purpose Clauses

regularly surfacing as conjunctions. English after, before, since and, importantly in our context, to all serve both functions, depending on the larger construction they appear in. In fact, Huddleston (1984: 338–41) is not reluctant to suggest conflating prepositions and conjunctions into a single word class. The reason for these recurrent patterns of syntactic multifunctionality is almost always found in diachronic processes, as it is very common for temporal and especially spatial adpositions to develop into clausal conjunctions (cf. Kortmann 1997: 110f.; see also Chapter 4 below). This appears to have happened in Modern Greek (Indo-European: Greece), where the benefactive preposition gia ‘for’ has become a standard means for introducing purpose clauses when it is followed by the subjunctive marker na, with which it synchronically forms a complex conjunction. Similarly, purpose clauses in Tukang Besi are headed by the preposition ako ‘for’, but since the preposition governs a whole clause, not just an NP, one might argue that ako can actually be analysed as a conjunction: (59) Tukang Besi (Austronesian: Indonesia; Donohue 1999: 409) No-kulia [ako na-j(um)ari guru]. 3rl-study.at.university [purp 3irr-become.s.irr teacher ‘She studied so that she could become a teacher.’ From a qualitative point of view, it will certainly be enlightening to exemplify the primary marking strategies found in purpose clauses. The major and wellknown marking types are fairly straightforward in terms of structure (their semantics will be dealt with in more detail soon) and can thus be illustrated briefly. (60) exemplifies a purpose clause introduced by an adverbial (i.e. purposive) conjunction; (61) highlights a purpose clause with an adposition, and in (62)–(63) subordinating suffixes are employed: (60) Persian (Indo-European: Iran; Mahootian 1997: 39) Eynæk-æm-o zæd-æm [ta be-tun-æm behtær be-bin-æm]. glasses-1sg-obj hit-1sg [purp subj-be.able-1sg better subj-see-1sg ‘I put my glasses on so I could see better.’ (61) Lango (Nilo-Saharan/Nilotic: Uganda; Noonan 1992: 245) Dákô òmàkò gwɛ̂nò [mɛ̂ àcámâ]. woman 3sg.catch.prf chicken [for eat.msd.1sa ‘The woman caught a chicken for me to eat.’ (62) Kolyma Yukaghir (Yukaghir: Siberia; Maslova 2003: 432) Terike-die tāt [šej-re-din] uke-s'. old.woman-dim conn [enter-caus-purp go.out-prfv.intr.3sg ‘The old woman went out in order to bring it in.’



Chapter 3.  The grammar of purpose

The purposive suffix in (62) is called a ‘Supine’ form by Maslova, a term often associated with Latin grammar, where the Supine inflection is used in purpose clauses and some sentential complements with similar semantics (cf. Rubenbauer and Hofmann 1995: 200f.). Sometimes subordinating suffixes with adverbial function also go by the name of ‘converbs’ (cf. Haspelmath and König 1995) and are glossed as such in the reference grammars: (63) Evenki (Tungusic: Siberia; Nedjalkov 1997: 52) [D’ev-de:-n] iri-che-n. [eat-conv-3sg.poss cook-pst-3sg ‘She prepared meals for him to eat.’ In addition to these, some frequently occurring primary markers are nominalizations and infinitives, but since these are inextricably bound to the verb, they have already been discussed in the context of deranked verb forms. What remains to be exemplified, then, are some extraordinary and hence less wide-spread marking patterns. This will take us to the boundary regions of the morphosyntactic variation space. In some purpose clauses, the primary marker is not strictly speaking an adverbial formative, but rather a more general tense-aspect-mood (TAM) morpheme. This is attested for a total of five constructions, i.e. two in Kayardild and one in Hausa, Supyire and Amele, respectively. It is generally the case that the semantics of the TAM marker in question is broadly compatible with the semantics of purpose. Thus in Kayardild, the TAM marker is a potential or desiderative mood suffix, while in Amele it is a relative future tense suffix: (64) Kayardild (Tangkic: Queensland/Australia; Evans 1995: 489) Ki-l-da karna-ja minal-i, [karn-marri-wu rajurri-ju]. they-pl-nom burn-rl scrub-mdl.loc [grass-prvt-mdl.prop walk-pot ‘They are burning off the scrub, so (they) can walk about unimpeded by grass.’ (65) Amele (Trans-New Guinea; Madang: Papua New Guinea; Roberts 1987: 58) Uqa [lotoc oso faj-igi-a bili] wau-g ben 3sg [clothing indf buy-3sg-fut dur stomach-3sg.poss big taw-en. stand-3sg.rem ‘He stood there proudly about to (in order to) buy some clothing.’ A cross-linguistic rarity in the expression of purpose is constituted by participles. It is found, for instance, in Ute. The following example actually contains two

75

76

A Typology of Purpose Clauses

­ urpose clauses. One is of the canonical type, marked by a nominalizing suffix p and a future tense marker. The other, much more infrequent purpose clause occurs sentence-medially and is expressed by a participle: (66) Ute (Uto-Aztecan: Colorado, Nevada; Southern Ute Tribe 1980: 266) Durángo-vwaa-cawí-kwa-i [sivąątu-ci wųnáy-kwa-ʁa], tųká-naró'a-paa-ci. Durango-at-to-go-imm [goat-obj sell-go-ptc eat-buy-fut-noml ‘He is going to Durango to sell a goat, in order to buy food.’ It is probably not an accident that present participles like the one found in Ute are cross-linguistically rather rare in the encoding of purposive relations: Present participles tend to be associated with simultaneous activities; they are commonly found in so-called depictive clauses. This is holds for Ute as well for English, as the following example and its translation demonstrate: (67) (Ute: Southern Ute Tribe 1980: 265) Ta’wà-ci 'u káa-ʁa wųųka-pųgá. man-sbj he sing-ptc work-rem ‘The man worked singing.’ Purposive events, by contrast, necessarily stand in a sequential relationship to the main clause situation, for which reason they usually resist being expressed by present participles. A number of languages exhibit purpose clauses which are formed periphrastically, i.e. in a larger constructional context. Supyire, a Gur language of the NigerCongo stock, has a rich inventory of different purpose clauses, the most complex of which is displayed in the following multi-clausal construction: (68) Supyire (Gur: Mali; Carlson 1994: 586) Pi na wyīge tùrù [bà pi gú m ` -pyì they prog hole.def dig.impf [like they pot fp-do sí lwɔhɔ ta mέ]. subj water get like ‘They are digging the hole in order to get water.’ In this construction, the whole purpose clause consists of an initial potential clause headed by pyi ‘do’, ‘be’, followed by a same-subject subjunctive clause, with the entire structure being framed by a comparison clause with the circum-clausal conjunction bà…mέ. The resulting construction could thus more literally be translated as ‘They are digging a whole as if they were to get some water’, which is certainly a noteworthy way of expressing purpose. In five languages, the purpose clause is marked by a quotative form, e.g. a verb, conjunction or discourse connector indicating direct speech. We will return to



Chapter 3.  The grammar of purpose

this phenomenon at other stages of the investigation, and simply provide an illustration here: (69) Kobon (Trans-New Guinea/Madang: Papua New Guinea; Davies 1981: 38) Yad ñin rɨb-em [hagape au-ag-aŋ a g-em] 1sg hand cut-ss.1sg [blood come-neg-imp3sg quote do-ss.1sg ñimagö wam pɨdöŋ gɨ-pin. hand bind strong do-prf.1sg ‘I cut my hand and bound it tightly so that it would not bleed.’ The construction contains a medial same-subject clause, typical of Papuan languages, which precedes a quotative predicate. In fact, this predicate functions almost like a conjunction (thus being classified as a primary marker), and we shall see in Chapter 4 that it is not at all unusual for utterance verbs to grammaticalize into (purposive) subordinators. Grammaticalization has ultimately also been at work in the evolution of socalled purposive auxiliary constructions. They are attested for five languages (Nkore-Kiga, Yagua, Tetun, Trumai and Jamul Tiipay) and in each case they are only one of a number of available constructions for forming purpose clauses. We have encountered the Purposive Auxiliary in Jamul Tiipay in connection with argument structure splits, so we will recruit the Fehan dialect of Tetun for illustration. While (70a) shows how at(u) is used in its canonical auxiliary function, where it can loosely be translated as ‘would’, (70b) demonstrates its use as a primary gestalt feature of a purposive construction: (70) Fehan Tetun (Austronesian: West Timor; Van Klinken 1999: 221–2) a. At ó bá m-ó na’in sia futu manu!? aux 2sg go 2sg-with noble pl fight(.cocks) bird ‘Would you go and cockfight with nobles?’ b. Sia mai at ukun rai Timor. 3pl come aux rule earth Timor ‘They (the various colonizers) came in order to rule Timor.’ According to the reference materials, at(u) inherently carries irrealis modality and retains of its original lexical meaning the notions ‘be about to’, ‘intend to’ and ‘want to’, or can synchronically simply be translated like a purposive conjunction ‘in order to’. Finally, a really unusual marking pattern is found in a negative purpose (or ‘lest’) construction in Duuŋidjawu. In this construction, ‘lest’ is coded suprasegmentally by simply juxtaposing two clauses while leaving an audible pause between them:

77

78

A Typology of Purpose Clauses

(71) Duuŋidjawu (Pama-Nyungan: Australia; Kite and Wurm 2004: 102) Madja nja-ø (pause) gundu gaye-nge. dem look.out-imp (lest) canoe tip.over-impf ‘Look out there lest the canoe tips over.’ Notice that this marking pattern would not be too outlandish for positive purpose. As we shall see in §3.4.3 below, purposive relations are often inferred from asyndetically coordinated (i.e. juxtaposed) clauses. In such cases, we would expect that the expression of negative purpose features at least an overt negative marker somewhere in the clause. In Duuŋidjawu, however, the reverse is true: Positive purpose is overtly marked by a verbal suffix, whereas negative purpose is coded, not by overt morphological material or any surface indication of negative semantics, but by an intonation break. This pattern is not found in any other language in the sample. Now that the general types of primary markers have been sketched, one may relate them to two typologically important parameters. The first question would be for a correlation between the structural type of marker (conjunction, affix, adposition etc.) and the constituent order type of the languages in which these markers occur. The intuition here is that, as revealed by Bybee et al.’s (1990) study, “grammatical morphemes (e.g., adpositions, clitics, particles, and auxiliaries) following the [verb] stem are more likely to become affixes than are grammatical morphemes preceding the stem” (Harris and Xu 2006: 510). Since the position of such grammatical morphemes is known to be correlated with the basic word order type, we would expect to find a synchronic correlation between word order at the clause-level and the morphological type of grammatical formatives. In order to shed light on this issue, the 225 purpose markers were re-allocated to purely structural (i.e. morphosyntactic) groups since their specific semantic content is not immediately relevant to questions of constituent order. TAM markers, for example, were assigned to affixes or particles, depending on their morphosyntacTable 17.  Types of purpose markers and basic constituent order (N = 225) Constituent-order type Other OV-languages (SOV, OVS, OSV) VO-languages (SVO, VSO, VOS) Total

Structural type of primary marker Conj. Affix

Adp.

Particle Clitic Aux.

None

10 17

25 63

0 25

1 6

0 1

0 0

2 4

38 116

34

8

14

1

1

3

10

71

61

96

39

8

2

3

16

225

Total



Chapter 3.  The grammar of purpose

1.0

tic make-up. By this procedure, the ‘no marking’ group is also slightly enhanced because it now also subsumes suprasegmental marking and bare verb stems (e.g. zero-marked infinitives). We will have much more to say about constituent order typology later on in the book, but for now we just distinguish three major word order types, i.e. verb-object (VO), object-verb (OV) and ‘other’ (flexible or mixed) languages. The distribution of the different structural types of purpose markers over the major word order categories is shown in Table 17. For the three major marking strategies, i.e. conjunctions, affixes and adpositions, the size of the data is sufficient to make some comparative quantitative observations. First, purposive conjunctions are twice as frequent in VO languages as in OV languages (34 :17). Second, purposive affixes are overwhelmingly more frequent in OV languages than in VO languages (63: 8). Third, adpositions are somewhat more common in OV languages, and a more specific count reveals that all adpositions found in OV languages are postpositions, while all adpositions found in VO languages are prepositions (in line with the well-known word order correlations, cf. Dryer 1992). Apparently, then, the structural type of purpose marker employed is sensitive to the general constituent order of the language in which it occurs. This is confirmed by a randomized Chi-squared test performed on the distribution of the three major marking types (χ² = 51.82, df = 4, p < .001, B = 100,000, Cramer’s V = 0.364), and the skewing is illustrated in the spine plot in Figure 11. What is more, conjunctions, affixes and adpositions differ considerably with regard to their position within the purpose clause. For ease of exposition, let us

0.6

0.8

adposition

0.2

0.4

affix

other

OV

VO

0.0

conjunction

Figure 11.  Correlation between structural type of marker and constituent order (N = 196)

79

A Typology of Purpose Clauses

Table 18.  Position of conjunctions, affixes and adpositions in the purpose clause Conjunctions

Affixes

Adpositions

Total

S-SUB SUB-S Other

8 51 2

74 2 20

21 12 6

103 65 28

Total

61

96

39

196

1.0

call the primary marker or clause-linking device a ‘subordinator’, which we can simply abbreviate as SUB. Its position in the purpose clause can then either be at a clause boundary— at the left periphery introducing the clause (SUB-S) or at the right periphery concluding it (S-SUB) — or occupy a certain position in the centre of the clause (a pattern we shall call ‘other’). Table 18 quantifies the extent to which different types of markers prefer a certain position in the clause. These data suggest that conjunctions are strongly preferred in clause-initial position, while affixes most frequently show up clause-finally. For adpositions we can detect a split again: all of the 21 adpositions occurring in S-SUB position are postpositions, while the 12 adpositions in the SUB-S patterns are uniformly prepositions. The distribution in Table 18 is statistically highly significant (χ² = 111.98, df = 4, p < .001, B = 100,000, Cramer’s V = 0.534), and its skewing is best illustrated by a spine plot again (Figure 12).

0.6

0.8

adposition

0.2

0.4

affix

conjunction S-SUB

SUB-S

other

0.0

80

Figure 12.  Correlation between structural type of marker and its position in the clause (N = 196)



Chapter 3.  The grammar of purpose

Uniting the results summarized in Tables 17 and 18, we can conclude that conjunctions are predominantly found as introducers of purpose clauses in VO languages. Affixes, by contrast, are mostly terminating elements of purpose clauses in OV languages. Prepositions generally behave like conjunctions, whereas postpositions resemble affixes in terms of word order. The position of the subordinator will be analysed more closely in §3.5.2, in connection with the position of the purpose clause. The second typological parameter which may exert an influence on the distribution of the primary markers is geography. Just as with the deranking data, we shall thus examine whether the preference for a particular marking type in purpose clauses is closely associated with any of the geographic macro-areas. To this end, all 225 attested primary markers were coded for their type and for the macroarea in which they appear, which yielded the following results (Table 19). Again, only the major marking types (conjunctions, affixes, adpositions) allow for quantitative statements. In some macro-areas, no particular or significant preference for one marking type over the others is observable (Oceania, Mesoamerica). For Africa, there is a skewing in favour of conjunctions, which account for 45% of all purpose markers (18∕40) and are more than twice as frequent as either affixes or adpositions. In Eurasia, affixes form the majority of purpose markers (21∕52 = 40.4%), while conjunctions and adpositions occur with about equal frequency. It should be noted, however, that Eurasia has an enormous geographic extension and also covers very heterogeneous stocks (from Indo-European to Tibeto-­Burman). In Australia, we get the most strongly asymmetric situation: affixes account for 73.3% of all purpose markers (23∕30), while the other types constitute very small minorities. North America, likewise, shows an affix preference (15∕28 = 53.6%), yet it is not as strong as the Australian one. In South America, finally, affixes also account for more than half of all purpose markers (18∕33 = 54.5%), yet conjunctions take a firm and well-established second rank (10∕33 = 30.3%). Comparing now the marking types, we can see that conjunctions are overall most frequent in Africa (18∕61 = 29.5%), while they Table 19.  Geographical distribution of different types of primary markers Conj.

Affix

Adp.

Particle Clitic

Aux.

None

Total

Africa Eurasia Australia Oceania North America Mesoamerica South America

18 12 3 10 7 1 10

8 21 23 11 15 0 18

8 13 2 8 2 2 4

2 2 0 4 0 0 0

0 0 0 1 0 1 0

1 0 0 1 1 0 0

3 4 2 3 3 0 1

40 52 30 38 28 4 33

Total

61

96

39

8

2

3

16

225

81

82

A Typology of Purpose Clauses

are very infrequent in Australia and North America (Mesoamerica cannot be taken into account because of its low overall contribution to the database). Affixes, the largest class of purpose markers, are most frequently found in Australia and Eurasia, with both of the Americas not lagging far behind. African languages, however, seem to disprefer affixing morphology. This conforms to the exceptional role played by Africa with regard to inflectional morphology more generally: The data in Bickel and Nichols (2005a) suggest that Africa displays the greatest internal diversity in how inflectional material is morphologically related to the stem. While in all other macro-areas, concatenative strategies conspicuously outnumber all others, African languages employ a fairly heterogeneous mixture of concatenative, isolating, tonal and mixed strategies for inflectional categories (Figure 13). Purposive adpositions, finally, are common in the languages of Eurasia, while being relatively infrequent in the remaining areas. I shall point out again, however, that Eurasia is really a mixed bag of genetically and geographically unrelated languages, so there may well be internal skewings for particular types of markers. In sum, the choice of the structural type of purpose marker is not only skewed in relation to the basic constituent order type of a language, but also with respect to its geographical distribution. The overall areal skewing of conjunctions, affixes and adpositions is highly significant (χ² = 34.57, df = 12, p < .001, B = 100,000, Cramer’s V = 0.297). Most striking in this context proved to be the strong preference for conjunctions in African purpose clauses, and the overwhelming dominance of affixes in Australian and, somewhat less pronounced, North American languages. Leaving particular marking strategies for now, we will close this section by examining the morphophonological shape of primary markers in purpose clauses.

1. Exclusively concatenative [122] 2. Exclusively isolating [16] 3. Exclusively tonal [3] 4. Tonal/isolating [1] 5. Tonal/concatenative [2] 6. Ablaut/concatenative [5] 7. Isolating/concatenative [13]

Figure 13.  Fusion of selected inflectional formatives (adapted from Bickel and Nichols 2005a: 86)



Chapter 3.  The grammar of purpose

Table 20.  Morphological shape of purpose markers Morphological complexity

Absolute frequency

Relative frequency (in %)

Monomorphemic Polymorphemic No morphological marking

193 16 16

85.8 7.1 7.1

Total

225

100.0

This is meant to be a purely descriptive overview.17 In terms of morphology, purposive markers strongly tend to be monomorphemic rather than polymorphemic. Judging from the synchronic shape of the marker and the degree of detail provided by the reference grammar, nearly 86% of the markers considered were classified as (synchronically) monomorphemic, while only about 7% were unambiguously polymorphemic (Table 20). Among this smaller polymorphemic class, we find, for example, a complex purposive conjunction in Hmong Njua: (72) Hmong Njua (Hmong Mien: China; Harriehausen 1990: 242) John dlha [txhaj le tau noj mov]. John run [purp eat rice ‘John is running in order to eat rice.’ English, too, has a complex purposive marker: so that is usually analysed as a single conjunction, although clearly being decomposable into two morphemes, as evidenced by the fact that the complementizer that can optionally be omitted. This combination of a resultative or consequential marker (so) and a complementizer, yielding a morphologically complex purposive conjunction, is also attested in other languages. Lealao Chinantec has a finite purpose clause whose structure parallels that of English so that clauses: (73) Lealao Chinantec (Oto-Manguean: Oaxaca/Mexico; Rupp 1989: 50) vó:M]. Hme:H ziúL [duʔM hąM ʔiL-tiá:L do.imp good [so that int-endure.iii3 long ‘Do it well so that it will last a long time.’ 17.  Notice, however, that this property of purpose clauses also has an interesting theoretical dimension. Kortmann (2001) points out that for the class of adverbial conjunctions, there is an inverse correlation of formal (i.e. morphophonological) complexity and both syntactic and semantic multifunctionality. More precisely, he claims that “the lower the degree of morphological complexity” of an adverbial conjunction (AC), “the higher is the likelihood that the AC additionally belongs to other word classes” (Kortmann 2001: 845). Although I will also discuss patterns of multifunctionality of purpose markers in this study (cf. § 3.4.2 below), Kortmann’s hypothesis will not be further pursued here.

83

84

A Typology of Purpose Clauses

Such complex conjunctions should be distinguished from constructions in which ‘so’ or a semantically similar morpheme occurs as a correlative place-holding element in the main clause, as in Hungarian: (74) Hungarian (Uralic: Hungary; Kenesei et al. 1998: 51) Péter az-ért olvasott, [hogy ne alud-j-on el]. Peter that-for read [that not sleep-subj-3sg prfv ‘Peter was reading so that he would not fall asleep.’ In this case, the place-holding element is not part of a more complex conjunction ‘so that’, but indicates the topic position of the purpose clause. English has what appears to be a similar pair of constructions: (75) a. Phoebe is working hard so that she can go to the concert. b. Phoebe is working so hard that she can go to the concert. In both cases, the proposition of the purpose clause is roughly the same. And yet it would be hard to decide whether so fulfils the same function in both constructions. I suggest using the following bracketing notation to show what the difference might be: (76) a. Phoebe is working hard [so that she can go to the concert]. b. Phoebe is working [so hard that she can go to the concert]. In (76a) we are really dealing with a full-fledged canonical purpose clause, modifying the proposition Phoebe worked hard. In (76b), by contrast, the whole clause governed by so functions as an adjunct of manner modifying Phoebe worked (cf. also Huddleston and Pullum 2002: 968–9, who analyse the latter structure as a PP). If the alleged correlative element is a demonstrative rather than a manner adverb, as in the Hungarian example, it can be quite reliably assumed that the whole purpose clause is being replaced or anticipated and that the purposive conjunction is simply the one introducing the purpose clause. As a further point, it is fascinating to observe that some languages owe their complex purposive markers – notably conjunctions – not to diachronic processes internal to the language, but to borrowing from either a neighbouring language or the more prestigious language in bilingual communities. Yaqui, a Cahitic language of the Uto-Aztecan stock, for example, did not traditionally have a purposive conjunction, using an adverbial suffix instead. Under prevailing Spanish influence, however, the foreign complex conjunction para que was borrowed into Yaqui. It has been reanalysed as a monomorphemic conjunction, parake, and now normally introduces purpose clauses:



Chapter 3.  The grammar of purpose

(77) Yaqui (Uto-Aztecan: Sonora/Mexico; Lindenfeld 1973: 84) Ame-u-te ča-čae-ne [parake bempo itom bit-ne]. them-to-we freq-call-exp [purp they us see-exp ‘We will keep calling them in order for them to come and see us.’ As Thompson and Longacre (1985: 205) point out, there are a number of possible reasons why languages may borrow conjunctions into their own inventory of syntactic markers. Backgrounding the sociolinguistic dimension of language dominance and prestige for the moment, we might also entertain the possibility of structural factors being at work. More specifically, the data in the reference grammar suggest that before borrowing parake from Spanish, Yaqui had a slightly disadvantageous marking strategy for purposes from a functional point of view. The indigenous purposive marker is -kai, which is still available synchronically: (78) Yaqui (Lindenfeld 1973: 82) Inepo in koarto-wi kibake-k [supe-te-bae-kai]. I my room-to go-rl [shirt-put.on-int-purp ‘I went to my room in order to get dressed.’ It turns out, however, that -kai is highly polysemous: it also functions to mark sentential complements, and it can be used to evoke a fairly heterogeneous class of adverbial meanings other than purpose. Thus “it most often carries an undifferentiated meaning of Subordination” (Lindenfeld 1973: 81). As a result, the exact semantic relationship between the two situations involved has to be inferred from a wide array of possible interpretations.18 From this point of view, the borrowing of parake could be motivated by a drive for transparent, i.e. iconic, form-function patterns: one form corresponding to one meaning. Although the example from Yaqui is unique in the sample, it certainly holds as a more general tendency that morphologically complex purpose markers, especially conjunctions, are commonly reanalysed and reduced versions of formerly two or more independent formatives. The morphological two-way contrast between mono- and polymorphemic markers can be reconsidered in a more fine-grained fashion on the phonological level. I coded all 225 purpose markers for the number of syllables they contain, and cross-classified this feature with the structural type of marker. This allows us assess the typical phonological complexity of purposive conjunctions, affixes, 18.  According to Bisang (2007), economical structures which can evoke a range of possible interpretations open up a layer of ‘hidden complexity’ in language, specifically in assigning the correct meaning in on-line processing.

85

86

A Typology of Purpose Clauses

Table 21.  Constructions with a monophonemic primary purpose marker Language

Macro-area

Purpose marker

Kayardild Nama (Khoekhoe) Somali Supyire Wolaytta Georgian Kannada Jamul Tiipay Amele Boumaa Fijian Imonda Tukang Besi Yimas Huallaga Quechua Mapuche Yagua

Australia Africa Africa Africa Africa Eurasia Eurasia North America Oceania Oceania Oceania Oceania Oceania South America South America South America

-n ñ ú i~e -w -o -i -i, -x -a i -m i -k -q -m -y

a­ dpositions, auxiliaries etc. The results are displayed in Figure 14 below. As can be gleaned from the overall distribution, primary purpose markers are not only most commonly monomorphemic, but also monosyllabic. In other words, they are on average as short as a grammatical marker can be. In fact, purposive constructions from 16 languages even employ a monophonemic (!) marker. Table 21 shows that these languages are geographically dispersed, so this cannot be a micro-areal effect due to borrowing. The more specific distributions in Figure 14 suggest that while purposive affixes can be polysyllabic, they strongly tend to be monosyllabic (71∕96 = 74%). Purposive conjunctions, by contrast, are equally frequently mono- and bisyllabic (each 24∕61 = 39.3%) but, most notably, they account for the lion’s share of markers that are trisyllabic (13∕19 = 68.4%). Adpositions, finally, pattern similar to conjunctions again. The overall distribution of the three major structural types (conjunctions, adpositions and affixes) is significantly skewed under a randomization test (χ² = 25.84, df = 4, p < .001, B = 100,000, Cramer’s V = 0.258). A final remark in this context concerns the class of conjunctions again. Kort­ mann (2001: 852) claims that – at least for Europe – there is a correlation between the morphophonological complexity of adverbial conjunctions and the constituent order type of the language. He hypothesizes that VO languages tend to have more complex conjunctions than OV languages. At least in terms of syllable structure, this does not hold cross-linguistically. A comparison of the 17 conjunctions occurring in OV languages and the 34 conjunctions found in VO languages (cf. Table 17



Chapter 3.  The grammar of purpose

above) shows that while conjunctions are overall more frequent in VO languages, they are not necessarily more complex. On the contrary, the average number of syllables in OV conjunctions is 2.8, as opposed to 1.8 syllables on average for VO conjunctions. A trisyllabic conjunction is found, for example, in the OV language Sanumá: (79) Sanumá (Yanomam: Venezuela, Brazil; Borgman 1990: 99) [ĩ naha sa ku pio salo], sa hu pä-o ke. [rel like 1sg say purp 1sg go merely-pnct imm.pst ‘I merely went in order that I might say that.’ Sanumá speakers use the complex conjunction pia salo ~ pio salo, which consists of two morphemes: pio ‘intend’ and salo ‘result/cause’. Since the whole formative is glossed by Borgman (1990: 99) as having the meaning ‘in order to’, one might well say that it has reached the status of a conjunction. Conjunctions in OV languages, therefore, need not be morphophonologically less complex than in VO languages. We have now completed our survey of primary markers in purpose clauses and proceed to what we called ‘secondary gestalt features’, i.e. any additional cues in a purposive construction.

80 70 60 50 40 auxiliary (N = 3) clitic (N = 2) particle (N = 8) adposition (N = 39)

30 20 10

conjunction (N = 61)

0 monosyllabic bisyllabic (N = 126)

affix (N = 96) trisyllabic

polysyllabic (N = 83)

tetrasyllabic

Figure 14.  Phonological (syllable) structure of different types of primary markers

87

88

A Typology of Purpose Clauses

3.4.2  Secondary gestalt features and the semantics of purpose Surveying our 218 purpose clause constructions for the gestalt features they contain apart from a primary marker (including those that have no genuine clauselinking morphology but still a unique other feature), we can classify these ‘secondary markers’ into salient semantico-syntactic categories again. In analogy to Figure 10, we represent this classification graphically (Figure 15). Before we look at the individual classes in more detail, it should be mentioned that Figure 15 represents a considerable reduction of the variation space in purpose clauses since it only captures recurrent categories. By opening up a fairly large class of ‘other’ features, we lose at this point a number of language-specific idiosyncrasies that cannot all be described individually. The ‘other’ group comprises, for instance, logophoric pronouns, evidential and attitudinal markers, specialized noun phrases (‘way’, ‘manner’) or copulas, reflexive possessor arguments, correlative placeholders, etc. What we thus try to capture in Figure 15 is a generalization across secondary features that are at least found twice in the sample and cannot readily be reduced to a construction-specific quirk in a particular language.19 Having said that, we can turn to the groups defined in Figure  15. Some of them allow for even further clustering. Nominalizations and infinitival verb forms, if secondary cues in the purposive construction, both serve as a means of deranking and have already received ample illustration in this context. Similarly, all the TAM markers found here are drawn from semantic domains that harmonize well with the notion of purpose, and they, too, have already been exemplified in some detail. This leaves us with a small and semantically very different class, i.e. that of allative and dative case markers. The labels ‘allative’ and ‘dative’ are taken directly from the reference grammars, so they may well subsume language-specific categories with different semantic maps, especially at their boundaries. But the core function they share is the expression of directed motion in allatives, and the encoding of recipient participants (i.e. typically the addresses of ‘give’ predicates) in datives. Even though these two categories are rather small in comparison to the other groups in Figure 15, there at least two reasons why they should not be treated as language-specific idiosyncrasies. First, it should be noted that case marking here does not refer to argument marking for the purpose of indicating syntactic functions in the clause, i.e. subject, 19.  The attentive reader may wonder why some of the data given here deviate from earlier counts. For instance, the counts for TAM categories differ from the ones listed in §3.2.2. The answer is that the present table takes stock of invariant gestalt features that are always present in the respective purposive construction. That is, if a future tense marker is attested in an example of a particular construction, it is listed here only if it is a characteristic of all examples of that construction. Luckily, many authors explicitly mention all such gestalt features of each purposive construction.



Chapter 3.  The grammar of purpose

22

nominalized verb form or masdar

15

infinitival verb form (including bare stems)

12

future or sequential tense-aspect markers 9

irrealis, hypothetical or potential mood 6

desiderative or optative mood 2

intentional mood

12

subjunctive mood 2

allative marker

8

dative marker

28

other

102

no additional element 0

10

20

30

40

50

60

70

80

90

100

110

Figure 15.  Secondary gestalt features in purpose clauses (N = 218)

direct object etc. Rather, the case marker appears as a unique feature of purpose clauses, often on the verb, as in Kannada: (80) Kannada (Dravidian: India; Sridhar 1990: 73) Praka:ʃ jarmanige [enjiniyaring o:duvudakka:gi] ho:gidda:ne. Prakash Germany.dat [engineering study.msd.dat.for go.n:pst.prf.3sg:m ‘Prakash has gone to Germany to study engineering.’ In this construction, the purpose clause is governed by a postposition and its subordinate status is signalled by a nominalized verb form (msd). Since the masdar essentially turns the verbal expression into a nominal one, the case marking appropriate for the postposition ‘for’ is added to the nominalized verb form. It is in this way that the dative suffix appears on the purposive verb in Kannada. Notice, however, that a dative or allative case affix on the verb is not always governed by an adposition and so its appearance needs to be motivated on independent grounds. At any rate, the crucial point is that case marking on purposive verb forms is universally skewed towards allative and dative case. In other words, (81) If a language uses case marking on the predicate in purpose clauses, the odds for dative or allative marking will be higher than for any other type of case. With the Kannada construction in (80) above, we also find the second – and much stronger – argument for the significance of (allative and) dative case marking in

89

90

A Typology of Purpose Clauses

Table 22.  Allative- and dative-oriented polysemy of purpose markers Purpose marker identical to…

Absolute frequency

Relative frequency (%)

locative marker locative and allative marker allative marker allative and benefactive marker dative/benefactive marker dative/goal marker n.i.a no marker of this domain

1 4 7 4 41 6 6 149

0.5 1.8 3.2 1.8 18.8 2.8 2.8 68.3

Total

218

100.0

a

  n.i. = no information on multifunctionality available

purpose clauses. Even if we considered the dative morpheme in (80) a structural requirement of the governing postposition ‘for’, we would still need to explain why this postposition is itself recruited from the domain of recipient-benefactive marking. In fact, if we examine all primary markers in our sample that function as a genuine purpose marker (i.e. can be glossed ‘purp’), it turns out that they are often formally identical to morphemes of the domains ‘locative-allative’ and ‘dative-recipient/benefactive’, respectively. In the sample of 218 constructions, such overlaps are distributed as set out in Table 22. If we count all primary purposive markers which simultaneously function as allative or dative markers, and add to this the number of constructions in which dative and allative case markers are secondary gestalt features in purpose clauses (cf. Figure 15 above), we arrive at a total of 72 out of 218 purpose constructions (33%) which contain an allative or dative marker. This figure even outnumbers the occurrence of special TAM forms: recall from Figure 15 that specific tense or mood markers surface in 41 constructions, and adding to this the five instances in which a TAM morpheme acts as the purpose marker itself (cf. Figure 10 above), we get a total of 46∕218 = 21.1%. That is, in about one fifth of all constructions does a purpose clause receive special TAM marking, but in about one third of all cases, it displays allative- or dative-related morphemes. Importantly, this does not appear to be a genealogical or macro-areal effect. The overlaps are attested in historically unrelated languages and also widely dispersed. Figure 16 shows all languages from the present sample for which the available material provides clear evidence that at least one purpose clause construction shows traces of allative-dative marking (N = 39). The specific distributions should not be overrated, though, as many language descriptions simply do not offer enough information on patterns of polysemy. Their absence in the map cannot, therefore, be taken to exclude the possibility that allatives and datives are (or have been historically) involved in the construction of purpose clauses.



Chapter 3.  The grammar of purpose

Figure 16.  Languages with traces of allative and/or dative marking in at least one purpose clause construction (N = 39)

The genealogical and areal independence of this phenomenon suggests that we are dealing with a statistical universal trend here that would need to be motivated. Given that the notions in question are recurrently coded by the same linguistic material, they appear to be systematically related in such a way that language users are tempted to extend an existing form-function mapping to one or more of the other functions. In cognitively-oriented typology, this hypothesis has been formulated as a tendency for certain functions to be contiguous in conceptual space (cf. also Chapter 2 for this idea). The functions are linked by a chain of similarity or what Wittgenstein (1958: 66) called a ‘family resemblance’. For allatives, datives and purpose morphemes, these contiguities have been noted in the literature from various phenomenological angles. Thus Blansitt (1988) develops a one-dimensional map for the connection of locative, allative and dative markers, and Haspelmath (2003) offers a two-dimensional cartography of cross-linguistic functions of the dative (Figure 17). In a similar vein, Heine (1991) maps out the functions of the Dative case in two Nilo-Saharan languages, Ik and Kanuri (Figure 18). Thus, according to Heine, the Dative suffix -(k)e in Ik has as its most basic function that of a directional locative (i.e. allative) (see (82)). Predicative possessor

External possessor

Direction

Recipient

Beneficiary

Purpose

Experiencer

Judicantis

Figure 17.  Conceptual space for typical dative functions (adapted from Haspelmath 2003: 213)

91

92

A Typology of Purpose Clauses

ALLATIVE GOAL

PLACE

PURPOSE

BENEFACTIVE

REASON

DATIVE

MANNER

TIME

POSSESSION

CONDITION

clause embedding Figure 18.  Functions of the Dative case in Ik and Kanuri (adapted from Heine 1991: 131) (Functions which are not printed in bold letters are confined to Ik.)

(82) Ik (Nilo-Saharan; Kuliak: Uganda; Heine 1991: 132) K’e-esá ntsa awá-ke. go-fut he home-dat ‘… and he will go home.’ In addition, the affix can denote human goals, specifically recipients and beneficiaries: (83) Ik (Heine 1991: 132) 'jɔ-ɔt-ɔsá' emá ńc-i-ke. roast-ven-pass meat I-dat ‘Meat has been roasted for me.’ Finally, the dative suffix also appears on verbs in purpose clauses: (84) Ik (Heine 1991: 133) Bɛr-ɛs-ɔ lo'dúrú-iké-e ni ot-és-í-e e'dí. build-fut-cop granary-pl-gen rel pour-fut-opt-dat grains ‘… and they will build granaries to store crops in.’ As we shall see, the historical pathways that lead to such synchronic overlaps are very intricate and cannot usually be captured by linear developments from one of those markers to the next (cf. §4.2.6 for details). But what needs to explained at this point is how the contiguities are motivated. In keeping with our commitment to cognitive-functional explanations, we will argue that the family resemblance is rooted in experiential correlations between directed motion, purpose, reception and benefaction, and the goal-orientation inherent in all of these notions. It will be conducive to look at individual overlaps separately because several experiential



Chapter 3.  The grammar of purpose

scenarios and cognitive processes are involved and thus need to be made explicit. We will begin by looking at the relationship between purpose and directed motion, i.e. allative marking in purpose clauses. A nice illustration of their interaction can be found in the following example from Maori: (85) Maori (Austronesian; Oceanic: New Zealand; Bauer 1993: 66) E haere ana ahau ki te taaone [ki te hoko mai i teetahi koha]. tns move tns 1sg to the town [purp the barter hither obj a.spec gift ‘I am going to town to buy a present.’ The matrix clause contains the allative preposition ki ‘to’ in its basic directional function. In the purpose clause, ki reappears as the very marker that introduces the subordinate clause and establishes the purposive relation between the two propositions. In this latter function, it usually occurs with the definite article, with which it might be said to have formed a complex purposive conjunction. Notice that English to in its most basic function is an allative preposition, but has also become a very frequent purpose marker. In a most recent publication, Rice and Kabata (2007) argue that the metaphorical extension to purpose is, in fact, the most widespread pattern of polysemy that allative markers develop: “The single most prevalent cohort sense of an allative is to mark purpose” (Rice and ­Kabata 2007: 472).20 A closely related pattern also attested in our sample is a conflation of locative, allative and purposive meanings into one formative. Tzutujil, a Mayan language of Guatemala, has two prepositions, pa(n) and ch(i), both of which govern infinitival purpose clauses. (86) is an example containing pa in this function: (86) Tzutujil (Mayan: Guatemala; Dayley 1985: 380) Ja nata7 b'enaq [pa tikoj chiij]. the my-father b3-has-gone [purp plant cotton ‘My father has gone to plant cotton.’ The more basic uses of pa include the spatial and directional prepositional meanings ‘in’, ‘on’ and ‘to’. Similarly, ch(i) has the locative meaning ‘at’ and the directional meaning ‘to’. They thus provide further evidence for the semantic extension 20.  It should be noted that Rice and Kabata’s analysis is based on a fairly idiosyncratic sample of 44 languages, which are not too-well balanced in terms of genealogical and geographical independence. One may criticize, for instance, that their sample contains nine Indo-European languages, many of which are closely related, as well as three relatively similar Oceanic languages, and only very few languages from South America. In this light, the results obtained may well be biased by a sampling error. Notice that in my database, a direct relationship between allative and purpose can only be attested for 15 languages, i.e. 18.8% as compared to Rice and Kabata’s 46.3%.

93

94

A Typology of Purpose Clauses

from allative to purposive meanings of one and the same formative. As we noted at the beginning of this section, allative marking can also occur elsewhere in the clause, i.e. not as the purposive marker as such but as an additional gestalt feature: (87) Basque (isolate: Spain, France; Hualde and Ortiz de Urbina 2003: 742) Zer ez dute egiten Espainiako euskualdunek […] eskuara beren what not aux do.impf Spain.rel Basques.erg   Basque their sor-mintzaia galtzerat ez uzteagatik? birth-language lose.noml.all not let.noml.det.purp ‘What will the Basques from Spain not do not to let the Basque language, their native tongue, die?’ In this case, the subordinate situation ‘let lose (‘die’)’ is marked by purposive nominalization and an additional allative marker on the subordinate predicate. Finally, a very strong piece of evidence for the close association of purpose and directed motion can be found if we look beyond the purpose clause into the predicate of the matrix clause. On closer inspection, it turns out that the matrix verbs of many purpose constructions in our sample are semantically restricted in a systematic way. Table 23 captures those restrictions. The table has to be read as follows. In 144 out 218 constructions, there is no particular restriction on the semantics of the main verb, hence it can in principle be any type of verb. This does not come unexpected since the conceptual structure of purposes does not logically constrain the nature of the superordinate situation. In many cases, however, it turns out that the main verb is typically from the domain of motion (12.8%). This is usually pointed out explicitly by the authors of the reference materials. In yet other cases, there is no such statement, and we have to rely on the examples attested. In 19 such constructions (8.7%), all purpose clauses exemplified contain a superordinate verb of motion. That is, the grammar does not provide any instance of a particular purpose clause whose associated matrix does not contain a motion verb. The most severe restriction holds for 21 constructions (9.6%) for which it is explicitly stated that they must occur with a motion verb. Finally, for a tiny minority (1.4%) the grammar did not offer any Table 23.  Semantic restrictions on the superordinate verb (N = 218) Semantic type of main verb

Absolute frequency

any copula verb any, though motion typical only motion attested must be motion n.i.

144 3 28 19 21 3

Relative frequency (%) 66.1 1.4 12.8 8.7 9.6 1.4

Total

218

100.0



Chapter 3.  The grammar of purpose

examples of purpose clauses, so we do not know what verbs it can co-occur with. Of the above list, the most constrained type deserves some further comments. We will call purposive constructions in which the superordinate verb is a motion verb ‘motion-cum-purpose’ constructions (a term adopted from Aissen 1984). At first it should be noted that such constructions are not confined to particular geographical areas or language families. They occur in languages from virtually all macro-areas, i.e. Africa (Noon, Supyire, Krongo), Eurasia (Basque, Kolyma Yukaghir, Lezgian, Semelai, Tibetan), Australia (Kayardild, Tiwi, Wardaman), Oceania (Imonda [3], Tukang Besi, Acehnese, Boumaa Fijian), North America (Lakhota) and South America (Huallaga Quechua, Hixkaryana, Warao). We will focus on some selected constructions in order to discuss their morphosyntactic characteristics. In Krongo (Kadugli: Sudan), purpose clauses are canonically formed by an infinitival clause which contains a genitive prefix; Reh (1985: 350) claims that this prefix has come to function as a purposive conjunction (which is very unusual for genitives cross-linguistically). Such a canonical Krongo purpose clause is given in (88): (88) Krongo (Kadugli: Sudan; Reh 1985: 350) N-àdúkwà àʔàŋ tìisí [máa-ní k-áatú òshóonì-ìcí]. 1/2-take.prfv I grass [gen-3sg loc-people finish.roof.inf-dtr ‘I have gathered some grass so that the people can finish the roof.’ In addition to this pattern, however, purposive relations can also be expressed by what Reh calls a ‘sentential complement of motion verbs’. This construction is only available for the most general motion verbs ‘come’ and ‘go’, and does not contain the characteristic genitive marker: (89) Krongo (Reh 1985: 351) M-áa cáaw òmúno-ŋ éekwàarà. conn.f-cop go.inf (dat).inf-call-tr chief ‘And she goes to call the chief.’ Notice that the infinitive glossing on the main verb ‘go’ is not a peculiar feature of this construction, it regularly occurs in other types of clauses (including main clauses) as well. The construction as a whole, however, is different from the standard way of expressing purpose in Krongo. Similarly, speakers of Hixkaryana have to make a choice between “two basic types of purpose clause: motion purpose and nonmotion purpose” (Derbyshire 1979: 28). Specifically, motion purpose clauses are marked by either of two derivational suffixes, -so or -xe, which are directly attached to the verb stem. Non-motion purpose clauses, by contrast, are usually marked by a purposive postposition and more complex nominalizing morphology on the verb. Compare:

95

96

A Typology of Purpose Clauses

(90) Hixkaryana (Carib: Brazil; Derbyshire 1979: 29) [Romararn ho wewe y-ama-xe] nomohtxow. [my.field in tree iii-cut-mot.purp they.came ‘They have come to fell the trees in my field.’ (91) Hixkaryana (Derbyshire 1979: 29) Rohtxe ymyako Nonato, [ohxe rownknr hor]. my.medicine he.gave.it Nonato [good my.sleep.noml purp ‘Nonato gave me my medicine so that I would sleep well.’ Another interesting case comes from Tiwi, an Australian indigenous language which has recently undergone dramatic and rapid structural changes due to heavy contact with English for the past 70 years. In Traditional Tiwi, purpose clauses were commonly introduced by a subordinating conjunction, followed by a finite clause with an overt subject. Almost all purpose clauses recorded in the author’s corpus of Modern Tiwi, by contrast, neither have a conjunction nor an overt subject, but in all cases a superordinate verb of motion. Moreover, the purpose clause tends to be very short and – in comparison to other adverbial clauses – can probably not be regarded as fully balanced. This motion-cum-purpose construction, then, would be the only deranked (‘non-finite’) construction in Modern Tiwi. (92) and (93) demonstrate the difference between purpose clauses in Traditional and Modern Tiwi: (92) Traditional Tiwi (Tiwian: Northern Territory/Australia; Lee 1987: 302) Yu-wa-nguwunu-kuwuna [ngini a-ma-ti-ri-kurruwa]. he.pst-frust-us-chase [purp he.npst-subj-irr-conn-catch ‘He chased us (but without success) in order to catch us.’ (93) Modern Tiwi (Lee 1987: 303) Ngi-ntu-wuriyi [kulalaa warta]. we-pst-go [hunt bush ‘We went hunting in the bush.’ What all of the motion-cum-purpose constructions outlined have in common is a relatively high degree of integration into the main clause, as compared to their non-motion purposive counterparts. They are usually more economical by virtue of containing less linguistic material, and are often described by authors as complements rather than adjuncts. As such, they are more closely integrated into the main clause because they fill a syntactic argument position. One indication for this syntactic function is that motion-cum-purpose clauses in Hixkaryana, for example, can precede the main verb, just like direct objects (OVS), while canonical purpose adjuncts are always postposed after the main clause. Furthermore, in



Chapter 3.  The grammar of purpose

Kiowa some purpose and complement clauses, but not other adverbial clauses, can be tightly integrated into the object position of the main clause by a process called ‘Raising Incorporation’. Crucially, purpose clauses allowed in this construction contain a motion verb, while in other purposive constructions all kinds of verbs are possible: (94) Kiowa (Kiowa-Tanoan: Kansas; Watkins 1984: 229, 238) a. purpose clause in the ‘Raising Incorporation’ construction Gyà-kí·+-kɔ·tɔ+tót. 1sg.agt: 3sg.obj-meat+buy+send.prf ‘I sent him to buy meat.’ b. ordinary purpose clause with sequential switch-reference markers [Èm-sɔ·-gyà· mɔn [èm-tò·+dô·gù-gɔ. [3sg.refl]-sit-prf probably [3sg.refl]-shoe-put = on-impf-ss ‘He sat down probably in order to put on his shoes.’ If tightly packed, i.e. economical, expression is a function of predictability (as argued earlier in this book), then the observation that reduction of purpose clauses is particularly common in motion contexts suggests that there is a close experiential correlation between directed motion and purpose. Cognitive linguists have, in fact, argued that this is precisely the case. They maintain that we experience from a very young age onwards that the realization of purposes is often contingent on a change of location, such as going to the market to buy some bread. Even for toddlers, achieving purposes such as getting something to drink or to play with often requires them to reach physical destinations. In other words, the achievement of particular purposes is often tied to a specific location. This experiential correlation gives rise to two conceptual operations which compete, as it were, in the explanation of the allative-purpose polysemy across the world’s languages. On the one hand, Lakoff and Johnson (1980, 1999) argue that the experiential association of purposes and destinations provides the very basis for the conceptualization of purposes in the first place. The underlying assumption here is that highly abstract domains of our experience, such as time, cannot readily be conceptualized in their own terms (cf. Evans 2004 for illustration). Instead, mental models of such abstract concepts can be constructed by relating them to a “well-developed frame of reference” (Gelfand and McCusker 2002: 298), typically a domain of physical or bodily activity that correlates with the ‘target’ concept in everyday experience. The conceptual domain of time, for example, is assumed to piggyback on the source domain of (orientation and motion in) space. We thus build up conceptual knowledge about time by unconsciously importing conceptual structure (e.g. image schemas) from the mental models we have of spatial experience. In verbalizing temporal concepts, then, we often recruit linguistic

97

98

A Typology of Purpose Clauses

­ aterial denoting spatial relations. We thus talk about things happening on time m or at a particular time, about the following day, about the past being behind us, etc. By virtue of time and space being repeatedly co-activated in experience, a ‘binding’, i.e. “permanent neural connection” (Lakoff and Johnson 1999: 46) emerges between them. Along the same lines, Lakoff and Johnson propose that purposes are also conceptualized metaphorically. Specifically, they hypothesize that since the fulfilment of purposes is so often contingent on (movement to) specific locations, a ‘primary metaphor’ arises “naturally, automatically, and unconsciously through everyday experience” (Lakoff and Johnson 1999: 46). In other words, the two conceptual domains are regularly conflated and give rise to the metaphor purposes are destinations. Once the conceptual structure of purpose is developed in spatial terms, it is, of course, applied to the conceptualization of purposes more generally, not just to those involving directed motion. On this account, the conflation of allative and purpose markers is grounded in the fact that purposive events are construed as destinations (i.e. goals of movement). Therefore, an already existing goal marker is pressed into service for verbalizing such complex or more abstract goals. The linguistic result is an innovation in language use, to the effect that a goal marker previously confined to NPs is extended to VPs, as we saw in our initial example (85) from Maori. The alternative account21 does not deny the close experiential correlation between purposes and destinations. On the contrary, it starts from the same assumption but offers a more dynamic account of how this correlation is exploited in language use. Specifically, it builds on the notion of a semantic frame as developed by Fillmore (e.g. Fillmore 1982). Frames can be defined as “coherent region[s] of human knowledge” in which “certain concepts ‘belong together’ because they are associated in experience” (Croft and Cruse 2004: 7, 14). For example, in contexts of directed motion, such as (95) I’m going to the river. the NP designating the location activates a conceptual frame including the (desired) activity typically associated with that location, e.g. bathe, fish, wash etc. The contiguity of location and purpose in the specific event frame thus allows the hearer of (95) to infer the purposive action. Such shifts of attention within a frame ultimately result in metonymic innovations in language use: a verb of directed motion is associated directly with the accomplishment of an action. Put differently, the action comes to stand for the location, and as a result, the grammatical marker previously flagging the destination comes to flag the purposive action. The out21.  I am very grateful to Fernando Zúñiga and an anonymous reviewer for emphasizing the importance of this alternative, i.e. metonymic, explanation.



Chapter 3.  The grammar of purpose

come is essentially the same paradigmatic extension from, say, allative-prepositional to purposive-conjunctional ki in Maori. The crucial difference between the metaphorical and the metonymic account is that the latter does not hinge on the assumption of profound conceptual similarities between purposes and destinations (i.e. that they are categorized in very similar ways). It rather claims that specific destinations are linked with specific purposes in everyday experience and can hence be inferred from contexts in which only the destination is made explicit. What then takes place in language use is a ‘meta-analysis’ in Croft’s (2006) sense: the purposive event is inferred as the salient part of the message and on ensuing usage events, the purposive part simply replaces the location after the directional expression. Notice, in this context, that the motion-cum-purpose constructions discussed above do not necessarily fall into this category. On the contrary, these tightly-packed constructions are specifically grammaticalized for purposive motion contexts and hence can often dispense with an overt purpose marker altogether. What we have been dealing with in the present section, by contrast, is the origin of overt purpose markers, which I have shown to lie in the experiential correlation of motion and purpose. Once created, such purpose markers can spread through the linguistic system and become analogically extended to all kinds of purposive contexts (‘rule generalization’). The motion-cum-purpose clauses, however, continue to fill their narrow though frequently tapped ecological niche. In the present discussion, they hence only served as evidence for the highly frequent co-occurrence of purpose and directed motion. We will now turn our attention to the overlap of purpose and ‘dative’ morphology, as exemplified by (96) from Abun. In this example, the dative (more precisely: benefactive) marker wa is first illustrated in its prepositional function, and secondly in its role as a purposive ‘conjunction’: (96) Abun (West Papuan: Indonesia; Berry and Berry 1999: 205) a. An fro nu yo wa men. 3sg prepare house det.indf for 1pl ‘She prepared a house for us.’ b. Marta tot su-gato gum do brer gwat ma [wa men git sugit mo]. Marta cut thing-rel name comp brer bring dir [for 1pl eat food on ‘Marta cut some brer (leaves) and brought them back for us to eat our food on.’ This area of polysemy is even more intricate since the term ‘dative’ has been used to refer to a cluster of individual participant roles which may, but need not be instantiated in the same way across different languages. Consequently, the attested overlaps between dative and purpose markers may have developed along different pathways, depending on the participant cluster from which they originate. Given

99

100 A Typology of Purpose Clauses

that there have been important advances in the recent literature on this topic, we will elucidate the relevant pathways in Chapter 4 and concentrate on conceptual motivations here. A fairly straightforward metaphorical relationship exists between allative and recipients since the latter can be conceptualized as the “human endpoints of a physical transaction” (Rice and Kabata 2007: 479–80). Since recipients often benefit from this transfer, it is intuitively plausible that (i) recipients and benefactives are often clustered for syntactic purposes, e.g. by a ‘dative’ morpheme, and that (ii) allatives can develop recipient and benefactive senses. Now, given that purposes are also often derivative of allative marking and, moreover, share the same goal-orientation as recipient-benefactives, we are not surprised to find synchronic overlaps which include all three notions. Yagua is a case in point: (97) Yagua (Peba-Yaguan: Peru; Payne and Payne 1990: 274) [Jáásiy-janu-jų] núúdya-jitǫǫ-jásiy. [cut.grass-inf-all 1ex-arrive-pst.prox1 ‘We arrived there to cut grass.’ As the glossing reveals, the postpositional suffix which renders a clause purposive is the allative marker jų. Precisely this marker also indicates benefactive meaning on NPs. (98) demonstrates the paradigmatic relationship clearly: (98) Yagua (Payne and Payne 1990: 379) a. ray-jų 1sg-all ‘for me’ (benefactive) b. jimyiy-jada-jų eat-inf-all ‘for eating’ (purposive) Imonda, similarly, has grammaticalized an all-embracing goal marker: (99) Imonda (Border: Papua New Guinea; Seiler 1985: 162) Tëla-l-na falgõ i-õb-n [iam maga-na uõl-m]. husband-noml-poss bow clf-pl-pst [later what-instr shoot-gl ‘She got her husband’s bow in order to shoot with it later.’ As we shall see later on, such cases are notoriously difficult to analyse historically since it is possible, indeed likely, that the various senses are not all related in a linear fashion (i.e. from allative to recipient-benefactive to purpose). But the metaphorical notion of ‘goal’ at least unites these markers conceptually and explains why they could be found on one or perhaps two closely related grammaticalization clines. A different explanation needs to be invoked for overlaps between benefactive and purposive morphemes that do not involve an intermediate allative or recipi-



Chapter 3.  The grammar of purpose 101

ent stage. In these cases, it is enlightening to realize that benefaction implicates the notion of purpose. As Rice and Kabata (2007: 481) argue, “when one acts for the benefit of another (or out of malevolence), he or she is usually acting purposefully.” This implicature can again be exploited in language use, in much the same way as allative marking in purpose clauses. More precisely, benefactive adpositional phrases or case-marked NPs are ‘meta-analysed’ (in Croft’s sense), so that the implied purposive component becomes a salient interpretational feature, as a result of which the formative in question is extended to the coding of purposive events. Such a paradigmatic extension may underlie the following marking pattern in Lavukaleve: (100) Lavukaleve (Solomons East Papuan; Terrill 2003: 161, 439) a. a-ham 3sg.m.obj-ben ‘for him’ b. a-vala-ham 3sg.m.obj-pull-ben ‘for pulling it/to pull it’ Metaphor theorists would probably be inclined to analyse the construction in (100b) as a reflection of construing a purposive event as a benefactive enterprise. Metonymically-oriented analyses, by contrast, would locate the origins of the purposive extension in the ‘context-induced reanalysis’ (Heine et al. 1991) sketched above. The latter may ultimately be better-equipped to deal with the Abun construction we saw in (96). In particular, the purpose clause in (96b) has the structure [for X to do Y], and for such constructions it may be argued that the initial for X used to form a constituent of the main clause, as in (101) S[I bought this book [for you]] S[to read]. Over time, the close association of the benefactive participant with the purposive event of the second clause may have led to a metonymic (i.e. part-whole) inference from participant to event, so that the benefactive participant becomes directly associated with the purposive event. As a result, the erstwhile preposition is reanalysed as a conjunction and can, crucially, then also introduce purpose clauses whose subject argument is left implicit (this does not apply, of course, to the English construction, but to Lavukaleve). In other words, the idea is that a benefactive marker emancipates from its corresponding NP via metonymic association with a purposive event and grammaticalizes into a purposive conjunction in its own right. It appears, therefore, that metonymic processes fit particularly well with the usage-based model underlying this study, as they emphasize the role of ­individual

102 A Typology of Purpose Clauses

usage events, i.e. cognitively grounded reanalyses of grammatical material in a specific context. This is not to deny the importance of metaphorical conceptualization, of course, but given the attention that is paid to metonymy in grammaticalization processes more generally (cf. Hopper and Traugott 2003: 92), it is probably safe to conclude that metonymy also plays a central role in the evolution of purpose markers. In Schmidtke-Bode (to appear [a]), I actually argue that metonymies have a yet more vital role to play in the expression of purpose. They are prominently involved in so-called purposive NPs, i.e. structures in which a benefactive-marked noun phrase functions as an economical shortcut for a complex purposive event. Such NPs commonly have an instrumental semantic role in the achievement of the implicit purpose (as in [102] from Yidiɲ) or, somewhat curiously perhaps, denote atelic temporal extension (as in [103] from Turkish): (102) Yidiɲ (Pama-Nyungan: Queensland/Australia; Dixon 1977: 342) Gawu: guga daŋgan buda:gu. tree.abs skin.abs take:off.imp blanket.purp ‘Take the Gawu: bark off [the Gawu: tree] for a blanket (i.e. to use it as a blanket).’ (103) Turkish (Turkic: Turkey; Kornfilt 1997: 263) Bir hafta-lığ-ın-a gel-dı-m. one week-abstr.n-cmpm-dat come-pst-1sg ‘I have come for one week (i.e. to stay here for one week).’ We will refrain from dwelling on those patterns here; the reader may kindly be referred to the above article for a detailed discussion. In sum, then, recourse to important notions from cognitive linguistics has provided us with powerful tools to motivate the frequent occurrence of allative and ‘dative’ (i.e. recipient and benefactive) markers in purpose clauses, both as primary and secondary gestalt features. We have anchored the use of these markers in arguably universal correlations of directed motion and purpose in everyday experience. The abundant existence of distinct motion-cum-purpose clauses constitutes first-rate evidence for the high frequency with which these correlations are verbalized in discourse. Under the assumption that language is shaped by our domain-general cognitive apparatus, the destination-purpose association give rise to particular conceptualizations of purposive action (‘conceptual metaphor’) and to inference processes in which a specific purpose is interpreted as a contextually salient association of a particular location (‘conceptual metonymy’). Both processes can result in paradigmatic extensions of goal-oriented nominal markers (allative or recipient relations) to purposive VPs. Apart from directed motion, such inference processes can also proceed from the notion of benefaction, which usually



Chapter 3.  The grammar of purpose 103

implies purposeful action. Again, it is probably a contextual reanalysis that leads to the paradigmatic extension of benefactive markers to verbal environments. The more precise diachronic processes, especially also their structural dimensions, will be outlined in §4.2.6.

3.4.3  ‘No marking’: Purposive inferences and ambiguities The final part of this section is devoted to those constructions whose purposive interpretation is not directly coded by overt linguistic material, but inferred instead from various cues in the discourse-pragmatic context of the construction. The cases to be discussed thus cover first and foremost the 14 instances which we previously assigned to the class of ‘no marking’ in Figure 10, but also a few others in which, despite overt marking, inferencing still plays an important role. Not surprisingly, these constructions are not uniform in their morphosyntactic properties, so that several construction types can be distinguished. The first type of construction that commonly invites a purposive interpretation is the well-known serial verb construction. In our sample, serial verbs are found in Mandarin Chinese (2), Semelai, Acehnese and Tukang Besi. As these languages are all located in East and South East Asia (particularly Indonesia), the purposive serial verb construction has to be regarded as a phenomenon with strong areal skewing. According to Payne (1997: 307), serial verbs reflect the tightest integration of two verbal situations, up to encoding different facets of a single event (cf. also Durie 1988: 3). For this reason, it is often mandatory that serial verbs share their notional subject argument, which holds for all serial constructions in our sample. Likewise, there are often tight constraints such that no linguistic material can be inserted between the two verbs in a series, as in Semelai (Kruspe 2004: 371). Importantly, the semantic interpretation of serial verbs is often non-compositional in the sense that the meaning of the whole verbal complex is more than just the sum of the meanings of the component parts. Thus in the following example from Mandarin Chinese (Sino-Tibetan: China), the serial verb invites different ­readings: (104) Mandarin (Chinese: China; Li and Thompson 1981: 595) Woˇ maˇi piào jìn-qu. I buy ticket enter-go ‘I bought a ticket and went in.’ [temporal sequence] ‘I bought a ticket to go in.’ [purposive] Notice that only the wider discourse context can clearly disambiguate this construction: while it is a logical requirement to buy a ticket in order to get in, only in the temporal sequence interpretation is there an entailment that the subject actually did go in. On the purposive reading, this is merely an implicature that could

104 A Typology of Purpose Clauses

easily be cancelled by the subsequent discourse. Sometimes the ­propositional d ­ ifference between the two interpretations can be more dramatic. In Tukang Besi, for instance, a serial verb construction can be ambiguous between a temporally simultaneous (i.e. depictive) and a purposive (i.e. posterior) situation: (105) Tukang Besi (Austronesian: Indonesia; Donohue 1999: 184) Te anabou iso no-wila no-kee-ngkee kua wunua. core child yon 3rl-go 3rl-rdp-hop all house ‘The child went hopping to the house.’ [depictive] ‘The child went to the house in order to hop.’ [purposive] The fact that most of the examples of serial verbs in our corpus have as their first part a verb of motion supports the association of motion and purpose outlined in the previous section. In fact, serial verbs in Semelai must contain a motion verb in order to evoke a purposive interpretation. If they do not, the verbal complex is usually understood resultatively, i.e. entailing that the second situation was actually realized. Thus compare (106a) and (106b): (106) Semelai (Austro-Asiatic: Malaysia; Kruspe 2004: 371–2) a. ‘purposive motion serial verb construction’ Ki = lɔh yɔk knɔn. 3a = go.across fetch offspring ‘He went across to fetch the child.’ b. resulative serial verb Ki = sayor cin. 3a = stew.with.vegetable be.cooked ‘She stewed the vegetable so it was cooked.’ It should finally be pointed out that the four languages having purposive serial verbs also possess at least one alternative construction in which purpose is overtly coded and thus made explicit. The second class of constructions comprises structurally heterogeneous cases in which overt marking of purposive semantics is dispensed with if it can be inferred from other morphemes in the complex sentence whose semantic properties are good predictors of purpose. According to Hawkins (2004: 44), such “stereotypical correlations […] maximize the ease of processing enrichments to a specific [meaning] in performance.” For example, in a number of constructions in our sample, the characteristic purposive conjunction can be omitted in performance, thus yielding a clause that is not overtly marked for purpose. Crucially, however, the omission of the conjunction is strictly confined to those cases where the matrix verb denotes motion. This pattern is attested in Hausa, Tiwi, Modern Greek and Punjabi. Similarly, in Noon, the purposive conjunctions doo or en ɗanaa are



Chapter 3.  The grammar of purpose 105

left out only if the verb of the main clause is a motion verb and if, additionally, the subjects of both clauses are co-referential: (107) Noon (Niger-Congo/Atlantic: Senegal; Soukka 2000: 284) Dí koor-in kaan-fu [ø ki-këñɗ-oh ɓo']. we-ex pass-prf house-rel.2sg [(purp) inf-greet-dur people ‘We passed by your house to greet people.’ In this case, the infinitival prefix, usually a secondary gestalt feature, has to function as the marker which constructs the complex sentence node in processing and attaches the subordinate clause; in addition, the matrix verb of motion functions as a processing enrichment because of the close correlation of purposes and destinations outlined previously. In a similar fashion, constructions that canonically do not have any purpose marker are often restricted to motion verbs in the main clause, such as purpose clauses with a bare verb stem in Tiwi, Lakhota, Kiowa and (possibly) Maori.22 Apart from motion verbs, various TAM markers can act as processing enrichments in purpose clauses without an overt purpose marker. In Nkore-Kiga, the subjunctive marker on the verb can help establish a purposive relation in the occasional absence of the conjunction ngu (Taylor 1985: 29), and in Lakhota, the specific combination of imperative matrix clause and future tense subordinate clause renders an otherwise unmarked clause combination ­purposive. The third type of purposive inference arises whenever a particular construction is multifunctional and purpose merely constitutes one of a restricted number of competing (adverbial) interpretations. This situation is commonly due to the ambiguity or vagueness of the subordinator and is attested for a number of constructions in the sample.23 As can be seen in Table 24, purpose markers and constructions can be potentially ambiguous or vague with respect to the adverbial meanings set out in Table 24. Some of these markers are clearly macro-functional 22.  In Maori, such deranked purpose clauses are characterized precisely by a bare “verb stem, non-introduced” (Bauer 1993: 66). However, there is a particle ai usually following the verb stem in purpose clauses, but it is not clear whether this marker can function as a reliable cue for constructing the complex sentence structure. Its distribution is very complex and its status as a subordinator can be debated (cf. Bauer 1993: 50). In this light, Maori infinitival purpose clauses may or may not fall into the present category of relying on processing enrichments, such as motion verbs. 23.  In the domain of grammar, it can at times be extremely difficult to draw a clear-cut dividing line between polysemous markers on the one hand (i.e. cases of conventionalized multifunctionality), and those with vague semantics on the other (i.e. cases of what Gil [2001] calls ‘macro-functionality’). All that matters for present purposes is to recognize that a marker simultaneously covers two functions, without committing ourselves to a specific semantic analysis of that marker (cf. also Haspelmath [2003: 231] on this point, who makes the important observation that semantic maps are also neutral to the distinction between polysemy and vagueness).

106 A Typology of Purpose Clauses

Table 24.  Adverbial multifunctionality of purpose markers (selected cases) Language

Marker

Functions

Tümpisa Shoshone Tümpisa Shoshone Kiowa Kiowa Lakhota Lakhota Noon Koyra Chiini Qiang Yaqui Tetun

-si(n) -sü -nɔ -gɔ c'a h'écel bi hal -xua -kai hodi

sequential marker purpose and result sequential marker (different subject) sequential marker (same subject) purpose, cause and result purpose, cause and result time and purpose (‘until’) time, purpose and result (‘until’) purpose and cause general subordinator (‘that’) coordinating conjunction (‘and’, time, purpose)

(i.e. vague) and hence can only be interpreted purposively by means of pragmatic inferencing. In Tümpisa Shoshone, for instance, purpose can be inferred from a non-finite sequential clause like the one in (108): (108) Tümpisa Shoshone (Uto-Aztecan: California, Nevada; Dayley 1989: 356) Hüüppiammü toya mantu mi'a [tüpanna tso'ikwasi]. women mountain to go [pinenut.p pick.seq(sub) ‘Women go to the mountains to pick pinenuts.’ The inference proceeds most likely along the lines of ‘things that occur together or in immediate temporal succession are often causally related’. Hence the sequential marker is open to causal and purposive interpretations, depending on the larger discourse context. As we can see in the list above, potential ambiguity between purposive, causal and resultative interpretations of a clause is not uncommon, and they will be discussed in their proper diachronic contexts in Chapter 4. A very interesting inferential pattern is found in the Fehan dialect of Tetun. In this language, a purposive interpretation can be induced by the coordinating linking verb hodi, which essentially behaves like the English conjunction and: (109) Fehan Tetun (Austronesian: West Timor; Van Klinken 1999: 306) Nia karian n-odi n-a-to'o naha bá uma laran. 3sg work 3sg-co 3sg-make-enough baggage to house interior ‘He works to supply things for inside the house.’ Although this pattern appears to stand out as a typological ‘rarum’ at first sight, we should set it into perspective by recognizing how frequently a coordinate ‘and’construction in familiar languages like English actually implies much more than the logical operation A ∪ B. Everyday sentences such as



Chapter 3.  The grammar of purpose 107

(110) John got up and closed the window. (111) She handed him the knife and he sliced up the bread. strongly implicate a purposive relationship between the two situations. In a recent quantitative analysis of Chafe’s famous Pear Stories (Chafe 1980), Croft found that typical purposive relations in the narrative were only coded by canonical English purpose clause constructions “approximately half of the time. In the other half, a coordinate structure is used, which almost always contains an overt conjunction” (Croft to appear [a]: 29), i.e. and. The extraordinary functional potential of English and has been discussed repeatedly and from various angles (cf. Posner 1980, Culicover and Jackendoff 1997, Stefanowitsch 1999, Carston 2002: ch.3), and hence it would not be too surprising if speakers from languages other than English also invited their interlocutors to make purposive inferences from coordinate conjunction or juxtaposition. These are patterns that usually exist alongside a canonical purpose clause and it is only these latter constructions that generally find entrance to a reference grammar. A  laudable exception is Kornfilt’s grammar of Turkish, where she mentions that the coordinating enclitic = DA ‘and’ “doesn’t just have the meaning of ‘and’, but also has a variety of shadings of meaning determined by the context” (Kornfilt 1997: 109): (112) Turkish (Turkic: Turkey; Kornfilt 1997: 110) Once iş-im-i bitir-eyim = de sinema-ya gid-eyim. first work-1sg-acc finish-1sg.opt = and cinema-dat go-1sg.opt ‘Let me finish my work first and (then) go to the movies.’ (‘…so that I can go to the movies.’) Asyndetic coordination with purposive inferences is found, for example, in Pirahã (Mura: Brazil; cf. Everett 1986: 265) and in an example from the literature. Longacre (1985: 261) discusses asyndetic coordination in (Chicahuaxtla) Trique, an OtoManguean language spoken in southern parts of Mexico: (113) (Chicahuaxtla) Trique (Oto-Manguean: Mexico; Longacre 1985: 261) [ža5h]. a. Ri3ki23 si3 ča3 gave he tortilla [will.eat.I ‘He gave me tortilla for me to eat.’ b. Ni5ko35ʔ ne3h po3li3sya23 [gi4da3ʔa34 ni3] followed the police [will.seize they [ga4ča4 ni3 du3gwa2gaʔa3]. [will.put they jail ‘The police followed him in order to seize him and put him in jail.’

108 A Typology of Purpose Clauses

Notice that the occurrence of the anticipatory mood translated as English ‘will’ is not a distinct gestalt feature of purpose clauses; it readily occurs in independent main clauses as well. Therefore, a construction like ‘he will-grab it, he will-eat it’ can be construed as both ‘He will grab it and eat it’, or ‘He will grab it in order to eat it’. Subtle hints that bias the hearer towards a purposive interpretation can be found in some coordinate constructions in Kewa (Engan: Papua New Guinea). Canonical coordinate clauses in Kewa are arranged iconically: the clause encoding the temporally (and logically) anterior situation precedes the clause encoding the subsequent situation. In some coordinate clauses, however, the order is reversed, and it is these clauses that strongly suggest purposive interpretations: (114) Kewa (Engan: Papua New Guinea; Franklin 1971: 110) [Go raí méá-ina] ípu. [this axe get-co come-imm ‘Come and get this axe.’ The intuition that the clause is meant to be interpreted purposively is supported by the fact that all adverbial clauses, but not standard coordinate clauses, usually precede their main clause in Kewa. The initial position of the temporally posterior clause in the above example may thus prime the reader for an adverbial (here: purposive) interpretation of the construction. I would like to emphasize that the present discussion should not be taken to imply that every coordinate clause construction can yield a purposive interpretation. It appears to me that it is rather a specific combination of linguistic features that strongly invites this inference. Tentatively, I would suggest the following features as a prototypical constellation in this scenario. First, the aspectual type construal of each clause (cf. Croft to appear [b]) should be dynamic rather than stative, as the semantic oddity of (115) shows: (115) ?I knew the answer and ( = so that) I could tell him. Notice, however, that if the antecedent clause contains a stative verb of suppressed motion, a purposive interpretation is perfectly possible: (116) I stayed/remained in the club in order to continue to enjoy all privileges. Second, the semantic role of the antecedent subject is typically animate and agentive in order to be able to act intentionally and retain a high degree of control over the dependent situation: (117) ?I fell off the ladder and broke my leg. [not interpreted purposively] Third, if both clauses contain the same subject, they satisfy a typical feature of purpose clauses, as we saw earlier. If they do not, the purposive inference is less readily drawn:



Chapter 3.  The grammar of purpose 109

(118) ?John went outside and we started smoking. In this case, it is not very likely that John went outside intentionally for the others to be able to smoke, at least one would be hard-pressed to draw this inference from and alone; the more salient reading seems to be a purely resultative one, such that being able to smoke is, in fact, construed as a result of John’s departure, but not necessarily one intended by John (cf. also Jones 1991: 46–7 for a related point). From the languages of our sample, Semelai can express purposive relations by asyndetic coordination, but only if the subjects of the two clauses are co-referential: (119) Semelai (Austro-Asiatic: Malaysia; Kruspe 2004: 374) Ki = lagət, jʔɔy rakit, ʔluc sbɒrɒŋ. 3a = break.off make raft pass.to otherside ‘He broke (it) off to make a raft to pass to the otherside.’ This would lend support to the hypothesis of shared subject arguments in purposive coordination. Unfortunately, we do not have sufficient cross-linguistic data in order to fully explore the extent to which languages make use of purposive inferencing from coordination. Finally, I shall at least mention the remaining types of constructions in which purpose is inferred rather than coded. Huallaga Quechua (Quechuan: Peru) has a construction for expressing negative purpose without any negative morpheme. ‘Lest’ semantics is thus inferred from the context. Georgian (Kartvelian: Georgia) has a very productive construction of direct quotation which is “widely used, particularly in speech, in place of other, more specific constructions” (Hewitt 1995: 615). Purpose seems to be one of these more specific meanings evoked by direct quotes: (120) Georgian (Kartvelian: Georgia; Hewitt 1987: 28) Mo-nadir-e-eb-i c'a-mo-vid-nen, da-v-i-sven-o-t-o. hunters.nom they-came-away-aor let-us-rest-aor.subj-they-said ‘The hunters came away in order to rest.’ Again, there is no overt marking of purposive semantics, yet this is precisely what the clause appears to convey. This example, then, concludes the section on inferring purposive relations in the absence of overt linguistic indications thereof, and it also rounds off the detailed survey of purpose markers and cross-linguistically typical gestalt features of purpose clauses. We will now turn to properties relating to the purpose clause in its entirety, i.e. some semantic and pragmatic characteristics of purposive constructions and their position vis-à-vis the main clause.

110 A Typology of Purpose Clauses

3.5  Positioning patterns of purpose clauses While the aforementioned coding properties of purpose clauses had already been discussed to varying degrees in the typological literature, the position of purpose clauses vis-à-vis their associated matrix clauses is less well-studied. An important precursor is Greenberg (1963). In his famous essay on constituent order typology, Greenberg investigates purpose clauses of 30 languages and arrives at the following conclusion: (121) Universal 15 (Greenberg 1963: 84) In expressions of volition and purpose, a subordinate verbal form always follows the main verb as the normal order except in those languages in which the nominal object always precedes the verb. The order of purpose and main clause contrasts with the linear order in conditional constructions, in which the subordinate protasis always precedes its apodosis (Universal 14). Greenberg’s explanation for this tendency is very straightforward, invoking the notion of iconicity: “Universals 14 and 15 possibly illustrate the same principle. The order of elements in language parallels that in physical experience or the order of knowledge.” (Greenberg 1963: 103) The objective of this section is to test Greenberg’s proposed universal against a larger empirical basis and to evaluate Greenberg’s explanation in light of recent advances in the contemporary functional-typological literature. I will show that the descriptive part of the alleged universal can still be subscribed to, while the explanation will have to be refined. Specifically, I claim that (i) processing factors play an important part in the overall distribution of final and initial purpose clauses, while (ii) the strong cross-linguistic tendency to place purpose clauses after the matrix clause is also motivated to a large extent by discourse-pragmatic factors.

3.5.1  Data and analysis We will approach the positioning patterns of purpose clauses at the constructionspecific level first. The central question here is whether individual purposive constructions show particular preferences in their position with respect to the main clause. Perhaps counter-intuitively, there is quite a wide range of both logical and actually attested possibilities for these preferences. First of all, there are complex sentence constructions in which the position of the purpose clause is entirely flexible. In other cases, the purpose clause must appear in a particular position, either rigidly preceding or rigidly following the main clause. Some constructions centreembed the purpose clause in the sense that the main clause is split into two parts by the intervening purpose adjunct. A case in point would be Warao:



Chapter 3.  The grammar of purpose

(122) Warao (isolate: Venezuela; Romero-Figueroa 1997: 18) Ine [ure buha-kitane] nao-a–e. 1sg [malanga dig.out-noml come-pnct-pst ‘I came to dig out malanga.’ In this example, the purpose clause is embedded between the subject NP and the predicate VP of the main clause. As we shall see below, this pattern is very common for non-finite, particularly nominalized, purpose clauses. Although in the example quoted here the purpose clause literally occupies the middle position in the sentence, there is usually a tendency for the purpose clauses of such discontinuous constructions to strive towards one or the other end of the sentence. Another class is formed by those purpose clauses which normally occur in one position vis-àvis the main clause, but can be shifted under some marked pragmatic conditions. Korean and Lavukaleve, for example, have purpose clauses which follow the main clause only if they provide an afterthought to the information of the preceding main clause. Compare the following sentences from Lavukaleve: (123) Lavukaleve (Solomons East Papuan; Terrill 2003: 440) a. [Kiu-ham hide fi] a-vo. [die-purp thus 3sg.n.foc 1sg.s-come ‘I have come like this to die.’ b. Ma-kalem na hano ve-m. [Siria-ham]. 3pl.poss-father.m sg.m.art then go-sg.m [catch.fish-purp ‘Their father had gone. To catch fish.’ Finally, the question for positioning patterns is not really applicable to a tiny minority of cases because they express purpose by a special main-clause construction that resists being analysed in full-fledged biclausal terms. An overview of the distribution of the above-mentioned ordering patterns is given in Table 25. Table 25.  Attested positioning patterns of purpose clauses (N = 218) Position type

Absolute frequency

Relative frequency (in %)

flexible rigidly postposed postposed or discontinuous non-rigidly postposed rigidly preposed preposed or discontinuous non-rigidly preposed NA

31 125 5 6 29 17 3 2

14.2 57.3 2.3 2.8 13.3 7.8 1.4 0.9

Total

218

100.0

111

112

A Typology of Purpose Clauses

Table 26.  Preferred positioning patterns of purpose clauses (N = 218) Position type

Absolute frequency

Relative frequency (%)

flexible postposed preferred preposed preferred other

31 136 49 2

14.2 62.4 22.5 0.9

Total

218

100.0

As a first impressionistic result, we can ascertain that in 57.3% of all cases, the purpose clause must be placed after the main clause. Given the rather large amount of logically possible patterns, this is a very strong tendency. In order to obtain more insightful and manageable categories, I  conflated some of the above values into one category. By this procedure, we can basically distinguish four conventionalized preference types for the position of purpose clauses within complex sentences. They are outlined in Table 26. On the construction-specific level, then, we can discern a strong preference for postposing purpose clauses after the main clause (62.4%). If we want to evaluate whether this preference arises as a result of particular languages contributing more than one postposed purpose clause, we have to replicate the above procedure on the level of languages (N = 80). Here we can essentially distinguish whether the majority of purpose clauses within a language is pre- or postposed, whether purpose clauses are mostly flexible, or whether the language shows a mixed behaviour (the same amount of pre- and postposing constructions, or a mixture of preposed, postposed and flexible purpose clauses). Astonishingly, we get almost exactly the same distribution, as visualized in Figure 19. As we can see, in the majority of cases (64%), languages have conventionalized preferences for final purpose clauses. That is, even if there is more than one purpose clause construction in such a language, the majority of these constructions occurs sentence-finally.

7; 9%

4; 5%

post preferred

18; 22% 51; 64%

pre preferred flexible mixed

Figure 19.  Preferred positioning patterns of purpose clauses (N = 80)



Chapter 3.  The grammar of purpose

However, the fact that purpose clauses tend to be postposed is not necessarily a revealing finding per se. It is well-known among typologists that the order within complex linguistic constructions often correlates with the basic constituent order of the respective language. Specifically, it has been argued that the order of verb and direct object correlates with the internal order of adpositional phrases, noun phrases and even complex sentences (cf. Greenberg 1963, W. Lehmann 1973, Vennemann 1973, 1974 and 1976, Hawkins 1983, Dryer 1992). While Vennemann captures these cross-categorial implications by the generalization that all of these categories contain a head and a dependent element, and that languages tend to be either head-initial or head-final with respect to these categories, Dryer slightly refined this proposal by claiming that these categories uniformly have a branching and a non-branching element, and that languages tend to be either consistently left-branching or right-branching for ease of processing. On both accounts, socalled OV languages would be expected to have postpositions after a branching NP, genitive phrases that precede the head noun, and subordinate clauses which, by virtue of being dependent and branching elements, precede their associated main clause, which acts as the head of a complex sentence construction. So-called VO languages are most likely to show the reverse patterns. Although many languages are not perfectly consistent in terms of cross-categorial harmony, the patterns suggested have a very strong empirical basis (e.g. Dryer 2005b) and do represent statistical language universals. In this light, sentence-final purpose clauses would be expected to be the default pattern in VO languages, whereas OV languages would be thought to prefer sentence-initial purpose clauses. This is, mutatis mutandis, what Greenberg’s Universal 15 predicts. It can be tested if we correlate the preferred positions of purpose clauses in each language with the basic constituent order of the language in question. Just as in previous analyses of word order (cf. §3.4.1 above), we shall distinguish VO languages (comprising SVO, VSO and VOS), OV languages (comprising SOV, OVS and OSV), and ‘other’ languages (i.e. those that are organized in a mixed or pragmatically determined [‘flexible’] way). The results of this cross-classification are summarized in Table 27 and can be brought out more clearly by a spine plot (Figure 20). Table 27.  Basic constituent order and the preferred position of purpose clauses (N = 80) Constituent-order type

Position of purpose clauses Flexible

Mixed

Post

Pre

Total

OV VO Other

2 0 5

4 0 0

15 25 11

18  0  0

39 25 16

Total

7

4

51

18

80

113

1.0

114 A Typology of Purpose Clauses

0.6

0.8

pre Pre

0.2

0.4

post Post

flexible other (N = 16)

OV (N = 39)

VO (N = 25)

0.0

mixed

Figure 20.  Basic constituent order and the preferred position of purpose clauses (N = 80). The positioning patterns correspond to the shadings in the following way: preposing = black; postposing = light grey; mixed = white; flexible = dark grey.

The distribution in Figure 20 is highly significant under a randomization test (χ2 = 44.29, df = 6, p < .001, B = 100,000, Cramer’s V = 0.526), and it is obvious at a glance where the strong skewing comes from. All VO languages in the sample postpose their purpose clauses, in keeping with the expected word-order correlations, but the mirror image is not borne out by the data: OV languages do not uniformly prepose purpose clauses, but instead exhibit pre- and postposing with about equal frequency. This split is also found at the level of individual OV languages, as they constitute the only word-order type whose purposive constructions can be mixed, some being pre- and some being postposed. Interestingly, when languages do not generally constrain their constituent order in syntactic ways (i.e. ‘other’ languages), purpose clauses are also predominantly postposed, and not as flexible as we would expect. Since the data are extensive enough and come in relatively few categories this time, we can also investigate whether this structural skewing holds for all major geographical areas. Based on an argument by Dryer (1989), Cysouw (2005: 573) advocates that it is “of the uttermost importance that typological correlations are investigated as to their areal distribution. What is needed is both statistical significance and areal independence.” To this end, all macro-areas were investigated individually with regard to positioning patterns, resulting in the display in Figure 21. The uniform trend for postposing in VO is indeed independent of areal factors. Also, OV languages mostly turn out to be ‘torn’, as it were, between pre- and post-

Figure 21.  Geographical distribution of significantly skewed positioning patterns. In order to enhance the readability of this graph, North America and Mesoamerica were conflated into one areal category. Each of the individual spine plots conforms to the design of Figure 20 above. Specif-

ically, the order of the bars from left to right is always ‘other’-OV-VO. The shadings are black at the top for preposing, (light) grey for postposing, white for mixed ordering patterns, and dark grey at the bottom for flexible position.

Chapter 3.  The grammar of purpose 115

116 A Typology of Purpose Clauses

posing, thus reflecting the aggregate trend to different degrees (compare, for instance, Eurasia and North America in this regard). Interestingly, the word order correlations are perfectly borne out in Oceania and New Guinea, but they are even more strongly overruled in the languages of Australia. On this continent, all OV languages have postposed purposes clauses. Notice that Greenberg’s universal is not in any way jeopardized by these findings; he phrased Universal 15 carefully enough to accommodate all of the exceptions. If we assume, however, that languages tend to have consistent branching directions in complex constructions for ease of processing, then the rather large amount of purpose clauses that follow their main clause in OV languages is certainly anomalous. It would appear, therefore, that other factors can override the drive for cross-categorial harmony in positioning patterns. In what follows, we shall investigate some possible candidates for these factors.

3.5.2  Explanation and discussion As mentioned at the beginning, Greenberg’s explanation for the sentence-final position of purpose clauses rests solely in iconic motivation: postposed purpose clauses reflect the principle of iconicity of sequence since both temporally and logically, the main clause action has to be carried out first before the intended result expressed in the purpose clause can be achieved. Iconicity was introduced briefly in Chapter 2 as an important explanatory concept in the cognitive-functional literature; it will now be examined in more detail. To begin with a standard definition, a linguistic sign is said to be iconic if its form directly reflects some properties of its denotatum (this goes by the name of isomorphism), or if the internal formal structure of a linguistic construction reflects the configuration of elements in the corresponding semantic structure (a relationship known as diagrammatic iconicity). Iconicity, then, establishes a motivating link between language-specific formal expressions and semantic notions found in regions of the universal conceptual space. Therefore, it appears to be well-suited as a tool of explanation in cross-linguistic research. Iconicity has been invoked in explanations for a diverse range of linguistic constructions (see Haiman [ed.] 1985 for an array of case studies and Pusch 2001 for a more general overview), and, interestingly, it has also been claimed that iconicity is ultimately conducive to efficient language processing: “All other things being equal, a coded experience is easier to store, retrieve, and communicate if the code is maximally isomorphic to the experience.” (Givón 1985: 189) Despite this rather long and successful tradition of iconicity in the functional literature, however, the concept is not uncontroversial and has most recently come under attack, even from otherwise functionally-minded



Chapter 3.  The grammar of purpose

linguists. In particular, Haspelmath (2008c) has argued that some coding asymmetries for which iconicity had often been invoked as a motivating force can be captured more accurately in terms of frequency effects (i.e. economy). Of course, Haspelmath’s critique does not imply that frequency and economy should entirely replace iconic accounts of language structure. In some cases, frequency does not appear to play any role, as when forms that semantically or conceptually belong together appear syntactically contiguous (e.g. aspect markers that commonly occur closer to the verb stem than tense markers, cf. Bybee 1985: 13). It is thus rather important to distinguish between several ‘types’ of iconicity (e.g. iconicity of distance) and evaluate each of those types with regard to its explanatory adequacy. If we focus on the type relevant to ordering distributions in complex sentences, i.e. iconicity of sequence, recent empirical data give us reason to believe that iconic ordering influences the processing of complex sentences, indeed. Diessel’s (2008) corpus study shows that the preferred positioning patterns of temporal adverbial clauses in English vary with the specific type of temporal information they encode. Moreover, a stepwise logistic regression model reveals that iconicity of sequence excels in predicting these individual preferences: as compared to other independent variables, iconicity can account for the lion’s share in the underlying variation of each clause type. In addition, Diessel also shows that pre- and postposed temporal clauses adhere to iconic ordering to significantly different degrees, and his explanation suggests that, just as Givón theorized, iconicity is an important factor in the on-line processing of complex sentences. Although this should make us optimistic, there is an inherent problem with iconicity if applied to purpose clauses. While it is certainly iconic for the purposive clause to follow the matrix clause because the former spells out the result of the matrix action, it could also be (and indeed has been) claimed that even preposed purpose clauses are iconic: after all, the conceptualization of the purpose necessarily precedes the action (intention > action > result, cf. Hwang 1995). On this account, both ordering patterns look iconic in some respect, but this, of course, makes iconicity void as an explanatory principle; it simply shifts the burden to explaining why one of the two iconic conceptualizations is apparently preferred across the world’s languages. Perhaps, then, a potential pitfall lies in postulating a priori that a surface structure which looks iconic is also iconically motivated. For the position of purpose clauses within complex sentences, this means that before we can claim that iconicity motivates sentence-final position, we should have sufficient reasons to exclude other factors that may favour this particular linear order. Let us consider the possible factors in turn. The important processing factor of consistent branching directions across all complex syntactic constituents of a language has already been ruled out because

117

118

A Typology of Purpose Clauses

of the significant amount of postposed purpose clauses in OV languages. Another factor that has been invoked for explaining the postposing or right extraposition of clauses is their length. There seems to be abundant evidence that speakers of many languages shift particularly long or ‘heavy’ elements towards the right end of the constituent of which they are part (De Smedt 1994, Wasow 1997). Subordinate clauses, on this account, are often heavy elements that would be expected at the end of the complex sentence. If valid from a cross-linguistic point of view, this principle would lead to extraposition in OV languages and thus nicely capture the great amount of postposed purpose clauses. On the face of it, some data in our sample would appear to corroborate this intuition. Compare the following examples from Punjabi: (124) Punjabi (Indo-European: India; Bhatia 1993: 74–5) a. Mãi [shikaago vekhaN] giaa. 1sg [Chicago see-inf.obl go-pst.m ‘I went to see Chicago.’ b. Ó ne kamar TaTolii [ki rakam Thiik bánnii 3sg erg waist.f.sg feel-pst.f.sg [purp money fine tie-ptc.pst.f.sg páii sii jaa náii]. sit-pst.f.sg was or neg ‘He felt his waist (to see) if the money was tied safely (to his waist).’ Whereas in (124a), the purpose clause is very short and placed before the main verb, the one in (124b) is rather lengthy and appears after the whole main clause. Notice, however, that we are dealing with two entirely different constructions here: the first is a non-finite (infinitival) purpose clause with an equi-deleted subject, while the second is a finite purpose clause with an overt purposive conjunction and an explicit subject. Each of the two constructions is maximally adapted to its own ecological niche: The economical construction is used for highly predictable same-subject, preferably motion-matrix contexts, and due its reduced status, it comes close to verbal complements and can also appear in the preverbal position canonically associated with such sentential objects. The more elaborate construction, by contrast, needs overt subjects due to non-coreferentiality and hence also more roundabout expression. It then cannot easily be incorporated into the main clause and is thus postposed. Purpose clauses in Punjabi, therefore, are ‘mixed’ with regard to their positioning patterns. Even if length were to be invoked for the construction in (124b), it could not account more generally for OV languages in which all purpose clauses are rigidly postposed, no matter whether they are long or short, finite or non-finite, same-subject or different-subject. In these cases, there simply is no option of preposing a purpose clause, even if it is short. This



Chapter 3.  The grammar of purpose 119

­appears to hold, for instance, in Ungarinjin (Pama-Nyungan: Australia), Yaqui (Uto-Aztecan: Mexico) and Epena Pedee (Chocó: Colombia). What is more, in these and quite a few other languages, purpose clauses are not longer on average than temporal and conditional clauses, but are still postposed, while temporal and conditional clauses preferably precede the main clause. Therefore, it is unlikely that length or heaviness are highly ranked motivations for postposing purpose clauses after the main clause. A more important factor for the position of adverbial clauses has recently been brought to our attention. In a cross-linguistic study of 40 languages, Diessel (2001) shows that there is a significant correlation between the position of the subordinating element in adverbial clauses and the position of the entire adverbial clause vis-à-vis the main clause. More precisely, Diessel argues that if a language employs clause-final adverbial subordinators, the adverbial clause precedes the main clause. If, by contrast, a language employs clause-initial subordinators, the position of adverbial clauses tends to be mixed: they occur before and after the main clause. This generalization entails that in the majority of cases, the subordinating element occurs right in between the clauses that it combines. Hawkins (1994, 2004) argues that such a linear arrangement of constituents facilitates the on-line parsing of complex sentences. Specifically, Hawkins suggests that the subordinator constitutes an important processing node which helps the hearer to recover that the unit to be parsed is a subordinate (adverbial) clause; at the same time, this provides a cue to the immediate constituent (IC) structure of the sentence as a whole. That is, once the hearer encounters an (adverbial) subordinator, it is immediately recoverable that the whole structure to be parsed is a complex sentence (IC ‘construction’), and that the constituent currently parsed needs to be integrated with a matrix clause (‘attachment’). Notice, therefore, that while (adverbial) subordinators are ‘mother node constructing categories’ (MNCCs), the associated main clause does not carry any overt cue to the complex sentence structure. Hawkins then goes on to claim that efficiency plays an important role in on-line processing. On his view, processing is facilitated if IC construction and attachment can be made by the hearer as early as possible. This can be captured quantitatively in so-called ‘constituent recognition domains’ (Hawkins 2004: 32), i.e. roughly the amount of linguistic elements (words or morphemes) that must be parsed “in order to recognize […] all ICs of a phrase once the parser has recognized the mother node M of the phrase.” (Diessel 2005: 456) Let me illustrate this line of reasoning with a concrete example, a complex sentence with a postposed purpose clause from Noon:

120 A Typology of Purpose Clauses

(125) Noon (Niger-Congo/Atlantic: Senegal; Soukka 2000: 283) M IC1

IC2

dii] sub[doo ɓuwaa mín-u ki-hot]. [sit-imp here [purp people can(aux)-pl inf-see main sub ‘Sit here so that people can see.’ main[Yug-aa

In this example, the sentence begins with the matrix clause, but there is no element in the structure that allows the hearer to recognize that this is, in fact, IC1 of a complex sentence (hence the broken line between M and IC1). The subordinate clause then begins with an initial conjunction, and it is at this point that the mother node M can be constructed. As indicated by the bold line, the conjunction allows for the construction of the mother node and for the recognition that a second IC will now be attached to M. From this perspective, it is optimal to have the conjunction as a linking morpheme between IC1 and IC2. In other words, once the last word of the main clause has been processed, it really only needs one additional morpheme, i.e. the subordinator, until construction and attachment can take place. The IC-to-morpheme ratio is thus 2: 2 = 1, a ‘minimal domain’ in Hawkins’ (2004) terminology. Now, imagine the adverbial clause were preposed, with the conjunction remaining in clause-initial position. In this case, there would be an immediate signal at the beginning of the sentence that a complex-sentence M will have to be constructed; crucially, however, this leads to the on-line expectation that a main clause will have to follow, but its attachment to M is now delayed until after the entire subordinate clause has been processed. Here is an example of an English temporal clause to illustrate this principle: M

(126) IC1

IC2

[When I went to the cellar without the light switched on], the door slammed behind me. In this sentence, the attachment of IC2 needs 11 (!) words (or 12 morphemes) after the construction of IC1 and M; the constituent recognition domain (2:12 = 0.16) is thus fundamentally different from the 1: 1 relation we had before. This is, of course, not necessarily problematic (as the ample attestation of preposed adverbial clauses in English demonstrates), but the structure in (126) is simply less efficient than the



Chapter 3.  The grammar of purpose

one in (125). On this account, preposed purpose clauses should preferably have their subordinator at the end of the clause, as in Abkhaz: (127) Abkhaz (Northwest Caucasian: Georgia; Hewitt 1989: 42) [S-y°əza də-z-bà-r+c] [à-kalak’ (a-)ax’ s-co-yt’]. [my-friend him-I-see-purp [art-town it-to I-go-fin ‘I am going to town to see my friend.’ Here, too, the IC-to-morpheme ratio is 2 : 2 = 1 again, the optimal or most efficient value in Hawkins’ matrix. If Diessel’s (2001) hypothesis is correct and efficient IC construction and attachment were, indeed, crucial for the position of adverbial clauses, then we would expect that purpose clauses precede their matrix clause if they are marked by a clause-final subordinator and follow if they are marked by a clause-initial subordinator. If there is a significant bi-directional correlation to this effect, then online processing pressures would account for particular positioning patterns of purpose clauses. In order to test this hypothesis, we first determine the precise position of each marker. Figure  22 illustrates which positioning classes of subordinators can be distinguished and how often the different marking types fall into these categories. These figures demonstrate that most purpose clauses place their primary marker at the clause-boundary, either in clause-initial (33.0%) or clause-final (51.4%) position. Only rarely do purposive markers embrace the purpose clause as a circumfix (0.9%), or occur in clause-medial position (2.3%). The category ‘flexible or 120 100 80 60 40 20 0

S-SUB

SUB-S

circum

medial

112

72

2

6

flexible or no marking other 9

17

Figure 22.  Position of the primary marker (‘subordinator’) in the purpose clause (N = 218)

121

122 A Typology of Purpose Clauses

Table 28.  Position of the purpose marker and the purpose clause Position of the subordinator

Position of the purpose clause Flexible

Post preferred Pre preferred Other

Total

No marker SUB-S S-SUB Circum Medial Flexible or other

1 7 17 0 4 2

13 65 48 2 2 6

2 0 47 0 0 0

1 0 0 0 0 1

17 72 112 2 6 9

Total

31

136

49

2

218

other’, finally, covers those 4.1% of all cases in which the marker is either flexible with regard to its position in the clause or occurs on the main clause (e.g. as a clitic). Once those positioning classes are in place, we can correlate them with the position of the whole clause in which they occur. The results are displayed in Table 28. This highly significant distribution (χ2 = 88.19, df = 15, p < .001, B = 100,000, Cramer’s V = 0.367) has several interesting frequency asymmetries to offer, which can be formulated as implicational generalizations. (i) If a language has no overt primary marker of subordination, it tends to place the purpose clause after the corresponding matrix clause (13∕17 = 76.5%). That is, in the absence of any construction-and-attachment marking, the most natural position is for the purpose clause to follow its matrix. (ii) If a language has a clause-initial purpose marker (SUB-S), the purpose clause follows the associated matrix clause (65∕72 = 90.3%). This confirms Diessel’s predictions and can neatly be captured with Hawkins’s (2004) ‘minimize domains’ account. (iii) If a language generally prefers preposed purpose clauses, the purpose marker almost always occurs in clause-final position (47∕49 = 95.9%). Again, this finding is harmonious with Diessel’s and can also be explained by processing considerations. However, the reverse implication does not hold: If a language has clause-final purpose markers, both preposed and postposed purpose clauses are attested with about equal frequency (47∕112 and 48∕112 = 42% and 43%, respectively). According to Diessel, we would expect clause-final subordinators only in preposed purpose clauses. As a result, we get the surprising finding that although the subordinator in postposed purpose clauses often occurs at the beginning of the purpose clause (65∕136 = 47.8%), it commonly also occurs at the very end of it (48∕136 = 35.3%). A visual summary of those implications is provided in Figure 23. In the case of postposed purpose clauses with a final subordinator, it appears that the processing preference for early immediate constituents is again overridden by some other factor since the domain between the subordinating ‘linking’ element and the associated main clause is maximized, not minimized. Cru-



Chapter 3.  The grammar of purpose 123

1.0

cially, these cases resemble the adverbial clauses in Diessel’s (2001) study which have initial subordinators yet are still not postposed. Diessel shows that this particular pattern affects only adverbial clauses with a specific discourse-pragmatic function, chiefly temporal and conditional clauses. These clause types commonly set a situational framework for the interpretation of the main clause and this unique discourse function motivates their position before the main clause (cf. also Chafe 1984). In information-structural terms, conditional and especially temporal clauses resemble (‘scene-setting’) topics (cf. Lambrecht 1994: 125, Thompson and Longacre 1985: 231–2) or themes (Halliday 1967, Fries 1983) in that they provide presupposed (‘given’) information as a “point of departure of the clause as a message” (Halliday 1967: 212). Building on this line of research, I suggest that purpose clauses can be conceived of as the mirror image of conditional and temporal clauses in discourse. Purpose clauses do not usually set an interpretative frame or thematic ground for the ensuing main clause, but tend instead to provide new information. Their discourse-pragmatic function, then, should favour purpose clauses in rhematic (focus) position rather than thematic (topic) position, and this is exactly what we find in the data: purpose clauses are overwhelmingly ­frequently postposed, even if they contain a clause-final subordinator and hence require a comparatively high processing effort. An illustrative example comes from ­Ngi­yambaa (see 128).

0.6

0.8

pre

0.2

0.4

post

no marker

0.0

flex SUB-S

S-SUB

Figure 23.  Major correlations between the positions of the primary marker and the purpose clause (N = 218)

124 A Typology of Purpose Clauses

(128) Ngiyambaa (Pama-Nyungan: Australia; Donaldson 1980: 284)

IC-to-morpheme ratio = 2∕4 = 0.5

[ŋadhu yana-ɲa] [ŋindu-dhan guɽuŋa-giri]. [1sg.nom go-prs [2sg.nom-evid swim-purp ‘I am going [expressly] so that you can swim.’ The claim that information-structural pressures heavily influence the position of purpose clauses is a prime example of anchoring quantitative typological data in patterns of language use. Consequently, the claim is amenable to the performancegrammar-correspondence (PGCH) perspective sketched in §2.2.2. Under the assumption that grammatical systems conventionalize or ‘freeze’ preferred choices in language performance, the PGCH approach encourages us to seek ‘converging evidence’ from grammatical and usage data. Starting with the performance data first, we are lucky to be able to draw on a valuable piece of discourse-analytical work by Thompson (1985). Her study of infinitival purpose clauses in written English discourse provides clear evidence ex negativo for the rhematic function of purpose clauses. Thompson sampled a corpus of written texts comprising three narratives, two procedural texts and one M. A. thesis, from which she extracted all infinitival purpose clauses (N = 185) to investigate their function in the wider discourse context in which they occur. Unlike balanced so that clauses, infinitival purpose clauses (to, in order to) can occur in both positions relative to the matrix clause in English. While the majority of the clauses analysed follow the associated matrix clause, there is a considerable amount of preposed purpose clauses in the data. Note that since English is a rigid VO language, preposed subordinate clauses with initial subordinators violate all processing pressures we have seen so far: they are non-harmonic by virtue of being preposed dependents, they are non-iconic in the standard interpretation and they create inefficient IC2-attachment domains. Thompson argues convincingly that these apparent violations go hand in hand with a fundamental shift in the function of the purpose clause. Canonical final purpose clauses are limited to the (narrow) function of modifying the proposition in the preceding matrix clause. They provide the very motive for the main clause action and as such typically introduce new, focal information into the ongoing discourse. When the purpose clause is moved to the beginning of the complex sentence, however, it loses these rhematic characteristics and adopts the discourse-organizing function of other initial adverbial clauses: “The initial purpose clause helps to guide the attention of the reader, by signalling, within the portion of text in which it occurs, how the reader is expected to associate the material following the purpose clause with the material preceding it. The final purpose clause does not play this role.” (Thompson



Chapter 3.  The grammar of purpose 125

1985: 61, emphasis in original). Thompson demonstrates that in such initial purpose clause constructions, it is usually the case that the preceding discourse establishes a set of expectations on the part of the reader. The purpose clause then spells out a problem connected to these expectations, and the ensuing discourse (i.e. the main clause and often even further material) typically provides the solution to this problem. Frequent encounters with such constructions can be made in instruction manuals or cooking recipes, but also in narratives, as in the following example: (129) The Brendan Voyage (cited in Thompson 1985: 62) Brendan was rushing madly farther and farther out to sea. To slow her down we streamed a heavy rope in a loop from the stern and let it trail in the water behind us to act as a brake […]. The preceding discourse context creates a set of expectations – a boat crew being in a dangerous situation – and the purpose clause then names a problem connected to this set of expectations: ‘How can they slow the boat down?’ The remaining material, i.e. the ensuing main clause, provides the solution. Initial purpose clauses, therefore, constitute a link in an expectation chain and thus provide a framework for the interpretation of the main clause which follows them, a discourse function not unlike that fulfilled by thematic (‘scene-setting’) temporal and conditional clauses. Because of this broader orientation function, initial purpose clauses also typically have more than just the immediately following main clause in their scope. Consider the following example from an instruction manual: (130) Scope of initial purpose clauses (Thompson 1985: 64) To true a blade, hold the steel firmly in the left hand, thumb on top of handle. Hold the hand slightly away from the body. Hold the knife in right hand, with the point upward. Place the heel of the blade against the far side […]. Taking a text with 70 initial purpose clauses, Thompson quantified that the average number of clauses that the purpose clause has in scope amounts to 3.8. Final purpose clauses, by contrast, have a scope of 1.0 clause across the entire database (i.e. the main clause). It seems that initial purpose clauses are less tightly integrated with their matrix clause than final purpose clauses. In written language, this is also evidenced by the differential distribution of commas: Thompson finds that 78.6% of all initial purpose clauses are separated from the matrix clause by a comma; this only holds for 1.5% of all final purpose clauses, precisely because they serve the function of locally modifying the content of the matrix clause. The orthographic difference is highly significant under a Fisher exact test (p < .001), with a strong effect size (χ2 = 162.92, df = 2, B = 100,000, Cramer’s V = 0.567). Interestingly, the special discourse-organizing function also allows initial purpose clauses to attach to the main clause at a different functional level. While final ­purpose clauses

126 A Typology of Purpose Clauses

­immediately relate to the propositional content of the matrix clause (and hence attach to it at the so-called ‘representational ­level’ in Functional Grammar terminology, cf. Hengeveld 1989: 156), initial purpose clauses often relate to the whole speech act performed by the matrix clause (and hence attach to it at the so-called ‘interpersonal layer’). The following examples, taken from a query into infinitival purpose clauses in the British Component of the International Corpus of English (ICE-GB), illustrate this phenomenon: (131) ICE-GB W1B-029 #115: 4 To sum up, I see no reason yet to be too concerned. (132) ICE-GB S1B-074 #395: 3:A To be honest on that kind of car it’s not really worth it. This pragmatic function cannot be fulfilled by final purpose clauses in English, which supports the view that canonical (i.e. final) purpose clauses never serve to orientate the hearer by setting a thematic ground for the interpretation of the main clause. For this reason, Thompson argues that initial and final purpose clauses should be regarded as two separate grammatical constructions in English, each associated with a specific discourse-pragmatic function. This idea is central to the relatively young theory of Construction Grammar, and current efforts by construction grammarians are geared towards elaborating Thompson’s discourse study into a full-fledged quantitative corpus linguistic analysis of initial adverbial clauses in English (cf. Hampe 2009, but also Verstraete 2004). In sum, there is convincing evidence from naturally occurring discourse that the default position of purpose clauses is after the matrix clause since they fulfil the focal, rhematic function of motivating the event in the matrix clause. It is only when this discourse function is substantially different that purpose clauses move to the front of the complex sentence and assume a thematic role in the ongoing discourse. Under the PGCH, we would expect grammars to conventionalize such pressures from discourse. Indeed, some reference grammars provide specific data to this effect. For example, in West Greenlandic, a typical OV language, it is the default case for adverbial clauses to precede their associated main clause, and the subordinator is usually a clause-final verb suffix. This is perfectly harmonious with the processing considerations articulated above. Fortescue (1984: 56) states, however, that in West Greenlandic the “ordering of adverbial elements relative to the superordinate verb depends very much […] on whether or not they represent essential or new information in the sentence.” As a result, almost all temporal and conditional clauses in West Greenlandic precede their main clause, while purpose, causal and result clauses often deviate from the expected OV pattern in being postposed. The postposing of causal and purposive clauses as a class is, in fact,



Chapter 3.  The grammar of purpose 127

a more widespread phenomenon in OV languages. Where the two clause types share the same or very similar morphosyntax, as in Hausa (Jaggar 2001: 637–40), their common positioning pattern may be due to ‘freezing’ of position during the grammaticalization process. That is, when Hausa dòmin clauses spread from causal to purposive constructions, their already sentence-final position might just have been taken over. However, when the two clause types are represented by different constructions, as in Supyire (Carlson 1994: 586) or Tepehua (Kung 2007: 609), their extraposition is noteworthy because it suggests that causal and purpose clauses share important functional characteristics that deviate from other adverbial clauses. Importantly, these cannot reduce to iconicity in this case since causeeffect or reason-result clauses are everything but iconic if postposed; after all, the cause temporally and logically precedes the effect. What seems to be more compelling is that both purpose and causal clauses express non-topical, foregrounded information in the sentence and are, therefore, located in rhematic positions (for independent discourse-pragmatic evidence on causal clauses, cf. Ford and Mori 1994, Wang 2006, among many others). In §2.2.2 we also introduced the idea that languages can develop special constructions that are sensitive to the information-structural potential of the elements in the clause. From this perspective, it is interesting to observe that some languages open their focus constructions, such as cleaving, preferably (or categorically) to purpose and causal clauses, but much less readily to other types of adverbial relations. This situation is found, for instance, in Ndyuka and Paumarí: (133) Ndyuka (Creole: Suriname; Huttar and Huttar 1994: 116) [A fu kaabu alisi] a be go a goon. [cop for scratch rice 3sg ant go loc field ‘It was to cultivate her rice plants that she went to the field.’ (134) Paumarí (Arauan: Brazil; Chapman and Derbyshire 1991: 233) [O-ka-'ojomo-'i-vini kaimoni mani] ida o-na-jiri-vini hi-ja. [1sg-trns-learn-sub purp cop dem.n 1sg-caus-write-sub aux-imm ‘For the purpose of me learning, I’m writing it.’ ( = ‘It is in order to learn it that I’m writing it.’) The English translations also exhibit a cleft construction, which typically consists of a copular clause containing the highlighted information unit and a relative clause. For English, Huddleston contends that “prototypically the information contained in the relative clause is assigned subordinate status because it is assumed to be known to the addressee, or readily inferrable from what is known.” (Huddleston 1988: 186) The information in the purpose clause, by contrast, is new and noteworthy, and can therefore receive special prominence in a cleaving construction. But

128 A Typology of Purpose Clauses

perhaps the best piece of typological evidence corroborating Thompson’s research comes from languages which grammaticalize her intuition that final and initial purpose clauses are actually separate construction types. This situation has been reported, for example, for Awa Pit. Again a typical OV language, Awa Pit generally prefers preposed adverbial clauses but postposes most purpose clauses. According to Curnow (1997: 365), this position is typical of adverbial elements that “give additional information which is not used to establish a framework” in the unfolding discourse. Crucially, if a purpose clause is supposed to adopt such a framing function (as in Thompson’s study), it appears in a slightly different construction. Specifically, it needs to be marked by a special topic morpheme normally associated with temporal and conditional clauses, and it must, as a rule, occur in sentenceinitial position: (135) Awa Pit (Barbacoan: Equador, Colombia; Curnow 1997: 280) [Maas yal sa-na = na], akkwan ti waa-y. [new house make-inf = top many stick there:is-nonlocut ‘To make a new house, there is much wood (much wood is needed).’ The fact that the topic marker is categorically ruled out on canonical (i.e. postposed) purpose clauses suggests that those clauses fulfil the opposite function in discourse, namely a focal role. This difference is also grammaticalized in Barbareño Chumash, an indigenous language of California (cf. Wash 2001: 399–421). Notice that both Awa Pit and Chumash are essentially spoken languages, and so Thompson’s observation for written language appears to be of more general validity and point to a genuine communicative pressure on the grammar of purpose clauses. To conclude this paragraph, we can safely state that Greenberg’s Universal 15 stands the test of a larger empirical investigation. A monofactorial explanation in terms of iconicity, however, was argued to be problematic since it is not clear, to begin with, what an iconic patterning would look like for purpose clauses. For this reason, we have developed a multifactorial explanation for the position of purpose clauses. On this account, there are several, partially competing pressures on language use, which may be resolved in different conventionalized positioning patterns across the world’s languages. On the one hand, a drive for harmonic ordering patterns across different types of constituents motivates the initial position of purpose clauses in OV languages. Crucially, however, if this general processing preference is not overridden by other functional pressures, then the purpose clause must contain a clause-final subordinator. In this way, minimal recognition domains ensure the efficient construction of the complex sentence in on-line processing. On the other hand, a very persistent and, in fact, the dominant cross-linguistic pattern is for the purpose clause to follow its associated matrix clause, even if this creates non-harmonic branching directions and maximal recognition domains in on-



Chapter 3.  The grammar of purpose 129

line processing. In order to motivate these cases, we took a closer look at findings from quantitative discourse analysis, which suggest that canonical purpose clauses are unlike thematic or framing adverbial clauses in not providing backgrounded, presupposed or topical information that guides the interpretation of the matrix clause. Their rhematic function is reflected cross-linguistically by a strong association with the focal position in the sentence. Many languages do simply not tolerate a different position for purpose clauses, while others have developed special constructions that must be employed when a purpose clause is used thematically, i.e. in preposed position. Incidentally, the preferred position of purpose clauses on the discourse-pragmatic account corresponds to the one predicted by the standard interpretation of iconicity (action > result). But this may turn out to be a side-effect of genuine information-structural pressures that can be observed and measured on independent empirical grounds. If iconicity is to be maintained as a an explanatory factor, the general scenario would be one in which multiple forces jointly work into the same direction, reinforcing each other and leading to a significant constituent order asymmetry in the world’s languages (cf. also Hawkins 2004: 223).

3.6  Semantic and pragmatic idiosyncrasies in purpose clauses The final section of this chapter will be concerned with some selected semantic and pragmatic issues that need to be addressed in a typology of purpose clauses. No attempt will be made to provide an exhaustive account of pertinent phenomena; we shall rather focus on typological data from our sample that relate to interesting aspects of the meaning of purposive constructions.

3.6.1  Avertive (‘lest’) constructions According to our initial definition in §2.3, a semantic relation of purpose holds between two situations, one of which is performed with the intention of bringing about the other. While this suggests that the result state is inherently desirable from the point of view of the intender, the purpose of an action sometimes also consists in preventing an undesirable event from happening. In such cases, the matrix clause typically encodes that precautions are taken so as to avoid an ‘apprehension-causing’ situation in the subordinate clause (Lichtenberk 1995: 298). Two examples from English are given below: (136) We took an umbrella so that we wouldn’t get wet on the way. (137) You should mark the date in the calendar so that you won’t forget Mary’s birthday this year.

130 A Typology of Purpose Clauses

Purpose clauses of this semantic type may also be called negative purpose clauses. Cross-linguistically, it is very common for negative purpose clauses to be expressed in the same way as their canonical positive counterparts, with simply a negative marker being added to the basic proposition (cf. above will/would + not). In quite a few languages, however, the semantic distinction between positive and negative purpose is grammaticalized, to the effect that both types of purpose clause employ a distinct grammatical marker or construction. English, too, has such a primary morpheme for negative purpose, although it is judged to be “archaic and very formal” by standard reference grammars (Greenbaum and Quirk 1990: 323): (138) BNC G17 501 Alexei formed a smile as he spoke, lest she detect a hint of the offence in his words. (139) 1526 TINDALE Mark xiii, OED Online Take hede lest eny man deceave you. Morphemes encoding negative purpose have been called ‘avertive’ or — somewhat Anglocentrically — ‘lest’ markers and, analogously, the clauses they head are ‘avertive’ or ‘lest’ clauses. In a comprehensive typology of purpose clauses, it is certainly interesting to ask how frequently languages grammaticalize avertive meaning and what morphosyntactic characteristics are shared by avertive clauses. The present section will try to answer these questions. Although ‘lest’ constructions have to be regarded as constructions in their own right, they have always been included in the previous analyses since they are, first and foremost, (a special kind of) purpose clauses. If we isolate them from our corpus of 218 purposive constructions, we obtain a sub-corpus comprising 20 ‘lest’ constructions distributed over 19 languages. In other words, 19 out of 80 languages (23.8%) have distinct avertive constructions, alongside at least one canonical (positive) purpose clause. Their geographic distribution is shown in Figure 24. While avertive constructions are found in all macro-areas, the map suggests that there are certain geographical skewings. The most evident asymmetry can be detected between the large Eurasian area, with scarce occurrences of ‘lest’ clauses, and the comparatively small Australian and Oceanic areas, which are host to the majority of all ‘lest’ clauses in our sample (11∕20 = 55%). By contrast, a bias for particular structural language types, e.g. in terms of constituent order, cannot be discerned, since ‘lest’ clauses are found across all constituent order types. This is not surprising, however, because the existence of avertive clauses is motivated primarily in semantic terms and hence we would not expect them to be found in certain word order types only. Similarly, the avertive marker itself is not restricted to a particular syntactic category: ‘lest’ markers can be conjunctions, adverbial suffixes,



Chapter 3.  The grammar of purpose

Figure 24.  Languages with ‘lest’ constructions (N = 19)

postpositions, TAM markers, particles and auxiliaries. Here are two illustrative examples of this wide formal variation: (140) Wardaman (Gunwinyguan: Northern Australia; Merlan 1994: 295) Yirrb-a me yiwarlng bujun warlad. remove-ptcl.suff aux clothes.abs lest sick ‘Take off your clothes lest you get sick.’ (141) Yagua (Peba-Yaguan: Peru; Payne and Payne 1990: 416) … naada-y-numaa dáátya jiy-daasaada.   3du-aux.neg-now know 2sg-mother.in.law ‘… so that your mother-in-law won’t know.’ Sometimes an avertive marker is accompanied by an overt negative morpheme in the ‘lest’ clause, thus emphasizing the notion of negative purpose. This is the case in Slave: (142) Slave (Athapaskan: Canada; Rice 1989: 1262) [Daniel yegúh ʔále ch'á] goghádehk'a. [Daniel 3opt.find 4.neg lest 1sg.threw ‘I threw it so Daniel wouldn’t find it.’ It is only when the matrix verb is inherently negative that the negative morpheme in the ‘lest’ clause can be omitted in Slave. Notice that in terms of position, the ‘lest’ clause in (142) also adheres to the typical cross-linguistic patterns since the sentence-initial position of the negative purpose clause conspires with the clause-final position of the subordinator to yield minimal recognition domains.

131

132 A Typology of Purpose Clauses

1.0

What is remarkable from a structural point of view is that ‘lest’ clauses turn out to show a somewhat different behaviour as compared to positive purpose clauses with regard to the major morphosyntactic domains we have investigated. In fact, they sometimes provide serious exceptions to the statistical universals uncovered for purpose clauses. With respect to verbal deranking, positive and negative purpose clauses are not yet different enough to achieve statistical significance. If we go back to our earlier counts of deranking (cf. Table 7 above) and now calculate the percentage of deranked verb forms for positive and negative purpose clauses separately, we see that balancing in negative purpose clauses (9∕20 = 45.0%) slightly outranks balancing in positive clauses (57∕193 = 34.7%), but the difference in proportions is not sufficiently different from chance (p = .46 in a Fisher exact test). The picture changes, however, with regard to argument-structural configurations. The first parameter we analysed earlier was the overt expression of the notional subject. Again, we can split up our aggregate findings (cf. Table 12 above) and count the preferences for implicit subjects separately for positive and negative purpose clauses. Again leaving out non-manageable cases (‘NA’, ‘no illustration’) in this analysis, we obtain a significant difference between the two clause types (χ2 = 8.93, df = 2, p = .01, B = 100,000, Cramer’s V = 0.204), as Figure 25 illustrates at a glance. Notably, the proportion of implicit subjects is much smaller in avertive clauses than in positive purpose clauses, which, conversely, increases the amount of explicit subjects significantly.

0.6

0.8

explicit subject

both possible

positive purpose clauses

0.0

0.2

0.4

implicit subject

negative purpose clauses

Figure 25.  Overtness of subjects in positive and negative purpose clauses (N = 214)



Chapter 3.  The grammar of purpose 133

Table 29.  Co-reference relations of the subject in positive and negative purpose clauses (in percent) Purposive subject co-referential with matrix… Construction type Different Mixed Subjects

Subjects and objects Objects

Obliques

Total (%)

Positive PC Negative PC

13.7 10.5

0 5.3

100.0 100.0

  5.8 36.9

24.2 31.5

54.2 10.5

2.1 5.3

This difference leads us to suspect that the two clause types contract different co-reference relations with the participants of the matrix clause (cf. §3.3.1 above). Again we can distinguish the same co-reference patterns as before (Figure 9) and compare the relative frequencies with which positive and negative purpose clauses establish each of those co-reference relations. Table 29 displays the results of this analysis. Although the quantity of ‘lest’ clauses is quite small overall, we will still assume that some significance can be attached to the relative frequencies given here. We then realize at a glance that positive and negative purpose clauses are quite different. In particular, it appears that the overwhelming ‘same-subject’ preference of positive purpose clauses (54.2%) is not shared by avertive constructions (10.5%), but that the latter instead prefer entirely different, i.e. non-co-referential subjects (36.9% as opposed to only 5.8% in positive clauses). The difference in the proportions given in Table 29 is highly significant if we submit the absolute frequencies underlying the distribution to a randomized Chi-squared test (χ2 = 37.12, df = 5, p < .001, B = 100,000, Cramer’s V = 0.421). In analogy to the association plot used in Figure 9 above, we will employ a similar graphical tool here that allows visualizing where the significant skewing comes from. The so-called mosaic plot in Figure 26 reflects the original contingency table, with the area of the rectangles being proportional to the observed frequencies in each cell; and just as in an association plot, the shading of the boxes indicates which cells make an individually significant contribution to the overall Chi-square value of the table. The darker the shading, the more the observed frequency in a cell deviates from the expected frequency (given the margin totals). Rectangles with broken lines indicate a lower frequency than expected (negative residuals), while solid lines capture positive residuals. Figure 26 demonstrates that the most dramatic skewing of the table lies in the significant positive association of ‘lest’ clauses and the different-subject ­pattern (cf. the black box in the upper-right corner). What is significantly underrepresented, by contrast, is a same-subject preference in avertive constructions, given our ­general expectations for purpose clauses. (The positive skewing for ­obliques

134 A Typology of Purpose Clauses

subjects and objects Objects objects Obliques obliques

Standardized Residuals:

subjects

<–4 –4:–2 –2:0

0:2

mixed

>4

different

negative purpose clauses

2:4

positive purpose clauses

Figure 26.  Mosaic plot for co-reference relations of the subject of positive and negative purpose clauses. Shaded (or extended) mosaic plots were proposed by Friendly (1994) and can be ­produced using the mosaicplot function in R. Shadings reflect individual significance levels: Residuals between 0 < |x| < 2 are not individually significant (white boxes), residuals between 2 < |x| < 4 are significant at α = .05 (grey shading), and residuals from |x| > 4 are significant at α = .0001 (black shading)

should not be taken as particularly meaningful since the absolute frequency under­lying it is only N = 1). Why are positive and negative purpose clauses fundamentally different in this respect? In keeping with the usage-based approach, such argument-structural patterns reflect the most frequently talked-about experiential scenarios involving avertive relations. Recall that avertive constructions encode the desired prevention of an unpleasant occurrence. In order to understand the typical participant constellations in such cases, I collected some representative examples of avertive situations from the reference materials (Table 30). All examples are translated into English, but reflect the original constellation of participant roles and syntactic functions. As can be seen, the conceptual situation associated with negative purpose clauses is prototypically such that an agent is acting (on some entity) in order to



Chapter 3.  The grammar of purpose

Table 30.  Translations of typical avertive constructions from the sample Exemplar sentence (143) Circle around so the moose won’t smell you. (144) Take off your clothes lest you get sick. (145) Theyi are careful with themj lest theyj end up being spoiled. (146) Put the meat in the smoke, so it doesn’t get eaten by flies. (147) I spoke in a whisper, so that other people could not hear it/ me. (148) Put it down so that it won’t fall. (149) Don’t shake him lest he wake up. (150) Don’t behave like that lest your father beat you! (151) Don’t you do that for they might hit or kill you. (152) I tried very hard for this letter not to get into the hands of the police.

Source (Language, page in the reference grammar) Slave, 1262 Wardaman, 295 Supyire, 587 Kayardild, 510 Hixkaryana, 30 Hausa, 641 Martuthunira, 250 Ngiyambaa, 285 Lakhota, 259 Kannada, 74

prevent some undesired situation from happening to her or to this entity. Two scenarios are possible here: (i) the agent in the main clause performs an action in order to prevent herself from getting into an unpleasant situation; this scenario can involve same subjects (ex. [144]) or different subjects ([143], (147], [150], [151]); (ii) the agent of the main clause is acting on some entity in order to prevent that entity from getting into an unpleasant situation; this would call for a different subject preference ([145], [146], [148], [149], [152]). We can see here again how perspective-taking (MacWhinney 1977; cf. also the discussion of OPCs in §3.3.2) enters the verbalization of multipropositional scenarios: In avertive clauses, speakers typically seem to start with the perspective of the main clause agent, but then shift to the perspective of either the negatively affected participant or the force acting on him. In both cases, the new perspectivizer only coincides with a matrix participant under certain semantic-role conditions. In particular, I would like to propose two complementary implicational universals: (153) If a language has a semantic agent as subject in avertive clauses, then it is not a shared participant of both main and avertive clause. (154) If a language has a co-referential subject argument in avertive clauses, then it is a semantic patient in the avertive clause. (153) predicts that active-{S} or {A} arguments in the avertive clause encode ‘third parties’, as it were, that do not correspond to matrix clause participants. (154) predicts that if the subject argument of the ‘lest’-clause takes up a matrix participant again, it can only be a patientive-{S} or {derived-S} argument in the avertive clause. The following example ties in with the latter prediction:

135

136 A Typology of Purpose Clauses

(155) Kayardild (Tangkic: Northern Australia; Evans 1995: 510) Ngada jaa-nangku wida-wu, [yarbuth-inja ba-yii-nyarra-nth]. 1sg.nom enter-neg.pot hole-mdl.prop [snake-obl bite-m-lest-cmp.obl ‘I won’t put my hand in that hole, or I’ll get bitten by a snake.’ ( = ‘I won’t put my hand in that hole lest I get bitten by a snake.’) In this example, the subordinate subject is deleted under co-referentiality with the matrix subject. Crucially, however, the oblique case suffix on yarbuth ‘snake’ indicates the agent of a passive clause; the equi-deleted grammatical subject is thus a derived-S argument and understood to be the semantic patient of the avertive situation. The sceptical reader may (correctly) object that the generalizations in (153) and (154) are tentative and a heuristic at best since they were established on the basis of less than 20 languages. Indeed, it is possible to find counterexamples even in this small database. (156) is precisely such a case in point: (156) Martuthunira (Pama-Nyungan: Western Australia; Dench 1995: 249) Ngayu puni-layi-rru [nyina-wirri manhamanha-ma-l.yarra nhuwala-a]. awkward-caus-sim 2du-acc 1sg.nom go-fut-now [be-lest ‘I’ll go now lest I’ll be making it awkward for you.’ In this example, the patient of the avertive clause is marked by accusative case and hence clearly not the subject of the subordinate clause. The subject participant is rather deleted because it is co-referential with the matrix subject. Oddly enough (by [154]), it is also the agentive {A} of the avertive clause. The question, then, is whether the typological generalizations made above could be due to sampling error or whether they really reflect usage frequencies in avertive constructions. This is again a good problem set for relating typological data to richer performance records from individual languages. In order to seek converging evidence, therefore, I retrieved all occurrences of ‘lest’ from the British National Corpus (BNC), a computerized 100-million-word database of authentic spoken and written English. This yielded 410 instances in total, several of which had to be discarded because they turned out to be false positives.24 The remaining final corpus boiled down to N = 259 genuinely avertive ‘lest’ constructions. For the analysis of argument-structural configurations, I created a random subsample of 100 instances, which were then coded for the semantic roles and syntactic functions of their subjects and co-referential arguments. Notice that English ‘lest’ construc24.  False positives are those instances of ‘lest’ clauses in the corpus that do not encode avertive situations, but are instantiations of either the conditional (‘in-case’) or complementizer meaning of lest, e.g. Clarissa was horrified lest Charles thought of identifying himself with such a call. [BNC ACE 240].



Chapter 3.  The grammar of purpose 137

tions cannot leave their subject implicit; they are always balanced and hence the question of economical coding does not arise in the first place. However, we have claimed more generally that identity of subjects is far less common in avertive than in positive purpose clauses. In the BNC sample, a simple two-way contrast between same-subject and different-subject patterns yields a ratio of 37: 63. This supports the universal trend but at the same time also shows the subject-sharing pattern to be more than three times as frequent as in the typological sample (cf. Table 28 again). With regard to semantic roles, we get the classic PGCH situation that the typological prototype also comes out as the most frequent performance option, but that the intra-linguistic data show much more variability than the cross-linguistic data afford. More precisely, there are 47 data points in the corpus in whose subject is a semantic agent. By our prediction in (153), the typical case in performance should now be for this agent not to correspond to a matrix participant, but rather to shift the perspective to a ‘third party’. This prediction is borne out by the data, but perhaps not as clear-cut as we would have hoped for. In Figure 27, I list and exemplify the co-reference pattern attested for agentive subjects, with an indication of their relative frequencies in our subsample. The downward arrow on the left signifies decreasing usage frequencies and hence an increasing distance from the typological prototype. Apart from the fact that the prototypical cross-linguistic association is reflected in the highest usage frequency, it is evident that the corpus also reveals patterns that are underrepresented or even unattested in the typological corpus. Similarly, our complementary generalization in (154) is confronted with a larger variation space in the intra-linguistic data. We contended that if a ‘lest’-clause subject e­ ncodes a participant shared between matrix and subordinate clause, then this (a) Subject not co-referential with any participant in the matrix clause (38.3%) Joe was still careful to keep their meetings secret [lest some kind friend wrote about them to Terry]. (BNC G16 2385) (b) Subject co-referential with matrix patient (34%) He cut the remark out of the final programme [lest it should offend the listeners].  (BNC BLY 1232) (c) Subject co-referential with matrix agent (23.4%) I relinquish my rights to redeem [lest I impair my own inheritance].  (BNC KNA 128) (d) Subject co-referential with an oblique role in the matrix (4.3%) They should never fly together in the same plane lest it (crash and) kill both heir and second-in-line to the throne. (BNC A7H 667) Figure 27.  Co-reference patterns of agentive avertive subjects in the BNC (N = 47)

138 A Typology of Purpose Clauses

participant will typically take on a patient-like semantic role in the ‘lest’-clause. There are 33 instances in the corpus in which the participant coded by the ‘lest’subject has no role to play in the matrix clause, i.e. in which there is no participant sharing with the ‘lest’-subject. This leaves 67 cases whose semantic role needs to be analysed. In broadest terms, i.e. simply distinguishing between agent-like and patient-like macro-roles, the ratio between agents and patients turns out to be 29: 38 (i.e. 44% macro-agents versus 56% macro-patients). Patients constitute a fairly heterogeneous class, though, comprising also experiencers and what Huddleston and Pullum (2002: 230) call ‘themes’, e.g. lest he make such a nauseous figure (BNC A06 826). In other words, while non-agentive roles are certainly preferred, in keeping with the typological finding, it is not that uncommon to find analogues to the Martuthunira example in (156), with an agentive but still co-­ referential subject in the ‘lest’ clause (cf. also pattern [c] in Figure 27 above). In sum, corpus-linguistic evidence suggests, indeed, that languages conventionalize syntactic structures in proportion to their degree of preference in performance. While some languages allow for a wide range of options in performance, others have altogether eliminated minority patterns from the grammar. As a result, typological hierarchies of grammatical conventions (A>B>C>D), “with cutoff points and declining frequencies of languages” (Hawkins 2004: 6), correspond to preference rankings in performance in languages which allow multiple options (A, B, C, D). In our specific case, English corpus data testify precisely to such multiple options, but there is a notable decline in usage frequency the further one gets away from the typological prototype. Based on converging evidence, it is thus safe to conclude that negative purpose clauses are significantly different from their positive counterparts with respect to argument-structural properties. In addition, however, there are also other parameters on which the two clause types differ markedly. One such parameter concerns the semantic restrictions on the matrix verb. Recall from our earlier discussion in §3.4.2 that many positive purpose clauses are typically associated with a matrix verb of motion, up to the categorical constraint that anything other than motion verbs is not licensed by a number of purposive constructions (cf. ­Table 23 again). We argued elaborately that this constraint is grounded in the ubiquitous experiential correlation between purposes and destinations, which often gives rise in language use to the reanalysis of allative or goal markers as purposive morphemes. Interestingly, both typological tendencies are absent from the sample of ‘lest’ clauses: In none of the 20 constructions do we find a similar restriction on the matrix verb, and not a single one of them exhibits traces of allative or dative morphology either. That is, the negative purpose marker itself is never polysemous with those notions (suggesting that it is historically derived from other sources), and additional gestalt features in ‘lest’ clauses also resist being recruited



Chapter 3.  The grammar of purpose 139

from goal-oriented semantic domains. Again one might argue that this may reflect suboptimal sampling conditions rather than pointing to genuine semantic restrictions. In the following, we will thus conduct a second corpus-based investigation, this time exploiting the fact that modern quantitative corpus linguistics has developed sophisticated methods for semantic analyses of grammatical constructions. The purpose of this intra-linguistic excursus is two-fold: on the one hand, we will seek converging evidence for the striking absence of motion contexts in avertive constructions; on the other hand, the methodology used to gather this evidence will also provide a more principled insight into the semantics of ‘lest’ constructions more generally and give us a clearer idea of the contours of avertive meaning. We thus return to our earlier exhaustive query of ‘lest’ clauses from the BNC, whose manual inspection yielded a total of 259 genuinely avertive instances of the construction. This time, the tools to be applied require us to consider all of the 259 instances. I inspected them manually, lemmatized their respective matrix verb and coded, using a three-way distinction, whether it designates directed motion (go to, enter), ‘simple’ motion (move) or non-motion situations. In terms of crude raw frequencies, 247 of the 259 sentences (95.4%) do not involve a motion context; it is only 8 matrix clauses that exhibit a verb of directed motion (3.1%) and 4 that evidence simple (undirected) motion (1.5%). However, even though the absolute figures are already telling and unequivocal in this case, we can go one step further and examine, in a more fine-grained way, the specific ‘association patterns’ (Biber et al. 1998: 4) of particular verbs or verb classes with the ‘lest’ construction. In a series of publications, Gries and Stefanowitsch have developed a family of quantitative techniques to investigate the interaction of verbs and the grammatical constructions they appear in (Stefanowitsch and Gries 2003, 2005, Gries and Stefanowitsch 2004). This family goes by the name of ‘collostructional analysis’, and its basic idea is to compute, for each verb that appears in a particular grammatical construction, whether its occurrence in this constructional environment is significantly more or significantly less frequent than expected given the verb’s overall frequency in the corpus. Thus operating with conditional rather than absolute frequencies, collostructional analysis filters out, as it were, those verbs that are frequent in the construction in question simply because they are very frequent verbs in general, such as be, have, make, get and many others. In this way, we can obtain a much closer approximation of which verbs are characteristic of the construction and thus serve as typical ‘profile equivalents’ (Croft 2001: 257). As applied to our data, collostructional analysis requires the following raw data: the frequency of a particular verb (lemma) in the matrix clauses of ‘lest’ constructions, the frequency of this verb in all other constructions, the total number of ‘lest’ clauses in the corpus, and the total number of lexical verbs in the entire corpus.

140 A Typology of Purpose Clauses

Table 31.  Contingency table for a collostructional analysis of cut as a matrix verb of ‘lest’ clauses cut

Other verbs

Row totals

Matrix of ‘lest’ clause

3

256

259

All other clauses

18,508

7,195,981

7,214,489

Column totals

18,511

7,196,237

7,214,748

Taking the verb cut as an example, Table 31 illustrates how this information (highlighted in grey ­shading) can be arranged schematically in a contingency table. Gathering these pieces of information is not straightforward and hence deserves some explanation. The frequency of the relevant construction in the corpus had already been determined earlier by counting all purposive uses of ‘lest’ clauses (cf. above, N = 259). Lemmatization of the matrix verbs then allows recording how often each verb type occurs in the main clause of the construction (for cut, N = 3). Crucially, we also need to know how often cut is used in all other (i.e. competing) environments, so we need to determine its frequency as a lexical verb in the BNC. Fortunately, the compilation by Kilgarriff (1996) provides this information for all verbs that occur at least 800 times in the BNC (e.g. cut, N = 18,511). For all others, I had to determine this frequency myself, applying Kilgarriff ’s procedure of counting all inflectional forms of each verb (e.g. blast, blasts, blasted, blasting). Finally, we also need to determine the total number of all lexical verb forms in the corpus. For this procedure, we can sum up all verb tags in BNC (as suggested by Stefanowitsch and Gries 2003: 218) and subtract from them the number of auxiliary uses of be, do, have, go and use (e.g. have been, going to, used to).25 This yields a sum total of 7,214,748 words, as indicated by the shaded lower right cell of the table. The remaining cells can now be filled in by subtraction, resulting in the full picture provided by Table 31. We are now in a position to evaluate whether a lexical verb such as cut is significantly ‘attracted’ to the matrix of ‘lest’ clauses. This is achieved by applying an association test to the bold-marked cells of Table 31. On statistical grounds, Stefanowitsch and Gries (2003: 218) advocate the use of the distribution-free FisherYates-Exact test (henceforth FET). They argue that the p-value of the FET can be interpreted as a measure of the ‘association strength’ between a given lexeme and a grammatical construction: Depending on whether the lexeme is found statistically more often or less often than expected in the construction, it is said to be particularly ‘attracted’ or ‘repelled’ by that construction. To stay with our example, cut 25.  Auxiliary verb uses were determined by retrieving all combinations of be, have, do, go and use with a lexical verb, in whichever inflectional form.



Chapter 3.  The grammar of purpose 141

exhibits a significant attraction of p = 0.02977 under a FET and thus qualifies as a significant ‘collexeme’ of the ‘lest’ environment (since p < .05). Crucially, the most interesting results are only revealed when collostructional analysis is repeated for all verbs occurring as matrix verbs of ‘lest’. Using Gries’ (2007) Coll.analysis 3.2 algorithm, I obtained an ordered list of all those matrix verbs, ranked according to their p-value and hence their relative degree of attraction to or repulsion by the lest-matrix. Table 32 lists the ‘top-ten’ verbs with the strongest degree of attraction. Table  32 allows for several interesting observations. First, the verbs most strongly attracted to the lest-matrix, beware and guard, are those with the closest fit to our hypothesized meaning of an avertive construction. They capture the precautious nature of the event in the matrix clause: beware denotes a general state of alertness to act thoughtfully or exercise caution, while guard admonishes to keep a protective eye on a valuable object or abstract entity (e.g. guard their own position [bnc crf] or guard her name [bnc jy1]). Chain and some uses of keep are, in fact, more specific subtypes of this protective sense since they literally prevent someone or something from being taken away (e.g. keep it locked away in the closet [bnc ka8]) or being disclosed (e.g. keep their meetings secret [bnc g16]). Temper, propitiate and domesticate also form a semantically related cluster of verbs since they all refer to the process of taming a character, creature or other entity of a feral nature (e.g. to temper conviction lest it degenerate into fanaticism [bnc k57]). Finally, it is interesting to note that add and restress are mostly meta-linguistic verbs in this construction, denoting that a remark is added to or emphasized in the discourse in order to prevent the listener from drawing false conclusions or making erroneous assumptions. Interestingly, 6 of the 7 instances of add are found in sentences with preposed, i.e. initial ‘lest’ clauses, such as (157). Table 32.  Collexemes most strongly attracted to the matrix of ‘lest’ clauses Collexeme

CollStrength

beware guard chain add temper restress hurry keep domesticate propitiate

6.91 4.46 4.39 4.15 4.15 4.14 4.00 3.28 3.08 3.06

Note: CollStrength directly reflects the FET p-value for each verb, but since p-values tend to be very small for the most attracted lexemes, they were logarithmically transformed for ease of interpretation: The higher the CollStrength value reported here, the higher the degree of attraction. All of the values in this table are ≥ 3, which represents a p-value of p < .001***.

142 A Typology of Purpose Clauses

(157) BNC B1J 1899 Lest you think that I am too good to be true, let me at once add that I am not in the slightest inclined to gamble. These clauses, just like initial positive purpose clauses (cf. §3.5 above), have a discourse-organizing function in that they anticipate a (false) impression or problem arising from the previous discourse, only to refute it by subsequent additions or qualifications. For this reason, preposed ‘lest’ clauses, just like other preposed purpose clauses, often have more than one ensuing clause in their scope. But, lest we digress, let us return to the matrix verbs. It should have become clear that the procedure of collostructional analysis is able to abstract away from the raw co-occurrences of lexemes and provide us with a list of verbs that are distinctive for the construction. This explains, for instance, why the verb make, which appears in ‘lest’ matrices just as often as keep, is not worth considering here: Given its high overall corpus frequency, make does simply not occur frequently enough in ‘lest’ constructions to be considered distinctive for this environment. By contrast, the verbs in Table 32 form systematic classes whose distinctive occurrence in ‘lest’ clauses is well-motivated given the constructional semantics. Importantly, collostructional analysis also supplies a list of the significantly repelled lexemes, i.e. those that occur considerably less frequently in the construction than expected. Table 33 displays the seven most repelled lexemes. What is most striking in this list is that it contains the two most general and frequent motion verbs, come and go, which are often so distinctive for positive purpose clauses that languages grammaticalize a separate motion-cum-purpose construction, as we saw earlier. Negative purpose clauses, by contrast, appear to resist contexts of (directed) motion, even though it must be said that go is not significantly repelled by the α < .05 criterion in our corpus (hence the broken line in Table 33). At any rate, however, it is telling that the prominence of motion verbs in positive purpose clauses does not carry over to avertive constructions, whose matrix clauses can, but generally disprefer to, exhibit a verb of motion. Table 33.  Collexemes most strongly repelled by the matrix of ‘lest’ clauses Collexeme

CollStrength

be have say get come

30.45 9.11 2.70 2.54 1.57

look go

1.04 1.02



Chapter 3.  The grammar of purpose 143

Now, the sceptical reader might object again that ‘lest’ clauses in English are heavily biased towards formal written (or even archaic) contexts and that this stylistic markedness might simply be the reason why comparatively profane words like come or go do not show up frequently in this construction. In order to dispel those doubts, I repeated all of the above analyses for the ‘unmarked’ negativepurpose construction in English, i.e. so that … not V. The exhaustive retrieval of this construction from the same corpus (i.e. the BNC) was achieved by querying all subordinate conjunctions so that which govern a clause that also contains a negative marker (n’t, not) and — to exclude most result clauses — a modal auxiliary (since we saw earlier that modality is a relatively reliable cue distinguishing the two clause types in English). This yielded a total of N = 685 hits, which were again inspected manually and coded for the above aspects. Notice that a fairly large number of false positives (chiefly further result clauses) had to be excluded from the sample, so that the ultimate corpus comprised 493 negative purpose tokens. Again, they were analysed in context and checked for a motion component. Overall, the percentage of motion contexts increases only slightly to 8.9% (44∕493), 37 instances of which involve directed motion (i.e. the typical context of positive purpose clauses). The general dispreference for motion is thus replicated. What is more, amongst the directed-motion verbs, there is a clear tendency for very ‘local’ motion, specifically of parts of the body rather than genuine locomotion. More precisely, the verb turn is prevalent in this group (37.8% of all directed motion tokens), and most often it refers to the body (or the head/face) turning away from someone: (158) BNC A0F 601 That brought tears to my eyes and I turned away so that Jeff wouldn’t be able to see them. In fact, this example turns out to be very typical. An extension of collostructional analysis, the so-called ‘co-varying collexeme analysis’ (Stefanowitsch and Gries 2005), allows us to examine systematically which matrix verbs co-occur with which subordinate verbs in the construction, and its application to this dataset reveals that the pair turn—see is the most distinctive collexeme set of the construction (CollStr = 9.49***).26 It is followed by a conceptually closely related pair, in which someone lowers their voice so that someone else may not hear their talk (CollStr = 6.01***). More generally, if directed motion occurs in this construction, it is typically the very local motion of (parts of) the body within the current 26.  Note that, more generally, the application of co-varying collexeme analysis to our two datasets does not yield particularly striking results since most of the V-V combinations are fairly unique, i.e. occur only once or twice. This, in turn, leads to very similar collostruction strengths and precludes a further detailed investigation.

144 A Typology of Purpose Clauses

Table 34.  Collexemes most strongly repelled by the matrix of ‘so-that-not’ clauses Collexeme

CollStrength

be have do say go get come

77.67*** 15.70*** 7.65*** 7.64*** 3.82*** 3.16*** 2.72***

setting of the conversational interaction. Rather rarely is there any large-scale locomotion involved. This is again confirmed by the collostructional analysis, which returns both come and go amongst the most strongly repelled collexemes of the ‘so-that-not’ matrix (cf. Table 34). Since this set of verbs mirrors our earlier set almost perfectly, we can safely conclude that the objection to motion verbs is a significant performance preference of negative purpose clauses, which is reflected in the different lexical cooccurrence constraints that languages conventionalize across the world. It seems, then, that there is no experiential correlation of negative purpose and particular locations; avertive situations are typically not contingent on directed motion. In the absence of such a regular correlation in language use, no allative or goal-marking material is regularly available, so that inferences from goal to purpose marking simply have no chance of arising. This, in turn, is evidenced by the typological data, as was noted earlier. In sum, the usage-based perspective taken here suggests that there is cross-linguistic convergence (in the evolutionary sense of Chapter 2) in the “verbalization of experience” (Croft to appear [a]), i.e. in the symbolization of avertive situations.

3.6.2  Further issues Apart from avertive constructions, there are some minor semantic and pragmatic issues pertaining to purpose clauses, all of which are restricted to only a subset of our sample. Epena Pedee has separate constructions for ‘immediate’ and ‘ultimate’ purpose: (159) Epena Pedee (Chocó: Colombia; Harms 1994: 158–9) a. immediate purpose Čé-hi [ne-k hó-de]. come-imp [gnr-eat-purp ‘Come here to eat!’



Chapter 3.  The grammar of purpose 145

b.

ultimate purpose (in combination with immediate purpose): [nása k ha-it hée]. Wã-hí [hk hára hr-de] go-pst vine hunt-purp [fish.trap weave-for ‘He went to search for vines, so he could weave a fish trap.’

Such constructions may be interesting from a Construction Grammar perspective since although the two clauses in (159) are propositionally equivalent in a truthconditional approach, they are each associated with a specific construal of a purposive situation and hence occupy separate ecological niches in the language system (cf. Goldberg 1995: 3 for the theoretical impact). From a pragmatic perspective, it should be mentioned that some purpose clause constructions are conventionally paired with a specific illocutionary function. We encountered this phenomenon in connection with preposed, discourseorganizing purpose clauses in §3.5. Such ‘speech act purpose clauses’ are notoriously difficult to investigate across languages since grammatical descriptions do not usually take such pragmatically constrained constructions into account. I found an example from Modern Hebrew, though, which is interesting for two reasons: First, while canonical purpose clauses in Hebrew are flexible with regard to their position vis-à-vis the matrix clause, speech-act purpose clauses must occur sentence-initially (Glinert 1989: 356). In keeping with our earlier hypothesis, this is a reflex of their topical function in discourse. Second, the morphosyntactic make-up of the construction, too, is sensitive to this thematic function. Specifically, it blends an infinitival verb form typical of same-subject purpose clauses with a conditional (rather than a purposive) conjunction. This is an overt morphosyntactic indication of the exceptional discourse-pragmatic status of the purpose clause: (160) Modern Hebrew (Afro-Asiatic/Semitic: Israel; Glinert 1989: 356) Im liyot dúgri, […] if be.inf frank   ‘To be frank, (I really don’t like your new shirt.)’ Another phenomenon pertaining to illocutionary force is that purposive constructions lend themselves well to being formulated as directives. In other words, the speaker commands the agent of the matrix clause to engage in purposive action. This is particularly frequent with avertive clauses, but also common for positive purpose clauses. Some languages have even grammaticalized distinct ‘purposive imperative constructions’, such as Lakhota (Buechel 1939: 259) and Paumarí: (161) Paumarí (Arauan: Brazil; Chapman and Derbyshire 1991: 228) I-dioni-'iana-ø mahija o-kamitha-vini! 2sg-repeat-again-imp purp 1sg-hear-trns ‘Repeat it so that I can hear it!’

146 A Typology of Purpose Clauses

According to the authors, the purposive conjunction or particle mahija occurs only if the main clause is marked for imperative mood. More generally, pragmatic issues certainly constitute an underexplored area in the typology of purpose clauses. But since grammatical descriptions increasingly incorporate discoursepragmatic perspectives on morphosyntactic phenomena, it is to be hoped that f­ uture research will fill this gap. With these selected semantic and pragmatic issues addressed, we have now completed our documentation of purpose clause constructions in the world’s languages. While not all phenomena could be dealt with in detail, I  hope that the most important and quantitatively significant typological characteristics of purpose clauses have been uncovered, discussed and explained.

chapter 4

Purpose clauses in the syntactic and ­conceptual space of complex sentences

After having documented the morphosyntactic unity and diversity of purpose clauses across the world’s languages, we will now broaden our perspective in order to characterize the place of purpose clauses in the overall system of complex sentences. While such systems vary immensely from language to language, we will see nevertheless that purpose clauses occupy a very special place in the syntactic space of complex sentence constructions in many languages. We shall unfold this space by systematically relating purpose clauses to all other syntactic functions commonly encoded in complex sentences. The first part of this chapter will thus be concerned with the synchronic relationship between purpose and (other types of) adverbial clauses (§4.1.3), complement clauses (§4.1.4) and relative clauses (§4.1.5). Cutting across these clause types is the syntactic status of purpose clauses, i.e. the distinction between coordinate and subordinate constructions, which we will briefly consider at the very beginning of this section (§4.1.2). In the second part of this chapter, I will argue that the special status of purpose clauses in complex sentence systems is rooted in functionally motivated diachronic developments within such systems. Specifically, it will be shown that purpose clauses provide important source constructions for the grammaticalization of certain types of complement, relative and other adverbial clauses (§4.2.2–3), as well as being intimately linked to the development of a number of tense-aspect-mood constructions (§4.2.4–5). Finally, we will have to ask where purposive constructions themselves come from (§4.2.6). Let us begin, however, with a general characterization of the phenomenon of complex sentences.

4.1  The synchronic status of purpose clauses 4.1.1  The conceptual and syntactic space of complex sentences From the functional-typological perspective underlying the present study, we can approach the phenomenon of complex sentences in two principal ways. On the one hand, we could characterize the conceptual or functional space of complex sentences, i.e. the cognitive, semantic and discourse-pragmatic relations holding

148 A Typology of Purpose Clauses

between the two situations encoded by a multiple clause construction. On the other hand, we can characterize various dimensions of the syntactic relation holding between the two clauses involved. The conceptual space of complex sentences is organized by the question of how the connection between the two encoded situations is construed by the speaker. In principle, one can construe both situations as equally important or salient; in this symmetrical situation, neither of the two situations is assigned a more prominent status than the other. In Langacker’s (1991) terms, both situations have an autonomous profile. If, by contrast, the construal is asymmetrical, one of the two situations is conceptually more prominent, with the other losing its autonomous profile and being construed in the perspective of the dominant situation. This distinction nicely corresponds to the pragmatic or information-structural notion of assertion (Lambrecht 1994). In actual discourse, the cognitively more prominent situation is often associated with the primary communicative information the speaker intends to convey, i.e. what s/he intends to assert. Taken together, the two criteria appear to provide a universally applicable framework for distinguishing between coordination and subordination in complex sentences: “By subordination will be meant a situation whereby a cognitive asymmetry is established between linked SoAs [states-of-affairs], such that the profile of one of the two ([…] the main SoA) overrides that of the other ([…] the dependent SoA). This is equivalent to saying that the dependent SoA is (pragmatically) non-asserted, while the main one is (pragmatically) asserted” (Cristofaro 2003: 33). From this perspective, we shall investigate whether purposive relations between two situations are typically construed as symmetrical or asymmetrical, as reflected in conventionalized preferences for either co- or subordination. In addition to the cognitive status attributed to events, the conceptual space of complex sentences is structured further with respect to propositional act functions. For example, one event can make inherent reference to another event, as in complementation relations. An act of thinking, for instance, is intrinsically connected to another proposition, i.e. the content of the thought. A different propositional act function is that of modification. That is, a proposition can be construed as modifying either another situation (as in adverbial relations) or an entity that is part of the other situation (as in typical relative relations). These different propositional act functions, then, seem to underlie the traditional classification of complex sentence types. Each of these functional types can then be characterized more specifically in terms of the semantic relationship between the two situations. By this criterion, adverbial relations are subdivided into temporal, conditional, concessive, purposive and quite a few other relations. Similarly, complement-taking predicates can be grouped into a limited set of classes with respect to the semantic relationship they establish with the complement situation, such as modal, desid-



Chapter 4.  Purpose clauses and complex sentences 149

erative, perception, knowledge or utterance predicates. Relative relations, too, can be set up in a fairly wide semantic range. Finally, different combinations or ‘clusters’ of the conceptual distinctions introduced so far are typically associated with specific discourse-pragmatic functions. For example, if a temporal adverbial relation is construed asymmetrically, i.e. expressed by subordinating morphosyntax, then it often organizes the current stretch of discourse in terms of a figure-ground (Talmy 1975, Reinhart 1984) or foreground-background structure (Tomlin 1985, Thompson 1987). That is, those clauses often set an interpretive ground for the main (figuring) event. If construed symmetrically, the same combination of semantic content and propositional act function could result in a coordinate structure (e.g. sequential coordination), and in this case it would be considered, not as a figure-ground pattern, but as a c­ omplex-figure relation (cf. Croft 2001: 336ff.): neither situation is taken to provide a ground or reference point for the interpretation of the other. As can be seen, the conceptual space of complex sentences is multidimensional, and in order to appropriately determine the position of purpose clauses within this conceptual space, we will have to appeal to all of these dimensions. The syntactic space of complex sentences has traditionally been thought to be monodimensional, with a binary distinction being made between syntactically coordinate and syntactically subordinate constructions (e.g. Lyons 1968). From a typological point of view, it is now well-known that many language-specific structures defy an analysis in terms of this binary contrast (see, for instance, the discussion of non-embedded dependent clauses in Van Valin 1984). Consequently, the syntactic space of complex sentences has been re-organized into a multidimensional continuum of gradable oppositions (e.g. Haiman and Thompson 1984, Stassen 1985, Lehmann 1988). Although the details between individual proposals vary, the continua are similar in basically spanning a region between syntactically autonomous, fully independent sentences on the one end and syntactically downgraded or desententialized clauses on the other. The opposition between balanced and deranked verb forms we encountered earlier captures perhaps the essential dimension of this continuum. After all, a deranked verb form can by default be taken to be an indication of asymmetrical construal of two events and thus always corresponds to conceptual subordination. (Notice, though, that deranked verb forms need not necessarily indicate syntactic subordination in the traditional binary sense. A so-called ‘medial’ verb form in clause chaining is often deprived of TA marking and hence deranked, but can show conjunct illocutionary scope, which is ruled out for subordinate clauses, cf. Bickel 2006b). Needless to say, this very brief overview cannot do justice to the vast research into typological parameters of clause combining, but it will suffice to sensitize the reader to the key dimensions and issues in the organization of complex sentence systems. It thus only

150 A Typology of Purpose Clauses

provides the background against which purpose clauses will now be analysed in more detail.

4.1.2  Coordination and subordination: (A)symmetry in purpose clauses Our initial definition of purpose constructions — roughly: clause combinations with conventionalized purposive semantics — was syntactically relatively unconstrained in that it did not require a special structural type of clause linkage. Specifically, it did not make reference to any notions in the syntactic space of complex sentences, such as coordination or subordination. In this section, we will take a very brief look at how purposive constructions are classified in the individual language-specific systems of complex sentences. Although, of course, the criteria for status assignment vary from language to language, most grammars make a fairly clear-cut and transparent distinction as to whether purpose clauses belong to a paradigm of ‘subordinate’ (i.e. clearly deranked or in some other respect dependent) clauses, to ‘coordinate’ (i.e. syntactically fully independent) clauses, or to some other category (e.g. cosubordination). We are not interested here in the specific criteria that led to such classifications in individual languages (cf. §3.2.1 on deranking for answers to this question). We rather take the classifications for granted and use them as a heuristic to determine whether purposive relations are preferably construed as conceptually symmetrical or asymmetrical. The underlying assumption is, again, that syntactic status provides an indication of cognitive status. The issue is actually very straightforward and does not call for much discussion. 197 of our 218 constructions (90.3%) were classified as syntactically ‘subordinate’ by the authors of the reference materials. That is, if a purpose relationship is conventionally associated with a clause-combining construction, then it will be found among the syntactically dependent constructions of a language. This, in turn, could be interpreted as purposive relationships being preferably construed as asymmetrical. In light of various discussions in Chapter 3, it appears quite plausible why this should be the case. Notably, we argued that purpose clauses express a mental-state relation with the matrix event (‘agent-binding’), which makes them conceptually tightly integrated with this event. Syntactic subordination (e.g. by deranking) is a linguistic response to this conceptualization. Nevertheless, we do occasionally find purpose clauses that have been classified as belonging to the set of coordinate, i.e. fully independent clauses. Note that we are not dealing here with additive coordination that potentially allows for purposive inferences. We saw above that this is, in fact, a very frequent implicature of coordination. In other words, we target constructions in which such implicatures have been conventionalized to a considerable degree. Such ‘purposive coordin-



Chapter 4.  Purpose clauses and complex sentences

ate clauses’ are attested, for instance, in Duuŋidjawu, Wambaya, Kiowa, Lakhota, Acehnese, Boumaa Fijian and Semelai, i.e. notably certain parts of the Pacific Rim (they are completely absent from the South American languages in our sample, though). Their geographical distribution is thus heavily skewed. Here are two representative examples of purposive coordination: (162) Coordinating conjunction ngaba in Wambaya (West Barkly: Northern Australia; Nordlinger 1998: 224) Angbardi-j-ba gurl baba-wuli-janka ngaba wurlu gulug-ba. build-th-fut du.imp elder.sibling-du-dat co 3du.s sleep-fut ‘Build (a windbreak) for (your) brothers so they can sleep.’ (163) Coordinating conjunctions c’a and h’écel in Lakhota (Siouan: USA; Buechel 1939: 254) ‘Adverbial clauses introduced by conjunctions expressing […] purpose […] cease to be subordinate clauses and become principal clauses as part of a compound sentence. The reason for this construction is this: the corresponding Lakota conjunctions are coordinating conjunctions..’ Finally, some authors note that a particular purpose clause construction cannot readily be handled with a binary distinction. Thus we find, for instance, a medial (i.e. non-embedded) purpose clause in Kobon and Kewa clause-chaining. Serial verbs, finally, form a different category altogether (and were also discarded in our discussion of deranking). In sum, conventionalized purposive coordination and other (intermediate) forms of syntactic clause linkage do exist, but are definitely much less common than clearly subordinated purpose clauses.1

4.1.3  Purpose clauses and other adverbial clauses There is little disagreement in the theoretical and typological literature that purpose clauses constitute a particular semantic type of adverbial clause (i.e. a clausal modifier of a proposition). As such, they often form a system with temporal, conditional, causal, concessive and other clauses. If we systematically compare purpose clauses to these other adverbial relations in all of our 80 languages, two tendencies strike the eye: first, there are recurrent overlaps in the coding of purpose and some other adverbial functions, notably reason and result; second, in quite a few languages, purpose clauses differ from the entire set of adverbial clauses in 1.  Note that for other types of adverbial clauses, notably causal constructions, the affinity with coordinate linking devices is much stronger, a finding which Diessel and Hetterle (to appear) investigate typologically and explain by a specific discourse-pragmatic function of many causal clauses, i.e. the explanation of a controversial statement.

151

152 A Typology of Purpose Clauses

Table 35.  Multifunctionality of purpose clauses (N=80) Purpose shows overlaps with…

Absolute frequency

reason or cause clausesa result clauses temporal clauses conditional clauses manner clauses

11 11 5 5 3

Note: If a purpose marker serves simultaneously in both reason/cause and result functions, each of them was counted separately. a  The authors of most reference grammars do not systematically distinguish between reason and cause. Verstraete (2008) and Wierzbicka (1998: 184) show, however, that there are important semantic differences between these two notions.

a number of important and diverse structural respects and are thus remarkably different from the ‘system’ of adverbial relations. We will deal with each of these issues in turn. Probing into patterns of multi-functionality first, we can observe that purpose clauses as a whole or their characteristic purposive marker are frequently polysemous with respect to one of the functions displayed in Table 35. The affinity to both reason/cause and result clauses to purpose was, in fact, part of our initial definition of purposive relations: a purpose was conceived of a reason formulated in terms of an intended result (cf. §2.3). Hence it is not surprising that languages tend to cover these notions by a single marker; from a functional point of view, this would evidence an overriding of paradigmatic iconicity by economic forces (‘Minimize Forms’, Hawkins 2004: 38). Let us look at purpose and result clauses first. It was noted previously that these two clause types differ as to whether the realization of the subordinate situation is actually entailed (result) or just implicated (purpose). In fact, finite purpose clauses often need to be marked for irrealis or hypothetical mood in order to be distinguished from result clauses, as in English so that clauses (cf. §3.2.2). Chalcatongo Mixtec, too, has the same marker for purpose and result, the complementizing proclitic xa=. While result clauses tend to be marked by the realis inflection, purpose clauses commonly lack this inflection or are alternatively marked for potential mood: (164) Chalcatongo Mixtec (Oto-Manguean: Mexico; Macaulay 1996: 157) a. Ni-xàʔá=ri [xa=ní-žee seʔé=rí]. compl-give=1 [purp/res=compl-eat.rl child=1 ‘I gave (food) so that my child could eat.’ (elicited: ‘I gave food to my baby.’) b. Ni-xãã=rí žaʔa [xa=sáʔa náa=ri molí]. compl-buy=1 chile [purp/res=make mother=1 mole ‘I bought chiles so that my mother could make mole.’



Chapter 4.  Purpose clauses and complex sentences

In other languages, such as Koyra Chiini, the two constructions are completely identical: they are both introduced by the temporal conjunction hal ‘until’, and both even carry a subjunctive marker. The essential semantic difference can, therefore, only be made overt by subsequent linguistic material: (165) Koyra Chiini (Nilo-Saharan: Mali; Heath 1999: 297) a. A-a haŋga a–a dira [hal a ma soroku guusu di 3sg.s-impf follow 3s.s-impf walk [until 3sg.s subj fall pit def ra]. loc ‘It (= the animal) just kept on walking so that it fell into the pit.’ b. Ay faani [hal ay ma duu wuraa] 1sg.s dig [until 1sg.s subj get gold (mere ay na gar ga). (but 1sg.s neg find 3sg.p ‘I dug in order to get some gold (but I didn’t find it).’ In West Greenlandic, similarly, result clauses are indistinguishable from purpose clauses and hence need disambiguating morphology, such as a demonstrative taamaattumik ‘therefore’ at the beginning of the subsequent clause (Fortescue 1984: 67). Notice that English infinitival constructions, which in the adverbial domain normally code purpose clauses only, can also exhibit some completely nonintentional readings and instead receive a resultative interpretation.2 (166)–(167) are taken from Huddleston and Pullum (2002: 1224), while (168) was found in a typological essay by Dryer (1997): (166) I went around the corner, just to run into my old friend Jack. (167) She opened the curtains to see that the ground was covered in snow. (168) Over the past thirty years, linguists have postulated hypotheses regarding what is a possible human language, only to have some other linguist demonstrate the existence of an exception to the hypothesis. In these examples, it is fairly obvious that the situation encoded by the ‘purpose’ clause is not intended by the matrix agent (on the contrary, actually, especially in [168]). Such ‘non-intentional’ purpose clauses are comparatively rare to find cross-linguistically; in fact, the only examples I detected come from Martuthunira (Pama-Nyungan: Australia; Dench 1995: 253) and Modern Greek:

2.  ‘Result‘ is understood in a rather broad sense here, i.e. as a situation whose occurrence has been made possible (or at least been facilitated) by the event in the matrix clause.

153

154 A Typology of Purpose Clauses

(169) Modern Greek (adapted from Holton et al. 1997: 464) Tης πήρε τα λεφτά, 3sg.f.gen take.3sg.aor art money.n.acc [για να μην της τα επιστρέψει ποτέ]. [for subj not 3sg.f.gen 3.n.acc return.3sg.prs never ‘He took the money from her, never to return it.’ Although it is not quite clear what leads to the loss of purposive semantics in these specific cases, I will propose, in §4.2.3, a mechanism that may account for the development from purpose to result clauses more generally. Reason and/or causal clauses also often share a single marker or construction with purpose. Thompson and Longacre (1985: 185) argue that these clauses have a common canonical semantic function: they provide an explanation for the matrix clause action.3 This is reflected, for instance, in Qiang, where a causal postposition is used to mark purpose and causal clauses: (170) Qiang (Tibeto-Burman: China; LaPolla 2003: 240) [The: stuaha tɕhə-(s ŋuə)-χua-ȵi], (qupu) dzigu̥] me ɹᶎ. [3sg food/rice eat-noml cop-because-adv 3sg money look.for ‘In order to live, s/he seeks to earn more money.’ This postposition4 is also the benefactive marker in Qiang, and so the pattern illustrated here is completely parallel to Turkish, where the dative postposition için ‘for’ marks both purpose and causal clauses (Kornfilt 1997: 73f.), and to Amele nu ‘for’ with exactly the same distribution (Roberts 1987: 58–9). Structural overlaps between purpose and reason/cause clauses are, in fact, found all across the world, from the Mandarin serial verb construction over Lezgian and Hausa to Kiowa, ­Lakhota, West Greenlandic, Hixkaryana and Paumarí. The only macro-area where this tendency is not observable is Australia, for reasons to become clear below.5 3.  Wierzbicka (1998: 184) demonstrates that from a conceptual point of view, because is a simpler notion than purpose since the latter contains the former as a semantic primitive. “Generally speaking, one could expect that simpler configurations of semantic primes have a greater chance of being grammatically encoded in many languages than more complex ones” (Wierzbicka 1998: 185). Interestingly, however, Wierzbicka claims that purpose clauses are much more frequently grammaticalized in languages than causal clauses (she does not provide any quantitative evidence for this claim, though). As a possible explanation, she suggests that purpose clauses (or the notion of purpose more generally) play a central role in human life, and this factor may also be a drive for developing a grammatical construction. 4.  Note that LaPolla (2003: 240) labels χua(-ȵi) a postposition, even though it appears as a bound marker here. 5.  Where purpose and result clauses are coded differently, they often show structural parallelism in the sense that while purpose markers tend to have goal-oriented markers, causal caus-



Chapter 4.  Purpose clauses and complex sentences

The remaining noteworthy patterns of multifunctionality exist between purposive and temporal, and purposive and conditional marking, respectively. The temporal morphemes in question typically denote general temporal sequence (e.g. Mandarin Chinese, Kiowa, Tümpisa Shoshone) or posteriority (cf. [165] again), but (perhaps somewhat unexpectedly) simultaneity is also an attested overlap (Ute, Barasano). Conditional morphemes, finally, can show up in both positive purpose clauses (e.g. Nkore-Kiga, Wardaman) and avertive constructions (e.g. Amele). For the latter overlap, we will encounter a possible diachronic explanation later on. While patterns of multi-functionality and other structural parallels are not unusual among adverbial clauses and hence not confined to purpose, there is some decisive cross-linguistic evidence that makes purpose clauses appear to occupy a special place in the adverbial clause system. In language after language, the coding and/or behavioural properties of purpose clauses differ conspicuously from those of all other adverbial clauses. We have seen some of these differences before, and will give them more prominence by assembling and quantifying them here: (i) In 16 languages (16∕80 = 20%), purpose clauses are the only semantic type of adverbial clause that is expressed by a non-finite (deranked) construction. (ii) In 12 languages (12∕80 = 15%), purpose clauses, but none of the other adverbial clauses, employ an infinitive or an otherwise highly integrated construction. In §3.2, (i) and (ii) were related to Cristofaro’s (2003) ‘Adverbial Deranking Hierarchy’, in which purpose clauses outrank all other adverbial relations with respect to deranked verb forms. (iii) Unlike many other adverbial clauses, purpose clauses often leave their subject (and sometimes also their object) argument unexpressed. This tendency has been captured quantitatively in the ‘Adverbial Argument Hierarchy’ (Cristofaro 2003), on which purpose clauses, again, outrank all other types of adverbial clauses. (iv) In 9 languages (9∕80 = 11.3%), purpose clauses are the only type of adverbial clause that either rigidly or consistently follows the associated main clause. (v) Finally, some of these properties combine to define a special status of purpose clauses. In Ungarinjin (Wororan: Western Australia), for example, purpose clauses are the only non-finite adverbial clauses and the only adverbial clauses that are consistently postposed after the matrix clause (Rumsey 1982).

es often have source-oriented morphemes, e.g. Wardaman: ‘he’s coming from having spent his money’ = ‘because he has spent/wasted all his money’ (Merlan 1994: 284). This demonstrates that the same principles of grammaticalization are at work in the adverbial domain (case marker/adposition > conjunction).

155

156 A Typology of Purpose Clauses

In the course of this study, I have tried to motivate why purpose clauses can show such anomalous behaviour. We saw, firstly, how the conceptual structure of purposes leads to a unique combination of predetermined time reference, inherent modal orientation and, most importantly, a mental-state relation that is crucially absent from the most widely recognized types of adverbial clauses. In fact, Cristofaro claims that the “semantic difference between purpose relations and the other adverbial relations is greater than the difference between any of the other adverbial relations” (Cristofaro 2003: 179). The semantic components of purpose clauses jointly motivate the morphosyntactic properties listed in (i)–(iii) above. Along other dimensions of the conceptual space of complex sentences, purpose clauses also behave differently from many adverbial clauses. For example, typical adverbial relations tend to be construed as figure-ground structures, with the adverbial clause providing a thematic framework for the interpretation of the main clause. Purpose clauses, by contrast, do not generally act as such scene-setting devices, as was shown in §3.5. This explains, for example, why in Imonda (Border: Papua New Guinea), all adverbial relations, except for purpose, can be expressed by a so-called ‘topic clause’: whenever the topic construction is chosen, the clause is taken to provide a situational framework for the main clause (cf. Seiler 1985: 8, 198–205). Purpose clauses never take the form of topic clauses. Their reluctance to be construed as scene-setting ground structures also explains our finding in (iv) above: purpose clauses generally strive for a sentence-final (i.e. rhematic) position. This position, crucially, clashes with a typical figure-ground interpretation. Croft (2001: 326), too, notes that purpose clauses are somewhat special in tending towards a complex-figure construal (a feature they share with result clauses, though). This may also reflect their information-structural profile: I argued that purpose clauses do not usually convey presupposed or given but new information, and as such may well make an important contribution to the current discourse topic. This observation ties in neatly with Kazenin’s (1994: 93) argument that purpose clauses occupy a mid-way position between foregrounding and backgrounding elements in discourse. In this respect, purpose clauses are similar to sentential complements, which are also not typically asserted yet often carry foregrounded information. In sum, purposive relations deviate from typical adverbial relations in a number of important respects. It may seem contradictory at first glance that purpose clauses tend to be syntactically much more integrated and deranked (or put differently: more clearly subordinated) than other adverbial clauses, yet at the same time defy the typical figure-ground construal often associated with adverbial subordination. But this simply underlines the importance of a multidimensional approach to the variation space of complex sentences, especially one that keeps semantic and discourse-pragmatic aspects apart. Of course, the present findings should not be taken to exclude that other types of adverbial clauses, especially less well-studied ones



Chapter 4.  Purpose clauses and complex sentences 157

Table 36.  Subordinate clause system in Lavukaleve (Terrill 2003: 423) Subordinate clause type

Cross-referencing system

ADVEBRIAL

split-ergative Anterior split-ergative Potential split-ergative Surprise as for main clauses as for main clauses but with Agreement Suffix

PURPOSIVE RELATIVE

Subtype

Overt marker of clause type -ge -le -meon -ham none

such as manner or concessive-conditional constructions, could also show a unique cluster of both semantic and discourse-pragmatic properties. But what is certainly exceptional about purpose clauses is the accumulation of such properties and the fact that purposive constructions occasionally form a separate class of clause combining altogether. This phenomenon seems to be particularly common in Australia and parts of Oceania. In Ngiyambaa, for example, all other adverbial relations are encoded by either a finite clause suffixed by -ba or a so-called ‘circumstantial clause’ (in other Australian languages, these have been called ‘generalized subordinating clauses’, cf. Merlan 1982: 9). Purpose clauses, by contrast, have distinct purposive inflections, modal marking and co-referential argument constraints. In Martuthunira, similarly, purpose clauses are distinct constructions, while all other types of adverbial clauses are formally realized by relative clauses (Dench 1995: 240ff.). A very instructive example also comes from Lavukaleve. In this language, the system of subordinate clauses is divided into relative, adverbial and purpose clauses (see Table 36). As can be seen, purpose clauses are distinguished from the set of adverbial clauses (among other things) by virtue of operating according to a different agreement system. The fact that many Australian and some Oceanic languages set purpose clauses apart from other adverbial clauses might also explain why in these areas, systematic patterns of multi-functionality between purpose, reason and result are much rarer than in other regions (cf. above). We conclude at the end of this section that purpose clauses, while certainly being adverbial modifiers, form a very distinctive construction type within this class and can sometimes even develop in such a way that they form an entirely separate class on structural grounds. In the next section, we will see that at the same time, purpose clauses often show synchronic affinities to a functionally rather different group of constructions, namely sentential complements.

4.1.4  Purpose clauses and complement clauses A recurrent observation in the study of reference grammars is that purpose clauses display unmistakable structural similarities to complement clauses, i.e. sentential

158 A Typology of Purpose Clauses

objects (and sometimes subjects) of a superordinate verb. As a matter of fact, a careful investigation shows that in 62 languages in our sample (62∕80 = 77.5%), at least one purpose clause construction shares some of its morphosyntactic properties with (certain kinds of) sentential complements, up to being completely identical with them. Conversely, in only 18 languages are purpose and complement constructions fully distinct. This is certainly a very strong universal tendency that by itself deserves further attention. Moreover, it turns out that in almost all cases in which purpose clauses deviate dramatically from the set of adverbial clauses (cf. the previous section), their overlap with sentential complements is particularly pronounced. Needless to say, such structural affinities can take a number of different shapes, but there are some dominant patterns of commonalities between purpose and complement clauses, which shall be outlined in the following. In quite a few languages, purpose and complement clauses have partial overlaps resulting from sharing a certain number of linguistic features. Commonly, the purposive marker (e.g. a conjunction or affix) is also used in certain kinds of complements. For example, the Tzutujil preposition ch(i) ‘at’, ‘to’, ‘in order to’ introduces purpose clauses, but is also described by Dayley (1985: 392) as a ‘complementizer’: (171) Tzutujil (Mayan: Guatemala; Dayley 1985: 383, 392) a. Xe7el [chi qach’ejyiik]. b3:pl.arrived [purp our.being.hit ‘They arrived to hit us.’ b. Xinb’ij chaawe [chi nqaajo7 serwéesa]. b3.a1.told to.you [comp b3.a1:pl-want beer ‘I told you that we want beer.’ In Paumarí, similarly, the purposive morpheme kaimoni can be used to mark desiderative complements, which (in contrast to purpose clauses) take a so-called non-thematic suffix (nth) on the subordinate verb: (172) Paumarí (Arauan: Brazil; Chapman and Derbyshire 1991: 230) Ihai hida isai-a bi-nofi-ja [bi-kidafi-ki kaimoni]. medicine dem.f child-erg 3sg-want-imm [3sg-swallow-nth purp ‘It is medicine the child wants to swallow.’ A somewhat different pattern of partial overlap arises when purpose and complement clause share, not the same primary marker (e.g. conjunction), but an identical verb form. For Yagua, for instance, we saw in (97) above that purpose clauses have an infinitival verb form, to which an allative morpheme is suffixed. In complement clauses, the infinitive can also be used, but it appears without any additional primary marking:



Chapter 4.  Purpose clauses and complex sentences 159

(173) Yagua (Peba-Yaguan: Peru; Payne and Payne 1990: 336–7) a. Sa-sííy-maa [jíy-jimúútya-janu-jų`-níí]. 3sg-run-prf [cor1-help-inf-all-3sg ‘He has run to help him.’ b. Jiy-vąąra-rà [jųnúúy-janu]? 2sg-want-inan [look-inf ‘Do you want to look?’ Finally, partial correspondence also exists when typical TAM features of purpose clauses also show up in (some) complements, such as desiderative markers or potential-hypothetical-irrealis markers in Modern Greek or Acehnese. What is much more common, however, is actually total structural overlap between at least one purposive construction and at least one type of complement clause.6 46 (!) such languages are discernable in our sample, or 47 if we add English as our standard of comparison. As the following map shows, such complete overlaps are not restricted to particular macro-areas, although there may well be some areal clusters (Figure 28). One interesting example should suffice to illustrate this pattern. In Babungo (Niger-Congo: Cameroon), purpose clauses do not only share a common conjunction with complement clauses, but also so-called logophoric pronouns which refer back to the subject of the main clause. In fact, Schaub (1985: 39) claims that

Figure 28.  Languages with at least one instance of identical purpose and complement clauses (N=46) 6.  Authors of some reference grammars have recognized this affinity between the two clauses types in question, and have hence introduced the label ‘purposive complement’ for purpose clauses which are coded by material typically found in complement clauses. We will return to this issue below.

160 A Typology of Purpose Clauses

“there are no purpose adverb clauses, but noun clauses are used to express purpose” (but cf. below on this point). Thus the following two sentences are structurally i­ dentical: (174) Babungo (Niger-Congo/Bantoid: Cameroon; Schaub 1985: 39) ŋw jwì féen [lāa yì gwía yítēe nyə̏]. he come.prfv here [comp he.log cut.impf palmtree this ‘He has come here in order to cut this palmtree.’ (175) Babungo (Schaub 1985: 30) ŋw z [lāa v´ tíi yì]. he hear.prfv [comp they call him.log ‘He heard that they called him.’ It should be pointed out that the above-made observations refer to overlaps between individual purposive constructions and individual complement clause constructions, i.e. neither all purpose clause constructions nor all different complement clause constructions in a language need to be affected by such overlaps. However, we can also encounter languages in which the whole class of either purpose clauses or sentential complements is expressed by the other class. On the one hand, there are languages which distinguish between adverbial clauses proper and adverbial meanings expressed by complement clauses. Importantly, these latter adverbial meanings typically (and sometimes solely) include purpose. This pattern applies to Babungo above, and another representative example comes from Slave. In her description of the language, Rice sets up a class of adverbial clauses characterized by conjunctions which “indicate time, causation, and condition” (Rice 1989: 1052) and a distinct class of complements marked by what she calls a ‘purposive complementizer’: (176) Slave (Athapaskan: Canada; Rice 1989: 1259) [Sedá newozo gha] sesodá dawehya. [1sg.eye 3opt.be.good comp 1sg.glasses 1sg.have.located ‘I wear glasses to protect my eyes.’ Crucially, this complementizer also appears in non-purposive complements: (177) Slave (Rice 1989: 1261) [Tįch’a yú dats’ehlee gha] gonezo. outside clothes nspec.hangs.pl.p comp area.is.good ‘It would be good to hang the clothes out.’ Care should be taken here not to misinterpret such a synchronic pattern. While it looks at first glance as if Slave does not possess a distinct set of purpose clauses,



Chapter 4.  Purpose clauses and complex sentences 161

using complements instead, Rice herself (pers. comm.) comments that the purposive marker is historically most likely the older one, with the non-purposive complement structure being diachronically derived from it. Therefore, on the other hand, it may well be that languages lack a distinct class of sentential complements at some point in their history, pressing purpose clauses into the service of fulfilling this function. This is reflected in the synchronic description of Wardaman, for example. Although Merlan (1994) operates with the term ‘complement’, the constructions to which she applies this term are fundamentally different from what is commonly understood to be a complement clause, i.e. a sentential object. Examples of sentential objects are, in fact, extremely rare to find in the reference grammar, but where they appear, they carry a purposive inflection: (178) Wardaman (Gunwinyguan: Northern Australia; Merlan 1994: 278) Ngan-jewu-rri [ginye-yi-yin mayi-wu]. 3sg/1sg-show-pst [cook-noml.purp food-dat ‘He showed me how to cook (the) food.’ In order to express sentential objects, the only alternative to such ‘purposive complements’ is the use of coordinated main clauses. It thus really appears that Wardaman lacks a structurally well-established class of subordinate complements, and purpose clauses can be recruited to fill this gap. Dixon (2006: 6) claims that this phenomenon represents a more general and quite common typological pattern. In sum, the evidence for a close structural relationship between purpose and complement clauses is overwhelming. There is, of course, a highly plausible functional underpinning of this mutuality. Although we have seen that purpose and complement clauses can — in the extreme case — wholly substitute for one another, it is much more typical for purpose clauses only to be used for a restricted set of sentential complements. Specifically, of the semantic complement types commonly distinguished (cf. Noonan 1985), it is chiefly manipulative-jussive and desiderative relations, less frequently also modal, perception-knowledge and utterance verbs, that tend to be coded by purpose clauses.7 In Kolyma Yukaghir, for instance, purpose clauses are coded by an infinitival suffix called the ‘Supine form’; this construction can be used for some sentential complements, notably those whose subject is co-referential with a participant in the main clause, and in which the situation is non-realized (hypothetical, irrealis): 7.  Languages differ vastly with regard to the specific semantic types of verbs they group together under a particular complementation strategy, and many reference grammars do not distinguish systematically between different types at all, so it is quite difficult to come up with a reliable semantic map that portrays all specific overlaps between purpose and complement clauses. Therefore, I will concentrate on the most prominent patterns here, refraining also from making any precise quantitative statements.

162 A Typology of Purpose Clauses

(179) Kolyma Yukaghir (Yukaghir: Siberia; Maslova 2003: 433) Met-in tet čilge kej-k [qaŋsā čiččī-din]. I-dat your branch give-imp.2sg [pipe clean-purp ‘Give me your branch, for me to clean the pipe.’ (180) Kolyma Yukaghir (Maslova 2003: 415) [Čumu lejdī-din] erd’ie-j. [all know-purp want:intr:3sg ‘He wants to know everything.’ The functional overlap of such desiderative complements with purpose clauses is evident: both involve what we have called agent-binding, i.e. a participant’s will or desire to bring about a certain situation, and in this sense both clause types express a mental-state relation and show a high degree of semantic integration, which is not characteristic of other complement clauses (e.g. utterance predicates). Furthermore, just as in purpose clauses, the time reference of desiderative complements is predetermined to be to the future, which entails that the realization of the desired situation is hypothetical at the moment of speech. The same observations apply to jussive-manipulative relations (‘order’, ‘ask’, ‘tell so. to do sth.’). Both classes could be subsumed under the pragmatic label of ‘directives’ (cf. Searle 1969). It is, therefore, not surprising that in her study of subordination systems, Cristofaro finds that both manipulative and desiderative complements show exactly the same behaviour as purpose clauses with respect to verb deranking, particularly non-specification of predetermined TA information and dependent moods, and in regard to argument omission, as the following implicational hierarchies ­demonstrate: (181) Subordination Deranking Hierarchy (Cristofaro 2003: 229) Phasals, Modals > Desideratives, Manipulatives, Purpose > Perception > Before, After, When, A relativization, S relativization > Reality condition, Reason, O relativization, Knowledge, Propositional attitude, Utterance, Indirect object relativization, Oblique relativization (182) Subordination Argument Hierarchy (Cristofaro 2003: 230–1) Modals, Phasals, A relativization, S relativization > Desideratives, ­Manipulatives, Purpose > Perception > Before, When, After, Reason, Utterance, Propositional Attitude, Knowledge, Reality condition What these hierarchies also show is that modals are often included in the discussion of complementation, although they are often deprived of their sentential status and have become highly grammaticalized auxiliary-verb complexes, as English I must go to the doctor’s. Interestingly, purpose clauses can also show structural overlaps with precisely such modal constructions. For example, in Yidiɲ, which



Chapter 4.  Purpose clauses and complex sentences 163

we have encountered multiple times before, the suffix used to mark purposive subclauses can also be concatenated to a main verb to yield deontic necessity: (183) Yidiɲ (Pama-Nyungan: Australia; Dixon 1977: 344) Gana ŋayu giʃa gunʃi:na bama:gu. try I.sa quick.abs return.purp person.all ‘I must try to hurry back to the people.’ The same process can be found in Ngiyambaa, and Palmer (1986: 83) claims, in fact, that this is a characteristic of Australian languages. However, it turns out that languages of other macro-areas and genetic affiliations, too, use purpose markers as deontic modals in main clauses. In Huallaga Quechua, for instance, the purposive suffix also extends into the domain of deontic necessity: (184) Huallaga Quechua (Quechuan: Peru; Weber 1989: 113) Kanan-qa simri wamra-nchi-kuna-ta iskuyla-man chura-y-paq. now-top always child-1incl.poss-pl-obj school-gl place-inf-purp ‘Now, we always ought to put our children into school.’ More generally, the realm of purpose and deontic modality share the property of a hypothetical result state and someone’s will or desire for it to be obtained. This can be the desire of the agent himself, as in (184), resulting in agent-binding, or the speaker’s, resulting in deontic statements such as You must hand in your paper by April 15. It is, therefore, not surprising to find languages with highly polyfunctional markers covering both domains. Boumaa Fijian has indeed developed an extremely versatile marker me which functions as a purposive coordinating conjunction, as a complementizer in desiderative clauses, sometimes with “jussive overtones” (Dixon 1988: 291), and as a standard way to form imperatives, i.e. clearly deontic expressions. In all of these cases, me can roughly be translated into English should, and it thus provides a good example of Gil’s (2001) concept of ‘macro-functionality’. Unfortunately, the database does not evidence further examples of this phenomenon. Quotative markers and utterance complements, finally, are also found in purposive constructions and here, too, a functional motivation can be invoked. Sri­ dhar (1990: 72) proposes that purposes can be conceptualized as “verbalized selfsuggestions” in Kannada, as reflected in the following construction: (185) Kannada (Dravidian: India; Sridhar 1990: 73) [EmTia:rnalli u:tappa tinno:Na anta] mane biTTu horaTa. [MTR.loc a.dish eat.sugg quot house leave leave.pst.sg.m ‘He left home to eat utappa [a pancake-like snack] at the MTR. (lit. ‘He left home, saying “Let me/us eat utappa at the MRT.”’)

164 A Typology of Purpose Clauses

The literal translation of this example neatly shows the link between direct quotation and purpose. Since this link is merely an implicature at the initial stage of such periphrastic constructions, it heavily relies on pragmatic inferences: an action paired with a statement suggesting the realization of a subsequent event is inferred to be purposive (cf. also §3.4.3 above). In Kannada, this implicature has already become very prominent, so that the above construction is now more or less conventionally associated with purpose. In yet other languages, there has been further progression, to the effect that quotative markers and utterance verbs have actually grammaticalized into purposive conjunctions. We will deal with this issue in §4.2.6. As a final point, it should be mentioned that purpose and complement clauses also behave similarly with respect to their discourse-pragmatic functions. While both clause types are usually construed asymmetrically, in the sense that neither a purpose nor a complement clause represents the asserted part of the complex sentence in which it occurs, both clauses typically encode non-presupposed, new or important information in the ongoing discourse. Thus in a sentence like (186) She said [that Tom had an important business meeting today]. the complement clause may well encode the information most relevant for the current discourse topic (cf. Erteshik-Shir and Lappin 1983 for the so-called ‘lie test’, which proves that although the complement clause is non-asserted, it is the most dominant information-structural part of the complex sentence). In this way, both purpose and complement clauses have an intermediate status between foregrounding and backgrounding clause types. To conclude, there are several well-motivated, i.e. functional, overlaps between purpose and (some) complementation relations, which may lead to diachronic extensions of one clause type to the other, either partially or with regard to the entire construction. On a more cautious interpretation, one may thus argue that the conceptual overlaps between the two domains act as selection pressures on diachronic change, which then leads to non-accidental structural similarities synchronically. A more radical interpretation of the synchronic facts is that purpose clauses “can be regarded as intermediate between the domains of complementation and adverbial modification, because they have to be defined on the basis of features from both domains” (Verstraete 2008: 779). Verstraete summarizes those features as laid out in Table 37. With regard to formal features, Verstraete’s argumentation is that since purpose clauses are adjuncts and hence not — unlike complements — governed by a matrix verb that denotes a mental state (‘think’, ‘want’ etc.), other formal devices need to indicate the mental-state relation characteristic of purpose clauses. This is done, for example, by recruiting clause linkers from the domain of direct speech



Chapter 4.  Purpose clauses and complex sentences 165

Table 37.  The intermediate position of purpose clauses (adapted from Verstraete 2008: 779) Meaning Clause type

Complement Purpose Other adverbiala

Typical adverbial

Typical complement

Formal strategies Mental state predicate in main clause

Mental state of main clause Circumstances participant of main clause event

+

Conjunction: reported speech

Dependent clause: mood

+

+

+/−

a  Other adverbial clauses here comprise the ‘usual suspects’, but not clauses of reason (as opposed to cause) and intended endpoint (both of which pattern in the same way as purpose clauses).

and thought or by overt moods that signal agent-binding. It is on the conceptual level, then, that purpose clauses need to be defined by features from both domains, and the recurrent marking strategies reflect this conceptual ‘categorization problem’ accordingly.

4.1.5  Purpose clauses and relative clauses Palmer (1986: 178) already notes a formal affinity between purpose and relative clauses, a phenomenon he terms ‘relative purpose’. In order to approach this issue, let us take a look at Mapudungun. In this language, all subordinate clauses are nominalized by adding a suffix to the subordinate verb. Purpose and relative clause are coded differently here: while purpose clauses are marked by an instrumental nominalizer, relative clauses are marked by a so-called objective ­nominalizer: (187) Mapudungun (Araucanian: Chile; Smeets 1989: 268) Elu-a-y-u permisio ta-mi give-irr.p-ind-1nsg-du-dat.a permission the-2sg.poss amu-tu-a-m. go-iter-irr-noml(instr) ‘I will give you permission so that you can go back.’ (188) Mapudungun (Smeets 1989: 268) Elu-a-y-u permisio ta-mi give-irr.p-ind-1nsg-du-dat.a permission the-2sg.poss amu-tu-a-el. go-iter-irr-noml(rel) ‘I will give you permission to go back.’

166 A Typology of Purpose Clauses

What is instructive about these examples is their propositional content and the form of the English translations. Other (i.e. contextual) things being equal, the difference in meaning is minimal, which suggests a substantial conceptual overlap between purposive and some kinds of relative relations. What is more, in the relative clause interpretation ([188]), English itself uses a genuine purpose clause — the to-infinitive construction. Unlike Mapudungun, then, English shows also formal overlap between purpose and relative clauses (at least in this particular non-­finite construction). In our sample, 14 languages (14∕80 = 17.5%) exhibit such formal similarities. Interestingly, these fall into a very limited number of distinct types: (i) Both purpose and relative clauses are marked by a general subordinator or complementizer similar to English that; (ii) purpose and relative clauses share a peculiar verb form (either an infinitive, a participle or a nominalization) not usually or rarely found in other subordinate clauses; (iii) purpose clauses are formally indistinguishable from (infinitival or reduced) relative clauses. These formal classes are distributed across the sample as shown in Figure 29. An example of an identical subordinator used for both purpose and relative clauses comes from Ma’di, in which purpose clauses carry the same general subordinator -le ~ -lε as object relative clauses: (189) Ma’di (Nilo-Saharan/Central Sudanic: Sudan, Uganda; Blackings and Fabb 2003: 202) Ká kì máŋgà āzī-lé gбà sùkúrù ní. 3 pl mango (n)-sell-purp (n)-pluck school ben ‘They are picking mangoes to sell for a school.’

1. [6] Identical subordinator (‘that’) 4. [3] Identical verb form 6. [5] Identical structure

Figure 29.  Languages with formal similarities between purpose and relative clauses (N=14)



Chapter 4.  Purpose clauses and complex sentences 167

(190) Ma’di (Blackings and Fabb 2003: 201) Ilí [ágз rì ςà ŋз-lέ] rì lзtʃī. knife man def poss (n)-break-sub def sharp ‘The knife which the man broke is sharp.’ Purpose and relative clauses sharing the same verb form are found, for instance, in Tümpisa Shoshone (Uto-Aztecan: California, Nevada), where both constructions can be marked by the infinitive suffix -nna. Interestingly, no other type of adverbial clause is formed by -nna, but a number of sentential complements are. We will see below that there is actually a diachronic mechanism which can account for such patterns. Complete identity between the two clause types can principally take either of two forms. In the first type, a purpose clause is expressed by genuine relativization. In this case, purpose clauses are governed by a head noun. In Somali, for instance, they follow the head noun si ‘way’, ‘manner’: (191) Somali (Afro-Asiatic/Cushitic: Somalia; Saeed 1999: 221) Waxaa gurigoodii soo galay jiirarkii, [si ay cunno uga wáxa(a) house.their all entered mice.the [way they food for.from xadaan]. steal ‘The mice went to their house in order to steal food from it.’ In the second type, a purpose clause is also used for relativization. This appears to be particularly common with infinitival or nominalized constructions, as in the English translation of (188) above. Synchronically, the distinction between the two clause types may be so blurred that we get a structurally ambiguous sentence. Thus in Krongo (Kadugli: Sudan), a so-called ‘purposive VP’ lacks an overt subject and is hence ambiguous between a purposive and a relative reading: (192) Krongo (Kadugli: Sudan; Reh 1985: 349) N-átàasà àʔàŋ bìitì [má-t-áwwá-ŋ ħállà]. 1/2-want.impf I water [gen-inf-wash-tr metal pot ‘I want some water [to wash the pot].’ OR: ‘I want [some water [to wash the pot]].’ The ambiguity can be resolved by inserting into the purpose clause an overt ­subject: (193) Krongo (Reh 1985: 349) N-átàasà àʔàŋ bìitì [máa-ní t-áwwá-ŋ ħállà]. 1/2-want.impf I water [gen-3sg inf-wash-tr metal pot ‘I want some water for her to wash the pot / so that she can clean the pot.’

168 A Typology of Purpose Clauses

In English, too, such ambiguities can arise: (194) Robert gave me a beer to drink. (195) Tina gave me a book to read. The to-infinitive invites two closely related, but structurally distinct readings. In the first possible interpretation, the infinitival construction functions as an NP modifier, providing descriptive (and relatively redundant) information about the NP. We find this structure unambiguously in sentences like (196): (196) [Two other books [to read on holiday]] were lent to me by Tina. Here, the syntactic position of the alleged purpose clause inevitably classifies it as an NP modifier. In (197), by contrast, the same clause can only be interpreted as an adjunct: (197) Tina lent me them [to read on holiday]. In this case, the pronoun cannot possibly be further modified, for which reason the subclause is best described as a purpose adjunct. Drawing on previous work by Faraci (1974) and Kirkpatrick (1982), Jones (1991: 48–50) suggests the following criteria for distinguishing between infinitival purpose clauses from relative clauses in English: (i) Purpose clauses in relativizing function cannot have a proper noun or pronoun as antecedents: *John [to talk to her] came along. (ii) In English, only relativizing purpose clauses, but not adverbial purpose clauses, can occupy the syntactic position immediately following the subject. Put differently, a purpose clause in this position is always interpreted adnominally. (iii) Purpose clauses in relativizing function presuppose their own truth, to the effect that the content of the relative clause cannot be questioned by a polar interrogative: (198) A:   Did the guy [to fix the sink] get here yet? B: *No, he’s here to put the phone in. Adverbial purpose clauses, by contrast, are allowed in this construction: (199) A: Did you bring him [to fix the sink]? B: No, he’s here to put the phone in. The last property also ties in with our previous finding that purpose clauses tend to occupy focus position because they introduce new or rather foregrounded information, while (restrictive) relative clauses are commonly associated with presupposed or backgrounded information (cf. also Kazenin 1994).



Chapter 4.  Purpose clauses and complex sentences 169

In view of these subtle differences despite an overall formal similarity, we may ask what the conceptual overlap between the two clause types consists in. To start with, both relative and purpose clauses serve the propositional act function of modification: relative clauses modify a head noun, while purpose clauses modify a verbal situation. More specifically, relative clauses are predominantly used to restrict the referential potential, i.e. the extension, of a nominal argument in a clause (non-restrictive relative clauses being a notable exception). Now such a restriction may well be made in terms of the purpose that an NP serves. This holds for animate as well as inanimate referents, although with the latter purposive modification may indicate their very raison-d’être: (200) [The man to fix the sink] has just arrived. (201) I bought [a gadget to voice-record my emails]. In (201), a relative clause is actually needed to describe the designated purpose and hence the unique property of the gadget in question so as to distinguish it from thousands of other gadgets currently in use. In other words, such purposive relative clauses can serve the function of ‘ad-hoc categorization’ in the flow of discourse (cf. Barsalou 1983). Sometimes, however, there are severe restrictions on such categories. For example, Basque speakers can use nominalized purpose clauses to relativize on NPs, yet only if the information contained in the purpose clause is characteristically associated with the NP, i.e. if it can be construed as a permanent property of the NP in question. Thus a standard purpose clause like (202) Basque (isolate: France, Spain; Hualde and Ortiz de Urbina 2003: 795) Ertzainak etorri dira [jendeekin bazkaltzeko]. policemen arrive aux [people.com lunch.noml.purp ‘The policemen came to have lunch with people.’ can only be used to relativize an NP like ‘policemen’ if it spells out a property or purpose generally attributed to policemen: (203) Basque (Hualde and Ortiz de Urbina 2003: 793, 795) a. [Trafikoa zaintzeko] ertzainak heldu dira. [traffic direct.noml.purp policemen arrive aux ‘Policemen to direct the traffic have arrived.’ *[Jendeekin bazkaltzeko] ertzainak etorri dira. b.   [people.com lunch.noml.purp policemen arrive aux ‘The policemen to have lunch with people arrived.’ In general, what the English and Basque examples testify to is both a functional and a semantic analogy between purpose and relative clauses: both nominal refer-

170 A Typology of Purpose Clauses

ents and verbal situations can be modified with regard to the (conventional) purpose they serve. However, especially with animate NPs the notion of purpose cannot always be felicitously invoked, as in the following examples, partly taken from Keenan (1985: 169): (204) The man [to see] is Fred. (205) The man [to be appointed Prime Minister] must be an eloquent person. (206) The first one [to cross the line] will get a prize. I would like to suggest that such reduced relative clauses share with ordinary purpose clauses their intrinsic future or modal orientation. That is, Fred is the man that will/must/shall/is supposed to be seen in (204), and the man in (205) is desired/ nominated/most likely to be Prime Minister in the future. Apparently, this hypothetical element suffices to allow for purpose clauses being pressed into the service of relativization. We may thus be justified to conclude that such purpose clauses can be conceived of as source constructions that are employed in a target environment, which hints already at the diachronic mechanisms to be expounded in the next section.

4.2  The diachronic development of purpose clauses 4.2.1  Grammaticalization and the dynamic nature of grammar One of the major arguments of the previous sections was that purpose clauses, unlike many other adverbial clauses, exhibit regular formal commonalities with complement clauses, relative clauses, the domain of deontic modality and direct speech, all of which were shown to be conceptually well-motivated. Many of these relationships reveal themselves as principled patterns of multi-functionality, to the effect that a purposive marker or construction is simultaneously used to express one or more of the above-mentioned functions. Of course, such synchronic patterns are ultimately grounded in individual speakers’ economical choice to encode two functions by the same or a very similar structure. In order for these economical preferences to become a conventionalized part of the grammar of a language, they have to propagate through both the speech community and the language system over a certain amount of time. Therefore, diachronic change is “the key mediating mechanism” (Haspelmath 2008b: 103) between linguistic innovations in language use and synchronically observable conventions of a grammar. As was mentioned in §2.2.2, the fundamental process in the evolution and further development of grammatical constructions is grammaticalization. The basic — and



Chapter 4.  Purpose clauses and complex sentences

grossly oversimplified — tenet of research in grammaticalization is that the morphosyntactic elements that characterize a synchronic grammatical system are historically derived from independent lexical items, chiefly nouns, verbs and demonstratives, and that grammatical markers and constructions then keep being subject to dynamic forces of language change. That is, once a lexical item has assumed a certain grammatical function, it may still be susceptible to further grammaticalization in the sense of being extended to other grammatical contexts. As is well-known, such functional (or behavioural) changes are usually accompanied by both syntagmatic and paradigmatic formal changes, such as phonetic attrition, loss of syntactic independence (when free words become affixes) etc. (cf. Lehmann 1995 for a more detailed classification). Grammaticalization is thus best conceived of as a family of related diachronic processes which prototypically, though by no means necessarily, co-occur or interact with each other. The literature on grammaticalization is vast (cf. Hopper and Traugott 2003 and Fischer 2007 for overviews), and many of the processes involved have been investigated in great detail, including the conceptual mechanisms underlying them (e.g. Heine et al. 1991, Bybee 2003a, b). In linguistic typology, too, grammaticalization has played an important role. In particular, it has been suggested that the semantic maps used in typology to capture synchronic patterns of multi-functionality also allow for hypotheses about the historical pathways in which these patterns emerged (cf. Haspelmath 2003: 233ff.). This entails that many of these pathways are very similar from one language to another. Ample evidence for this hypothesis can be found, for example, in the recent compilation of recurrent grammaticalization channels by Heine and Kuteva (2002). In the following paragraphs, I will thus look at purpose clauses and the complex sentence systems they are part of from the perspective of grammaticalization. This implies a construction-based view of grammaticalization (cf. Wiemer and Bisang 2004) since I assume that (i) purpose clauses can provide a constructional environment in which processes of grammaticalization can take place and that (ii) purpose clause constructions, by virtue of being specific form-function pairings, can themselves be regarded as ‘grams’ that have developed from other constructions. As the present book is not a detailed case study in historical linguistics, but rather supposed to be a broader overview of what we currently know about purpose clauses, no attempt will be made to provide exhaustive accounts of all grammaticalization channels and the way they have been discussed in the literature. I will rather concern myself with some important pathways in the development of purpose clauses, providing a rough and certainly incomplete sketch of a number of widely attested changes in which purpose clauses act as the input to or constructional environment for processes of grammaticalization (§4.2.2–5) and of the historical sources of purposive constructions themselves (§4.2.6). My main point is that the recognition of these

171

172 A Typology of Purpose Clauses

historical processes will help us understand the synchronic structural properties of purpose clauses and their peculiar status in complex sentence systems.

4.2.2 Constructional intraference: From purpose to complement (and relative) clauses Let us begin with the most striking constructional overlap we encountered above, that of purpose and complement clauses. We saw that the close relationship between the two clause types was significant from both a quantitative and a qualitative point of view since it represents a highly frequent and widespread typological pattern that, in the extreme form, is manifested in the complete identity of the two constructions. Similar observations led Haspelmath (1989) to conclude that there is, in fact, a “universal path of grammaticization” from purpose clauses to infinitival complements. Haspelmath starts out from the insight that what appears to be a “semantically empty” (Haspelmath 1989: 288) infinitive such as English to + inf is actually by no means that deprived of meaning.8 For one thing, the preposition to is originally (and still synchronically) an allative preposition, and the infinitive in some complement clauses actually carries a modal meaning, namely precisely the desiderative or manipulative-jussive function we saw earlier. Haspelmath shows that this combination of features can be found in language after language, and he hypothesizes that it represents a synchronic reflection of a diachronic development from purpose clauses to infinitival complement clauses. What happens is thus that a purpose marker (conjunction, affix etc.) or the whole purposive construction is extended to a functionally closely related domain, a process that Croft (2006: 85) calls ‘intraference’. It is ultimately based on speakers’ perception of a functional analogy. Thus in the first stage of intraference, a purpose clause encroaches onto the domain of manipulative and desiderative complements (Haspelmath’s “irrealis-directive” group), to which it bears the closest conceptual relation. Again by analogy, the construction is extended to those complements that still have one crucial feature in common with purpose: “irrealis-potential” complements. Once such functional extensions have taken place, the infinitival construction may more widely be recruited as sentential complements and further intraference takes place, to the effect that, finally, both “realis-non-factive” (propositional attitude and utterance) complements and “realis-factive” (cognition) complements come to be used with the same structure. As Haspelmath (1989: 299) notes, “these modality meanings are increasingly different from the purposive meaning, showing a progressive weakening of the rather concrete original meaning.” Thus the central 8.  Notice that even typological experts such as Thompson and Longacre (1985: 172) claim that English to is a “grammatical morpheme with no lexical meaning”, as opposed to conjunctions such as before, when, if, which they claim to carry lexical content.



Chapter 4.  Purpose clauses and complex sentences 173

element of this grammaticalization process is what has traditionally been called ‘semantic bleaching’ or ‘desemanticization’ (Lehmann 1995). Since this concept has recently come under attack (cf. the discussion in Croft 2003: 261–3), we should probably speak more correctly of a functional shift from purposive to (infinitival) complement, which is accompanied by a gradual loss of the more concrete purposive-intentional meaning. Of course, no purposive form has to progress to the ultimate endpoint of the cline, but such cases are attested and hence we can summarize the complete potential cline as in Figure 30. Purposive

Irrealisdirective

Irrealispotential

Realisnon-factive

Realisfactive

Figure 30.  Semantic grammaticalization of the infinitive (adapted from Haspelmath 1989: 298)

For English, the direction of this grammaticalization process has recently been confirmed: The evidence that the purpose adjunct was the earliest function in which the to-infinitive appeared is, first, that its Gothic parallel, the du-infinitive, appears exclusively as purpose adjunct (Köhler 1867), never as verb complement; and secondly, the form of the to-infinitive points to it having started out as a to-PP, and one of the main functions of the to-PP in O[ld] E[nglish] is that of purpose adjunct.  (Los 2005: 28)

Although authors of reference grammars do not often have access to far-reaching historical records of the language in question (particularly in oral cultures), a few do actually explicitly support our hypothesis. For example, in Huallaga Quechua, certain types of complement clauses are formed by a genuine purpose clause: (207) Huallaga Quechua (Quechuan: Peru; Weber 1989: 26) [Aywa-na-n-paq] ni-sha. [go-noml-3poss-purp say-3prf ‘He told him to go.’ This structure is used for a fairly wide range of complements up to the ‘realis-nonfactive’ group (cf. Weber 1989: 26 for a list of verbs). According to Weber, the “purposive complement construction” is a grammaticalized version of the adverbial purpose clause. Semantically, it has lost its strong purposive element, and syntactically, “the purpose structure has simply become the grammatical mechanism for forming the complement” (Weber 1989: 206). In this light, a number of previously cited examples, such as (171) from Tzutujil, may now be ‘dynamicized’ and can be given a historically directed interpretation. More generally, I suspect that many instances of ‘purpose-complement-identity’ in our sample have come about by the same diachronic mechanism, yet the absence of explicit etymological information

174 A Typology of Purpose Clauses

in many cases makes it impossible to quantify the extent to which Haspelmath’s proposal can actually be called a statistical universal. For this reason, I am grateful for some experts’ opinions on the subject. In particular, Keren Rice has provided valuable comments on Slave. As the examples (176) and (177) above have shown, one could be tempted to interpret Slave as a language which recruits complement clauses for the expression of purposive relations. However, Rice (pers. comm.) rather believes the complementizer gha to have developed from the purposive conjunction gha, which in turn is a grammaticalized form of the postposition gha ‘for’. This would provide support for the pathway outlined above, despite the reverse synchronic impression. In addition, the examples from Slave also provide us with a case in point that such intraference and concomitant grammaticalization need not be confined to infinitival purpose clauses, since the Slave complements are finite clauses. Haspelmath, observing the same pattern for finite purpose clauses in Greek, also concludes that “the grammaticization from purposive meaning to realis modality complement functions […] is an even more general phenomenon, of which the evolution of the infinitive is just a special case” (Haspelmath 1989: 308). An interesting repercussion of many grammaticalization processes is socalled ‘reinforcement’ or ‘renewal’, during which a highly grammaticalized element, due to its functional shift or desemanticization, is strengthened again by adding a new element that re-emphasizes the original meaning. Haspelmath cites Modern Greek as an example: the original purposive marker hina, phonologically reduced to ná, was extended to realis-non-factive complements; as a result, it was felt to be insufficient to express the notion of purpose, for which reason the preposition giá was added as reinforcement. The synchronic effect of this process is the complex purposive conjunction we saw in (169). The process of renewal may now also explain why Yaqui borrowed a purposive conjunction from Spanish after the indigenous purposive suffix -kai had grammaticalized into a more general complementizer (cf. example [77] in §3.4.1). Finally, English to could be analysed from the same perspective. During the late Old and especially Middle English period, the to-infinitive was increasingly extended to a wide range of syntactic environments (cf. the many raising and control structures governed by both verbs and adjectives in present-day English). One might hypothesize that the original purposive function needed reinforcement, and that the emergence of in order in front of the to-infinitive in the early 17th century fulfilled precisely this function: (208) Early Modern English (1609 Bible (Douay): Exod. vi. 27, OED online) These are they that speak to Pharao, king of Egypt, in order to bring out the children of Israel from Egypt. When purpose clauses develop into complement clauses, they shift their function in the conceptual space of complex sentences from purposive adjunct



Chapter 4.  Purpose clauses and complex sentences 175

to a  genuine argument of the matrix clause verb. This has been translated into syntactic terms by Matthiesen and Thompson (1988) by the opposition between ‘hypotaxis’ (characterizing adverbial clauses) and ‘subordination’ (characterizing sentential object complements). The distinguishing feature invoked here is [± embedding]. A rather interesting observation is that purpose clauses in their genuine adverbial function (i.e. before any grammaticalization process) already show some syntactic indications of subordination, while other adverbial clauses are exclusively hypotactic. First, Hopper and Traugott (2003: 183) argue that fully subordinated constructions display a higher “level of incorporation” than hypotactic clauses. This is essentially what we have argued all the way along: in several respects, purpose clauses are more integrated with their associated main clause than any of the other adverbial relations. Second, some purpose clauses even appear to have proper argument status because the verb they modify is felt to be incomplete without the purpose clause, as Van Valin and LaPolla (1997: 125) claim to be the case for English use: (209) I used a knife [to cut the bread]. (210) ?I used a knife. The latter sentence can only felicitously occur if the purposive situation was mentioned in the previous discourse. Hence the occurrence of a purpose clause is obligatory rather than optional, which is a typical behavioural property of arguments. A similar example is the purposive use of the verb be, which would in some cases be ungrammatical without the purpose clause: (211) Have you been [to see Daddy]? In sum, there seem to be both conceptual and in some contexts also syntactic similarities between purpose clauses and sentential complements. They can give rise to reanalysis or constructional intraference, to the effect that a synchronic look at a grammar may reveal erstwhile purpose clauses to be at work in the domain of complementation.9 The exact same mechanism is at work when infinitival (or otherwise reduced) purpose clauses come to serve as relative clauses. We have seen above that some relative clauses in Krongo, Basque and English are actually indistinguishable from purpose clauses, and I have suggested that some functional features of purpose clauses, either their potential to modify an entity or their inherent ­modality, ­enable 9.  A particular construction type that is often included under the label of complementation is the causative construction. In this domain, too, we commonly find purpose markers or remnants of purposive constructions (e.g. in Swahili, Basque or Modern Greek). These are documented and discussed extensively in Song (2005) and will not be taken up again here.

176 A Typology of Purpose Clauses

this analogical coding (or intraference). We know from above that the English to‑infinitive originated in purposive function. Even in Old English times, however, it could already also be used as a relative clause: (212) Old English (c893 K. ÆLFRED Oros. III. xi. §3, OED online) Onne seo leo bring his hungreum hwelpum hwæt [to etanne]. ‘One of his lyons brings his hungry puppies/whelps something to eat.’ Needless to say, I do not have comparable historical data from Krongo or Basque, but in light of what we observed for complement clauses, it is highly plausible that (reduced) purpose clauses can be extended to the functionally analogous domain of purposive relativization. Such cases need to be sharply distinguished, however, from those instances of grammaticalization in which a relativizer is extended to the domain of purpose (cf. §4.2.6 below), which not only reverses the directionality, but also proceeds along a different developmental path. In sum, Croft’s notion of constructional intraference and the ensuing processes of grammaticalization have provided us with suitable tools to describe how purpose clauses have come to occupy their midway position in the variation space of complex sentences. Once again, the argument is not that purpose clauses are not proper adverbial clauses. They are, yet their specific conceptual and syntactic properties cluster so close to complementation relations in the space of complex sentences that they are often recruited for serving in this domain (provided we assume an economical behaviour of language users, of course). The synchronic outcome is frequently a grammatical system which can trick us into thinking that complement clauses have been extended to purposive contexts, but we have ample reason to believe that it often works the other way around.10 With the methodological toolkit of the typologist rapidly expanding, it has now also become possible to measure synchronic overlaps in precise ways and to develop graphical models of variation spaces based on statistically significant groupings in the data (e.g. by multidimensional scaling). Those quantitative approaches will certainly be a focus in future work on grammaticalization (cf. Hetterle in prog.). The following sec10.  A highly interesting, but admittedly much rarer, case is the spread of a purposive infinitive within the adverbial domain. Schulte (2007) demonstrates, for instance, how the prepositional infinitive construction in the Romance languages (e.g. a da ‘to[ALL] give[INF]’ in Romanian) has spread across several adverbial functions, including temporal, substitutive and concessive contexts. Schulte argues that the analogical extension (or constructional intraference, as we called it here) is sensitive to the purposive property of subject sharing: The more likely an adverbial relation shares its subjects in language use, the more prone it will be to analogical coding by a prepositional/purposive infinitive. But once the construction is firmly in place with some adverbial meanings, further extension is simply due to entrenchment of the pattern, as it can come to be used with adverbial relations that do not normally share their subjects (e.g. concessive clauses).



Chapter 4.  Purpose clauses and complex sentences 177

tions will test whether we can further exploit the diachronic dimension to account for other synchronic patterns of multi-functionality.

4.2.3  The purpose-reason-result cline(s) In §4.1.3, I argued that there is a well-motivated polysemy pattern of purpose-reason/cause and purpose-result, and sometimes all three functions are conflated in a single marker. One may wonder whether these relationships are also diachronically directed. Interestingly, both the sample data and previous findings in the literature yield conflicting results. As for the exceedingly widespread pair purpose-cause, we collected examples above (cf. Turkish, Amele and [170] from Qiang) in which a single marker covers purpose, cause and benefaction, and it is well-attested that benefactive adpositions are common sources for the grammaticalization of adverbial conjunctions, notably purpose, as we will see in more detail below. Now given that benefactive is not an attested source for cause (cf. Heine and Kuteva 2002: 328), we have to assume that the diachronic path proceeds from benefactive to purpose to causal markers. However, Luraghi (2005) has recently also provided evidence for several pathways from cause to benefaction and purpose and demonstrates that the exact grammaticalization channel depends on the specific spatial construal of the source domain (e.g. directionality versus location metaphors). Also, we should note that since the notion of purpose implies the notion of cause (cf. Wierzbicka 1998: 184, as discussed in footnote 3 on p. 154), it would certainly make sense if causal expressions were in some way more basic than purposive expressions, with the latter being diachronically derived from the former. In fact, Haspelmath (1989: 291) actually lists cause as a historical source of purpose, failing to provide concrete examples, though. In our sample, the description of Hausa suggests an instantiation of the cline cause > purpose. Jaggar (2001: 636) states that the causal conjunction dòmin ‘because’ “also indicate[s] purpose”, and his further choice of words strongly suggests that purposive dòmin is, in fact, derivative of the causal meaning. Furthermore, in cases where the polysemy pair purpose-cause actually also includes temporal and conditional meanings, as it does, for instance, in Kiowa (Watkins 1984: 237), I suspect indeed that the causal extension precedes the purposive extension since there is a very common pathway from temporal to both causal and conditional meanings (cf. English since (temporal/causal) or German wenn (temporal/conditional)), but not directly to purpose. In sum, it appears to me that the exact relationship in each individual case presumably depends on the larger grammaticalization trajectories that purpose and causal clauses are part of. If purpose, cause and benefaction are coded by a single marker, there are at least two scenarios, with causal meaning either historically anterior or posterior

178 A Typology of Purpose Clauses

to ­purpose. If, by contrast, a temporal or conditional sense is also involved, then purpose is likely an extension of the previously established causal meaning. The pair purpose-result is even more difficult since the notion of result has not yet entered the standard literature on grammaticalization (cf. its absence in Heine and Kuteva 2002). Again, my own findings are ambivalent. On the one hand, we have encountered purpose clauses that have come to be used without any intentional semantics involved, thus being described as result clauses in the literature (cf. the English and Greek examples in [166]-[169] above). One may argue that although purpose clauses do not logically entail the realization of the intended result, there is often a strong implicature that the result was actually achieved. It seems to me that this is the default case in natural language, and only if this expectation is not borne out, the speaker will have to add linguistic material in order to cancel the implicature: (213) John went to the supermarket to get some beer, but they were closed. It is well-known in the literature that recurrent conversational implicatures can become highly entrenched so that they actually become part of the semantics of the expression (cf. Hopper and Traugott 2003: 82). The semanticization of such inferences is typical of the earlier stages of grammaticalization. From this perspective, it would appear that the pragmatic implicatures in purpose clauses can indeed develop into semantic implications, thereby creating a genuine result clause. This would account for the patterns we saw in Koyra Chiini ([165]) and West Greenlandic above. On the other hand, conceptually the notion of purpose involves that of result, and hence we are not surprised to find complex purposive markers that are derived from a more basic result marker. In Sanumá (Yanomam: Brazil, Venezuela), for example, the complex purposive conjunction pia salo consists of the basic marker salo ‘result’ (also: ‘cause’) and the lexical verb pia ‘intend’, yielding a complex conjunction. Phonological attrition has taken place to the extent that some speakers have reduced the conjunction to pi salo (cf. Borgman 1990: 98). Therefore, the purposive marker is derivative of a more lexical result marker. To conclude, the intimate synchronic relationship between purpose, reason and result can be a consequence of a number of different trajectories of grammaticalization processes, which will have to be defined more clearly in future research on more extensive diachronic data.

4.2.4  From ‘motion-cum-purpose’ to TAM constructions In what is perhaps the textbook example of grammaticalization, a purpose clause provides a constructional environment for the evolution of auxiliary-verb con-



Chapter 4.  Purpose clauses and complex sentences 179

structions, most of which are associated with tense-aspect-mood semantics. Thus the English going-to-infinitive construction in (214) Take care, the bridge is going to collapse! has come to be used as way of expressing future time reference (although it is questionable whether it should be called a genuine ‘tense’ marker). This presentday grammatical complex of vp[aux[be going to] v[inf]] is the outcome of a grammaticalization process that originated in a complex sentence containing go as the main verb of the matrix clause and an associated subordinate purpose clause: (215) I am going (to the cinema) [to watch the latest James Bond movie]. In the absence of the directional adverbial in the matrix clause, the syntactic structure of the sentence was reanalysed as I [am going to] watch the movie and extended to contexts that did not actually involve either physical movement or purposive semantics (cf. [214] above), but are supposed to simply express a future event. Recent work in grammaticalization theory has considerably refined our understanding of the cognitive mechanisms that lead to such a rather dramatic functional shift. To begin with, the semantic reanalysis seems to depend heavily on the specific constellation of motion verb and purpose clause in a single linear string of discourse, and the inherent meanings that are ‘indexed’ by both of these elements. More precisely, purposive meaning always indexes future time reference since purposes can only be realized in the future (relative to the moment of speech). In other words, there is a unidirectional implicature from purpose to futurity (i.e. purposes are always future-oriented, but not every event in the future is intended). On the other hand, directed motion also unidirectionally indexes a passage of time. This is again one of the most basic experiential correlations, similar to the association of purposes and destinations. And again there are two suggestions for modelling this correlation. Metaphor theorists such as Lakoff and Johnson (1980) propose that time is conceptualized in terms of space in such a way that spatial concepts provide the very structure for building mental models of time. Specifically, we often conceptualize ourselves as ‘moving egos’ on a time line, and this is why we are ‘going’ to future events located on that line (cf. also Lakoff and Johnson 1999: ch.10 for elaborate discussion). On the metaphor account, this is precisely the reason why verbs of directed motion can co-occur with other verbs and receive a future time interpretation. Metonymic accounts, by contrast, are again less committed to particular conceptual representations and simply rely on the entrenched unidirectional correlation of motion and the passage of time and the resulting implicatures that come with motion verbs in language use. As the review in Evans and Green (2006: 714–33) shows, there is still some debate in the literature as to which conceptual account should be pre-

180 A Typology of Purpose Clauses

ferred. But, just as in our earlier discussion of purposes and destinations, both proposals proceed from a regular conflation of two experiential domains. Notice also that it is a language-specific question whether both the purpose-futurity and the motion-futurity inferences need to be invoked. In English, this seems crucial since the auxiliary construction includes the former purpose marker to; in other languages, there is “no overt purposive, in which case the future inference arises out of the directional verb and its associated aspect alone” (Hopper and Traugott 2003: 90), either via metaphor or metonymy (‘indexing’, ‘meta-analysis’). We will discuss such examples shortly. In addition to the semantic reanalysis, however, there was also a syntactic reanalysis from a biclausal structure to a monoclausal auxiliary-verb complex. As Fischer (2007) argues convincingly, this rebracketing requires the perception of a profound formal analogy on the part of language users. More precisely, the emerging construction would not have developed in the way it did if it had not been for already established auxiliary-verb complexes (“e.g. a modal verb such as will followed by a bare infinitive, expressing future intention” [Fischer 2007: 124]). In other words, the previous existence of productive [aux-v] templates in the grammar (of individual speakers) provides a fertile ground for the syntagmatic reanalysis of a biclausal structure. The new going-to-V construction is then drawn into the auxiliary-verb paradigm and develops, by analogical extension, a similarly wide array of productivity. A common effect during this extension process is that the original meaning continues to constrain the applicability and semantic nuances of the new construction in certain contexts, which becomes particularly noticeable if the construction contrasts in distribution with a competing pattern (such as present-day English will). This facet of grammaticalization has been described by Hopper (1991: 22) as ‘persistence’. Also, once reanalysis and analogical extension had taken place, the productive and routine use of the construction led to morphological coalescence and phonological erosion, by which modern English has gained the auxiliary gonna (cf. also Bybee 2003b). As the grammaticalization of the going-to-future in English is very well studied, we need not dwell on it any longer. We shall rather take a closer look at our sample languages to investigate similar cases of diachronic change. A grammaticalized construction largely parallel to English is found in Ma’di. In this language, a purpose clause marked by -rε can be combined with the verb mū ‘go’ to yield an expression of future time. That mū has really become a “grammatical verb” (Blackings and Fabb 2003: 430) is evidenced in the following example by the fact that the notion of physical movement is lost, as is the intentional component of purpose:



Chapter 4.  Purpose clauses and complex sentences 181

(216) Ma’di (Nilo-Saharan/Central Sudanic: Sudan, Uganda; Blackings and Fabb 2003: 431) ɔtʃɛ [ko-mū [mā tʃi-rε]] r kɔ-drà. rá. dog [3-ton-go [1sg (ton)-bite-purp def 3-(ton)-die aff ‘The dog that is going to bite me will certainly die.’ A slightly different pathway is instantiated by the Supine construction in Kolyma Yukaghir (cf. example [179] above). Here, a verb with the purposive Supine inflection can be combined with the already existing auxiliary l’e ‘be’ to yield a prospective aspect construction: (217) Kolyma Yukaghir (Yukaghir: Siberia; Maslova 2003: 179) Čarčaqan d’e tet-ul lek-tin l’e-je. C. ptcl you-acc eat-purp be-intr.1sg ‘Charchahan, I am going to eat you.’ It should be emphasized again that such changes do not imply that purpose clauses or markers develop into future time markers; rather, purpose clauses only provide the constructional environment which, if combined with a suitable main verb sense such as motion or activity in progress, can invite a semantic and structural reanalysis of the complex sentence. Finally, it must also be mentioned that grammaticalization does not have to proceed to the ultimate TAM stages of the proposed cline. Specifically, some of the ‘motion-cum-purpose’ constructions discussed in §3.4.2 only attest to formal reanalyses and shifts in grammatical status, but without being semantically reanalyzed as a TAM marker. It is in this light that we can now take up a construction from Chapter 3 again. In §3.3.1 we presented the ‘motion-cum-purpose’ construction in Tzotzil, whose striking peculiarity was the inability of the matrix verb of motion to cross-reference its subject argument. Recall the following example from above: (218) Tzotzil (Mayan: Mexico; Aissen 1984: 559) Ch-ba [s-man-ø chitom li] Xun=e. the Juan=def incompl-go [A3-buy-subj pig ‘Juan will go buy pigs.’ In a follow-up paper, Aissen (1994) notes that a closed class of 12 motion verbs behaves uniformly in this syntactic environment: whenever they combine with a purpose clause, they trigger the subjunctive marker in the subordinate clause and relegate the task of indexing the matrix subject to the purposive predicate. The account Aissen (1994) offers for this synchronic anomaly is essentially historical in nature. She argues that the present status of the 12 motion verbs in question is no longer that of a lexical verb, but of an auxiliary. In functional terms, this would

182 A Typology of Purpose Clauses

again amount to a prime example of grammaticalization. Just as in other languages, the motion verb is ‘bleached’, but, crucially, “bleaching takes the form of eliminating from the semantic structure of the verb everything but the trajectory it denotes […], including, therefore, its argument structure” (Aissen 1994: 658). A concomitant development typical of grammaticalization processes is phonological attrition (bat ‘go’ > ba; ay ‘go and return’ > a). In general, the motion verbs in question have slightly changed their phonological structure, having acquired the canonical (C)V shape of functional markers in Tzotzil (cf. the determiners li and ti, the complementizers ti and mi, the preposition ta and the negative morpheme mu; lexical verbs in (C)V, by contrast, are very rare.) Furthermore, it is only the semantically most general verbs of motion that enter the grammaticalization process; verbs encoding manner of motion, such as ‘jump’ or ‘fly’, are not susceptible to being reanalysed as auxiliaries. Moreover, the most general ba ‘go’ has reached the status of a future tense marker. This development can be seen most clearly in constructions where the former purpose clause would be incompatible with the literal meaning of ‘go’, such as a combination with another motion verb that denotes a semantically opposite type of movement (‘come’): (219) Tzotzil (Aissen 1994: 661) Ch-ba tal-uk. incompl-go come-subj ‘It’s going to come.’ This development parallels the Ma’di example seen before. But, again, grammaticalization does not have to proceed that far. It can also ‘stop’ at the level of a purposive auxiliary construction, whose chief deviation from the original ‘motioncum-purpose’ sentence is a promotion of the subordinate predicate into the matrix clause. A final example from Tzotzil illustrates a minimal pair of purposive imperatives, one with the lexical verb ʔoch ‘enter’, signalling imperative directly, and one with the auxiliary ʔoch ‘enter’, which has relegated the imperative marking to the purposive predicate, suggesting a complex VP in the second example: (220) Tzotzil (adapted from Haviland 1981: 225) a. ʔoch-an [av-a’iy li k’op=e]. enter-imp [A2-hear det word=[enclitic] ‘Entra para que escuches la discusión.’ [Come in so that you hear the discussion.] b. [ʔoch a’iy-o] li k’op=e. [enter hear-imp det word=[enclitic] ‘Entra a escuchar la discusión.’ [Come in to hear the discussion.]



Chapter 4.  Purpose clauses and complex sentences 183

Notice also that the Spanish translation reflects the higher degree of syntactic integration in the grammaticalized construction (finite purpose clause versus infinitive). In more theoretical terms, the Tzotzil data appear to be compatible with a historical scenario described by Croft (2001: 259–61). Croft argues that in the evolution of auxiliary-verb complexes, the former lexical verb typically ceases to be the head of the construction, i.e. its morphosyntactic and semantic locus. Consequently, some of its behavioural properties (e.g. triggering of indexation) can shift from the former matrix to the dependent verb. Other behavioural properties, such as the triggering of the subjunctive inflection, may still be retained, though, not unlike the persistence effects typical of grammaticalization. Importantly, this needs to be taken as a theoretical interpretation of the data rather than a descriptive account of the actual diachronic change. As such, it crucially hinges, for instance, on one’s conception of head and dependent elements; but it may nevertheless be an interesting stimulus for looking at historical data from Tzotzil (and related languages) from this perspective. We can round off this section by focussing on another ‘purposive auxiliary construction’. It is found in Trumai, a language isolate of Brazil. Similar to Tzotzil, Trumai has a whole set of motion verbs, such as laketsi ‘go for a walk /visit’, pita ‘go out with/for X’ or katsu ‘(go) towards river’. Many of these lexical verbs have developed into auxiliaries, which now form quite a large word class in its own right (as outlined in Guirardello 1999: ch. 4). In keeping with the above discussion, some of them can develop TAM semantics. Thus compare the lexical and the prospective-auxiliary meaning of laketsi in the following example: (221) Trumai (isolate: Brazil; Guirardello 1999: 147) a. Ina hen ha laketsi hulat oela-n. there then 1sg go.for.walk/visit beach centre-loc ‘Then, I had a walk in the centre of the beach.’ b. Ha k’ad-chachxo laketsi ka-in. 1sg hand-write/draw pros foc-tns ‘I am going (to begin) to write.’ The historical trajectories in the evolution of Trumai auxiliaries are revealing for the present discussion in two important respects. One the one hand, they provide evidence for our earlier remark that the evolution of a tense-aspect construction from a motion verb need not necessarily involve a genuine purpose clause. All that matters is that an event of directed motion inherently involves a passage of time; any activity after moving somewhere is thus in the relative future. When the two events are juxtaposed in discourse, forming so-called deictic serialized constructions (G. Anderson 2006: 307), the motion verb can be reanalysed as a future time marker. On a metaphorical account, such ‘motion towards

184 A Typology of Purpose Clauses

an event’ embodies the moving-ego metaphor; on the metonymic account, the motion verb indexes the future time reference of the second event and this implicature can become conventionally associated with the motion verb over time. As a mere temporal operator, it loses its former equal status in the serial construction and becomes a grammatical auxiliary (cf. also Heine and Kuteva 2002: 161–3 for many more examples). On the other hand, some of the motion verbs in Trumai developed into auxiliaries when combined with a purposive event, but they were not reanalysed as TAM markers. Specifically, the class of ‘directional verbs’ comprising notions such as ‘go towards river’ or ‘go uphill’ can be concatenated directly to a main verb and yield a typical case of a ‘motion-cum-purpose’ construction:

motion V

Experiential correlation PURPOSES & DESTINATIONS (metaphor or metonymy)

+V

Experiential correlation MOTION & PASSAGE OF TIME (‘Moving ego’ metaphor or metaanalysis/metonymy)

motion V

REANALYSIS

motion AUX

possible boundary of grammaticalization

Di

chonolaj-ik-on.

Unidrectional correlations [purpose-futurity [metonymy], motion-futurity [metaphor or metonymy]] META-ANALYSIS [INFERENCING]

FUTURITY tense-aspect AUX

tense-aspect CLITIC > AFFIX

yi

Trumai: huma-ktsu.

woman Y1 bathe-AUX.DIR ‘The woman went to take a bath’

INCOMFL-go trade-SUBJ-B1.SG PL ‘I’ll go [literally] to trade.’

Trumai: laketsi go.for.walk/ AUX.PROS

English: I am going [to visit Bill].

+V

motion-cum-purpose construction

Tzotzil: Ch-ba

+ purpose clause

SEMANTIC SHIFT (English, Spanish, some verbs in Tzotzil) PHONOLOGICAL REDUCTION

Legend:

form of the (original) lexical or grammatical construction semantics of a construction emerging grammatical construction direction of historical change LIGHT CAPITALS

linguistic processes involved in grammaticalization

DARK CAPITALS

conceptual processes involved in grammaticalization direction of conceptual mapping or inferencing process

Figure 31.  Purpose clauses in the grammaticalization of motion verbs



Chapter 4.  Purpose clauses and complex sentences 185

(222) Trumai (Guirardello 1999: 422) Di yi [ha hu’tsa-t’] huma-ktsu. woman yi [1sg see-noml bathe-aux.dir ‘The woman who I saw went to take a bath.’ The auxiliary ktsu derives from the directional verb katsu ‘(go) towards river’. As an auxiliary, it has undergone phonological reduction in some environments, as can be seen in the example. Another consequence of the grammaticalization process is that lexical verbs other than the directed-motion verbs cannot enter the above construction. If they are to be used in purposive relations, an ordinary purpose clause has to be recruited. It is a finite subordinate clause with a special ­purpose marker: (223) Trumai (Guirardello 1999: 401) Hai-ts chï(-in) k’ate yi wawa-ktsi [[hi otle] hak]. 1sg-erg foc(-tns) fish yi carry-dir [[2 roast purp ‘I brought fish for you to cook.’ As was mentioned earlier, the development of a special motion-cum-purpose (auxiliary) construction is evidence for economical coding of the most predictable scenarios in language use (i.e. the co-occurrence of motion and purpose). But it also shows that this kind of diachronic adaptation is in principle independent of the emergence of future TA-auxiliaries from motion verbs. In fact, the motion verbs entering the purpose-auxiliary construction are different and strictly kept apart from the motion verbs developing into prospective (or other TA) auxiliaries. For this reason, we have to conclude that there can be two distinct, logically independent grammaticalization channels, although it is certainly quite common for them to be intertwined. If they are, this usually takes the form of two successive stages of grammaticalization (motionlex-cum-purpose > motionaux–purpose > futureaux–main verb). Based on the examples discussed in this section,11 Figure 31 attempts to capture this complex diachronic picture graphically. It portrays the constructions we have encountered, the historical processes in grammatical change (particularly syntactic reanalysis and ensuing phonological effects), as well as some of the crucial cognitive mechanisms underlying them.

11.  This restriction is important since it is very likely that other language-specific paths of grammaticalization cannot be captured by the following graph. Figure 31 does thus not make any claims to universality.

186 A Typology of Purpose Clauses

4.2.5  From avertive (‘lest’) clauses to apprehensional epistemics The last grammaticalization channel to be explored is based on the observation that synchronically, negative purpose markers, i.e. what we have called avertive or ‘lest’ markers, also enter principled patterns of multifunctionality. Starting from English again, we can see that lest has at least three distinct synchronic uses. The first is the ordinary avertive function we encountered in §3.6.1. In some contexts, however, lest can also be paraphrased by in case: (224) BNC ALH 752 But, oh dear, there’s a noise; it’s my brothers at play; So I’ll hide with the cake, lest they wander this way. What both uses have in common is that the ‘lest’ clause encodes an apprehensioncausing situation, while the matrix clause spells out some precautions to be taken in order to avoid this situation. This reading can be extended to complements of predicates of fearing: (225) BNC AMC 469 Wendy and I were at once amused and embarrassed, but also concerned lest he be mown down by a passing train. As Lichtenberk (1995) has shown, this polysemy can be found in geographically and genetically independent languages, and it often goes even further, to the effect that ‘lest’ clauses can be used as independent main clauses. In such constructions, the ‘lest’ element comes to function as a so-called ‘apprehensional epistemic’, as it combines the apprehensional semantics of avertive clauses with the speaker’s ­assessment of the likelihood of the factuality of the proposition. An example from our sample comes from Wardaman: (226) Wardaman (Gunwinyguan: Northern Australia; Merlan 1994: 295) Yilama ye-0-ngegba yi-ngarr-gurlmi bujun. maybe 3-3sg-die-prs irr-1incl.pl-lose lest/if ‘I think he’s dying, we may lose him.’ In our sample, similar patterns cluster in the macro-areas of Oceania and Australia (Amele, Martuthunira, Ngiyambaa, Tukang Besi and Boumaa Fijian). According to Lichtenberk (1995), the synchronic overlaps can be captured by recourse to grammaticalization. Specifically, he suggests that there is a general historical development along the following cline (Figure 32). precautioning (purpose, in case)

fear

Figure 32.  Grammaticalization of avertive clauses

apprehensional-epistemic



Chapter 4.  Purpose clauses and complex sentences 187

Lichtenberk (1995: 319) argues that the implication of a situation being undesirable, or indeed feared, becomes explicit by virtue of clauses encoding such situations as subordinate to predicates of fearing. […] Over time, through metonymy based on cooccurrence of linguistic forms, the notion of apprehension comes to be associated with the lest element, and the predicates of fearing become expendable. When lest clauses come to be used without being embedded under predicates of fearing, the lest element becomes an apprehensional epistemic. It signals both the possibility of an undesirable situation and apprehension.

While it is not clear which developmental direction exists between the avertive and the ‘in case’ function, both most likely precede the fear and epistemic function. Incidentally, the fact that English lest is currently both formal in style and felt to be old-fashioned may explain why it has never reached the highly grammaticalized status of an epistemic marker. In sum, we have seen ample evidence that both positive and negative purpose clauses provide important constructional environments for the development of new grammatical markers or constructions. The very last question to be addressed in this chapter is now: Given that diachronic processes are responsible for the rise of grammatical constructions, where do purpose clauses themselves come from?

4.2.6  The origins of purpose clauses According to Hopper and Traugott (2003: 177), the most valuable insights into the grammaticalization of subordinate clauses can be gained if one does not focus primarily on the evolution of complex sentences ex nihilo, but rather if one pays close attention to processes that re-organize or re-structure complex sentence systems. The fact that purpose clauses often occupy an intermediate status between various types of complex sentences may suggest, on the one hand, that they are the source for intraference, as was argued above. On the other hand, however, their exceptional status may also be a result of purpose clauses having developed from other complex sentence constructions themselves. In order to pursue this question, we should carefully distinguish between the evolution of purposive markers (i.e. lexical items that have acquired a grammatical function) and purposive constructions in their entirety. Purposive markers, like many other grammatical items, tend to develop from lexical expressions, thus constituting typical examples of grammaticalization in the narrow sense. In light of the previous discussion (see especially §3.4.2), it should not come as a great surprise that the most common lexical sources of purpose markers are from the domain of directed motion or benefaction, motivated by the experiential correlations discussed at length above. Thus the patterns of multifunctionality containing allative, ‘dative’ (i.e. recipient-benefactive) and goal

188 A Typology of Purpose Clauses

elements can now be dynamicized and translated into a number of historical pathways. The most direct mapping from allative to purpose occurs when a verb of directed motion develops into a purposive marker. This pattern is described in Heine and Kuteva (2002) for lexemes denoting ‘come’ and ‘go’, both of which are also attested at least once in our own sample. Thus in Hmong Njua, the purposive conjunction lug is directly derived from the lexical verb lug ‘come’: (227) Hmong Njua (Hmong-Mien: China; Harriehausen 1990: 241) John yuav npaaj [lug muab rua Maria]. John buy flower [come give dir Maria ‘John is buying flowers in order to give them to Maria.’ In Amele, one is struck by the formal similarity between the lexical verb nu ‘go’ and the purposive preposition nu: (228) Amele (Trans-New Guinea/Madang; Roberts 1987: 58) Ija [sab faj-ec nu] h-ug-a. 1sg [food buy-inf purp come-1sg-hod ‘I came to buy food.’ (229) Amele (Roberts 1987: 48) Ija Naus nu-ug-a te-na. 1sg Naus go-2sg-imp 1sg>3sg-prs ‘I want Naus to go.’ According to Roberts (pers. comm.), there is no explicit evidence that the two are related, but the fact that nu is the semantically most general of all motion verbs in Amele predestines it to enter a grammaticalization channel. Apart from directed motion and purpose, nu is also used as a benefactive postposition ‘for’. This replicates the multiple-overlap situation we encountered in §3.4.2 above. We argued there that allative markers (here preceded by ‘go’ > all) frequently develop into

ALLATIVE PURPOSE REASON

INFIN

SUBORD

MAN

CONCEPUAL PERC

EXP

POSS

RECIPIENT

LOCATIVE

ADR

BOUND BEN

TIME ABL DUR

Figure 33.  Polysemous extensions of allatives in 44 languages (adapted from Rice and Kabata 2007: 490)



Chapter 4.  Purpose clauses and complex sentences 189

purpose markers by virtue of contextual (i.e. metonymic) reanalyses, possibly aided by metaphorical conceptualizations of purpose in the first place. Since allative marking also lends itself well to being applied to human goals (via metaphor), it is frequently also found for recipient marking. Recipients, in turn, are often beneficiaries and the latter may hence receive the same coding. Therefore, instead of a linear development from allative > recipient-benefactive > purpose, we are confronted with a much more differentiated diachronic network. On this account, the benefactive marking in Amele would be historically independent of the purpose marker, linked only by a ‘family resemblance’ chain of inferences. This is, in fact, the perspective that has recently been confirmed by empirical research into the grammaticalization of allative morphemes. Rice and Kabata (2007), a study cited previously in this book, used the Fisher-Yates exact test to calculate the synchronic strength of polysemy patterns and inferred the most plausible diachronic pathways along which these patterns can arise. Their results are displayed in Figure 33. In view of this proposal and the data presented in Figure 15, it should thus be acknowledged that allative, purposive and benefactive morphemes may be historically related to one another in a “unilateral” (Rice and Kabata 2007: 497) rather than a straightforward linear way. Throughout our investigation, we have seen several examples that testify to such unilateral developments. The locative-allative preposition ch(i) in Tzutujil (cf. example [171] above) has developed into a purpose marker, but is distinct from the benefactive preposition rxin (cf. Dayley 1985: 311 for examples). English to originated as an allative marker and was extended to both purpose and recipients, but not to benefactives. For the latter, the preposition for is reserved, which has also come to encode purpose. More generally, it seems that when purpose, allative and recipient-benefactives overlap, the benefactive sense is dissociated from purpose, arising as a secondary extension of recipients (cf. also Schmidtke-Bode to appear [a]). On the other hand, however, it is also widely attested that benefactive markers form an important direct source domain for purpose (cf. Heine and Kuteva 2002: 55–7). We can find this phenomenon, for instance, in Lavukaleve (cf. [100] again), but also in constructions in Turkish, Lango, Ndyuka and Yaqui (cf. Schmidtke-Bode to appear [a] for details). This development does not, of course, contradict Rice and Kabata’s map since it is not in any way related to allative marking (or directed motion, for that matter). Instead, it draws directly on the profound connection between purposeful action and the beneficiary involved (to recapitulate, benefaction implicates the notion of purpose). I argued in §3.4.2 that this gives rise to metonymic inference processes which, in turn, pave the way for a paradigmatic extension from nominal-benefactive to verbal-purposive contexts. It is not difficult to retrace the structural processes involved in all of these diachronic extensions. First of all, motion verbs can develop into allative adpositions

190 A Typology of Purpose Clauses

and case affixes via well-known syntagmatic processes in grammaticalization (cf. Lehmann 1995). Particularly, these include what Myhill (1988) has called ‘clustering’, i.e. the syntactic attraction of the motion verb to the informationally more important or salient NP (go house > to house), until it becomes fixed in a specific order relative to the NP, i.e. until its rigidification as a pre- or postposition. The cognitive process underlying syntactic clustering is the formation of routinized chunks, “the process by which a frequently used sequence of words or morphemes becomes automated as a single processing unit” (Bybee 2003b: 603). Frequent repetition and further automation can lead to phonological coalescence, in Lehmann’s terms, to the effect that the adposition becomes affixed to the NP (cf. also Bybee and Thompson 1997 for such reduction effects of token frequency). In sum, grammaticalization leads to the emergence of either isolating or synthetic allative markers that, in combination with their associated NP, express motion towards a goal. When such markers are extended, via metonymy, to verbal contexts, the typical attractors would be purposive situations expressed by nominalized verb forms (cf. Genetti 1991 for a detailed study of this process in Newari). This is because nominalizations display both nominal and verbal properties: they “have an inherent dual nature, which can lead naturally to dual analysis” (Harris and Campbell 1995: 310). In other words, nominalized verb forms are easily construed as paradigmatic alternatives to goal NPs. In this process, the erstwhile case marker (adposition or nominal affix) shifts its word-class status to a conjunction or adverbial affix and may inherit the same distributional restrictions that apply to such elements in a given language. Essentially the same processes can be invoked for the development of purpose markers from benefactive morphemes. Here, too, nominalizations are typical target constructions in grammaticalization, but notice more generally that nominalized verb forms are not a necessary environment for paradigmatic intraference. Taking the construction in (100) as an example again, Lavukaleve does not show any overt signs of nominal status on the purposive verb; the benefactive morpheme is simply suffixed to a bare verb stem. Although directed motion and benefaction thus represent very common sources for purpose markers, they are by no means the only semantic domains to which purposive morphemes avail themselves. Our sample data suggest that, most generally, other lexical expressions giving rise to purposive reanalyses ‘index’ (in the metonymic sense) some conceptual ingredient of purposive semantics. In West Greenlandic, for instance, “purpose […] is expressed by derivational affixes of desire and futurity” (Fortescue 1984: 64). Thus the purposive suffix -qqu is ultimately derived from the lexical verb qqu, which appears in the grammar as ‘tell to’, ‘ask to’ and ‘want’, i.e. in typical desiderative and jussive contexts. These senses can also be shades of the meaning of a purpose conjunction in Acehnese:



Chapter 4.  Purpose clauses and complex sentences 191

(230) Acehnese (Austronesian: Indonesia; Durie 1985: 259) [Mangat rijang trôk], bu=bagah ta=jak. [purp fast arrive des=fast 1.incl=go ‘Let’s go quickly so that we can arrive faster.’ According to Durie (1985: 258), the conjunction mangat derives from a lexical predicate ‘be easy, nice’. Positive evaluation predicates conceptually index the notion of ‘desired’ properties and hence could be argued to carry a desiderative component, just like purpose. Futurity is involved in Abun, where the purposive conjunction su-bere (> subere) contains the lexeme bere ‘future, later’: (231) Abun (West Papuan: Indonesia; Berry and Berry 1999: 207) A me ri yo a grem 2sg see ginger.root when.irr 2sg put [subere ji mu kar Sem et]. [purp 1sg go go.get Sem subj ‘If you find some ginger root then keep it (here) so that I can go and get Sem soon (so that he can use the ginger root).’ Futurity is also the link when temporal conjunctions corresponding to ‘until’ are extended to purposive ‘so that’. This happened with hal in Koyra Chiini (cf. ex. [164] above) and with kóse in Khoekhoe (cf. Hagman 1977: 135). For this language, Heine and Kuteva (2002: 212) also report that a purpose marker derives from the noun !kèiï ‘matter, story’, as illustrated by the following example: (232) Khoekhoe (Khoisan: Namibia; Krönlein 1889: 206) Nesa ta ra mĩba [tsi !gũṅts nĩ !kèië]. this.f 1sg prog say.appl [2sg go.2.m.sg fut purp ‘I tell you this so that you go.’ The path from matter to purpose is, however, not attested by any other language in our sample. Avertive clauses, finally, often also owe their characteristic ‘lest’ marker to a lexical source. Thus for Tukang Besi (Austronesian: Indonesia), Donohue (1999: 399) notes that the conjunction bara ‘lest’ has the meaning ‘don’t’ in main clauses. This meaning is consonant with the avertive semantics of ‘lest’ clauses: precautions are taken so that an unpleasant situation can be avoided. This is also reflected in the origin of the ‘lest’ marker ada in the Austronesian language To’aba’ita, spoken on the Solomon Islands. Lichtenberk (1995: 304, 320) claims that ada is historically derived from a lexical verb meaning ‘see, look out, watch out’, i.e. perception verbs that can also have a connotation of warning. This certainly makes sense in the context of avertive situations.

192 A Typology of Purpose Clauses

What cannot be explained in terms of desiderative, future or avertive semantics is the use of quotative markers and constructions in purpose clauses. We encountered earlier an example of this structure from Kannada, where purpose is conceptualized as ‘verbalized self-suggestion’ (cf. [185] above). That is, in a sentence like (233) He left, saying ‘Let’s get something to eat.’ the verb of speech can be interpreted as a verbal link between an action and an intended subsequent situation. What has thus happened in some languages is that ‘say’ did not only grammaticalize into a quotative marker (cf. Saxena 1988 for this development), but further came to mark purpose clauses as well. In Lezgian, for example, the converbal forms luhuz and lahana of the lexeme luhun ‘say’ are very common ways of expressing a purposive relation between two situations. We encountered examples containing luhuz and lahana in other contexts in Chapter 3 (cf. example [58]). The historical pathway of their reanalysis is exactly as in Kannada: “‘A does B, saying: May C do D!’ gives rise to ‘A does B in order for C to do D’” (Haspelmath 1993: 393). Another language from the Caucasus, Abkhaz, has grammaticalized the past absolutive of the verb ‘say’, ħ°a, into a quotative particle marking direct speech complements. Since purposes can be formulated as verbalized wishes or statements of intention, it is not surprising that Abkhaz speakers have extended the direct speech complement to purposive environments: (234) Abkhaz (Northwest Caucasian: Georgia; Hewitt 1987: 39) [Warà waž°ə a-gàra (ø-)q’a-s-c’a-wà-yt’ ħ°a] [you.m now art.cradle (it.)prev.I.do.dyn.fin quot/purp a-žəy-ra w-à-la-ga-wa-yt’. art-smith-abs you-it-prev-begin-dyn-fin ‘You will now begin work as a smith in order to make the cradle.’ As Hewitt (1987: 39) demonstrates, the quotative particle ħ°a is reanalysed in such constructions as “simply […] a conjunction”. What is remarkable here is that the new purpose clause retains the morphosyntactic properties of the original construction, so we are dealing with a case of genuine constructional intraference rather than just a further grammaticalization of the marker. This leads to a very unusual synchronic picture: The new purpose clause retains the finiteness of the original direct speech complement, and hence forms the only finite adverbial clause in Abkhaz, while all other adverbial (as well as relative and most complement) clauses are generally non-finite. This was recognized as a typologically exceptional pattern in §3.2.1. At the same time, the reanalysis of ħ°a as a conjunction “is quite a revolutionary development for a language which does not (as yet) possess a class of true conjunctions” (Hewitt 1987: 39). This example demonstrates, there-



Chapter 4.  Purpose clauses and complex sentences 193

fore, that well-motivated diachronic processes, especially constructional intraference and grammaticalization, can help to understand synchronically anomalous patterns (cf. also Harris 2006 on this point). In this light, we can now also explain an unusual split construction from Chapter 3. You may recall that Jamul Tiipay has a finite purpose clause construction, to be used with first and third person matrix subjects, and a ‘purpose auxiliary construction’ for second person matrix subjects. For the sake of convenience, we will repeat the two pertinent ­examples here: (235) Jamul Tiipay (Hokan: Baja California; Miller 2001: 260, 274) a. Tom peyii paa [nya’wach varaax uuniiw-x-ich]. Tom here arrive [we.sbj card.pl compete-irr-purp ‘Tom came here so that we (he and I) could play cards.’ b. [Varaax m-uuniiw-x–m–i-ch] me-yiw-me-yu? [cards 2-compete-irr-2-say-ss 2-come-2-be ‘Did you come here to play cards?’ In the first construction, the verbal suffix -ich renders the clause purposive, with a first person subject in the matrix clause and a different subject in the purpose clause. In the second construction, purpose is indicated by the auxiliary -i, which derives from the lexical verb -‘i ‘say’, and a switch reference suffix signals that the subject of matrix and purpose clause are identical, they must be second person. This highly idiosyncratic person split and the striking similarity of the purposive markers can be captured in diachronic terms. Miller (2001: 275) suggests that the auxiliary construction, involving a grammaticalized form of the verb ‘say’, is in fact the older form of expressing purpose in Jamul Tiipay. Historically, it most likely evolved from a “multi-clausal construction” which “might once have had the literal meaning ‘saying that s/he would do …, (subject) did …’”, as illustrated in the following schema: (236) CLAUSE1 (inflectional prefix)-i-ch CLAUSE3 … …-say-ss … Miller further claims that “[i]n first and third person contexts, the sequence of -i ‘say’ followed by -ch ‘same-subject realis’ was frozen, reduced in stress, and reanalyzed as the purposive clitic -ich” (Miller 2001: 275). On this view, hearers would have perceived the purpose marker holistically, disregarding its former ‘same subject’ component. As a result, the new purpose clause is no longer required to have the same subject as its matrix. The loss of switch-reference also paves the way for positioning the purpose clause iconically: while other adverbial clauses containing an ss switch-reference marker must precede their associated main clause, purpose clauses can now be postposed, in keeping with their semantic

194 A Typology of Purpose Clauses

and discourse-­pragmatic function. It still remains a puzzle why the construction was not reanalysed in second-person contexts, yet the remainder of the suggested grammaticalization process neatly captures the synchronic peculiarities of the younger construction. Further examples of the development from direct speech or thought could be listed: (237) Kana (Benue-Congo: Nigeria; Ikoro 1996: 279) kɔ̄ ‘say’ > purposive connector k (238) Barasano (Tucanoan: Colombia; Jones and Jones 1991: 156) yi ‘do, say, think’ as linker evoking a purposive interpretation: [Kẽda-ro bãsi-heo-koã-a-to ĩ yi-gʉ] bʉ˜-re uka-gʉ [good-nom know-all.caus-ff-prox-impf.2 3.m think-m 2sg-p write-m ya-ha yʉ. do.prs-3 1sg ‘I am writing to you so that you might know all about (it) (lit. Thinking: “May he know it well,” I am writing to you).’ It thus appears that quotative constructions (i.e. direct speech and thought) provide important source constructions for the development of purpose clauses (cf. also Heine and Kuteva [2002: 265–7] for further illustration). The penultimate case of grammaticalization accounts for the overlap of purpose, complement and relative clauses in the form of a common ­complementizer comparable to English that. Here, however, the direction is generally not from purpose to complement and relative clause, but rather the other way around. For English that, for instance, it is well-documented that the original marker was a demonstrative in a main clause that pointed, as it were, to the contents of the subordinate clause. It was then reanalysed as a complementizer “that had a whole clause within its scope” (Hopper and Traugott 2003: 192): (239) Late Old English (c. 1000, ÆCHom I, 4 60.32, cited in Hopper and Traugott 2003: 192) Dyslic bið [þæt hwa woruldlice speda forhogie for manna herunge. foolish is [comp someone worldly goods despise for men’s praise ‘It is foolish to despise worldly goods in order to win the praise of men.’ The original constituent structure of this sentence would thus have been (240) [Foolish is that: [Despise worldly goods…]]. Once reanalysed from a cataphoric (discourse-deictic) demonstrative to a complementizer, that came to be used to form other subordinate constructions, notably complement and relative clauses, but also some adverbial clauses. For the lat-



Chapter 4.  Purpose clauses and complex sentences 195

ter, however, that was semantically too bleached to be used as a conjunction on its own, so it had to be reinforced by the resultative marker so, for instance, to yield the complex purpose and result conjunction so that. Similar processes are also observable in other languages. The Chalcatongo Mixtec proclitic xa=, for which we saw in (164) that it covers both purpose and result, is also found in complement clauses and ultimately goes back to a relative marker (cf. Macaulay 1996: 153). The Bulgarian complementizer če is reinforced by the allative preposition da to yield the purposive conjunction če da ‘so that’ (Heine and Kuteva 2002: 91). If we compare these examples to the ones from §4.2.2, where purpose clauses are extended to the domain of complementation and relativization, we can see that apparently two different pathways with exactly opposite directions are involved in the historical development of purpose clauses. Specifically, non-finite purpose clauses often give rise to infinitival complements and reduced relative clauses via constructional intraference, while a general subordinator introducing finite relative and complement clauses can also extend to the adverbial domain, resulting in finite purpose (and possibly reason and/or result clauses). In the literature, this development is well-described for the ‘classical’ Indo-European languages, notably Latin and Greek, in which “the pre-relative markers *io/kwo” develop “spatio-­temporal uses and/or a deictic, pronominal relative use […] which become further specialized semantically and grammaticalized syntactically as logical connectors of cause and result” and purpose (Pepicello 1982: 262; cf. also Jeffers and Pepicello 1979). Clearly, more historical data would be necessary to further substantiate this claim for other languages, but in view of the previous discussion it seems to have at least some empirical validity. A very last question concerns the rise of those purpose clauses that are found in the region between co- and subordination. In the literature on historical syntax, it has often been argued that there is a common diachronic development from parataxis to hypotaxis, i.e. from (loosely) conjoined clauses to more tightly integrated complex sentences consisting of a main and a dependent clause (cf. Hopper and Traugott 2003: 176). The reanalysis of English that could thus be conceived of as a transition from the loose succession of two independent clauses in discourse to a biclausal sentence involving a linking element. Similarly, it has been observed that speakers reanalyse the coordinating conjunctions of their language, notably the equivalents of English and, as subordinators establishing a more specific semantic relationship to the antecedent clause (cf. Harris and Campbell 1995: 290 for examples from Mingrelian and Heine and Kuteva 2002: 43 for an example from !Xun). Heine and Kuteva hypothesize that there might be a more general grammaticalization channel from ‘VP-and’ to subordinator. Apparently, what happens in such cases is that an implicated semantic relationship between two coordinated clauses (i.e. a pragmatic inference) is activated so frequently that it becomes part

196 A Typology of Purpose Clauses

of the meaning of a construction; in the above cases from Mingrelian and !Xun, the coordinating element has acquired a conditional, temporal or causal meaning. Given that purposive relations can also commonly be implicated by coordinated ‘and’-constructions (cf. §3.4.3), one may wonder if our sample also contains purposive markers that ultimately derive from a coordinating conjunction ‘and’. Unfortunately, I was not able to discover such a pattern, with one ostensible exception. Evenki has a purposive converb -dA (cf. example [17] in Chapter 3) and a coordinating enclitic -dA ‘and’: (241) Evenki (Tungusic: Siberia; Nedjalkov 1997: 87) Nungartyn tynive eme-cho-tyn, girki-l-tyn-da they yesterday come-pst-3sg friend-pl-3pl.poss-co tynive chagudu eme-cho-tyn. yesterday after come-pst-3pl ‘They came yesterday and their friends came the day before yesterday.’ The similarity seemed striking to me, yet I was cautious enough to consult Igor Nedjalkov, who contends (pers. comm.) that there is no convincing evidence for a historical relationship between the two -dAs. My conclusion, once more, is that the potential and the actual occurrence of purposive ‘and’-coordination require further investigation, but this necessitates more available data, comprising both historical records or reconstructions and language-specific synchronic corpora which give us an idea of how frequently such coordinate ‘and’-constructions are used with an intended purposive reading. The same applies to serializations. We saw that in Mandarin Chinese and other (South-)East Asian languages, serial verb constructions are a conventionalized means of expressing purposive relations. In languages with well-established purpose clauses, such purposive serializations also begin to emerge. The English colloquial phrase (242) Run go get me a newspaper. is a well-known example, and I have recently come across a French purposive s­ erialization as well: (243) J’ai réçu sa lettre hier et je viens constater les dégâts causés par la rupture de la conduite d’eau. ‘I received her letter yesterday and I (have) come to ascertain the damages caused by the burst pipe.’ In conclusion, it seems that there is some potential for syndetic coordination and simple juxtaposition of verbs or clauses to induce purposive interpretations, but unfortunately, we have too little data at present to estimate the extent of grammaticalization patterns drawing on these interpretations.



Chapter 4.  Purpose clauses and complex sentences 197

4.2.7  Summary: The developmental trajectories of purpose clauses In the previous sections, I have tried to argue that the peculiar status of purpose clauses in synchronic systems of complex sentences can at least in part be understood by recourse to diachronic processes. I have sketched what appear to be the most important historical pathways along which purpose clauses and purposive markers develop, and they can be summarized by the following schemata (Figures 34a–i): a.

recipient (> benefactive) (motion verb)

allative

(cause)

b.

benefactive

purpose

relativizer

discourse-deictic demonstrative

extension of marker or construction affecting most likely finite clauses

c.

conditional benefactive

temporal

purpose cause

d.

(relative clause)

constructional intraference affecting most likely infinitival (or otherwise reduced) clauses

complementizer

purpose cause result

complement clause

(result)

result + intention = purpose result

purpose conventionalization of pragmatic inference

e.

motion-cum-purpose

f. avertive construction g.

purposive auxiliary construction

fearing complement

quotative marker or construction (direct speech and thought)

(future tenseaspect marker)

apprehensional epistemic

purpose (‘verbalized self-suggestion’)

198 A Typology of Purpose Clauses

h.

lexical items

purpose marker

verbs of directed motion time (‘future, later, until’) desire (‘order, tell to, ask to, want’) avoidance (‘don’t, watch out’)

i.

negative purpose marker (‘lest’)

semanticization of conversational implicature coordinate ‘VP-and’ and serial verb constructions

purpose clause

Figure 34.  The developmental trajectories of purpose clauses

Needless to say, this list of diachronic pathways is not claimed to be authoritative, let alone complete. I have only been able to sketch the alleged historical processes in a coarse way, and future research will undoubtedly refine (or abandon) the ­scaffolding presented here. The potential of dynamicizing typology should, however, have become evident, and we can close with Haspelmath’s (1989: 303) statement that the “nature of grammatical categories is best understood from the point of view of the source items from which they developed, not the final point where they end up after a long history of grammaticization.”

chapter 5

Conclusion and outlook Purposeful action is a central characteristic of human behaviour; talking about intentions and purposes is, in turn, a communicative function fulfilled by all human languages across the speech communities of the world. The present study has provided a typological survey of an important grammatical means which serves this function, the purpose clause. In the spirit of current typological work, I have tried to combine three important research agendas: (a) the appreciation of the fascinating diversity of grammatical structures in the form of an empirically well-founded documentation of the phenomenon; (b) the uncovering of some (qualitatively and quantitatively) significant cross-linguistic generalizations across purpose clause constructions of genetically, geographically and typologically independent languages; (c) the explanation of universal tendencies in terms of the conceptual correlates of the phenomenon and the cognitive-functional principles underlying language use and language change. Rather than repeating all empirical results here, I will try to unite them by answering the question, What are typical purposes clause like, and why do they have the properties they have? Our construction-based approach has shown that languages typically develop more than one purposive construction, each of which is usually characterized by a unique arrangement of linguistic gestalt features, and associated directly with a specific semantic, pragmatic and discourse-pragmatic function. Across the sample investigated, five recurrent construction types could be identified: (i) Finite purpose clauses. Such clauses are typically marked by an overt purposive conjunction, adposition or affix, the particular choice of marker being dependent on the general morphological type and the constituent order of the language. More often than not, the purpose marker of finite purpose clauses is a complex unit, consisting of an intentional or resultative element (e.g. so) and a more general complementizer (e.g. that), the latter of which usually goes back to a cataphoric discourse-deictic demonstrative that, once reanalysed as a conjunction, comes to function as a complementizer in a wide variety of subordinate clause constructions. Finite purpose clauses are characteristically marked for mood, such as hypothetical or desiderative mood, and their semantic arguments are usually overtly realized in the syntax. (ii) Non-finite purpose clauses. Such clauses are typically nominalized or, even more commonly, infinitival clauses. Oftentimes, the infinitive marker itself

200 A Typology of Purpose Clauses

or the purpose marker of the construction historically derives from an allative and/or benefactive case marker (case affix or adposition). I argued that their reanalysis is based on the frequent experiential correlation of directed motion and purpose, and the cognitive processes (metaphor, metonymy) that operate on this correlation. Non-finite purpose clauses by default contain deranked verb forms and usually leave their subject, but sometimes also their object positions implicit. Especially the subject position is easily recoverable since the most common experiential pattern encoded in purpose clauses is for the agent of the goal-directed action to be identical to the person whose goal is referred to. Non-finite purpose clauses do thus not only represent the most economical way of expressing purpose, but also iconically reflect a high degree of integration of the purpose clause into its associated matrix clause. Non-finite purpose clauses, by virtue of being morphosyntactically reduced and behaving rather like clausal goals (i.e. NPs), often appear in the canonical position of object arguments. Because of their functional resemblance to sentential complements, non-finite purpose clauses are often reanalysed as complements by a process we called constructional intraference. From then on, they are typically subject to forces of grammaticalization. The same processes also apply when (non-finite) purpose clauses come to be used in the domain of ­relativization. Non-finite and finite purpose clause constructions can be seen as complementary. Whereas non-finite purpose clauses are used for the canonical same-subject situations, finite clauses can be employed whenever participants need to be made overt (e.g. different-subject situations), provided that a language has both construction types. Also, it should be emphasized that the two constructions have many properties in common. They share, for example, a common discourse-pragmatic function: both finite and non-finite purpose clauses provide a motivation for the action expressed by the matrix clause. As such, they typically introduce nonpresupposed or new information into the ongoing discourse. This property (and perhaps the drive for iconicity of sequence in language processing, as some would argue) motivate the universally preferred postposed position of purpose clauses, even when important processing factors such as consistent branching directions or minimal recognition domains would favour a different position in the sentence. Also, it should be noted that the historical pathways outlined for finite and nonfinite purpose clauses may well hold across both construction types. That is, finite purpose clauses may also develop into sentential complements, and their purpose marker might also be derived from an allative or benefactive expression. Moreover, both finite and non-finite purpose clauses can show synchronic similarities to clauses of reason and result. (iii) Motion-cum-purpose constructions, motion-purpose complements and purposive auxiliary constructions. Purpose clauses often encode the highly



Chapter 5.  Conclusion and outlook 201

frequent experiential pattern of moving somewhere in order to achieve a certain goal. As a result, many purpose clauses are actually governed by a matrix clause containing a verb of motion. What is more, languages may grammaticalize a distinct ‘motion-cum-purpose’ construction alongside a more generally applicable purpose clause. Such constructions are highly frequent in language use and hence show a particular propensity for economical coding (e.g. implicit arguments, nonfinite verb forms, omission of overt purposive conjunctions, tight integration into the matrix clause). Over time, they may further grammaticalize into monoclausal complex VPs, such as auxiliary-verb complexes, and come to serve new grammatical functions, most commonly as tense-aspect markers. The frequent association of motion and purpose more generally was also argued to be susceptible to metonymic inferences (and/or metaphorical mappings) from destinations to purposes, which are, in turn, responsible for allative-related marking in purpose clauses. (iv) Constructions with purposive inferences. We have reason to believe that a purposive interpretation is often intended in a coordinate ‘and’-, serial-verb or quotative construction. Two situations in immediate succession are likely to be causally related, and oftentimes the causal link rests in an intention to bring about the second situation. Diachronically, such pragmatic inferences — provided they are salient — may become part of the conventionalized meaning of a construction. We have seen that quotative predicates, in particular, can be reanalysed as purposive conjunctions. For obvious reasons, this development is more likely to result in finite purpose clauses. (v) Avertive (‘lest’) constructions. Some languages grammaticalize a distinct construction for the expression of negative purpose. Such avertive clauses encode that precautions are taken in order to prevent an entity from an apprehensioncausing situation happening to him/her/it. This experiential pattern differs in important ways from the typical experience encoded by positive purpose clauses, and hence the morphosyntactic properties of ‘lest’ clauses are conspicuously different from — in fact, sometimes the opposite of — those of positive purpose clauses. Here we find a propensity for different, overt subjects which are aligned with patient rather than agent roles, and no particular preference for motion verbs or allative expressions. However, avertive clauses, too, are also source constructions for processes of grammaticalization, especially for the rise of apprehensional-epistemic mood markers. In view of the details provided throughout the book, it should be obvious that the five construction types just summarized are vastly idealized and oversimplified generalizations. Yet in their quantitative-statistical underpinning, which has been a major methodological concern of this study, they do constitute prototypes of attested typological patterns. They now also allow us to claim that English ­purpose

202 A Typology of Purpose Clauses

clauses are rather typical from a cross-linguistic perspective. Not only are all five construction types and their characteristic properties instantiated in present-day English; we also find evidence for all major grammaticalization processes leading to their existence and further development. A central idea emerging from this cross-linguistic study is that purpose clauses constitute a very distinctive construction type that deviates in important ways from the archetypal characteristics shared by many adverbial clauses. The conceptual structure of purpose predetermines a number of important semantic ingredients of purpose, and is crucially defined by a mental-state relation between the purposive event and the agent of the main clause. Despite the high conceptual integration of the purposive situation into the matrix event frame, however, purpose clauses have their own information-structural value and resist the typical fi ­ gure-ground conceptualization characteristic for a number of adverbial relations (such as temporal conditional and concessive clauses). This unique and somewhat exceptional constellation of properties leads to a constant ecological pressure on purpose clauses in complex sentence systems. On the one hand, the fact that purposive relations share a number of relevant properties with many complementation relations acts as a centrifugal force: purpose clauses constantly serve as environments in which complement clauses develop and subsequently conquer the (entire) niche of complementation. The synchronic outcome can be a system in which purpose clauses do not seem to ‘belong’ into the adverbial domain but cluster instead closely to complementation. On the other hand, purpose clauses overlap in function (and some semantic features) with reason and result clauses, although they often exhibit a more downgraded morphosyntax and other properties not normally found with those clauses. Nevertheless, the three clause types typically form a cluster and constantly interact in diachrony, and these processes work as centripetal forces that keep purpose clauses in the adverbial domain. This dynamic way of thinking about the typological distribution of purposive constructions has helped us understand the peculiar position of purpose clauses in complex sentence systems, and also contributed to explaining some of the more ‘exotic’ patterns of variation we encounter in the typology of purpose clauses. Despite the ambitious goal of this book to provide a comprehensive typology of purpose clauses, it leaves a number of questions insufficiently addressed, and some might not have been discussed at all. For example, although I attempted to describe the geographical distribution of some important properties of purpose clauses and their correlations with other features, a full-fledged areal typology of purpose clauses remains to be worked out. As especially the work of Nichols (1992) and Bickel and Nichols (2003, 2005b, 2006) has demonstrated, important contributions to understanding typological distributions can be made if a perspective is taken from linguistic areas (e.g. typological enclaves, the Pacific Rim etc.) and



Chapter 5.  Conclusion and outlook 203

population science. Needless to say, the present sample was not well-designed to investigate such areal phenomena, but there is certainly fundamental research potential along these lines. After all, although I argued that the synchronic properties of purpose clauses are well-motivated from a functional point of view, notably in terms of functional pressures on individual utterances in language use and on language change, there remains the possibility that some of the skewings we observed may in fact be better understood as the results of population history, i.e. mutual contact, feature sharing and/or diversification due to migrations. Another important extension of the present research lies in developing and applying precise measurement and mapping techniques for the variation space of complex sentences. In Chapter 4, I argued that purpose clauses cluster in specific ways in this variation space, but the evidence for this hypothesis was essentially qualitative in nature. Modern methods of clustering and scaling (cf. Baayen 2008: ch.5, Cysouw 2007) provide complex tools for enhancing traditional (static) semantic maps in a way that captures variation along several grammatical dimensions at once and determines significant groupings in the data that reflect similarities and differences of language-specific constructions (‘multivariate typology’, cf. Bickel 2006b). However, such quantitative endeavours are crucially dependent on larger databases with rich information on especially those parameters that are currently understudied. Discourse-pragmatic aspects of purposive clause combining, the peculiarities of avertive constructions and the diachronic pressures on purpose clauses could thus be researched more successfully if more data could be ­accessed. In parallel to expanding the typological database, I also suggested turning to large-scale corpora of individual languages. If exploited with the rigour and the analytical sophistication of modern quantitative corpus linguistics (cf. Gries 2006, Baroni and Evert 2009), corpora can reveal principled patterns of usage preferences in many grammatical domains: they can help uncover the discourse-pragmatic properties associated with seemingly alternating orders, significant associations of verbs (or predicate classes) with grammatical constructions, or intricate constraints on co-reference relations (cf. also Schmidtke-Bode to appear [b] for an overview). On a usage-based approach to typological generalizations, such data are crucial to weighing factors that potentially contribute to the explanation of grammatical asymmetries in the languages of the world. Therefore, it is in the combination of cross-linguistic and intra-linguistic perspectives that I see a promising area of future research, especially for work that aims to make a contribution to the theory of human language. In view of the work that remains to be done, therefore, the present study can only claim to have worked out a typology of purpose clauses, not the typology of this fascinating construction type.

References Aissen, J. 1984. Control and command in Tzotzil purpose clauses. Berkeley Linguistics Society 10: 559–71. Aissen, J. 1994. Tzotzil auxiliaries. Linguistics 32: 657–90. Aitchison, J. 2001. Language Change: Progress or Decay? 3rd edn. Cambridge: Cambridge University Press. Anderson, G. D. S. 2006. Auxiliary Verb Constructions. Oxford: Oxford University Press. Anderson, S. R. 1976. On the notion of subject in ergative languages. In Subject and Topic, C. N. Li (ed.), 1–24. New York NY: Academic Press. Anderson, S. R. & Lightfoot, D. W. 2002. The Language Organ: Linguistics as Cognitive Physiology. Cambridge: Cambridge University Press. Baayen, H. 2008. Analyzing Linguistic Data. Cambridge: Cambridge University Press. Bach, E. 1982. Purpose clauses and control. In The Nature of Syntactic Representation, P. Jacobson & G. K. Pullum (eds.), 35–57. Dordrecht: Reidel. Bakker, D. To appear. Language sampling. In The Oxford Handbook of Linguistic Typology, J. J. Song (ed.). Oxford: Oxford University Press. Baroni, M. & Evert, S. (2009). Statistical methods for corpus exploitation. In Corpus Linguistics: An International Handbook, A. Lüdeling & M. Kytö (eds.), 777–802. Berlin: Mouton de Gruyter. Barsalou, L. W. 1983. Ad hoc categories. Memory and Cognition 11: 211–27. Bates, E. & MacWhinney, B. 1989. Functionalism and the competition model. In The Cross-Linguistic Study of Sentence Processing, B. MacWhinney & E. Bates (eds.), 3–73. Cambridge: Cambridge University Press. Bauer, W. 1993. Maori. London: Routledge. Bell, A. 1978. Language samples. In Universals of Human Language I: Method and Theory, J. H. Greenberg, C. Ferguson & E. Moravcsik (eds.), 123–56. Stanford CA: Stanford University Press. Berry, K. & Berry, C. 1999. A Description of Abun: A West Papuan Language of Irian Jaya. Canberra: Australian National University. Bhatia, T. K. 1993. Punjabi – A Cognitive Descriptive Grammar. London: Routledge. Biber, D., Conrad, S. & Reppen, R. 1998. Corpus Linguistics: Investigating Language Structure and Use. Cambridge: Cambridge University Press. Biberauer, T. (ed.). 2008. The Limits of Syntactic Variation. Amsterdam: John Benjamins. Bickel, B. 2003. Referential density in discourse and syntactic typology. Language 79: 708–36. Bickel, B. 2006a. Understanding typological distributions. Course taught at the Leipzig Spring School on Linguistic Diversity, Leipzig, 27–8 March 2006. Available online at . Bickel, B. 2006b. Towards a multivariate typology of clause linkage, with particular reference to non-embedded structures. Paper presented at the International Symposium on the Grammar and Pragmatics of Complex Sentences (LENCA 3), Tomsk, 27–30 June 2006. Available online at .

206 A Typology of Purpose Clauses Bickel, B. 2007. Typology in the 21st century: Major current developments. Linguistic Typology 11: 239–51. Bickel, B. 2008a. A general method for the statistical evaluation of typological distributions. Ms., University of Leipzig. Available online at . Bickel, B. 2008b. A refined sampling procedure for genealogical control. Sprachtypologie und Universalienforschung 61: 221–33. Bickel, B. to appear. Absolute and statistical universals. In The Cambridge Encyclopedia of the Language Sciences, P. C. Hogan (ed.). Cambridge: Cambridge University Press. Bickel, B. & Nichols, J. 2002. Autotypologizing databases and their use in fieldwork. In Proceedings of the International LREC Workshop on Resources and Tools in Field Linguistics, Las Palmas, 26–7 May 2002, P. Austin, H. Dry & P. Wittenburg (eds.). Nijmegen: MPI for Psycho­ linguistics. Bickel, B. & Nichols, J. 2003. Typological enclaves. Paper presented at the 5th Conference of the Association for Linguistic Typology, Cagliari, Italy. Available online at . Bickel, B. & Nichols, J. 2005a. Fusion of selected inflectional formatives. In The World Atlas of Language Structures, M. Haspelmath, M. S. Dryer, D. Gil & B. Comrie (eds.), 86–9. Oxford: Oxford University Press. Bickel, B. & Nichols, J. 2005b. Areal patterns in the World Atlas of Language Structures. Paper presented at the 6th Conference of the Association for Linguistic Typology, Padang, Indonesia. Available online at . Bickel, B. & Nichols, J. 2006. Oceania, the Pacific Rim and the theory of linguistic areas. Paper presented at the 32nd Meeting of the Berkeley Linguistics Society. Available online at . Bisang, W. 2007. On the evolution of complexity: Sometimes less is more in East and mainland Southeast Asia. Paper presented at the workshop on Language Complexity as an Evolving Variable, MPI Leipzig, 12 April 2007. Available online at . Blackings, M. & Fabb, N. 2003. A Grammar of Ma’di. Berlin: Mouton de Gruyter. Blansitt, E. L. Jr. 1988. Datives and allatives. In Studies in Syntactic Typology, M. Hammond, E. Moravcsik & J. Wirth (eds.), 173–91. Amsterdam: John Benjamins. Borgman, D. M. 1990. Sanuma. In Handbook of Amazonian Languages, Vol. 2, D. C. Derbyshire & G. K. Pullum (eds.), 15–248. Berlin: Mouton de Gruyter. British National Corpus (BNC). Information available online at . Brown, P. & Levinson, S. C. 1996. Politeness: Some Universals of Language Usage. Cambridge: Cambridge University Press. Buechel, E. 1939. A Grammar of Lakota: The Language of the Teton Sioux Indians. St. Louis MO: John S. Swift Co. Bybee, J. L. 1985. Morphology: A Study of the Relation between Meaning and Form. Amsterdam: John Benjamins. Bybee, J. L. 2003a. Cognitive processes in grammaticalization. In The New Psychology of Language, Vol. 2, M. Tomasello (ed.), 145–67. Mahwah NJ: Lawrence Erlbaum. Bybee, J. L. 2003b. Mechanisms of change in grammaticization: The role of frequency. In Handbook of Historical Linguistics, B. D. Joseph & R. Janda (eds.), 602–23. Oxford: Blackwell. Bybee, J. L. 2006. From usage to grammar: The mind’s response to repetition. Language 82(4): 711–33.



References 207

Bybee, J. L., Pagliuca, W. & Perkins, R. 1990. On the asymmetries in the affixation of grammatical material. In Studies in Diachronic Typology for Joseph H. Greenberg, W. Croft, K. Denning & S. Kemmer (eds.), 1–42. Amsterdam: John Benjamins. Bybee, J. L., Perkins, R. & Pagliuca, W. 1994. The Evolution of Grammar. Tense, Aspect, and ­Modality in the Languages of the World. Chicago IL: University of Chicago Press. Bybee, J. L. & Thompson, S. A. 1997. Three frequency effects in syntax. Berkeley Linguistics ­Society 23: 378–88. Carlson, R. 1994. A Grammar of Supyire. Berlin: Mouton de Gruyter. Carston, R. 2002. Thoughts and Utterances: The Pragmatics of Explicit Communication. Oxford: Blackwell. Chafe, W. (ed.). 1980. The Pear Stories. New York NY: Ablex. Chafe, W. 1984. How people use adverbial clauses. Berkeley Linguistics Society 10: 437–49. Chafe, W. 1994. Discourse, Consciousness and Time: The Flow and Displacement of Conscious Experience in Speaking and Writing. Chicago IL: University of Chicago Press. Chapman, S. & Derbyshire, D. C. 1991. Paumarí. In Handbook of Amazonian Languages, Vol. 3, D. C. Derbyshire & G. K. Pullum (eds.), 161–352. Berlin: Mouton de Gruyter. Chelliah, S. L. 1997. A Grammar of Meithei. Berlin: Mouton de Gruyter. Chomsky, N. 2000. New Horizons in the Study of Language and Mind. Cambridge: Cambridge University Press. Chung, S. & Timberlake, A. 1985. Tense, aspect, and mood. In Language Typology and Syntactic Description, Vol. 3: Grammatical Categories and the Lexicon, T. Shopen (ed.), 202–58. Cambridge: Cambridge University Press. Clark, E. 1971. On the acquisition of the meaning of before and after. Journal of Verbal Learning and Behaviour 10: 266–75. Cohen, J. 1969. Statistical Power Analysis for the Behavioral Sciences. New York NY: Academic Press. Cohen, J. 1994. The earth is round (p < .05). American Psychologist 49(12): 997–1003. Comrie, B. 1976. Aspect. Cambridge: Cambridge University Press. Comrie, B. 1985. Tense. Cambridge: Cambridge University Press. Comrie, B. 1998. Rethinking the typology of relative clauses. Language Design 1: 59–86. Comrie, B. 2005. Alignment of case marking of full noun phrases. In The World Atlas of Language Structures, M. Haspelmath, M. S. Dryer, D. Gil & B. Comrie (eds.), 398–405. Oxford: Oxford University Press. Cristofaro, S. 2003. Subordination. Oxford: Oxford University Press. Cristofaro, S. 2005. Purpose clauses. In The World Atlas of Language Structures, M. Haspelmath, M. S. Dryer, D. Gil & B. Comrie (eds.), 506–7. Oxford: Oxford University Press. Cristofaro, S. 2007. Deconstructing categories: Finiteness in a functional-typological perspective. In Finiteness: Theoretical and Empirical Foundations, I. Nikolaeva (ed.), 91–114. ­Oxford: Oxford University Press. Croft, W. 2000. Explaining Language Change: An Evolutionary Approach. London: Pearson. Croft, W. 2001. Radical Construction Grammar: Syntactic Theory in Typological Perspective. ­Oxford: Oxford University Press. Croft, W. 2003. Typology and Universals. 2nd edn. Cambridge: Cambridge University Press. Croft, W. 2006. Evolutionary models and functional-typological theories of language change. In The Handbook of the History of English,. A. van Kemenade & B. Los (eds.), 68–91. Oxford: Blackwell.

208 A Typology of Purpose Clauses Croft, W. To appear, a. The origins of grammaticalization in the verbalization of experience. Linguistics. Croft, W. To appear, b. Verbs: Aspect and Argument Structure. Oxford: Oxford University Press. Croft, W. To appear, c. Language structure in its human context: New directions for the language sciences in the 21st century. In The Cambridge Encyclopaedia of the Language Sciences, P. C. Hogan (ed.). Cambridge: Cambridge University Press. Croft, W. & Cruse, D. A. 2004. Cognitive Linguistics. Cambridge: Cambridge University Press. Croft, W. & Poole, K. T. 2008. Inferring universals from grammatical variation: Multidimensional scaling for typological analysis. Theoretical Linguistics 34: 1–37. Culicover, P. W. & Jackendoff, R. 1997. Semantic subordination despite syntactic coordination. Linguistic Inquiry 28(2): 195–217. Curnow, T. 1997. A  Grammar of Awa Pit (Cuaiquer). PhD dissertation, Australian National University. Cutler, A., Hawkins, J. A. & Gilligan, G. 1985. The suffixing preference: A processing explanation. Linguistics 23: 723–58. Cysouw, M. 2003. Against implicational universals. Linguistic Typology 7: 89–101. Cysouw, M. 2005. Quantitative methods in typology. In Quantitative Linguistics: An International Handbook, R. Köhler, G. Altmann & R. G. Piotrowski (eds.), 554–78. Berlin: Mouton de Gruyter. Cysouw, M. 2007. New approaches to cluster analysis of typological indices. In Exact Methods in the Study of Language and Text. Dedicated to Gabriel Altmann on the Occasion of his 75th Birthday, P. Grzybek & R. Köhler (eds.), 61–75. Berlin: Mouton de Gruyter. Dąbrowska, E. 2004. Language, Mind and Brain: Some Psychological and Neurological Constraints on Theories of Grammar. Edinburgh: Edinburgh University Press. Dąbrowska, E. 2008. Different speakers, different grammars: Individual differences in linguistic knowledge. Plenary talk at the conference Cognitive and Functional Perspectives on Dynamic Tendencies in Languages, University of Tartu, Estonia, May 2008. Daniel, M. 2007. Representative sampling and typological explanation: A  phenomenological ­lament. Linguistic Typology 11(1): 69–78. Davies, H. J. 1981. Kobon. Amsterdam: North-Holland. Dayley, J. P. 1985. Tzutujil Grammar. Berkeley CA: University of California Press. Dayley, J. P. 1989. Tümpisa (Panamint) Shoshone Grammar. Berkeley CA: University of California Press. De Saussure, F. 1922. Cours de Linguistique Générale. 2nd edn. Paris: Payot. De Smedt, K. J. M. J. 1994. Parallelism in incremental sentence generation. In Parallelism in Natural Language Processing, G. Adriaens & U. Hahn (eds.), 421–47. Norwood NJ: Ablex. Deacon, T. 1997. The Symbolic Species: The Co-Evolution of Language and the Human Brain. London: Penguin. Declerck, R. 1991. Tense in English: Its Structure and its Use in Discourse. London: Routledge. Dedrick, J. M. & Casad, E. H. 1999. Sonora Yaqui Language Structures. Tucson AZ: University of Arizona Press. Dench, A. C. 1995. Martuthunira: A Language of the Pilbara Region of Western Australia. Canberra: Australian National University. Denwood, P. 1999. Tibetan. Amsterdam: John Benjamins. Derbyshire, D. C. 1979. Hixkaryana. Amsterdam: North-Holland. Diessel, H. 1999. Demonstratives: Form, Function, and Grammaticalization. Amsterdam: John Benjamins.



References 209

Diessel, H. 2001. The ordering distribution of main and adverbial clauses: A typological study. Language 77(2): 433–55. Diessel, H. 2005. Competing motivations for the ordering of main and adverbial clauses. Linguistics 43(3): 449–70. Diessel, H. 2008. Iconicity of sequence: A corpus-based analysis of the positioning of temporal adverbial clauses in English. Cognitive Linguistics 19(3): 457–82. Diessel, H. & Hetterle, K. To appear. Causal clauses: A  cross-linguistic investigation of their structure, meaning, and use. In Linguistic Universals and Language Variation, P. Siemund (ed.). Berlin: Mouton de Gruyter. Dixon, R. M. W. 1977. A Grammar of Yidiɲ. Cambridge: Cambridge University Press. Dixon, R. M. W. 1988. A Grammar of Boumaa Fijian. Chicago IL: University of Chicago Press. Dixon, R. M. W. 1994. Ergativity. Cambridge: Cambridge University Press. Dixon, R. M. W. 2006. Complement clauses and complementation strategies in typological perspective. In Complementation, R. M. W. Dixon (ed.), 1–48. Oxford: Oxford University Press. Donaldson, T. 1980. Ngiyambaa: The Language of the Wangaaybuwan. Cambridge: Cambridge University Press. Donohue, M. 1999. A Grammar of Tukang Besi. Berlin: Mouton de Gruyter. Dryer, M. S. 1989. Large linguistic areas and language sampling. Studies in Language 13(2): 257–92. Dryer, M. S. 1992. The Greenbergian word order correlations. Language 68(1): 80–138. Dryer, M. S. 1997. Why statistical universals are better than absolute universals. Chicago Linguistic Society 33: 123–45. Dryer, M. S. 2005a. Order of adverbial subordinator and clause. In The World Atlas of Language Structures, M. Haspelmath, M. S. Dryer, D. Gil & B. Comrie (eds.), 382–3. Oxford: Oxford University Press. Dryer, M. S. 2005b. Relationship between the order of object and verb and the order of relative clause and noun. In The World Atlas of Language Structures, M. Haspelmath, M. S. Dryer, D. Gil & B. Comrie (eds.), 390–3. Oxford: Oxford University Press. Du Bois, J. A. 1985. Competing motivations. In Iconicity in Syntax, J. Haiman (ed.), 343–66. Amsterdam: John Benjamins. Durie, M. 1985. A Grammar of Acehnese. Dordrecht: Foris. Durie, M. 1988. Verb serialization and «verbal-prepositions» in Oceanic languages. Oceanic Linguistics 27: 1–23. Elman, J. L., Bates, E. A., Johnson, M. H., Karmiloff-Smith, A., Parisi, D. & Plunkett, K. 1996. Rethinking Innateness: A Connectionist Perspective on Development. Cambridge MA: The MIT Press. Erteshik-Shir, N. & Lappin, S. 1983. Under stress: A functional explanation of English sentence stress. Journal of Linguistics 19: 419–53. Evans, N. 1995. A Grammar of Kayardild. Berlin: Mouton de Gruyter. Evans, V. 2004. The Structure of Time. Amsterdam: John Benjamins. Evans, V. & Green, M. 2006. Cognitive Linguistics: An Introduction. Edinburgh: Edinburgh University Press. Everett, D. 1986. Pirahã. In Handbook of Amazonian Languages, Vol. 1, D. C. Derbyshire & G. K. Pullum (eds.), 200–325. Berlin: Mouton de Gruyter. Everett, D. L. & Kern, B. 1997. Wari’ – The Pacaas Novos Language of Western Brazil. London: Routledge.

210 A Typology of Purpose Clauses Faraci, R. 1974. Aspects of the Grammar of Infinitive and For Phrases. PhD dissertation, MIT. Farkas, D. 1988. On obligatory control. Linguistics and Philosophy 11: 27–58. Fillmore, C. J. 1982. Frame semantics. In Linguistics in the Morning Calm, The Linguistic Society of Korea (ed.), 111–37. Seoul: Hanshin. Fischer, O. 2007. Morphosyntactic Change: Functional and Formal Perspectives. Oxford: Oxford University Press. Foley, W. A. 1991. The Yimas Language of New Guinea. Stanford CA: Stanford University Press. Ford, C. E. & Mori, J. 1994. Causal markers in Japanese and English conversations: A cross-linguistic study of interactional grammar. Pragmatics 4: 31–61. Fortescue, M. 1984. West Greenlandic. London: Croom Helm. Frajzyngier, Z. & Shay, E. 2002. A Grammar of Hdi. Berlin: Mouton de Gruyter. Franklin, K. J. 1971. A Grammar of Kewa, New Guinea. Canberra: Australian National University. Friendly, M. 1994. Mosaic displays for multi-way contingency tables. Journal of the American Statistical Association 89: 190–200. Fries, P. 1983. On the status of theme in English: Arguments from discourse. In Micro and Macro Connexity of Texts, J. S. Petöfi & E. Sözer (eds.), 116–52. Hamburg: Buske. Gelfand, M. J. & McCusker, C. 2002. Metaphor and the cultural construction of negotiation: A paradigm for research and practice. In Handbook of Cross-Cultural Management, M. J. Gannon & K. L. Newman (eds.), 292–314. New York NY: Blackwell. Genetti, C. 1991. From postposition to subordinator in Newari. In Approaches to Grammaticalization, Vol. 2: Focus on Types of Grammatical Markers, E. C. Traugott & B. Heine (eds.), 227–56. Amsterdam: John Benjamins. Gil, D. 2001. Escaping eurocentrism: Fieldwork as a process of unlearning. In Linguistic Fieldwork, P. Newman & M. Ratliff (eds.), 102–32. Cambridge: Cambridge University Press. Givón, T. 1980. The binding hierarchy and the typology of complements. Studies in Language 4: 333–77. Givón, T. 1985. Iconicity, isomorphism and nonarbitrary coding in syntax. In Iconicity in Syntax, J. Haiman (ed.), 187–220. Amsterdam: John Benjamins. Givón, T. 1990. Syntax: A  Functional-Typological Introduction, Vol. 1. Amsterdam: John Benjamins. Givón, T. 1991. Syntax: A Functional-Typological Introduction, Vol. 2. Amsterdam: John Benja­ mins. Glinert, L. 1989. The Grammar of Modern Hebrew. Cambridge: Cambridge University Press. Goldberg, A. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago IL: University of Chicago Press. Goldberg, A. 2003. Constructions: A new theoretical approach to language. Trends in Cognitive Sciences 7(5): 219–24. Good, J. 2008. Introduction. In Language Universals and Language Change, J. Good (ed.), 1–29. Oxford: Oxford University Press. Gordon, R. G. Jr., (ed.). 2005. Ethnologue: Languages of the World, 15th edn. Dallas TX: SIL International. Available online at . Greenbaum, S. & Quirk, R. 1990. A Student’s Grammar of the English Language. London: Longman. Greenberg, J. H. 1963 [1966]. Some universals of grammar with particular reference to the order of meaningful elements. In Universals of Language, 2nd edn, J. H. Greenberg (ed.), 73–113. Cambridge MA: The MIT Press.



References

Greenberg, J. H. 2005. Genetic Linguistics, W. Croft (ed.). Oxford: Oxford University Press. Grice, H. P. 1975. Logic and conversation. In Syntax and Semantics, Vol. 3: Speech Acts, P. Cole & J. L. Morgan (eds.), 41–58. New York NY: Academic Press. Gries, S. T. 2006. Some proposals towards more rigorous corpus linguistics. Zeitschrift für ­Anglistik und Amerikanistik 54(2): 191–202. Gries, S. T. 2007. Coll.analysis 3.2. A Program for R for Windows 2.x. Available from the author upon request. Gries, S. T. & Stefanowitsch, A. 2004. Extending collostructional analysis: A corpus-based perspective on ‚alternations‘. International Journal of Corpus Linguistics 9(1): 97–129. Guirardello, R. 1999. A  Reference Grammar of Trumai. PhD dissertation, Rice University, ­Houston TX. Hagman, R. S. 1977. Nama Hottentot Grammar. Bloomington IN: Indiana University Press. Haiman, J. 1980. The iconicity of grammar: Isomorphism and motivation. Language 54: 565–89. Haiman, J. 1983. Iconic and economic motivation. Language 59(4): 781–819. Haiman, J. (ed.). 1985. Iconicity in Syntax. Proceedings of a Symposium on Iconicity in Syntax. Stanford, June 24–6, 1983. Amsterdam: John Benjamins. Haiman, J. & Thompson, S. A. 1984. “Subordination” in universal grammar. Berkeley Linguistics Society 10: 510–23. Halliday, M. A. K. 1967. Notes on transitivity and theme in English. Journal of Linguistics 3(1): 37–81 & 3(2): 199–244. Hampe, B. 2009. Non-iconic clause ordering and discourse function of sentence-initial adverbial clauses. Paper to be presented at ICLC 11. Berkeley, July 2009. Hargreaves, D. 1991. The conceptual structure of intentional action: Data from Kathmandu N ­ ewari. Berkeley Linguistic Society 17: 379–89. Harms, P. L. 1994. Epena Pedee Syntax.Arlington TX: SIL & University of Texas at Arlington. Harriehausen, B. 1990. Hmong Njua. Tübingen: Niemeyer. Harris, A. C. 2006. Explaining exuberant agreement. Paper presented at the conference on Rara and Rarissima: Collecting and Interpreting Unusual Characteristics of Human Languages. Leipzig, 31 March 2006. Harris, A. C. & Campbell, L. 1995. Historical Syntax in Cross-Linguistic Perspective. Cambridge: Cambridge University Press. Harris, A. C. & Xu, Z. 2006. Diachronic morphological typology. In Encyclopedia of Language and Linguistics, Vol. 3, 2nd edn, K. Brown (ed.), 509–15. Oxford: Elsevier. Haspelmath, M. 1989. From purposive to infinitive – A  universal path of grammaticization. F ­ olia Linguistica Historica X(1–2): 287–310. Haspelmath, M. 1993. A Grammar of Lezgian. Berlin: Mouton de Gruyter. Haspelmath, M. 1997. Indefinite Pronouns. Oxford: Oxford University Press Haspelmath, M. 1999. On the cross-linguistic distribution of same-subject and different-subject complement clauses: Economic vs. iconic motivation. Paper presented at the 6th International Cognitive Linguistics Conference (ICLC), Stockholm, July 1999. Available online at . Haspelmath, M. 2003. The geometry of grammatical meaning: Semantic maps and cross-linguistic comparison. In The New Psychology of Language, Vol. 2, M. Tomasello (ed.), 211–42. Mahwah NJ: Lawrence Erlbaum. Haspelmath, M. 2005. ‘Want’ complement clauses. In The World Atlas of Language Structures, M. Haspelmath, M. S. Dryer, D. Gil & B. Comrie (eds.), 502–9. Oxford: Oxford University Press.

211

212 A Typology of Purpose Clauses Haspelmath, M. 2007. Pre-established categories don’t exist: Consequences for language ­description and typology. Linguistic Typology 11: 119–32. Haspelmath, M. 2008a. Comparative concepts and descriptive categories in cross-linguistic studies. Ms., MPI Leipzig. Available online at . Haspelmath, M. 2008b. Parametric versus functional explanations of syntactic universals. In The Limits of Syntactic Variation, T. Biberauer (ed.), 85–107. Amsterdam: John Benjamins. Haspelmath, M. 2008c. Frequency vs. iconicity in explaining grammatical asymmetries. Cognitive Linguistics 19(1): 1–33. Haspelmath, M. 2008d. Creating economical morphosyntactic patterns in language change. In Language Universals and Language Change, J. Good (ed.), 85–107. Oxford: Oxford University Press. Haspelmath, M. & König, E. (eds.) 1995. Converbs in Cross-Linguistic Perspective. Berlin: Mouton de Gruyter. Haspelmath, M., König, E., Oesterreicher, W. & Raible, W. 2001. Preface. In Language Typology and Language Universals: An International Handbook, M. Haspelmath, E. König, W. Oesterreicher & W. Raible (eds.), v-vi. Berlin: Mouton de Gruyter. Haspelmath, M., Dryer, M. S., Gil, D. & Comrie, B. (eds.) 2005. The World Atlas of Language Structures. Oxford: Oxford University Press. Haviland, J. 1981. Sk’op sotz’leb: El Tzotzil de San Lorenzo Zinacantán. Mexico City: Universidad Nacional Autónoma de México. Haviland, J. 1993. The syntax of Tzotzil auxiliaries and directionals: The grammaticalization of ‘motion’. Berkeley Linguistics Society 19: 35–49. Hawkins, J. A. 1983. Word Order Universals. New York NY: Academic Press. Hawkins, J. A. 1994. A Performance Theory of Order and Constituency. Cambridge: Cambridge University Press. Hawkins, J. A. 2004. Efficiency and Complexity in Grammars. Oxford: Oxford University Press. Hawkins, J. A. To appear. The typology of noun phrase structure from a processing perspective. Linguistic Typology. Heath, J. 1984. Functional Grammar of Nunggubuyu. Canberra: Australian Institute of Aboriginal Studies. Heath, J. 1999. A Grammar of Koyra Chiini. Berlin: Mouton de Gruyter. Heine, B. 1991. The dative in Ik and Kanuri. In Studies in Typology and Diachrony: Papers Presented to Joseph H. Greenberg on his 75th Birthday, W. Croft, K. Denning & S. Kemmer (eds.), 129–49. Amsterdam: John Benjamins. Heine, B., Claudi, U. & Hünnemeyer, F. 1991. Grammaticalization: A  Conceptual Framework. Chicago IL: The University of Chicago Press. Heine, B. & Kuteva, T. 2002. World Lexicon of Grammaticalization. Cambridge: Cambridge University Press. Hengeveld, K. 1989. Layers and operators in functional grammar. Journal of Linguistics 25: 127– 57. Hetterle, K. In prog. Inter- and Intra-categorical Variation in Adverbial Clauses. PhD project, University of Jena. Hewitt, B. G. 1987. The Typology of Subordination in Georgian and Abkhaz. Berlin: Mouton de Gruyter. Hewitt, B. G. 1989. Abkhaz. London: Routledge



References 213

Hewitt, B. G. 1995. Georgian: A Structural Reference Grammar. Amsterdam: John Benjamins. Hewitt, B. G. 2005. Abkhazian Folktales. Munich: Lincom. Holton, D., Mackridge, P. & Philippaki-Warburton, I. 1997. Greek: A Comprehensive Grammar of the Modern Language. London: Routledge. Hopper, P. J. 1987. Emergent grammar. In Proceedings of the 13th Annual Meeting of the Berkeley Linguistics Society, J. Aske, N. Beery, L. Michaelis & H. Filip (eds.), 139–57. Berkeley CA: University of California Press. Hopper, P. J. 1991. On some principles of grammaticization. In Approaches to Grammaticalization, Vol. 1: Focus on Theoretical and Methodological Issues, E. C. Traugott & B. Heine (eds.), 17–35. Amsterdam: John Benjamins. Hopper, P. J. & Traugott, E. C. 2003. Grammaticalization. 2nd edn. Cambridge: Cambridge University Press. Horie, K. (ed.). 2000. Complementation: Cognitive and Functional Perspectives. Amsterdam: John Benjamins. Hualde, J. I. & Ortiz de Urbina, J. (eds.). 2003. A Grammar of Basque. Berlin: Mouton de Gruyter. Huddleston, R. 1984. Introduction to the Grammar of English. Cambridge: Cambridge University Press. Huddleston, R. 1988. English Grammar: An Outline. Cambridge: Cambridge University Press. Huddleston, R & Pullum, G. K. 2002. The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. Hummel, J. 1996. Linked bar charts: Analysing categorical data graphically. Computational Statistics 11: 23–33. Huttar, G. L. & Huttar, M. L. 1994. Ndyuka. London: Routledge. Hwang, S. J. J. 1995. Purpose clauses in Korean and iconicity. In Interfaces in Korean Linguistics: 1993 Ohio State LSA Workshop, C. Lee (ed.), 125–42. Seoul: Thaehaksa. Ikoro, S. M. 1996. The Kana Language. Leiden: Research School CNWS. International Corpus of English, British Component (ICE-GB). Information available online at . Jackson, H. 1995. Grammar and Meaning: A Semantic Approach to English Grammar. London: Longman. Jaggar, P. J. 2001. Hausa. Amsterdam: John Benjamins. Janssen, D. P., Bickel, B. & Zúñiga, F. 2006. Randomization tests in language typology. Linguistic Typology 10(3): 419–40. Jeffers, R. J. & Pepicello, W. J. 1979. The expression of purpose in Indo-European. Indogermanische Forschungen 84: 1–16. Jespersen, O. 1924. The Philosophy of Grammar. London: Allen and Unwin. Jones, C. 1991. Purpose Clauses. Dordrecht: Kluwer. Jones, W. & Jones, P. 1991. Barasano Syntax – Studies in the Languages of Columbia II. Arlington TX: SIL & University of Texas at Arlington. Kazenin, K. I. 1994. Split syntactic ergativity: Toward an implicational hierarchy. Sprachtypologie und Universalienforschung 47: 78–98. Keenan, E. 1971. Two kinds of presupposition in natural language. In Studies in Linguistic ­Semantics, C. Fillmore & T. Langendoen (eds.), 44–52. New York NY: Holt, Rinehart and Winston. Keenan, E. 1985. Relative clauses. In Language Typology and Syntactic Description, Vol. 2: Complex Constructions, T. Shopen (ed.), 141–70. Cambridge: Cambridge University Press.

214 A Typology of Purpose Clauses Keenan, E. & Comrie, B. 1977. Noun phrase accessibility and universal grammar. Linguistic Inquiry 8: 63–99. Keller, R. 1994. Sprachwandel. Von der unsichtbaren Hand in der Sprache. Tübingen: Francke. Kenesei, I., Vago, R. M. & Fenyvesi, A. 1998. Hungarian. London: Routledge. Khrakovskij, V. S. (ed.). 2005. Typology of Conditional Constructions. Munich: Lincom. Kibrik, A. E. 1985. The meaning-form correspondence in grammatical description. In Language Typology 1985. Papers from the Linguistic Typology Symposium, Moscow, 9–13 December 1985, W. P. Lehmann (ed.), 163–7. Amsterdam: John Benjamins. Kilgarriff, A. 1996. BNC Database and Word Frequency Lists. Available online at . Kirkpatrick, C. 1982. The transitive purpose clause in English. Texas Linguistic Forum 19: 99– 123. Kite, S. & Wurm, S. 2004. The Duuŋidjawu Language of Southeast Queensland. Canberra: Australian National University. Koptjevskaja-Tamm, M. 1993. Nominalizations. London: Routledge. Koptjevskaja-Tamm, M. 1994. Finiteness. In Encyclopedia of Language and Linguistics, Vol.3, R. E. Asher & J. M. Y. Simpson (eds.), 1245–1248. Oxford: Pergamon. Kornfilt, J. 1997. Turkish. London: Routledge. Kortmann, B. 1995. Adverbial participial clauses in English. In Converbs in Cross-Linguistic Perspective, M. Haspelmath & E. König (eds.), 189–237. Berlin: Mouton de Gruyter. Kortmann, B. 1997. Adverbial Subordination. A Typology and History of Adverbial Subordinators based on European Languages. Berlin: Mouton de Gruyter. Kortmann, B. 2001. Adverbial conjunctions. In Language Typology and Language Universals: An International Handbook, M. Haspelmath, E. König, W. Oesterreicher & W. Raible (eds.), 842–54. Berlin: Mouton de Gruyter. Kortmann, B. 2007. English linguistics today – its current situation and future perspectives. In English Studies Today: Recent Developments and New Directions, A. Nünning & J. Schläger (eds.), 23–40. Trier: Wissenschaftlicher Verlag. Krönlein, J. G. 1889. Wortschatz der Khoi-Khoin. Berlin: Deutsche Kolonialgesellschaft. Kruspe, N. 2004. A Grammar of Semelai. Cambridge: Cambridge University Press. Kung, S. S. 2007. A Descriptive Grammar of Huehuetla Tepechua. PhD dissertation, University of Texas at Austin. Lakoff, G. & Johnson, M. 1980. Metaphors We Live By. Chicago IL: University of Chicago Press. Lakoff, G. & Johnson, M. 1999. Philosophy in the Flesh: The Embodied Mind and its Challenge to Western Thought. New York NY: Basic Books. Lamberti, M. & Sottile, R. 1997. The Wolaytta Language. Köln: Köppe Verlag. Lambrecht, K. 1994. Information Structure and Sentence Form. Cambridge: Cambridge University Press. Langacker, R. W. 1987. Foundations of Cognitive Grammar, Vol. 1: Theoretical Prerequisites. Stanford CA: Stanford University Press. Langacker, R. W. 1991. Foundations of Cognitive Grammar, Vol. 2: Descriptive Application. Stanford CA: Stanford University Press. Langacker, R. W. 1999. Grammar and Conceptualization. Berlin: Mouton de Gruyter. LaPolla, R. J. 2003. A Grammar of Qiang. Berlin: Mouton de Gruyter. Lee, J. 1987. Tiwi Today – A Study of Language Change in a Contact Situation. Canberra: Australian National University.



References 215

Lee, M. 1988. Language, perception, and the world. In Explaining Language Universals, J. A. Hawkins (ed.), 211–46. Oxford: Blackwell. Lehmann, C. 1988. Towards a typology of clause linkage. In Clause Combining in Grammar and Discourse, J. Haiman & S. A. Thompson (eds.), 181–225. Amsterdam: John Benjamins. Lehmann, C. 1995. Thoughts on Grammaticalization. Munich: Lincom. Lehmann, W. P. 1973. A structural principle of language and its implications. Language 49: 47– 66. Leite, N. 1994. Master metonymy list. Ms, University of California at Berkeley. Li, C. N. & Thompson, S. A. 1976. Subject and topic: A new typology of language. In Subject and Topic, C. N. Li (ed.), 457–89. New York NY: Academic Press. Li, C. N. & Thompson, S. A. 1981. Mandarin Chinese: A Functional Reference Grammar. Berkeley CA: University of California Press. Lichtenberk, F. 1995. Apprehensional epistemics. In Modality in Grammar and Discourse, J. L. Bybee & S. Fleischman (eds.), 293–327. Amsterdam: John Benjamins. Lindenfeld, J. 1973. Yaqui Syntax. Berkeley CA: University of California Press. Longacre, R. E. 1985. Sentences as combinations of clauses. In Language Typology and Syntactic Description, Vol. 2: Complex Constructions, T. Shopen (ed.), 235–86. Cambridge: Cambridge University Press. Los, B. 2005. The Rise of the To-Infinitive. Oxford: Oxford University Press. Luraghi, S. 2005. Paths of semantic extension: From cause to beneficiary and purpose. In Historical Linguistics 2003, M. Fortescue, E. Skafte Jensen, J. E. Mogensen & L. Schøsler (eds.), 141–57. Amsterdam: John Benjamins. Lyons, J. 1968. Introduction to Theoretical Linguistics. Cambridge: Cambridge University Press. Maas, L. 2004. “Finite” and “nonfinite” from a typological perspective. Linguistics 42(2): 359–86. Macaulay, M. 1996. A Grammar of Chalcatongo Mixtec. Berkeley CA: University of California Press. MacWhinney, B. 1977. Starting points. Language 53(1): 152–68. MacWhinney, B. & Pléh, C. 1988. The processing of restrictive relative clauses in Hungarian. Cognition 29(2): 95–144. Mahootian, S. 1997. Persian. London: Routledge. Maslova, E. 2000. A dynamic approach to the verification of distributional universals. Linguistic Typology 4(3): 307–33. Maslova, E. 2003. A Grammar of Kolyma Yukaghir. Berlin: Mouton de Gruyter. Matthiesen, C. & Thompson, S. A. 1988. The structure of discourse and “subordination.” In Clause Combining in Grammar and Discourse, J. Haiman & S. A. Thompson (eds.), 275– 333. Amsterdam: John Benjamins. Merlan, F. 1982. Mangarayi. Amsterdam: North-Holland. Merlan, F. 1994. A Grammar of Wardaman. Berlin: Mouton de Gruyter. Meyer, D., Zeileis, A. & Hornik, K. 2003. Visualizing independence using extended association plots. In Proceedings of the 3rd International Workshop on Distributed Statistical Computing, Vienna, Austria, 2003, K. Hornik, F. Leisch & A. Zeileis (eds.). Available online at . Meyer, D., Zeileis, A. & Hornik, K. 2006. The strucplot framework: Visualizing multi-way contingency tables with vcd. Journal of Statistical Software 17(3): 1–48. Miller, A. W. 2001. A Grammar of Jamul Tiipay. Berlin: Mouton de Gruyter. Myhill, J. 1988. Categoriality and clustering. Studies in Language 12: 261–97.

216 A Typology of Purpose Clauses Murphy, G. L. 2002. The Big Book of Concepts. Cambridge MA: The MIT Press. Nedjalkov, I. 1997. Evenki. London: Routledge. Newmeyer, F. J. 2005. Possible and Probable Languages: A Generative Perspective on Linguistic Typology. Oxford: Oxford University Press. Newmeyer, F. J. 2007. Linguistic typology requires crosslinguistic formal categories. Linguistic Typology 11: 133–57. Nichols, J. 1992. Linguistic Diversity in Space and Time. Chicago IL: University of Chicago Press. Nichols, J. 2004. Samples for comparative grammar: A practical guide. Ms., University of California at Berkeley. Noonan, M. 1985. Complementation. In Language Typology and Syntactic Description, Vol. 2: Complex Constructions, T. Shopen (ed.), 42–140. Cambridge: Cambridge University Press. Noonan, M. 1992. A Grammar of Lango. Berlin: Mouton de Gruyter. Nordlinger, R. 1998. A Grammar of Wambaya, Northern Territory (Australia). Canberra: Australian National University. Oxford English Dictionary. Online version, accessible at . Palmer, F. R. 1986. Mood and Modality. Cambridge: Cambridge University Press. Payne, D. L. & Payne, T. E. 1990. Yagua. In Handbook of Amazonian Languages, Vol. 2, D. C. Derbyshire & G. K. Pullum (eds.), 249–474. Berlin: Mouton de Gruyter. Payne, T. 1997. Describing Morphosyntax. Cambridge: Cambridge University Press. Pepicello, W. J. 1982. On the sources of Indo-European conjunctions of purpose, cause and result. In Papers from the 5th International Conference on Historical Linguistics, A. Ahlqvist (ed.), 256–64. Amsterdam: John Benjamins. Perkins, R. D. 1989. Statistical techniques for determining language sample size. Studies in Language 13(2): 293–315. Perkins, R. D. 2001. Sampling procedures and statistical methods. In Language Typology and Language Universals: An International Handbook, M. Haspelmath, E. König, W. Oester­ reicher & W. Raible (eds.), 419–34. Berlin: Mouton de Gruyter. Posner, R. 1980. Meaning and use of sentence connectives in natural language. In Speech Act Theory and Pragmatics, J. R. Searle, F. Kiefer & M. Bierwisch (eds.), 169–203. Dordrecht: Riedel. Potter, S. 1969. Changing English. London: Deutsch. Pusch, C. D. 2001. Ikonizität. In Language Typology and Language Universals: An International Handbook, M. Haspelmath, E. König, W. Oesterreicher & W. Raible (eds.), 369–84. Berlin: Mouton de Gruyter. Pustet, R. 1993. Copulas. Universals in the Categorization of the Lexicon. Oxford: Oxford University Press. R Development Core Team. 2007. R: A  Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing. Available online at <www.r-project.org>. Reh, M. 1985. Die Krongo-Sprache (Nìinò Mó-Dì). Berlin: Dietrich Reimer Verlag. Reinhart, T. 1984. Principles of gestalt perception in the temporal organization of narrative texts. Linguistics 22: 779–809. Rice, K. 1989. A Grammar of Slave. Berlin: Mouton de Gruyter. Rice, S. & Kabata, K. 2007. Crosslinguistic grammaticalization patterns of the allative. Linguistic Typology 11: 451–514. Rijkhoff, J., Bakker, D., Hengeveld, K. & Kahrel, P. 1993. A method of language sampling. Studies in Language 17(1): 169–203.



References 217

Robert, C. P. & Casella, G. 2005. Monte Carlo Statistical Methods, 2nd edn. Berlin: Springer. Roberts, J. 1987. Amele. London: Croom Helm. Romero-Figueroa, A. 1997. A Reference Grammar of Warao. Munich: Lincom. Rubenbauer, H. & Hofmann, J. B. 1995. Lateinische Grammatik. 12th edn. Bamberg: Buchners, Lindauer, Oldenbourg. Ruhlen, M. 1991. A Guide to the World’s Languages, Vol. 1: Classification. Stanford CA: Stanford University Press. Rumsey, A. 1982. An Intra-Sentence Grammar of Ungarinjin, North-Western Australia. Canberra: Australian National University. Rupp, J. E. 1989. Lealao Chinantec Syntax. Dallas TX: University of Texas at Arlington Publications. Saeed, J. 1999. Somali. Amsterdam: John Benjamins. Saxena, A. 1988. On the grammaticalization of the verb say/thus: A typological study. Paper presented at the Symposium on Grammaticalization, Eugene OR, 12–15 May 1988. Schaub, W. 1985. Babungo. London: Croom Helm. Schmidtke-Bode, K. To appear, a. The role of benefactives and related notions in the typology of purpose clauses. In Benefactives and Malefactives: Case Studies and Typological Perspectives, F. Zúñiga & S. Kittilä (eds.). Amsterdam: John Benjamins. Schmidtke-Bode, K. To appear, b. The performance basis of grammatical constraints on complex sentences: A preliminary survey. In Clause Combining in Cross-Linguistic Perspective, V. Gast & H. Diessel (eds.). Berlin: Mouton de Gruyter. Schulte, K. 2007. What causes adverbial infinitives to spread? Evidence from Romance. Language Sciences 29: 512–37. Searle, J. R. 1969. Speech Acts. An Essay in the Philosophy of Language. Cambridge: Cambridge University Press. Searle, J. R. 2006. What is language: Some preliminary remarks. Ms., University of California at Berkeley. Available online at < http://socrates.berkeley.edu/~jsearle/articles.html>. Séguy, J. 1973. Atlas Linguistique de la Gascogne, Vol. 6: Notice Explicative. Paris: Centre National de la Recherche Scientifique. Seiler, W. 1985. Imonda. Canberra: Australian National University. Smeets, I. 1989. A Mapuche Grammar. PhD dissertation, Rijksuniversiteit te Leiden. Sohn, H. 1994. Korean. London: Routledge. Song, J. J. 2005. Periphrastic causative constructions. In The World Atlas of Language Structures, M. Haspelmath, M. S. Dryer, D. Gil & B. Comrie (eds.), 446–9. Oxford: Oxford University Press. Soukka, M. 2000. A Descriptive Grammar of Noon – A Cangin Language of Senegal. Munich: L ­ incom. Southern Ute Tribe. 1980. Ute Reference Grammar. Ignacio CO: Ute Press, The Southern Ute Tribe. Sridhar, S. N. 1990. Kannada. London: Routledge. Stassen, L. 1985. Comparison and Universal Grammar. Oxford: Blackwell. Stefanowitsch, A. 1999. The go-and-VERB construction in a cross-linguistic perspective: Imageschema blending and the construal of events. In Proceedings of the 2nd Annual High Desert Linguistics Society Conference, D. Nordquist & C. Berkenfield (eds.), 123–34. Albuquerque NM: High Desert Linguistics Society. Stefanowitsch, A. & Gries, S. T. 2003. Collostructions: Investigating the interaction between words and constructions. International Journal of Corpus Linguistics 8(2): 209–43.

218 A Typology of Purpose Clauses Stefanowitsch, A. & Gries, S. T. 2005. Co-varying collexemes. Corpus Linguistics and Linguistic Theory 1(1): 1–43. Sweetser, E. 1990. From Etymology to Pragmatics. Metaphorical and Cultural Aspects of Semantic Structure. Cambridge: Cambridge University Press. Talmy, L. 1975. Figure and ground in complex sentences. Working Papers on Language Universals 17: 65–75. Taylor, C. V. 1985. Nkore-Kiga. London: Croom Helm. Terrill, A. 2003. A Grammar of Lavukaleve. Berlin: Mouton de Gruyter Thompson, S. A. 1985. Grammar and written discourse: Initial vs. final purpose clauses in English. Text 5(1–2): 55–84. Thompson, S. A. 1987. “Subordination” and narrative event structure. In Coherence and Grounding in Discourse, R. Tomlin (ed.), 435–54. Amsterdam: John Benjamins. Thompson, S. A. & Couper-Kuhlen, E. 2005. The clause as a locus of grammar and interaction. Discourse Studies 7(4–5): 481–505. Thompson, S. A. & Longacre, R. E. 1985. Adverbial clauses. In Language Typology and Syntactic Description, Vol. 2: Complex Constructions, T. Shopen (ed.), 171–234. Cambridge: Cambridge University Press. Tomasello, M. 2003. Constructing a Language: A Usage-Based Theory of Language Acquisition. Cambridge MA: Harvard University Press. Tomasello, M. 2004. What kind of evidence could refute the UG hypothesis? Studies in Language 28(3): 642–5. Tomlin, R. 1985. Foreground-background information and the syntax of subordination. Text 5: 85–122. Tomlin, R. 1986. Basic Word Order: Functional Principles. London: Croom Helm. Tomlin, R. & Villa, V. 1994. Attention in cognitive science and second language acquisition. Studies in Second Language Acquisition 16: 183–203. Valentine, R. J. 2001. Nishnaabemwin Reference Grammar. Toronto: University of Toronto Press. Van Klinken, C. L. 1999. A Grammar of the Fehan Dialect of Tetun – An Austronesian Language of West Timor. Canberra: Australian National University. Van Valin, R. D. Jr. 1984. A typology of syntactic relations in clause linkage. Berkeley Linguistics Society 10: 542–58. Van Valin, R. D. Jr. 2001. Functional linguistics. In The Handbook of Linguistics, M. Aronoff & J. Rees-Miller (eds.), 319–36. Oxford: Blackwell. Van Valin, R. D. Jr. & LaPolla, R. 1997. Syntax: Structure, Meaning, and Function. Cambridge: Cambridge University Press. Vennemann, T. 1973. Explanation in syntax. In Syntax and Semantics 2, J. Kimball (ed.), 1–50. New York NY: Academic Press. Vennemann, T. 1974. Theoretical word order studies: Results and problems. Papiere zur Linguistik 7: 5–25. Vennemann, T. 1976. Categorial grammar and the order of meaningful elements. In Linguistic Studies Offered to Joseph Greenberg on the Occasion of his Sixtieth Birthday, A. Juilland (ed.), 615–34. Saratoga CA: Anma Libri. Verstraete, J.-C. 2004. Initial and final position for adverbial clauses in English: The constructional basis of the discursive and syntactic differences. Linguistics 42(4): 819–53. Verstraete, J.-C. 2008. The status of purpose, reason and intended endpoint in the typology of complex sentences, and its implications for layered models of clause structure. Linguistics 46: 757–88.



References 219

Wang, F. 2006. The information structure of adverbial clauses in Chinese discourse. Taiwan Journal of Linguistics 4: 49–88. Wash, S. 2001. Adverbial Clauses in Barbareño Chumash Narrative Discourse. PhD dissertation: University of California at Santa Barbara. Wasow, T. 1997. Remarks on grammatical weight. Language Variation and Change 9: 81–105. Watkins, L. J. 1984. A Grammar of Kiowa. Lincoln NB: University of Nebraska Press. Weber, D. J. 1989. A Grammar of Huallaga (Huánuco) Quechua. Berkeley CA: University of California Press. Weinreich, U., Labov, W. & Herzog, M. I. 1968. Empirical foundations for a theory of language change. In Directions for Historical Linguistics: A Symposium, W. P. Lehmann & Y. Malkiel (eds.), 95–195. Austin TX: University of Texas Press. Wiemer, B. & Bisang, W. 2004. What makes grammaticalization? An appraisal of its components and its fringes. In What Makes Grammaticalization? A Look from its Fringes and its Components, W. Bisang, N. P. Himmelmann & B. Wiemer (eds.), 3–20. Berlin: Mouton de Gruyter. Wierzbicka, A. 1998. Anchoring linguistic typology in universal semantic primes. Linguistic Typology 2(2): 141–94. Wittgenstein, L. 1958. Philosophical Investigations. Translated by G. E. M. Anscombe. 2nd edn. Oxford: Blackwell. Witzlack-Makarevich, A. 2007. Construction-centered database on grammatical relations. ­Paper presented at the 7th Biennial Meeting of ALT, Paris, September 25–8, 2007. Zipf, G. 1935. The Psychobiology of Language: An Introduction to Dynamic Philology. Cambridge, MA: The MIT Press.

Index of authors A Aissen, J.  58f., 95, 181f. Aitchison, J.  12 Anderson, G. D.  183 Anderson, S. R.  6 B Baayen, H.  203 Bach, E.  61 Bakker, D.  24 Baroni, M.  203 Barsalou, L. W.  169 Bates, E.  71 Bauer, W.  93, 105 Bell, A.  21 Berry, C.  99, 191 Berry, K.  99, 191 Bhatia, T. K.  118 Biber, D.  139 Biberauer, T.  9 Bickel, B.  6, 7, 8, 9, 16, 17, 18, 24, 65, 82, 149, 202, 203 Bisang, W.  85, 171 Blackings, M.  62, 166f., 180f. Blansitt, E. L.  91 Borgman, D. M.  87, 178 Brown, P.  18 Buechel, E.  41 Bybee, J. L.  11, 14, 15, 73, 78, 117, 171, 180, 190 C Campbell, L.  190, 195 Carlson, R.  76, 127 Carston, R.  107 Casella, G.  34 Chafe, W.  11, 107, 123 Chapman, S.  127, 145, 158 Chomsky, N.  6, 9 Chung, S.  43 Clark, E.  11 Cohen, J.  34, 38 Comrie, B.  1, 10, 19, 36, 43, 51

Couper-Kuhlen, E.  19 Cristofaro, S.  1, 27, 34, 35, 39, 42, 50ff., 60, 69ff., 148, 155, 156, 162 Croft, W.  6, 8, 11, 12, 13, 15, 17, 19, 30, 31, 98, 99, 101, 107, 108, 139, 144, 149, 156, 172, 173, 183 Cruse, D. A.  11, 12, 98 Culicover, P. W.  107 Curnow, T.  128 Cutler, A.  73 Cysouw, M.  6, 8, 33, 114, 203 D Dąbrowska, E.  6, 15 Daniel, M.  7 Davies, H. J.  77 Dayley, J. P.  60, 93, 106, 158, 189 De Saussaure, F.  17 De Smedt, K. J. M. J.  118 Deacon, T.  9 Declerck, R.  36 Dench, A. C.  54, 136, 153, 157 Derbyshire, D. C.  55, 95f., 127, 145, 158 Diessel, H.  vii, 15, 18, 117, 119, 121ff., 151 Dixon, R. M. W.  47, 66f., 102, 161, 163 Donaldson, T.  124 Donohue, M.  74, 104, 191 Dryer, M.  6, 21, 22, 73, 113, 114, 153 Du Bois, J. A.  14 Durie, M.  38, 46, 103, 191 E Elman, J. L.  6, 10 Erteshik-Shir, N.  164 Evans, N.  64, 75, 136 Evans, V.  11, 97, 179

Everett, D. L.  36, 107 Evert, S.  203 F Fabb, N.  62, 166f., 180f. Faraci, R.  168 Farkas, D.  61 Fillmore, C.  98 Fischer, O.  171, 180 Foley, W. A.  54 Ford, C. E.  127 Fortescue, M.  126, 153, 190 Frajzyngier, Z.  61 Franklin, K. J.  42, 108 Friendly, M.  134 Fries, P.  123 G Gast, V.  58 Gelfand, M. J.  97 Genetti, C.  190 Gil, D.  105, 163 Givón, T.  9, 11, 35, 43, 57, 116 Glinert, L.  145 Goldberg, A.  30, 145 Good, J.  15 Gordon, R. G.  21 Green, M.  11, 179 Greenbaum, S.  130 Greenberg, J. H.  5, 7, 18, 21, 23, 110, 113 Grice, H. P.  14 Gries, S. T.  34, 139ff., 203 Guirardello, R.  63, 183ff. H Hagman, R. S.  191 Haiman, J.  11, 43, 116, 149 Halliday, M. A. K.  123 Hampe, B.  126 Hargreaves, D.  47 Harms, P. L.  13, 144f. Harriehausen, B.  83, 188

222 A Typology of Purpose Clauses Harris, A. C.  78, 190, 193, 195 Haspelmath, M.  1, 6, 7, 12, 13, 14, 18, 24, 27, 47, 51, 57, 60, 70, 75, 91, 105, 117, 170, 171, 172f., 177, 192, 198 Haviland, J.  58, 182 Hawkins, J. A.  10, 11, 15, 71, 104, 113, 119f., 129, 138, 152 Heath, J.  153 Heine, B.  91f., 101, 171, 177f., 184, 188f., 191, 194f. Hengeveld, K.  126 Hetterle, K.  vii, 151, 176 Hewitt, B. G.  44f., 67, 109, 121, 192 Hofmann, J. B.  75 Holton, D.  154 Hopper, P. J.  15, 16, 102, 171, 175, 178, 180, 187, 194f. Horie, K.  1 Hualde, J. I.  37, 94, 169 Huddleston, R.  14, 27, 63, 74, 84, 127, 138, 153 Hummel, J.  48 Huttar, G. L.  127 Huttar, M. L.  127 Hwang, S. J. J.  117 I Ikoro, S. M.  37, 194 J Jackendoff, R.  107 Jackson, H.  18 Jaggar, P. J.  127, 177 Janssen, D. P.  33, 34 Jeffers, R. J.  195 Jespersen, O.  43 Johnson, M.  12, 18, 97f., 179 Jones, C.  1, 27, 62, 109, 168 Jones, P.  65, 194 Jones, W.  65, 194 K Kabata, K.  93, 100f., 188f. Kazenin, K. I.  1, 27, 67, 156, 168 Keenan, E.  46, 51, 170 Keller, R.  15 Kenesei, I.  37, 84 Kern, B.  36 Khrakovskij, V. S.  1 Kibrik, A. E.  18

Kilgarriff, A.  140 Kirkpatrick, C.  168 Kite, S.  78 König, E.  75 Koptjevskaja-Tamm, M.  35, 68 Kornfilt, J.  54, 102, 107, 154 Kortmann, B.  2, 61f., 74, 83, 86 Krönlein, J. G.  191 Kruspe, N.  103, 104, 109 Kung, S. S.  127 Kuteva, T.  171, 177f., 184, 188f., 191, 194f. L Lakoff, G.  12, 18, 97f., 179 Lambrecht, K.  123, 148 Langacker, R. W.  10, 15, 19, 20, 148 LaPolla, R. J.  19, 59, 154, 175 Lappin, S.  164 Lee, J.  96 Lee, M.  12 Lehmann, C.  52, 69, 149, 171, 173, 190 Lehmann, W. P.  113 Leite, N.  12 Levinson, S. C.  18 Li, C. N.  10, 62, 103 Lichtenberk, F.  129, 186f., 191 Lightfoot, D. W.  6 Lindenfeld, J.  85 Longacre, R. E.  20, 85, 107, 123, 154, 172 Los, B.  173 Luraghi, S.  177 Lyons, J.  149 M Maas, L.  35 Macaulay, M.  46, 152, 195 MacWhinney, B.  64, 71, 135 Mahootian, S.  37, 74 Maslova, E.  21, 74, 162, 181 Matthiesen, C.  175 McCusker, C.  97 Merlan, F.  131, 155, 157, 161, 186 Meyer, D.  56 Miller, A. W.  68, 193 Mori, J.  127 Murphy, G. L.  12 Myhill, J.  190

N Nedjalkov, I.  vii, 44, 75, 196 Newmeyer, F. J.  6, 9 Nichols, J.  vii, 7, 17, 21, 22, 23, 82, 202 Noonan, M.  52, 74, 161 O Ortiz de Urbina, J.  37, 94, 169 P Palmer, F. R.  45, 163, 165 Payne, D. L.  100, 131, 159 Payne, T.  19, 38, 100, 103, 131, 159 Pepicello, W. J.  195 Perkins, R. D.  21, 33 Pléh, C.  64 Poole, K. T.  8 Posner, R.  107 Potter, S.  12 Pullum, G. K.  14, 27, 63, 84, 138, 153 Pusch, C. D.  116 Pustet, R.  18 Q Quirk, R.  130 R R Development Core Team  34 Reh, M.  95, 167 Reinhart, T.  149 Rice, K.  vii, 131, 160–1, 174 Rice, S.  93, 100f., 188f. Rijkhoff, J.  21 Robert, C. P.  34 Roberts, J.  vii, 41, 75, 154, 188 Romero-Figueroa, A.  36, 111 Rubenbauer, H.  75 Ruhlen, M.  21, 23 Rumsey, A.  155 Rupp, J. E.  47, 83 S Saeed, J.  167 Saxena, A.  191 Schaub, W.  159f. Schmidtke-Bode, K.  102, 189, 203 Schulte, K.  176 Searle, J. R.  18, 19, 162 Séguy, J.  8

Seiler, W.  100, 156 Shay, E.  61 Smeets, I.  46, 165 Sohn, H.  36, 44 Song, J. J.  175 Soukka, M.  105 Southern Ute Tribe  31, 43, 76 Sridhar, S. N.  59, 89, 163 Stassen, L.  35, 149 Stefanowitsch, A.  107, 139ff. Sweetser, E.  12 T Talmy, L.  149 Taylor, C. V.  105 Terrill, A.  57, 101, 111, 157 Thompson, S. A.  10, 14, 19, 62, 85, 103, 123, 124f., 149, 154,

Index of authors 223 172, 175, 190 Timberlake, A.  43 Tomasello, M.  6, 15 Tomlin, R.  10, 11, 149 Traugott, E. C.  16, 102, 171, 175, 178, 180, 187, 194f. V van Klinken, C. L.  27, 60, 77, 106 Van Valin, R. D.  9, 19, 59, 149, 175 Vennemann, T.  113 Verstraete, J-C.  1, 27, 47ff., 58, 126, 152, 164f. Villa, V.  11 W Wang, F.  127

Wash, S.  128 Wasow, T.  118 Watkins, L. J.  97, 177 Weber, D. J.  68, 109, 163, 173 Weinreich, U.  7 Wiechmann, D. vii Wiemer, B.  171 Wierzbicka, A.  19, 152, 154, 177 Wittgenstein, L.  91 Witzlack-Makarevich, A.  67 Wurm, S.  78 X Xu, Z.  78 Z Zipf, G.  14 Zúñiga, F.  98

Index of languages A Abkhaz  25, 42, 49, 67, 68, 72, 121, 192 Abun  26, 43, 57, 99, 191 Acehnese  25, 38, 46, 95, 103, 151, 159, 191 Amele  25, 41, 75, 86, 154, 155, 177, 186, 188 Awa Pit  128 B Babungo  25, 159f. Barasano  26, 65, 155, 194 Basque  25, 37, 49, 94, 95, 169, 175 Bulgarian  195 C Chinantec (Lealao)  26, 47, 57, 83 Chumash (Barbareño)  128 D Duuŋidjawu  25, 78, 151 E English, 1, 24, 26, 27, 30, 45, 61, 62f., 83, 84, 93, 106–7, 108–9, 117, 124ff., 129f., 136ff., 153, 166, 168ff., 173ff., 177, 179ff., 184, 186, 189, 194ff., 196 Epena Pedee  13, 119, 144f. Evenki  25, 44, 68, 75, 196 F Fijian (Boumaa)  25, 47, 86, 95, 151, 163, 186 French  196 G Georgian  25, 44, 86, 109 German  177 Greek (Ancient)  195 Greek (Modern)  25, 74, 104, 154, 159, 174, 175

H Hausa  25, 75, 104, 127, 135, 154, 177 Hdi  25, 60f. Hebrew (Modern)  145 Hixkaryana  26, 55, 95f., 135, 154 Hmong Njua  25, 83, 188 Hungarian  25, 37, 84 I Ik  91f. Imonda  26, 42, 86, 95, 100, 156 K Kana  25, 37, 194 Kannada  25, 59, 86, 89, 135, 163, 192 Kanuri  91f. Kayardild  25, 64, 75, 86, 95, 135, 136 Kewa  25, 42, 108, 151 Khoekhoe  25, 57, 86, 191 Kiowa  26, 38, 97, 105, 106, 151, 154, 155, 177 Kobon  25, 57, 77, 151 Korean  25, 35, 44, 111 Koyra Chiini  25, 106, 153, 178, 191 Krongo  25, 95, 167, 175 L Lakhota  26, 41, 51, 95, 105, 106, 135, 145, 151, 154 Lango  25, 66, 74, 189 Latin  195 Lavukaleve  26, 57, 101, 111, 157, 189 Lezgian  25, 47, 66, 70, 95, 154, 192 M Ma’di  25, 62, 166f., 180f. Mandarin Chinese  25, 37, 62,

103, 154, 155, 196 Maori  25, 93, 99, 105 Mapudungun (Mapuche)  26, 46, 57, 72, 86, 165 Martuthunira  25, 54, 135, 136, 153, 157, 186 Meithei  25 Miao (see Hmong Njua) Mingrelian  195 Mixtec (Chalcatongo)  26, 46, 57, 152, 195 N Nama (see Khoekhoe) Ndyuka  26, 42, 127, 189 Newari (Kathmandu)  47, 190 Ngiyambaa  25, 124, 135, 157, 163, 186 Nishnaabemwin  26, 57, Nkore-Kiga  25, 77, 105, 155 Noon  25, 41, 42, 95, 104f., 106 Nunggubuyu  25 O Ojibwe (Eastern) (see Nishnaabemwin) P Paumarí  26, 126, 145, 154, 158 Persian  25, 74 Pirahã  26, 107 Punjabi  25, 72, 104, 118 Q Qiang  25, 51, 106, 177 Quechua (Huallaga)  26, 68, 86, 95, 109, 163, 173 R Romanian  176 S Sanumá  26, 87, 178

Semelai  25, 95, 103, 104, 109, 151 Slave  26, 57, 66, 131, 135, 160, 174 Somali  25, 57, 86, 167 Spanish  85, 184 Supyire  25, 49, 75, 76, 86, 95, 127, 135 Swahili  175 T Tepehua  127 Tetun  25, 42, 60, 77, 106 Tibetan  25, 95 Tiipay (Jamul)  26, 68, 77, 86, 193 Tiwi  25, 95, 96, 104 To’aba’ita  191 Trique (Chicahuaxtla)  107

Index of languages 225 Trumai  26, 57, 63, 77, 183ff. Tukang Besi  25, 74, 86, 95, 103, 104, 186, 191 Tümpisa Shoshone  26, 60, 106, 155, 167 Turkish  25, 54, 66, 102, 107, 154, 177, 189 Tzotzil  58f., 181f., 184 Tzutujil  26, 42, 93, 158, 173, 189

Wardaman  25, 95, 131, 135, 151, 155, 161, 186 Wari’  26, 36, 57 West Greenlandic  26, 126, 153, 154, 178, 190 Wolaytta  25, 42, 86

U Ungarinjin  25, 119, 155 Ute  26, 31, 43, 68, 76, 155

Y Yagua  26, 77, 86, 100, 131, 159 Yaqui  26, 84f., 106, 119, 174, 189 Yidiɲ  25, 66f., 102, 163 Yimas  26, 54, 86 Yukaghir (Kolyma)  25, 74, 95, 161f., 181

W Wambaya  25, 151 Warao  26, 36, 95, 111

X !Xun  195

Index of subjects A adverbial clause, adverbial relation  2, 42, 46, 48, 50, 52, 70f., 96, 106, 108, 119f., 124, 126f., 148, 151ff., 160, 195, 202 agent-binding  48, 49, 58, 69f., 150, 165 agreement (see indexation) alignment (see argument structure) allative  88ff., 93ff., 98, 102, 138, 158, 172, 176, 187ff., 195, 200 ambiguity  49, 63, 103ff., 153, 167f. analogy, analogical extension  12, 99, 172f., 176, 180 antipassive  66 apprehension  129, 141, 186f., 197, 201 areal distributions (see macroareas) argument structure  50ff., 132ff. alignment  9ff., 13, 58, 66f., 201 alignment split  66f., 68, 193 control pragmatic, actual  47, 58, 60ff., 69, 108 syntactic (equideletion)  52, 59, 62, 66f., 118, 136 (non-)obligatory control  62f. coreferentiality  52ff., 66f., 68, 69, 105, 109, 118, 133ff., 161, 176 object-gap purpose clause (OPC)  62ff., 200 referential density  18, 65 semantic roles  108, 134ff., 201 subject  31, 41, 50ff., 103, 108, 118, 132ff., 155, 161, 167, 176,

193, 200 syntactic ergativity  66f. aspect  42ff. aspectual type  108 (im)perfective aspect  43f. immediate aspect  43f. prospective aspect  43f., 183 assertion  148, 156, 164 atelic temporal extension  102 attention  11, 98, 124 attrition (see frequency effects > reduction (phonological)) autotypology  17 auxiliary  30, 45, 47, 68, 73, 77, 87, 140, 143, 162, 178ff., 193, 197, 200f. avertive clause  3, 77f., 129ff., 155, 186f., 191, 197, 201 B balancing, balanced verb form  34ff., 48, 96, 132, 137, 149, 155, 174, 199 benefaction, benefactive  60, 74, 90ff., 99ff., 154, 177, 187ff., 197, 200 borrowing  84f., 174 British National Corpus (BNC)  16, 130, 136f., 139ff., 143, 186 C categorization  12, 30, 32, 71, 165, 169 causal clause  36, 48, 50, 52, 106, 126f., 151f., 154f., 157, 160, 177f., 195, 196, 197, 200, 202 causative  175 chunking  190 clause-linking devices  71ff., 119ff., 158, 164 cleaving, cleft sentence  14, 127 clustering (in grammatical­

ization)  190 cognitive linguistics  11f., 28, 91ff., 97, 102, 116, 148, 199 collexeme  141ff. collostructional analysis (see statistical analysis) comparative concept  18 complement clause, complementation, complementizer  1, 2, 8, 41, 48, 51, 52, 57, 75, 83, 95f., 118, 136, 148, 152, 156, 157ff., 166, 167, 172ff., 182, 186f., 194f., 197, 199f., 202 complexity (of formatives)  83ff., 85 conceptual space  12f., 91, 116, 148ff., 174 concessive clause  36, 148, 151, 157, 176, 202 conditional clause  1, 50, 110, 119, 125f., 145, 151ff., 155, 160, 177, 196, 197 constituent order  11, 14, 78f., 86f., 110, 113ff., 130, 199 construal, conceptualization  11, 97f., 108, 117, 145, 148f., 169, 177, 202 construction  3, 8, 12, 29ff., 67, 70, 71ff., 126, 139ff., 144f., 171, 202 Construction Grammar  30, 71, 126, 145, 171 control (see argument structure) converb  36, 61f., 75 , 192, 196 convergent evolution  9 converging evidence  124, 136, 138f. coordination  3, 20, 27, 78, 106ff., 148, 150f., 161, 163, 195f., 198, 201 coreferentiality (see argument structure)

corpus  2, 8, 15, 16, 26, 31, 47, 57, 61, 96, 117, 124, 126, 136ff., 139ff., 143, 196, 203 correlative element, placeholder  84 cosubordination (see medial verb) D dative  88ff., 99ff., 102, 138, 154, 187ff. demonstrative  15, 18, 84, 153, 171, 194, 197, 199 deontic (see modality and mood) depictive clause  76, 104 deranking, deranked verb form  35ff., 48, 69, 88, 96, 118, 132, 149, 155, 162, 199f. desiderative (see modality and mood) discourse-pragmatic constraints  110, 111, 123ff., 144f., 151, 164, 200 E ecology, ecological organization, ecological niche  41, 69f., 99, 118, 145, 202 economy (see frequency effects) efficiency  14, 119ff. emergence, emergent grammar (see usage-based approach) entrenchment (see frequency effects) epistemic (see modality and mood) equi-deletion (see control) experiential correlation  92, 97, 102, 134f., 138, 144, 179, 200 F family resemblance  91 figure-ground organization (see information structure) finiteness (see balancing) focus (see information structure) foreground-background organization (see information structure) frame, frame semantics  98

Index of subjects 227 frequency effects economical coding  7, 41, 49, 57, 69, 96f., 102, 116, 118, 185, 200f. entrenchment  14, 178 predictability  14, 41, 57, 97, 104, 185 reduction (phonological)  85, 171, 174, 178, 182, 184f., 190, 193 usage frequency  2, 14, 43, 47, 57, 102, 117, 134ff., 139ff., 150, 190, 195, 201 Zipf ’s Law  14, 57 future (see tense) G generative linguistics  5ff., 62f. gerund (see nominalization) gestalt feature  30ff., 71ff., 88ff., 102, 138 goal, goal-orientation  18, 92, 100, 102, 138, 154, 187ff., 200 grammaticalization  15f., 18, 27, 42, 54, 60, 69, 77, 99, 100, 101f., 127f., 130, 142, 145, 154, 155, 162, 164, 170–98, 200f. Gricean maxims of conversation  14, 57 H hypothetical (see modality) I iconicity  11, 85, 108, 110, 116ff., 124, 127, 128, 200 imperative  145, 163, 182 implicational universal  6f., 42, 56, 66, 67, 89, 122, 135 implicature (see inference processes) incorporation  41, 97 indexation  37f., 48, 57, 58f., 63, 67, 157, 181f. inference processes  12, 98f., 103ff., 178, 179f., 195, 197f., 201 infinitive  36, 41, 59, 65, 68, 71ff., 79, 88f., 93, 95, 118, 124ff., 145, 153, 155, 158, 161, 166, 168, 172ff., 176, 179, 183, 195, 199f.

information structure figure-ground organization  149, 156, 202 focus  14, 123ff., 128f. foreground-background organization  149, 156, 164, 168 theme (see topic) topic  14, 123ff., 128f., 145, 156 innateness  6, 10 integration (semantic, syntactic)  35, 41, 48f., 69f., 96, 103, 125, 150, 162, 183, 195, 200ff. intentionality  18f. interlacing  52, 69 International Corpus of English (ICE)  16, 47, 126 intonation  78 intraference (see paradigmatic change) irrealis (see modality and mood) L ‘lest’ (see avertive clause) linearization, linear structure  10f., 108, 110ff., 131, 141f., 145, 155, 168, 193, 200 locative  90f., 93f., 177, 189 M macro-area  7, 22ff., 31f., 40, 81f., 90f., 114, 130, 154, 159, 163, 186 macrofunctionality (see multifunctionality) manner (adverbials)  84, 167 masdar (see nominalization) medial verb/clause  77, 149, 151 meta-analysis (in language change)  99, 101, 180 metaphor (conceptual)  12, 97ff., 100f., 179, 184, 189, 200 primary metaphor  98 metonymy (conceptual)  12, 98f., 101f., 179f., 184, 189, 190, 200 modality and mood  44ff. modal orientation and marking (general)  30, 49, 170, 172, 175

228 A Typology of Purpose Clauses modality and mood (cont.) deontic modality  2, 163 desiderative (optative)  45ff., 70, 75, 161f., 172, 190f., 199 epistemic modality  186f., 201 hypotheticality (irrealis)  31, 44ff., 70, 72, 75, 152, 159, 161, 163, 172, 199 potential mood  45f., 75, 152, 159 subjunctive  37, 44, 45f., 58, 74, 105, 153, 181 motion-cum-purpose construction  58f., 95ff., 99, 102, 181ff., 197, 200f. motion verbs  58f., 69, 94f., 108, 118, 138f., 142ff., 179f., 183f., 188, 198, 201 multidimensional scaling (see statistical analysis) multifunctionality  11, 12, 20, 41, 74, 83, 85, 90, 93, 97, 99, 105ff., 138, 152, 163, 171, 177, 186ff.

Correspondence Hypothesis (PGCH)  14ff., 124, 126, 136f. persistence (in grammatical­ ization)  180, 183 person marking (see indexation) perspective  64f., 135 polysemy (see multifunctionality) possessive marking  31, 68 potential (see modality and mood) predictability (see frequency effects) presupposition  46, 123ff., 168, 200 (on-line) processing  10f., 64, 71, 104f., 110, 113, 116, 117, 119ff., 190, 200 constituent recognition domain  119ff., 131, 200 construction and attachment  71, 119ff., 124

N negation  46, 130f., 143, 182, 198 negative purpose (see avertive clause) nominalization (including gerund, masdar)  36, 43, 57, 59, 65, 67, 68, 72f., 75f., 88f., 94ff., 111, 165, 166f., 169, 190, 199 non-finite (see deranking)

Q quotative construction  72, 76f., 109, 163f., 192, 197, 201

O object-gap purpose clause (see argument structure) optative (see modality and mood) Oxford English Dictionary (OED)  130, 174, 176 P Pacific Rim  41, 57, 151, 202 paradigmatic change (extension, intraference)  99, 101f., 172, 176, 186, 190, 192, 197, 200 participle  36, 43, 73, 75f., 166 passive  14, 58f., 136 Performance-Grammar-

R raising  52, 97 reanalysis (in grammaticalization)  84, 101f., 138, 175, 179, 183, 192, 194f., 197, 199, 201 reason clause  47, 177f. (but see also causal clause) recipient  88ff., 187ff., 197 referential density (see argument structure) reinforcement (renewal)  174, 195 relative clause, relativization  1, 2, 51, 127, 148, 157, 165ff., 175f., 194f., 197, 200 result clause  30, 45, 83, 87, 104, 106, 109, 126, 151ff., 157, 177f., 195, 197, 199.  200, 202 rule generalization  99 S sampling  20ff., 34, 136, 203

exhaustive sampling  33 probability sampling  24 random sampling  33 variety sample  24 semantic ‘bleaching’  173, 182, 195 semantic map  13, 91ff., 105 semantic roles (see argument structure) sequentiality  76 serial verb  35, 38, 103f., 154, 183, 196, 198, 201 speech act  48, 126, 145 statistical analysis association plot  55f., 133 binomial test  38, 51, 66 collostructional analysis  139ff. co-varying collexeme analysis  143 cluster analysis  203 effect size  34 Cohen’s g  38, 51, 66 Cramer’s V  34, 40, 55, 79, 80, 86, 114, 122, 125, 132, 133 Phi  34, 49 Fisher(-Yates) Exact test  33, 49, 125, 140ff., 189 logistic regression modelling  8, 117 Monte-Carlo simulation  34 mosaic plot  133f. multidimensional scaling  8, 176, 203 non-parametric significance tests  33 randomized Chi-squared analysis  33, 40, 55, 64, 79, 80, 86, 114, 122, 132, 133 residual analysis  55f. spine plot  48f., 79, 80, 114, 115, 123, 132 subject (see argument structure) subjunctive (see modality and mood) subordination (theoretical concerns)  16, 20, 27, 31f., 36f., 48, 64, 77, 80, 105, 113, 119f., 127, 148ff., 162, 166, 175, 186, 195 Supine  75, 161, 181

suprasegmental marking  73, 77f. switch-reference  193 T temporal clause  46f., 50, 52, 103f., 106, 108, 119, 125f., 151f., 153, 155, 160, 176, 177, 196 tense  42ff., 48 determined time reference  42, 49, 70, 156 future time/tense  31, 41, 43, 75, 105, 179ff., 190f., 197f., 201

Index of subjects 229 independent time reference  36 relative tense  36 time sphere  36 topic (see information structure) U Universal Grammar  6f., 9 universals (of language)  5ff. usage-based approach  1, 9, 14ff., 101f., 134, 144, 203 (but see also PerformanceGrammar-Correspondence Hypothesis)

V vagueness  105f. W The World Atlas of Language Structures (WALS)  7, 24, 39 word order (see constituent order) Z Zipf ’s Law (see frequency effects)

Typological Studies in Language A complete list of titles in this series can be found on the publishers’ website, www.benjamins.com 88 Schmidtke-Bode, Karsten: A Typology of Purpose Clauses. 2009. xii, 228 pp. 87 Cyffer, Norbert, Erwin Ebermann and Georg Ziegelmeyer (eds.): Negation Patterns in West African Languages and Beyond. 2009. vi, 368 pp. 86 Mahieu, Marc-Antoine and Nicole Tersis (eds.): Variations on Polysynthesis. The Eskaleut languages. 2009. ix, 312 pp. 85 Givón, T. and Masayoshi Shibatani (eds.): Syntactic Complexity. Diachrony, acquisition, neurocognition, evolution. 2009. vi, 553 pp. 84 Newman, John (ed.): The Linguistics of Eating and Drinking. 2009. xii, 280 pp. 83 Corrigan, Roberta, Edith A. Moravcsik, Hamid Ouali and Kathleen M. Wheatley (eds.): Formulaic Language. Volume 2. Acquisition, loss, psychological reality, and functional explanations. 2009. xxiv, 361 pp. 82 Corrigan, Roberta, Edith A. Moravcsik, Hamid Ouali and Kathleen M. Wheatley (eds.): Formulaic Language. Volume 1. Distribution and historical change. 2009. xxiv, 315 pp. 81 Corbett, Greville G. and Michael Noonan (eds.): Case and Grammatical Relations. Studies in honor of Bernard Comrie. 2008. ix, 290 pp. 80 Laury, Ritva (ed.): Crosslinguistic Studies of Clause Combining. The multifunctionality of conjunctions. 2008. xiv, 253 pp. 79 Abraham, Werner and Elisabeth Leiss (eds.): Modality–Aspect Interfaces. Implications and typological solutions. 2008. xxiv, 422 pp. 78 Harrison, K. David, David S. Rood and Arienne Dwyer (eds.): Lessons from Documented Endangered Languages. 2008. vi, 375 pp. 77 Seoane, Elena and María José López-Couso (eds.): Theoretical and Empirical Issues in Grammaticalization. In collaboration with Teresa Fanego. 2008. x, 367 pp. 76 López-Couso, María José and Elena Seoane (eds.): Rethinking Grammaticalization. New perspectives. In collaboration with Teresa Fanego. 2008. x, 355 pp. 75 Frajzyngier, Zygmunt and Erin Shay (eds.): Interaction of Morphology and Syntax. Case studies in Afroasiatic. 2008. v, 234 pp. 74 Kurzon, Dennis and Silvia Adler (eds.): Adpositions. Pragmatic, semantic and syntactic perspectives. 2008. viii, 307 pp. 73 Ansaldo, Umberto, Stephen Matthews and Lisa Lim (eds.): Deconstructing Creole. 2007. xii, 292 pp. 72 Næss, Åshild: Prototypical Transitivity. 2007. x, 240 pp. 71 Nedjalkov, Vladimir P. (ed.): Reciprocal Constructions. With the assistance of Emma Š. Geniušienė and Zlatka Guentchéva. 2007. xxiii, 2219 pp. (5 vols.). 70 Zúñiga, Fernando: Deixis and Alignment. Inverse systems in indigenous languages of the Americas. 2006. xii, 309 pp. 69 Aranovich, Raúl (ed.): Split Auxiliary Systems. A cross-linguistic perspective. 2007. vii, 277 pp. 68 Abraham, Werner and Larisa Leisiö (eds.): Passivization and Typology. Form and function. 2006. x, 553 pp. 67 Veselinova, Ljuba N.: Suppletion in Verb Paradigms. Bits and pieces of the puzzle. 2006. xviii, 236 pp. 66 Hickmann, Maya and Stéphane Robert (eds.): Space in Languages. Linguistic Systems and Cognitive Categories. 2006. x, 362 pp. 65 Tsunoda, Tasaku and Taro Kageyama (eds.): Voice and Grammatical Relations. In Honor of Masayoshi Shibatani. 2006. xviii, 342 pp. 64 Voeltz, F. K. Erhard (ed.): Studies in African Linguistic Typology. 2006. xiv, 426 pp. 63 Filimonova, Elena (ed.): Clusivity. Typology and case studies of the inclusive–exclusive distinction. 2005. xii, 436 pp. 62 Couper-Kuhlen, Elizabeth and Cecilia E. Ford (eds.): Sound Patterns in Interaction. Crosslinguistic studies from conversation. 2004. viii, 406 pp. 61 Bhaskararao, Peri and Karumuri Venkata Subbarao (eds.): Non-nominative Subjects. Volume 2. 2004. xii, 319 pp.

60 Bhaskararao, Peri and Karumuri Venkata Subbarao (eds.): Non-nominative Subjects. Volume 1. 2004. xii, 325 pp. 59 Fischer, Olga, Muriel Norde and Harry Perridon (eds.): Up and down the Cline – The Nature of Grammaticalization. 2004. viii, 406 pp. 58 Haspelmath, Martin (ed.): Coordinating Constructions. 2004. xcv, 578 pp. 57 Mattissen, Johanna: Dependent-Head Synthesis in Nivkh. A contribution to a typology of polysynthesis. 2003. x, 350 pp. 56 Shay, Erin and Uwe Seibert (eds.): Motion, Direction and Location in Languages. In honor of Zygmunt Frajzyngier. 2003. xvi, 305 pp. 55 Frajzyngier, Zygmunt and Erin Shay: Explaining Language Structure through Systems Interaction. 2003. xviii, 309 pp. 54 Aikhenvald, Alexandra Y. and R.M.W. Dixon (eds.): Studies in Evidentiality. 2003. xiv, 349 pp. 53 Givón, T. and Bertram F. Malle (eds.): The Evolution of Language out of Pre-language. 2002. x, 394 pp. 52 Güldemann, Tom and Manfred von Roncador (eds.): Reported Discourse. A meeting ground for different linguistic domains. 2002. xii, 425 pp. 51 Newman, John (ed.): The Linguistics of Sitting, Standing and Lying. 2002. xii, 409 pp. 50 Feigenbaum, Susanne and Dennis Kurzon (eds.): Prepositions in their Syntactic, Semantic and Pragmatic Context. 2002. vi, 304 pp. 49 Wischer, Ilse and Gabriele Diewald (eds.): New Reflections on Grammaticalization. 2002. xiv, 437 pp. 48 Shibatani, Masayoshi (ed.): The Grammar of Causation and Interpersonal Manipulation. 2002. xviii, 551 pp. 47 Baron, Irène, Michael Herslund and Finn Sørensen (eds.): Dimensions of Possession. 2001. vi, 337 pp. 46 Aikhenvald, Alexandra Y., R.M.W. Dixon and Masayuki Onishi (eds.): Non-canonical Marking of Subjects and Objects. 2001. xii, 364 pp. 45 Bybee, Joan and Paul J. Hopper (eds.): Frequency and the Emergence of Linguistic Structure. 2001. vii, 492 pp. 44 Voeltz, F. K. Erhard and Christa Kilian-Hatz (eds.): Ideophones. 2001. x, 436 pp. 43 Gildea, Spike (ed.): Reconstructing Grammar. Comparative Linguistics and Grammaticalization. 2000. xiv, 269 pp. 42 Diessel, Holger: Demonstratives. Form, function and grammaticalization. 1999. xii, 205 pp. 41 Frajzyngier, Zygmunt and Traci S. Walker-Curl (eds.): Reciprocals. Forms and functions. Volume 2. 2000. xii, 201 pp. 40 Frajzyngier, Zygmunt and Traci S. Walker-Curl (eds.): Reflexives. Forms and functions. Volume 1. 2000. xiv, 286 pp. 39 Payne, Doris L. and Immanuel Barshi (eds.): External Possession. 1999. ix, 573 pp. 38 Siewierska, Anna and Jae Jung Song (eds.): Case, Typology and Grammar. In honor of Barry J. Blake. 1998. 395 pp. 37 Giacalone-Ramat, Anna and Paul J. Hopper (eds.): The Limits of Grammaticalization. 1998. vi, 307 pp. 36 Newman, John (ed.): The Linguistics of Giving. 1998. xv, 373 pp. 35 Givón, T. (ed.): Grammatical Relations. A functionalist perspective. 1997. viii, 350 pp. 34 Givón, T. (ed.): Conversation. Cognitive, communicative and social perspectives. 1997. viii, 302 pp. 33 Fox, Barbara A. (ed.): Studies in Anaphora. 1996. xii, 518 pp. 32 Bybee, Joan and Suzanne Fleischman (eds.): Modality in Grammar and Discourse. 1995. viii, 575 pp. 31 Gernsbacher, Morton Ann and T. Givón (eds.): Coherence in Spontaneous Text. 1995. x, 267 pp. 30 Downing, Pamela A. and Michael Noonan (eds.): Word Order in Discourse. 1995. x, 595 pp. 29 Kahrel (PJK), Peter and René van den Berg (eds.): Typological Studies in Negation. 1994. x, 385 pp. 28 Givón, T. (ed.): Voice and Inversion. 1994. viii, 402 pp. 27 Fox, Barbara A. and Paul J. Hopper (eds.): Voice: Form and Function. 1994. xiii, 377 pp. 26 Lord, Carol: Historical Change in Serial Verb Constructions. 1993. x, 273 pp. 25 Svorou, Soteria: The Grammar of Space. 1994. xiv, 290 pp. 24 Perkins, Revere D.: Deixis, Grammar, and Culture. 1992. x, 245 pp.

23 Kemmer, Suzanne: The Middle Voice. 1993. xii, 300 pp. 22 Payne, Doris L. (ed.): Pragmatics of Word Order Flexibility. 1992. viii, 320 pp. 21 Downing, Pamela A., Susan D. Lima and Michael Noonan (eds.): The Linguistics of Literacy. 1992. xx, 334 pp. 20 Croft, William, Suzanne Kemmer and Keith Denning (eds.): Studies in Typology and Diachrony. Papers presented to Joseph H. Greenberg on his 75th birthday. 1990. xxxiv, 243 pp. 19:2 Traugott, Elizabeth Closs and Bernd Heine (eds.): Approaches to Grammaticalization. Volume II. Types of grammatical markers. 1991. xii, 558 pp. 19:1 Traugott, Elizabeth Closs and Bernd Heine (eds.): Approaches to Grammaticalization. Volume I. Theoretical and methodological issues. 1991. xii, 360 pp. 18 Haiman, John and Sandra A. Thompson (eds.): Clause Combining in Grammar and Discourse. 1988. xiii, 428 pp. 17 Hammond, Michael, Edith A. Moravcsik and Jessica Wirth (eds.): Studies in Syntactic Typology. 1988. xiv, 380 pp. 16 Shibatani, Masayoshi (ed.): Passive and Voice. 1988. xi, 706 pp. 15 Austin, Peter (ed.): Complex Sentence Constructions in Australian Languages. 1988. vii, 289 pp. 14 Hinds, John, Shoichi Iwasaki and Senko K. Maynard (eds.): Perspectives on Topicalization. The case of Japanese WA. 1987. xi, 307 pp. 13 Never published. 12 Nedjalkov, Vladimir P. (ed.): Typology of Resultative Constructions. Translated from the original Russian edition (1983). Translation edited by Bernard Comrie. 1988. xx, 573 pp. 11 Tomlin, Russell S.: Coherence and Grounding in Discourse. Outcome of a Symposium, Eugene, Oregon, June 1984. 1987. viii, 512 pp. 10 Ransom, Evelyn N.: Complementation: its Meaning and Forms. 1986. xii, 226 pp. 9 Bybee, Joan: Morphology. A Study of the Relation between Meaning and Form. 1985. xii, 235 pp. 8 Slobin, Dan I. and Karl Zimmer (eds.): Studies in Turkish Linguistics. 1986. vi, 294 pp. 7 Craig, Colette G. (ed.): Noun Classes and Categorization. Proceedings of a symposium on categorization and noun classification, Eugene, Oregon, October 1983. 1986. vii, 481 pp. 6 Haiman, John (ed.): Iconicity in Syntax. Proceedings of a symposium on iconicity in syntax, Stanford, June 24–26, 1983. 1985. vi, 402 pp. 5 Rutherford, William E. (ed.): Language Universals and Second Language Acquisition. 1984. ix, 264 pp. 4 Chisholm, William, Louis T. Milic and John A.C. Greppin (eds.): Interrogativity. A colloquium on the grammar, typology and pragmatics of questions in seven diverse languages, Cleveland, Ohio, October 5th 1981-May 3rd 1982. 1984. v, 302 pp. 3 Givón, T.: Topic Continuity in Discourse. A quantitative cross-language study. 1983. vi, 492 pp. 2 Haiman, John and Pamela Munro (eds.): Switch Reference and Universal Grammar. Proceedings of a symposium on switch reference and universal grammar, Winnipeg, May 1981. 1983. xv, 337 pp. 1 Hopper, Paul J. (ed.): Tense-Aspect. Between semantics & pragmatics. 1982. x, 350 pp.