New Perspectives in Role and Reference Grammar 1443833886, 9781443833882

Citation preview

New Perspectives in Role and Reference Grammar

New Perspectives in Role and Reference Grammar

Edited by

Wataru Nakamura

New Perspectives in Role and Reference Grammar, Edited by Wataru Nakamura This book first published 2011 Cambridge Scholars Publishing 12 Back Chapman Street, Newcastle upon Tyne, NE6 2XX, UK British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Copyright © 2011 by Wataru Nakamura and contributors All rights for this book reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the copyright owner. ISBN (10): 1-4438-3388-6, ISBN (13): 978-1-4438-3388-2

TABLE OF CONTENTS

Contributors............................................................................................... vii Editor’s Introduction................................................................................... xi Part 1. Inflectional and Derivational Morphology Towards a Realizational Approach to Morphology in RRG ........................ 2 Michael Boutin Case Syncretism in Typological Perspective: An RRG-OT Account......... 35 Wataru Nakamura Meaning Construction and Grammatical Inflection in the Layered Structure of the Irish Word: An RRG Account of Morphological Constructions............................................................................................. 64 Brian Nolan Part 2. Lexicon, Argument Structure, and Construction Romance Anticausatives: A Constructionist RRG Approach .................. 104 Rolf Kailuweit Se Incompatible Predicates in Spanish: An RRG Explanation ................ 134 Carlos González Vergara Three-Participant Events in Serial Verb Constructions and the Syntax-Semantics Interface......................................................... 143 Anna Riccio The Theoretical Importance of Constructional Schemas in RRG............ 168 Elke Diedrichsen

vi

Table of Contents

Part 3. Syntax-Semantics Interface PP Types in RRG: A Top-down Approach to Their Classification........... 200 Sergio Ibáñez Cerda Clause Linkage and Purpose Clauses in Southern Uto-Aztecan Languages................................................................................................ 217 Lilián Guerrero Modality in Taiwan Sign Language......................................................... 246 Chien-hung Lin and Jung-hsing Chang Part 4. Syntax-Information Structure Interface The Left Periphery and Focus Structure in Japanese............................... 266 Mitsuaki Shimojo Topic, Focus, and Word Order in the Acquisition of Spanish.................. 294 Antoinette Hawayek Part 5. Computational Applications of RRG UniArab: RRG Arabic-to-English Machine Translation.......................... 312 Brian Nolan and Yasser Salem Language Index ....................................................................................... 347 Subject Index ........................................................................................... 349

CONTRIBUTORS Michael Boutin Applied Linguistics Department Graduate Institute of Applied Linguistics 7500 W. Camp Wisdom Rd. Dallas, TX 75236 USA [email protected] Jung-hsing Chang Graduate Institute of Linguistics National Chung Cheng University 168 University Road, Min-Hsiung Chia-Yi, Taiwan 62102, R.O.C. [email protected] Elke Diedrichsen Institut für Geistes- und Kulturwissenschaften (IGK) Universitdt Vechta Driverstra_e 22 D-49377 Vechta Germany [email protected] Carlos González Vergara Departamento de Ciencias del Lenguaje Pontificia Universidad CatAslica de Chile Av. VicuAqa Mackenna, 4860. Campus San Joaqumn. Santiago de Chile, 7820436 [email protected] Lilián Guerrero Instituto de Investigaciones Filológicas Universidad Autónoma de México Ciudad Universitaria, 04510, México, D.F. [email protected]

viii

Contributors

Antoinette Hawayek Lingüistic -Departamento de Filosofia Universidad Autonoma Metropolitana San Rafael Atlixco 186 Iztapalapa, Mexico, D.F. 09340 Mexico [email protected] Sergio lbáñez Cerda Instituto de investigaciones Filológicas Universidad Nacional Autónoma de México (UNAM) Circuito Mario de la Cueva S/N, Ciudad de la Investigación en Humanidades, Ciudad Universitaria, Zona Cultural, Delegación Coyoacán, México, D. F. CP: 04510 [email protected] Rolf Kailuweit Romanisches Seminar Albert-Ludwigs-Universität Platz der Universität 3 D-79085 Freiburg im Breisgau Germany [email protected] Chien-hung Lin Graduate Institute of Linguistics National Chung Cheng University 168 University Road, Min-Hsiung Chia-Yi, Taiwan 62102, R.O.C. [email protected] Wataru Nakamura Center for the Advancement of Higher Education and Graduate School of International Cultural Studies Tohoku University 41 Kawauchi Aoba-ku, Sendai, Miyagi 980-8576, Japan [email protected]



New Perspectives in Role and Reference Grammar

Brian Nolan Department of Informatics School of Informatics and Engineering Institute of Technology Blanchardstown Blanchardstown Road North Blanchardstown Dublin 15, Ireland [email protected] Anna Riccio Dipartimento di Studi del Mondo Classico e del Mediterraneo Antico Università degli Studi di Napoli “L'Orientale” Palazzo Corigliano, Piazza S. Domenico Maggiore, 12 80134 Napoli, Italy [email protected] Yasser Salem Department of Informatics School of Informatics and Engineering Institute of Technology Blanchardstown Blanchardstown Road North Blanchardstown Dublin 15, Ireland [email protected] Mitsuaki Shimojo Department of Linguistics University at Buffalo, The State University of New York 609 Baldy Hall Buffalo, NY 14260-1030 USA [email protected]

ix

EDITOR’S INTRODUCTION WATARU NAKAMURA

Role and Reference Grammar [RRG] (Van Valin and LaPolla 1997; Van Valin 2005) is a functionalist framework with a particular emphasis on typological adequacy. RRG started out as a theory of linking and complex sentences (clause linkage) with a direct mapping between semantic and syntactic representations (Foley and Van Valin 1984) and has since developed into a brand of parallel architecture theory (Jackendoff 1997, 2002; cf. Sadock 1991; Bresnan 2000) with a set of linking rules that relates syntax, semantics, and discourse-pragmatics and a principled typology of complex sentences and grammatical relations as its hallmark (see Foley and Van Valin 1984; Van Valin 1990, 1993, 2005, 2008; Ohori 1992; Hasegawa 1996; Van Valin and LaPolla 1997; Bentley 2006; Kailuweit et al. 2008; Guerrero et al. 2009; and Narrog 2009, among others). What is notable about the RRG linking theory is that it allows discourse-pragmatics to play an important role in the linking. The present collection of papers, all of which were presented in the 10th International RRG Conference, held at University of California, Berkeley, in August 2009, represents a continued investigation in a number of important areas of linguistic theory including the linking between syntax and semantics, argument structure, the interaction between the lexicon and construction, the distinction between argument and adjuncts, control in complex sentences, and the interface between syntax and information structure. The papers not only exemplify analytical tools available in RRG, but also present notable advances in the component of morphology and the syntax-morphology and syntax-information structure interface. Essentially, the volume extends the parallel architecture of RRG to morphology, one of the previously neglected areas in RRG, and elaborates on the theory of syntax-semantics-information structure interface. The volume also has a computational dimension, for instance in applying the linking algorithm that relates lexical semantics and syntax to a machine translation system. The book is divided into five sections: inflectional and derivational morphology (Part 2), lexicon, argument structure, and constructions (Part

xii

Editor’s Introduction

 3), syntax-semantics interface (Part 4), syntax-information structure interface (Part 5), and computational applications of RRG (Part 6). Part 2 contains three papers, all of which address the question of what the RRG theory of morphology should include. Two of them, Michael Boutin’s and Wataru Nakamura’s paper, propose two related but distinct views of inflectional morphology. With an illustration from verbal conjugation paradigms in Bonggi (Austronesian), Boutin presents a lexeme-based, inferential-realizational model (e.g. Anderson 1992, Stump 2001), which derives inflectional affixes from applying word-formation rules to a full set of morphosyntactic features characterizing a cell within the paradigm. In contrast, Nakamura argues for a lexical-realizational view of inflectional morphology (e.g. Halle and Marantz 1993), which derives inflectional affixes from an OT-style constraint hierarchy, while allowing independent status to affixes (as well as stems) in the lexicon. Brian Nolan’s paper, “Meaning construction and grammatical inflection in the layered structure of the Irish word”, sketches a constructional approach to the derivational and inflectional morphology in RRG. Specifically, he proposes to extend the layered structure of the clausal and nominal structure (Van Valin and LaPolla 1997; Van Valin 2005) to the word-internal structure. Nolan characterizes a derivational affix as a construction that contains a skeletal structure for an input lexeme, treats both derivation and compounding with an illustration from the Irish derivational morphology, and proposes a division of labor between the constituent projection (derivational morphology) and operator projection (inflectional morphology). Part 3 contains four papers centering on the theory of argument structure and interactions between the lexicon and constructions. RRG has committed itself to establishing the lexical decomposition system whose main ingredients are the control structure and four-way aspectual classification à la Vendler (1967). Both Rolf Kailuweit’s and Carlos González Vergara’s paper focus on Romance anti-causative constructions formed from their transitive counterparts with the help of reflexivization. Kailuweit’s paper purports to be a semantic classification of the Romance anti-causative alternations. Kailuweit builds on Reinhart’s (2002) feature-based classification of transitive verbs and locates the Romance anti-causative constructions on an active-passive continuum (with respect to the role of undergoer) ranging from genuine reflexive to reflexive passive constructions. González Vergara argues that the basic function of Spanish reflexive clitic se is to promote the role of undergoer, while diminishing that of actor. Under this analysis, González Vergara explains why some Spanish verbs



New Perspectives in Role and Reference Grammar

xiii

cannot occur with se. The reason is that they already privilege the role of undergoer or have no macrorole argument in the first place. It is worthy of mention here that Kailuweit and González Vergara presuppose constructional schemata accommodating the set of language-specific generalizations in addition to the universal argument reduction rule that realizes a non-actor argument as the “subject” and reduces the number of core arguments by one. Anna Riccio’s paper, “Three-participant events in serial verb constructions and the syntax-semantics interface”, provides a typological survey of three-participant serial verb constructions and illustrates an RRG account of them. The serial verb constructions under Riccio’s examination overlap with ditransitive constructions, which have attracted a great deal of attention (since they involve an intriguing interaction of semantic roles, grammatical relations, and information structure (see Malchukov et al. 2010 and Dalrymple and Nikolaeva 2011), but Riccio argues that they also instantiate a particular type of clause linkage constructions, i.e. a nuclear or core juncture. Elke Diedrichsen’s paper, “The theoretical importance of constructional schemas in RRG”, capitalizes on the growing role played by constructional schemata for describing argument structure constructions (cf. Goldberg 1995, 2006) and examines the role played by macroroles in the linking. Diedrichsen argues that using constructional schemata beyond the realm of idiosyncratic, language-specific features (cf. Van Valin 2005) allows us to dispense with macroroles, a key notion in the RRG linking theory, and to accommodate the ambiguity and flexibility of constructions. Part 4 contains three papers. Sergio Ibáñez Cerda’s paper, “PP types in RRG”, recasts the familiar distinction between arguments and adjuncts in terms of a combination of semantic and syntactic considerations. Ibáñez Cerda builds on Jolly (1993), the first RRG account of preposition assignments, and proposes to classify prepositional phrases according to whether they encode a semantic argument of a verb or not ([±argument]), whether they belong to a core or not ([±core]), and whether they are predicative (i.e. have their own semantics) or not ([±predicative]). Given these three distinctions, Ibáñez Cerda proposes a logical typology of PP types, with illustrations drawn mainly from Spanish. Lilián Guerrero’s paper, “Clause linkage and purpose clauses in Southern Uto-Aztecan languages”, is an extension of the RRG account of control constructions, according to which aspectual, modal, and desiderative verbs (e.g. hope) involve actor control, while implicative and jussive verbs (e.g. order) involve undergoer control (Foley and Van Valin 1984; cf. Sag and Pollard 1991; Jackendoff and Culicover 2003). Guerrero investigates

xiv

Editor’s Introduction

 control phenomena outside the realm of complement clauses. Specifically, she selects purposive constructions in Southern Uto-Aztecan languages (e.g. Yaqui, Guarijío) as her target and argues that controllers in those constructions are determined by semantic and pragmatic factors. Chien-hung Lin and Jung-hsing Chang’s paper, “Modality in Taiwan Sign Language”, explores the syntactic distribution of modal expressions and its correlation with speakers’ subjectivity in Taiwan Sign Language. What is interesting about their account is that the modal expressions at the clause-final position are accompanied by non-manual features that are meant to convey stronger subjectivity. Part 5 comprises two papers, which target the syntax-information structure interface. Mitsuaki Shimojo’s paper, “The left periphery and focus structure in Japanese”, proposes to extend the scope of the RRG theory of information structure (focus structure; cf. Lambrecht 1994) by incorporating subordinate focus structure, detailed in Erteschik-Shir (2007). He focuses on various uses of the nominative marker -ga and topic marker -wa in Japanese. It is customary in Japanese linguistics to assume that -wa has two uses, topical and contrastive, while -ga has two uses, neutral and exhaustive-listing. Shimojo notes that both -wa and -ga have marked uses and derives their non-canonical uses from focus structure augmented by the subordinate focus structure. Finally, Antoinette Hawayek’s paper, “Topic, focus, and word order in the acquisition of Spanish”, applies the RRG view of focus structure to L1 acquisition of Spanish, with a focus on the acquisition of word order in intransitive and transitive clauses. The present volume ends with a paper with a computational application of RRG. Brian Nolan and Yasser Salem’s paper, “UniArab: RRG Arabic-to-English machine translation”, argues that the RRG linking algorithm is able to serve as a model for sentence comprehension and generation, for instance in machine translation.

References Anderson, Stephen R. 1992. A-Morphous Morphology. Cambridge: Cambridge University Press. Bentley, Delia. 2006. Split Intransitivity in Italian. Oxford: Oxford University Press. Dalrymple, Mary and Irina Nikolaeva. 2011. Objects and Information Structure. Oxford: Oxford University Press. Erteschik-Shir, Nomi. 2007. Information Structure: The Syntax-Discourse Interface. Cambridge: Cambridge University Press.



New Perspectives in Role and Reference Grammar

xv

Foley, William and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. —. 2006. Constructions at Work: The Nature of Generalization in Language. Oxford: Oxford University Press. Guerrero Valenzuela, Lilián, Sergio Ibáñez, and Valeria A. Belloro, eds. 2009. Studies in Role and Reference Grammar. México: The IIFL-UNAM Press. Halle, Morris and Alec Marantz. 1993. Distributed Morphology and the pieces of inflection. In Kenneth Hale and Samuel Jay Keyser, eds., The View from Building 20, 111-176. Cambridge, MA: The MIT Press. Hasegawa, Yoko. 1996. A Study of Japanese Clause Linkage: The Connective TE in Japanese. Stanford, CA: CSLI Publications. Jackendoff, Ray S. 1997. The Architecture of the Language Faculty. Cambridge, MA: The MIT Press. —. 2002. Foundations of Language: Brain, Meaning, Grammar, Evolution. Oxford: Oxford University Press. Jackendoff, Ray S. and Peter W. Culicover. 2003. The semantic basis of control. Language 79.3: 517-556. Jolly, Julia A. 1993. Preposition assignment in English. In Robert D. Van Valin, Jr., ed., Advances in Role and Reference Grammar, 275-310. Amsterdam and Philadelphia: John Benjamins. Kailuweit, Rolf, Björn Wiemer, Eva Staudinger, and Ranko Matasoviü, eds. 2008. New Applications of Role and Reference Grammar. Newcastle: Cambridge Scholars Publishing. Lambrecht, Knud. 1994. Information Structure and Sentence Form. Cambridge: Cambridge University Press. Malchukov, Andrej L., Martin Haspelmath, and Bernard Comrie, eds. 2010. Studies in Ditransitive Constructions: A Comparative Handbook. Berlin and New York: Mouton de Gruyter. Narrog, Heiko. 2009. Modality in Japanese: The Layered Structure of the Clause and Hierarchies of Functional Categories. Amsterdam and Philadelphia: John Benjamins. Ohori, Toshio. 1992. Diachrony in Clause Linkage and Related Issues. Unpublished Ph.D. Dissertation, University of California, Berkeley. Reinhart, Tanya. 2002. The theta system: an overview. Theoretical Linguistics 28.3: 229-290. Sag, Ivan A. and Carl Pollard. 1991. An integrated theory of complement control. Language 67.1: 63-113

xvi

Editor’s Introduction

 Stump, Gregory T. 2001. Inflectional Morphology: A Theory of Paradigm Structure. Cambridge: Cambridge University Press. Van Valin, Robert D. Jr. 1990. Semantic parameters of split intransitivity. Language 66.2: 221-260. —. ed. 1993. Advances in Role and Reference Grammar. Amsterdam and Philadelphia: John Benjamins. —. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press. —. ed. 2008. Investigations of the Syntax-Semantics-Pragmatics Interface. Amsterdam and Philadelphia: John Benjamins. Van Valin, Robert D., Jr., and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca, NY: Cornell University Press.



PART 1. INFLECTIONAL AND DERIVATIONAL MORPHOLOGY

TOWARDS A REALIZATIONAL APPROACH TO MORPHOLOGY IN RRG MICHAEL BOUTIN GRADUATE INSTITUTE OF APPLIED LINGUISTICS

Abstract Role and Reference Grammar has assumed a traditional morphemebased approach to verb morphology in which changes in verb classes are described as a combination of a base with a derivational morpheme which expresses the meaning of the derived class. This paper argues for a process-based approach in which changes in verb classes are explained in terms of changes in features. In the realizational approach to morphology described in this paper, Aktionsart classes are a key morphosyntactic feature of inflectional rules. The fundamental insight of processual approaches to morphology is that morphology is a set of relationships rather than a set of morphemes.

Keywords Realizational morphology, verbs, paradigm

1. Introduction1 The traditional view of the lexicon is that it is a list of the indivisible morphological units, or morphemes, in a language. In this view, the English word dogs consists of two morphemes, the root dog and the suffix -s. According to this view, roots and affixes are treated similarly in the lexicon with both being defined in terms of at least a phonological representation, a syntactic category, and a semantic representation. Role 1

I appreciate the helpful comments which I received from Debbie King on an earlier version of this paper.

Michael Boutin

3

and Reference Grammar [RRG] has inherited this traditional view of the lexicon in which lexical units are morphemes (both words and affixes). According to Van Valin (2005:161), “[I]t is necessary to think of the lexicon as having at least two parts, one the traditional storehouse of words and morphemes, and the second a “workshop” where lexical rules and other lexical processes can create new lexical forms which would not otherwise be stored.” A competing view of the lexicon is that lexical entries are lexemes, rather than morphemes. In this view, the English words dog and dogs are the singular and plural forms/shapes of the same lexeme DOG. The property “PLURAL” is a paradigmatic relationship between forms, not a unit listed in the lexicon (Spencer 1998:124). According to this view, affixes like -s are not lexical entries; instead, affixation is thought of as the result of an operation (Spencer 1998:124). Derived lexemes, like the adjective doggish, are present in the lexicon, but regular inflected forms, like dogs, are not in the lexicon, and neither are affixes. A number of morphologists have argued against lexicalist approaches to morphology in which inflectional affixes are assumed to have the same status as words, and have argued for realizational approaches in which the lexicon consists of lexemes, not morphemes. In realizational approaches to morphology, inflectional morphemes are replaced by rules which relate the form of an inflected word to its morphosyntactic representation (Anderson 1984:190). The primary purpose of this paper is to describe a realizational approach to inflectional morphology within RRG, and to show that an RRG lexicon need not contain inflectional morphemes. Section 2 introduces some basic morphological concepts, while §3 briefly summarizes some of the arguments against morpheme-based approaches to the lexicon. Section 4 provides an overview of semantic representations in an RRG lexicon, while §5 briefly describes syntactic representations in RRG. Section 6 introduces a paradigm-based approach to morphology, and §7 describes the linking between semantic and syntactic representations in RRG. Section 8 shows how a paradigm-based approach to morphology operates within the RRG linking system. Finally, §9 summarizes the implications of these findings for RRG. Most of the data for this paper comes from Bonggi, a Western Austronesian language spoken in the Kudat District of Sabah, Malaysia.

2. Basic Morphological Concepts “A lexeme is a word with a specific sound and a specific meaning. Its shape may vary depending on syntactic context” (Aronoff and Fudeman

4

Towards a Realizational Approach to Morphology in RRG

2005:42). Dog and dogs are two different word-forms of the same lexeme DOG.2 Dog occurs in contexts appropriate for a singular noun, and dogs in contexts appropriate for a plural noun. Lexemes are defined by (at least) three dimensions: phonological representation, syntactic category, and semantic representation (Spencer 2004:71). A lexical entry for DOG might look something like (1), where the syntactic dimension includes subcategory information and the semantic representation specifies the meaning. (1) DOG Phonological representation: Syntactic category: Subcategory: Semantic representation:

/dƥʪ/ N count noun dog (x) animate' (x), domesticated' (x), related-to-wolf' (x), natural-kind' (x)

Morphosyntactic categories are categories which are relevant to both morphology and syntax, including case, number, and gender for nouns, and tense, aspect, and modality for verbs. Each morphosyntactic category is associated with a set of morphosyntactic properties such as singular, plural, nominative, past, perfective, etc. Word-forms are assigned morphosyntactic features such as [Number:SG] and [Number:PL].3 Inflection involves the formation of word-forms from a single lexeme, such as singular dog and plural dogs from the lexeme DOG. The two word-forms dog and dogs realize the morphosyntactic features “singular form of DOG” and “plural form of DOG”. Derivation involves the creation of one lexeme from another. For example, the verb stem DOG2 meaning “to track like a dog” is formed by zero-derivation from the noun DOG1.4 The verb stem DOG2 can be inflected for tense (e.g., dogged) or aspect (e.g., dogging). Classical morphology was concerned with the arrangement of morphemes in a particular order. For example, dogs results from the concatenation of the two morphemes dog and -s. In this item-and2 Lexemes occur in caps, while word-forms occur in italics. See chapter 1 of Matthews (1974) for a detailed discussion of differences between lexemes and word-forms. 3 Morphosyntactic categories occur in bold italics, while morphosyntactic properties occur in small caps. 4 Zero-derivation is a word-formation process which changes the lexical category of a word without changing its phonological shape.

Michael Boutin

5

arrangement view (cf. Hockett 1954), affixes have the same status as words and are listed in the lexicon. This paper takes a word-and-paradigm or realizational approach to inflectional morphology, whereby complex words such as dogs result from the lexeme DOG being assigned the morphosyntactic feature [Number:PL] with the [z] in [dƥʪz] being an exponent of the feature [Number:PL].5

3. Arguments against Morpheme-based Approaches Anderson (1992), Stump (2001), and Spencer (2004) are among the morphologists who have argued for realizational approaches to inflectional morphology in which the lexicon consists of lexemes, not morphemes. This section summarizes some of their arguments. Readers are referred to their papers and references therein for elaboration of the arguments against morpheme-based lexicons. The form dogs consists of the root dog and a suffix -s. In the American Structuralist tradition associated with Bloomfield and Pike, dog and -s are morphemes which are the smallest meaningful components in a word. Under a morpheme-based theory, or lexical theory, dog and -s are both lexical entries. This means that dogs is no different structurally than the compound doghouse. Both affixes and compounds are bound. Furthermore, affixes and compounds cannot be distinguished on the basis of potential allomorphy. Affixes (such as the English plural suffixes) frequently exhibit phonologically-conditioned allomorphy, and Mathiassen (1996:537) provides evidence of allomorphy in Lithuanian compounds. The alternation of the English indefinite article a(n) is evidence that allomorphy is not restricted to affixes or compounds. Neither boundedness nor allomorphy can distinguish affixation from compounding. A morpheme-based approach treats morphemes as a linear string of phonemes which are attached to a base. However, morphosyntactic properties can be realized by suprasegmental features such as tone, stress, and nasalization. For example, Ngambay, a Nilo-Saharan language of Southern Chad, has both lexical and grammatical tone marking subject agreement properties (Ndjerareou et al. 2010). Morphological properties can also be realized by changes in stress pattern (e.g., cóntrast – noun vs. contrást – verb), ablaut (e.g., sing ~ sang ~ sung), and consonant mutation (e.g., house /haus/ – noun vs. to house /hauz/ – verb. 5

Exponents are markers of morphosyntactic features.

6

Towards a Realizational Approach to Morphology in RRG

Word-and-paradigm or realizational approaches to morphology stress the existence of non-concantenative phenomena. The process involves relating a basic form to a derived form by a set of phonological operations. Affixation or concatenation is treated the same as non-concatenative morphology. Other problems in a morpheme-based approach relate to how morphemes contribute to the meaning of words. Consider the Finnish data in (2) in which the lexeme TALO “house” is inflected for number and case. (2) talo talo-t talo-ssa talo-i-ssa talo-lla talo-i-lla

“house” “houses” “in the house” “in the houses” “at the house” “at the houses”

nominative singular nominative plural inessive singular inessive plural adessive singular adessive plural

The Finnish plural suffix -i occurs in all cases except nominative where the plural marker is -t. This means there would have to be two lexical entries meaning “plural”. How does the grammar know which plural marker to select when constructing a word form? In an item-andarrangement or concatenative approach, the morphotactics of the language first has to select the plural suffix and then the case suffix. The only way to get the right form would be to subcategorize the nominative case suffix so that it appears following -t. In a word-and-paradigm or realizational approach, -t is a portmanteau affix simultaneously conveying two features: [Number:PL] and [Case:NOMINATIVE]. Morphosyntactic properties can exhibit extended exponence as illustrated by the Swahili marking of negation in (4) where negation is marked by both h(a)- “NEG” and ku- “NEG.PST”.6 In negative clauses like (4), past tense is marked by ku- “NEG.PST”, whereas past tense is marked by li- “PST” in positive clauses like (3). (3) ø-simba a-li-m-shambulia CL9-lion 3SG.SBJ.AGR-PAST-3SG.OBJ.AGR-attack “The lion attacked the dog.”

m-bwa CL9-dog

(4) ø-simba h-a-ku-m-shambulia m-bwa CL9-lion NEG-3SG.SBJ.AGR-NEG.PST-3SG.OBJ-attack CL9-dog “The lion did not attack the dog.”

6

The negative prefix ha- is realized as [h] before a- “3SG.SBJ.AGR”.

Michael Boutin

7

While non-realizational theories assume that a morphosyntactic property has one exponent, realizational theories do not require that a single property be realized by at most one exponent per word (Stump 2001:4). Like American Structuralism, Distributed Morphology (Halle and Marantz 1993) and much of the work in Optimality Theory (McCarthy and Prince 1998) is morpheme-based. For that matter, most of work on the lexicon in RRG has also been morpheme-based.

4. Semantic Representation in an RRG Lexicon Because Bonggi nouns are not inflected for case, number, or gender, the remainder of this paper deals with verbs which involve both derivational and inflectional morphology.7 The primary mechanism in the RRG approach to semantics is a system of lexical representation involving lexical decomposition. The RRG system of lexical representation is based on the classification of predicates into Aktionsart classes; i.e., classes based on inherent aspectual properties (Van Valin 1993:34). Vendler (1967) devised a universal four-way semantic distinction between: 1) states, 2) accomplishments, 3) achievements, and 4) activities. The distinctive features of the four Aktionsart classes are shown in Table 1. Table 1: Distinctive features of basic Aktionsart classes State +static -telic -punctual

Accomplishment -static +telic -punctual

Achievement -static +telic +punctual

Activity -static -telic -punctual

These four Aktionsart classes correspond to classes of predicates which are encoded in the morphology of Bonggi. For example, the predicates in (5), (6), and (7) belong to different Aktionsart classes; however, all three predicates are derived from the root koriƾ “dry”.8

7 Van Valin and LaPolla (1997:184ff.) illustrate how the semantics of nominals described in Pustejovsky (1995) can be integrated within Role and Reference Grammar. 8 The Bonggi data is taken from unpublished texts and an unpublished dictionary. Bonggi has seventeen consonants /p t k b d ʪ ٧ s dٕ m n Є ƾ l ӫ y w/ and five

vowels /i u e o a/. The symbol g is used for /ʪ/ and r is used for flap /ӫ

8

Towards a Realizational Approach to Morphology in RRG

(5) Piasu ku ƾ-koriƾ.9 coconut 1SG.GEN ATTR.ST-dry “My coconut is dry.”

/m-/ + /koriƾ/ “ATTR.ST” “dry”

(6) Piasu ku koriƾ. coconut 1SG.GEN dry “My coconut is drying.”

/-m-/ + “ACL”

/koriƾ/ “dry”

(7) Sia ƾ-oriƾ piasu ku. 3SG.NOM ISA.AV-dry coconut 1SG.GEN “He is drying my coconut.”

/ƾ-/ + “ISA.AV”

/koriƾ/ “dry”

Example (5) illustrates an attributive stative predicate.10 States are static situations with no activity. Attributive states have the morphosyntactic feature [Akclass:ATTR.ST] which is realized morphologically as a prefix m-. As seen in Table 2, the prefix m- has several phonologicallyconditioned allomorphs. In (5), the morphosyntactic feature [Akclass:ATTR.ST] is realized as a velar nasal [ƾ] due to nasal assimilation.

9 The abbreviations and glossing conventions used follow the Leipzig Glossing Rules which are available at http://www.eva.mpg.de/lingua/files/morpheme.html. Underlying forms of roots and affixes are shown in phonemic brackets following each example. Infixes are marked by hyphens within phonemic brackets, but separated from their base by angle brackets in examples and glosses. A list of abbreviations follows the conclusion. 10 Bonggi has several subclasses of states.

Michael Boutin

9

Table 2: Sample attributive stative and accomplishment predicates Roots ayad iƾi odom basa٧ buka٧ panas puti٧ dalam doot sega٧ tuug kapal gia ramig

Attributive stative predicates m-ayad ATTR.ST-pretty m-iƾi ATTR.ST-crazy ATTR.ST-black m-odom ATTR.ST-wet m-basa٧ ATTR.ST-open m-buka٧ m-panas m-puti٧ n-dalam n-doot n-sega٧ n-tuug ƾ-kapal mi-gia ma-ramig

ATTR.ST-hot ATTR.ST-white ATTR.ST-deep ATTR.ST-bad ATTR.ST-red ATTR.ST-dry ATTR.ST-thick ATTR.ST-big ATTR.ST-cold

Accomplishment predicates kam-ayad ACL-pretty kim-iƾi ACL-crazy kom-odom ACL-black ACL-wet kam-basa٧ ACL-open kum-buka٧ kam-panas kum-puti٧ dalam doot sega٧ tuug kapal gia ramig

ACL-hot ACL-white ACL-deep ACL-bad ACL-red ACL-dry ACL-thick ACL-big ACL-cold

Example (6) illustrates an accomplishment verb. Accomplishments are non-punctual changes of state. They have the morphosyntactic feature [Akclass:ACL] which is realized morphologically as either a prefix km- or an infix -m-. As seen in Table 2, prefixes occur before vowel-initial roots and roots whose initial consonant is a bilabial (i.e., /b/ and /p/); infixes occur elsewhere. The prefix or infix vowel is epenthetic, being a copy of the initial vowel in the root. Example (7) illustrates an induced state of affairs in which an actor does something resulting in a change of state to an undergoer. Induced state of affairs can occur in actor or undergoer voice. Example (7) is in actor voice. The morphosyntactic features in (7) are [Akclass:ISA, Voice:AV, IF:DECL]. The features [Akclass:ISA, Voice:AV] are realized morphologically as a prefix ƾ-. As seen in Table 3, this prefix has several phonologically-conditioned allomorphs. 11 In (7), the morphosyntactic features [Akclass:ISA, Voice:AV] are realized as a velar nasal [ƾ] as a result of the coalescence of the prefix ƾ- “ISA.AV” with the initial consonant of the root koriƾ “dry”.

11 The prefix vowels in Table 3 are epenthetic, being a copy of the initial vowel in the root.

10

Towards a Realizational Approach to Morphology in RRG

Table 3: Induced states of affairs in actor voice Root ala elu bereit binasa pali٧

pesa٧ guab kakas kotop loput lomos sekat tedak tutuƾ

Stem ƾ-ala ƾ-elu m-ereit m-inasa m-ali٧

m-esa٧ ƾu-guab ƾ-akas ƾ-otop ƾo-loput ƾo-lomos n-ekat n-edak n-utuƾ

Gloss “defeat someone” “get someone drunk” “tear something” “break something” “burn someone” “break something” “split something open” “uncover something” “sever something” “snap (rope/chain)” “suffocate something” “uproot something” “puncture something” “burn something”

The verbs in (8), (9), and (10) are derived from the root dabu٧ “fall”. Example (8) illustrates an activity verb, (9) illustrates an achievement verb, and (10) illustrates an induced state of affairs in actor voice. (8)

Dolok kaa٧ na dabu٧. rain near now fall “Rain is about to fall.”

/-m-/ + /dabu٧/ “ACY” “fall”

(9)

Sia n-dabu٧. 3SG.NOM PFV-fall “She/he fell.”

/n-/ + /dabu٧/ “PFV” “fall”

(10) Sia i-ƾa-dabu٧ sou. 3SG.NOM PFV-ISA.AV-fall anchor “He dropped the anchor.”

/i-/ + /ƾ-/ + /dabu٧/ “PFV” “ISA.AV” “fall”

Activities are dynamic situations which are inherently temporally unbounded. They have the morphosyntactic feature [Akclass:ACY] which is realized morphologically as either a prefix m- or an infix -m- when the illocutionary force is non-imperative (i.e., declarative or interrogative). Table 4 lists some motion activity verbs whose illocutionary force is non-

Michael Boutin

11

imperative.12 As seen in Table 4, prefixes occur before vowel-initial roots and roots whose initial consonant is a bilabial; infixes occur elsewhere. The infix vowel is epenthetic, being a copy of the initial vowel in the root. In (8), the morphosyntactic feature [Akclass:ACY] is realized as an infix because the root begins with /d/. The infix vowel in (8) is a copy of the root-initial vowel. Table 4: Motion activity verbs with non-imperative illocutionary force Root ilaƾ upug uli٧ usag panu piit dua٧ loƾi luas selekei suak tindiaƾ tulak

Stem m-ilaƾ m-upug m-uli٧ m-usag m-panu m-piit dua٧ loƾi luas selekei suak tindiaƾ tulak

Gloss “ACY-lie.down” “ACY-sit.down” “ACY-return.home” “ACY-stand.up” “ACY-walk; go” “ACY-send” “ACY-descend” “ACY-swim” “ACY-exit” “ACY-ascend” “ACY-enter” “ACY-turn.at.intersection” “ACY-depart”

Achievements are puntual changes of state. They have the morphosyntactic feature [Akclass:ACH]; however, this feature is not morphologically marked. The prefix n- in (9) marks the morphosyntactic feature [Aspect:PERFECTIVE]. In RRG, verbs and other predicates are analyzed in terms of a lexical decomposition system in which state and activity predicates are basic and the other classes are derived from them (Van Valin 2005:42). The decompositional representations of predicates are called logical structures. Logical structures express the relationship between a predicate and its arguments. Table 5 shows the lexical representations for different types of Aktionsart classes (cf. Van Valin 2005:45).13

12

The imperative form of these verbs is the bare root. Operators like BECOME are presented in small caps, constants like predicate' are presented in boldface followed by a prime, and variables like x are presented in normal typeface.

13

12

Towards a Realizational Approach to Morphology in RRG

Table 5: Lexical representations for Aktionsart classes Aktionsart class State Accomplishment Achievement Activity Active Accomplishment Causative

Logical Structure predicate' (x) or (x, y) BECOME predicate' (x) or (x, y) INGR predicate' (x) or (x, y) do' (x, [predicate' (x) or (x, y)]) do' (x, [predicate1' (x, (y))]) & INGR predicate2' (z, x) or (y) Į CAUSE ȕ, where Į and ȕ are logical structures of any type

The generic logical structure [LS] for attributive stative predicates is shown in (11). The logical structure for the attributive stative predicate ƾkoriƾ “ATTR.ST-dry” in (5) is shown in (12), and the semantic representation [SR] for the clause in (5) is shown in (13).14 (11) Generic LS for attributive stative predicates: be' (x, [predicate']) (12) LS for ƾ-koriƾ “ATTR.ST-dry”:

be' (x, [dry'])

(13) SR for (5):

be' (piasu 1SG, [dry'])

The generic logical structure for accomplishment verbs with an underlying attributive stative predicate is shown in (14). The logical structure for the accomplishment verb koriƾ “dry” in (6) is shown in (15), and the semantic representation for the clause in (6) is shown in (16). (14) Generic LS for accomplishment verb with underlying attributive stative: BECOME be' (x, [predicate']) (15) LS for koriƾ “dry”:

BECOME

be' (x, [dry'])

(16) SR for (6):

BECOME

be' (piasu 1SG, [dry'])

Predicates which belong to the same class share the same generic logical structure. For example, all the attributive stative predicates in Table 2 have 14 Possessive NPs like piasu ku “my coconut” in (5) involve a possessive predication within the NP which would be captured in a more detailed semantic representation than (13). This paper ignores information focus structure. A richer semantic representation would include the activation status of arguments (Van Valin 2005:79-80).

Michael Boutin

13

the generic logical structure in (11), and all the accomplishment predicates in Table 2 have the generic logical structure in (14). The difference in meaning between predicates in the same class is captured by replacing the predicate' in the logical structure with a specific constant such as dry' in (12) and (15).15 As stated in §2, lexemes are defined by three dimensions: phonological representation, syntactic category, and semantic representation. The wordform ƾ-koriƾ “ATTR.ST-dry” is derived from the adjective root koriƾ “dry”. The lexeme ƽKORIƽ contains the information in (17) in its lexical entry (cf. the lexical entry for DOG in (1)). The semantic representation in (17) shows the logical structure of the predicate. (17) ƽKORIƽ Phonological representation: Syntactic category: Semantic representation:

/ƾkoriƾ/ Adj be' (x, [dry'])

Levin and Rappaport Hovav (1998:258) point out that lexical representations can be related in two ways. First, they can share the same lexical semantic template, but have a different constant. For example, the accomplishment verbs koriƾ “dry” in (15) and kam-ayad “ACL-pretty” in (18) share the same lexical semantic template, but have different constants, dry' and pretty'. The shared lexical semantic template is the generic logical structure for accomplishment verbs with an underlying attributive stative predicate shown in (14). All of the accomplishment verbs in Table 2 share the lexical semantic template in (14). (18) LS kam-ayad “ACL-pretty”:

BECOME

be' (x, [pretty'])

Second, lexical representations can contain the same constant, but have a different lexical semantic template. For example, ƾ-koriƾ “ATTR.ST-dry” in (12) and koriƾ “dry” in (15) share the same constant dry', but have a different lexical semantic template. The logical structure for the accomplishment verb koriƾ “dry” includes the operator BECOME which is not part of the lexical semantic template of stative predicates (cf. Table 5). Van Valin (2005:47ff.) argues that related verbs can be derived by lexical rules. For further discussion of the Aktionsart classes listed in Table 5, including tests for determining Aktionsart classes, readers are 15

Constants are English words since English is the semantic metalanguage used.

14

Towards a Realizational Approach to Morphology in RRG

referred to chapter 2 of Van Valin (2005). For detailed descriptions of other Aktionsart classes in Bonggi, see Boutin (2007) and Boutin (2009).

5. Syntactic Representation in RRG Section 4 provided an overview of semantic representations in an RRG lexicon, whereas this section briefly describes syntactic representations in RRG.

5.1. Constituent Projection “Every language makes a distinction between predicates and arguments, and every language distinguishes between NPs/PPs which are arguments of the predicate and those which are adjuncts” (Van Valin and LaPolla 1997:27). These distinctions in clause structure are illustrated in Figure 1.

Figure 1: Universal oppositions underlying clause structure

The primary syntactic constituents of a clause are the nucleus, which contains the predicate, the core, which includes the predicate and its arguments, and the periphery, which consists of non-arguments (adjuncts) of the predicate. This layered structure of the clause is illustrated in Figure 2.

Figure 2: Layered structure of the clause

RRG only recognizes one level of syntactic representation, which is the surface syntax. The morphosyntactic representation represents the actual form of the sentence, including the linear sequence of its constituent

Michael Boutin

15

elements and their morphological properties. 16 This is illustrated by the tree in Figure 3, which shows the constituent projection for (7), repeated here as (19).

Figure 3: Constituent projection for (19)

(19) Sia ƾ-oriƾ piasu ku. 3SG.NOM ISA.AV-dry coconut 1SG.GEN “He is drying my coconut.”

/ƾ-/ + /koriƾ/ “ISA.AV” “dry”

5.2. Operator Projection Each of the major layers of the clause (nucleus, core, and clause) is modified by one or more operators which include grammatical categories such as tense, aspect, modality, and illocutionary force. As shown in Figure 4, operators are represented in a distinct projection of the clause from predicates and arguments. 17 Aspect is a nuclear operator and illocutionary force is a clause-level operator.

16

According to Van Valin (2009:4), representation of the internal structure of words (or morphological representation) is part of the syntactic representation. However, the structure of words is very different from the structure of phrases and clauses. 17 Readers are referred to Van Valin and LaPolla (1997) and Van Valin (2005) for a more detailed description of syntactic representations in RRG.

16

Towards a Realizational Approach to Morphology in RRG

V | Aspect ĺ NUCLEUS | CORE | CLAUSE ĸ | SENTENCE

IF

Figure 4: Constituent and operator projections for (19)

RRG recognizes only one level of syntactic representation which is directly linked with the semantic representation of the sentence (Van Valin and LaPolla 1997:21). The general structure of an RRG-based theory of grammar is presented in Figure 5. SYNTACTIC REPRESENTATION Ĺ Linking algorithm Ļ SEMANTIC REPRESENTATION Figure 5: General structure of RRG (Van Valin and LaPolla 1997:21)

The heart of the grammar in RRG is the linking between semantic representations like (20) and syntactic representations like Figure 4 (Van Valin and LaPolla 1997:645). Before describing this linking system in §7, section 6 introduces a realizational approach to morphology. (20) SR for (19):

do' (3SG, Ø) CAUSE [BECOME dry' (piasu 1SG)]

Michael Boutin

17

6. Paradigm-based Approach to Morphology A morphological paradigm is a set of morphological contrasts that a given class of lexemes can make. Morphological paradigms are defined in terms of morphological categories (e.g., Number), their permissible values (e.g., PLURAL), and any co-occurrence restrictions. This section introduces a paradigm-based approach to Bonggi morphology in which morphological rules are formulated as operations on morphological expressions. Consider the subparadigm of induced states of affairs in (21)-(26). (21) ƾ-oriƾ

“ISA.AV-dry”

actor voice, imperfective aspect, non-imperative illocutionary force

(22) i-ƾ-oriƾ

“PFV-ISA.AV.dry” actor voice, perfective aspect, non-imperative illocutionary force

(23) po-ƾ-oriƾ “IMP-ISA.AV.dry” actor voice, imperative illocutionary force (24) kiriƾ-in

“dry-ISA.UV”

(25) koriƾ “dry” (26) kiriƾ-a٧

undergoer voice, imperfective aspect, non-imperative undergoer voice, perfective aspect, non-imperative

“dry-ISA.UV.IMP” undergoer voice, imperative illocutionary force

The word-forms in (21)-(26) are representative of simple causative verbs in Table 5. These verbs are described in §4 as induced state of affairs in which an actor does something resulting in a change of state to an undergoer. All induced states of affairs have a CAUSE operator in their logical structure (e.g., (20)). They are semantically transitive, having both an actor and an undergoer, either of which can be the subject. Examples (7) and (19) illustrate the verb ƾ-oriƾ “ISA.AV-dry” which is an induced state of affairs in actor voice. The actor voice (which occurs when the subject is the actor) has two prefix slots, one for aspect (perfective/ imperfective) and illocutionary force (imperative), and one for voice. The undergoer voice (which occurs when the subject is the undergoer) has a suffix slot for voice when the verb is imperfective, and an infix slot for aspect when the verb is perfective. Table 6 provides a subset of

18

Towards a Realizational Approach to Morphology in RRG

morphosyntactic categories which are associated with Bonggi predicates and a subset of morphosyntactic properties which are possible values for each category. Together, the categories and properties in Table 6 show some morphosyntactic features, such as [Akclass:ACL] and [Asp:PFV]. Table 6: Selected morphosyntactic features of Bonggi predicates Categories Aktionsart class

Voice Aspect Illocutionary force

Properties attributive state accomplishment achievement activity induced states of affairs actor undergoer perfective imperfective imperative non-imperative declarative

Abbreviation ATTR.ST ACL ACH ACY ISA AV UV PFV IPFV IMP NIMP DECL

Each Aktionsart class in Table 5 has a unique lexical representation with a unique meaning; however, a unique morpheme cannot be assigned to each Aktionsart class. All the verbs in (21)-(26) share the same logical structure: do' (x, Ø) CAUSE [BECOME dry' (y)]; however, they do not share the same stem. The three actor voice forms (i.e., (21), (22), and (23)) share a derived stem ƾ-oriƾ “ISA.AV-dry”; however, the three undergoer voice forms in (24), (25), and (26) do not share a derived stem. The choice between actor or undergoer voice is an option in the linking between syntax and semantics. Tense, aspect, modality, and illocutionary force are operators (cf. §5.2). A set of functions are needed to realize the features in Table 6. These functions are realization rules (RRs) like (27). (27) RR{Akclass:ISA, Voice:AV}, V ()

=

Following Stump (2001), the features to be realized and the lexical class that the function refers to are given as subscripts. The function maps a pair consisting of a form X and the complete set of features characterizing the final word form. The output is another form (e.g., a root + affix, or a stem + affix) and the same complete feature set. The variable ı stands for the complete feature set of the word being computed. The realization rule in

Michael Boutin

19

(27) states that induced states of affairs in actor voice are formed by adding ƾ- to a form X. Realization rules apply whenever the set of features which they realize is found as a subset of ı. The rule in (28) states that induced states of affairs in actor voice and perfective aspect are formed by adding i- to a form X. (28) RR{Akclass:ISA, Voice:AV, Asp:PFV}, V ()

=

When both rule (27) and rule (28) apply, (28) applies to the output of (27). Rules occur in distinct, extrinsically-ordered blocks. The ordering is defined by an index as seen in (29) and (30). (29) RRI,{Akclass:ISA, Voice:AV}, V ()

=

(30) RRII,{Akclass:ISA, Voice:AV, Asp:PFV}, V ()

=

Because (29) is in block I, it applies to a root. Rule (30) applies to the stem which is the output of block I rules. The realization rule needed to produce the imperative form in (23) is shown in (31) which is a block II rule.18 (31) RRII,{Akclass:ISA, Voice:AV, IF:IMP}, V ()

=

Because imperatives are always imperfective, imperfective is part of the complete feature set ı in (31). Rule (30) cannot apply to the output of (31) or vice versa, because the two rules belong to the same block. This is expected since the features perfective and imperfective are incompatible. The realization rules needed to produce the undergoer voice forms in (24), (25), and (26) are shown in (32), (33), and (34). (32) RRI,{Akclass:ISA, Voice:UV, IF:NIMP, Asp:IPFV}, V ()

=

(33) RRII,{Akclass:ISA, Voice:UV, Asp:PFV}, V ()

= 19

(34) RRII,{Akclass:ISA, Voice:UV, IF:IMP}, V ()

=

The rule in (32) belongs to block I and applies to roots producing new stems. Because the infix -i- does not co-occur with the undergoer voice 18

The prefix vowel in (23) is epenthetic, being a copy of the first vowel in the stem. 19 The rule in (33) inserts an infix after the initial consonant of consonant-initial roots. The prefix in- occurs before vowel-initial roots.

20

Towards a Realizational Approach to Morphology in RRG

suffix -on,20 the rule in (33) does not apply to the output of the rule in (32). Instead, the rules in block I apply vacuously, then rules (33) and (34) in block II apply to the output of the rules in block I producing forms like (25) and (26). In a realizational approach to morphology, a word’s association with a particular set of morphosyntactic properties licenses the introduction of those properties’ exponents (Stump 2001:2). Morphological rules establish a correspondence between the morphosyntactic properties and phonological forms. The rules replace a list or lexicon of grammatical morphemes (cf. Anderson 1984:158). “The crucial insight behind paradigm-based morphology is that once we have paradigms we don’t need (inflectional) morphemes. Inflected word forms are realizations of cells in paradigms” (Spencer 2004:72).

7. Linking in RRG The RRG linking system works both from semantics to syntax and from syntax to semantics. The linking between semantics and syntax is governed by the Completeness Constraint in (35) (Van Valin and LaPolla 1997:325). (35) Completeness Constraint All of the arguments explicitly specified in the semantic representation of a sentence must be realized syntactically in the sentence, and all of the referring expressions in the syntactic representation of a sentence must be linked to an argument position in a logical structure in the semantic representation of the sentence.

7.1. Linking from Semantics to Syntax The first step in linking from semantics to syntax is to construct the semantic representation of the sentence, based on the logical structure of the predicate (Van Valin 2005:136). Returning to the example in (7) and (19), the semantic representation is shown in (20), repeated here as (36). (36) SR for (19):

do' (3SG, Ø) CAUSE [BECOME dry' (piasu 1SG)]

The semantic representation in (36) shows the argument structure of the verb ƾ-oriƾ “ISA.AV-dry” in (19). Notice that the semantic representation 20

This is a general feature of Philippine-type languages like Bonggi.

Michael Boutin

21

makes no reference to semantic roles or grammatical relations (cf. Kroeger 2005:67-69). RRG uses two semantic macroroles, actor and undergoer. Actor refers to the entity which instigates, controls, or effects the action expressed by the verb. Undergoer indicates the entity affected by the action or state expressed by the verb (Walton 1986:45). The second step in linking from semantics to syntax is to determine the actor and undergoer assignments (Van Valin 2005:136). The information that is necessary for mapping from semantic arguments to syntactic arguments can be read off the semantic representations. The relationship between macroroles and argument positions in logical structures is captured in the Actor-Undergoer Hierarchy in (37) (Van Valin and LaPolla 1997:146). This double hierarchy states that the argument position that is leftmost on the cline will be the actor and the argument position that is rightmost will be the undergoer. This is the unmarked situation; marked assignments to undergoer are possible. (37) ACTOR-UNDERGOER HIERARCHY ACTOR

UNDERGOER

ņņņņņņņņņņņņņņņņņņņ>

Each verb class has a unique lexical representation (cf. Table 5). By including information about tense, aspect, and illocutionary force in semantic representations, all of the morphosyntactic features in Table 6 (with the exception of voice) can be determined from semantic representations. As stated in §7.1, the choice between actor or undergoer voice is an option in the linking from semantics to syntax. The inflectional features (e.g., aspect and illocutionary force) included in enriched semantic representations like (42) match the inflectional morphosyntactic features in realization rules. In other words, the realization rules in §6 are part of the system of linking from semantics to syntax. They produce the exponents of the feature sets found in each rule. Stump (2001) is concerned with inflectional morphology; yet, the features in Table 6 are a mixture of derivational and inflectional features. According to Stump (1998:13), “The structure of paradigms in a given language is determined by the inventory of morphosyntactic properties available in that language.” Stump’s morphosyntactic properties of verbs include the properties associated with the categories voice, aspect, and illocutionary force in Table 6, but exclude the properties associated with Aktionsart classes (cf. Stump 1998:28). The Aktionsart class properties (e.g., state, accomplishment, achievement, activity, etc.) belong to what Stump refers to as lexicosemantic properties. According to Stump (1998:2), lexicosemantic properties like “stative” determine the semantic composition. Stump and Spencer make a clear distinction between inflection and derivation. Derivation encodes lexicosemantic relations within the lexicon,

26

Towards a Realizational Approach to Morphology in RRG

while inflection encodes phrase-level properties and relations (Stump 1998:22). Like Stump, RRG views differences in Aktionsart classes as lexical. The different verb classes in Table 5 have different logical structures and different meanings. Differences in morphology which correspond to differences in Aktionsart classes are clearly derivational. On the other hand, operators like tense, aspect, modality, and illocutionary force are clearly inflectional (see §5.2). Differences in voice, which are accounted for by the principles in (37) and (38), result from different linking choices outside of the lexicon. Linguists widely assume that derivational morphological processes within the lexicon take place before inflectional processes. Given this view and an item-and-arrangement approach to morphology, one would assume that verbs which belong to the same class share the same stem. Consider the Bonggi stems in Table 7 which are inflected for perfective aspect. Table 7 shows that Bonggi has six distinct forms for marking perfective aspect: three prefixes i-, in-, and n- as seen in rows (a-c); two infixes and as seen in rows (e and j); and ablaut as seen in row (p). With the exception of ablaut, which is a suppletive form, both the phonological shape (/i/, /in/, or /n/) and the position (prefix or infix) are predictable. The shape of the inflected forms is conditioned by the phonology; however, the position of the inflected forms is conditioned by lexical semantics (i.e., Aktionsart class). Perfective aspect is always marked by a prefix for achievements (e.g., rows c and f in Table 7) and actor voice induced states of affairs (e.g., rows a, d, g, and i). Infixes can only occur with undergoer voice induced states of affairs (e.g., rows e, h, and j), activity verbs (e.g., rows n and o), and accomplishment verbs (rows k, l, and m). The position of the perfective aspect marker provides information about the possible verb class. In other words, part of the functional yield of the perfective marker is carried by the templatic position, rather than exclusively by the segmental make-up.

Michael Boutin

27

Table 7: Perfective allomorphs Root a.

ala

Inflected stem i-ƾala

b.

ala

in-ala

c. d.

tutuƾ tutuƾ

n-tutuƾ i-nutuƾ

e.

tutuƾ

tutuƾ

f.

pesa٧

i-pesa٧

h.

pesa٧

pesa٧

i.

titik

i-nitik

j.

titik

bereit

k.

odom

kmodom

l.

panas

kmpanas

m.

tikuƾ

tmikuƾ

n. o.

upug tindiaƾ

mupug tmindiaƾ

p.

mati

meti

g.

pesa٧

i-mesa٧

Gloss “defeat someone” “defeat someone” “burnt” “burnt something” “burnt something” “broken” “broke something” “broke something” “beat instrument” “tear something” “became black” “became hot” “became crooked” “sat down” “turn at intersection” “died”

Class & voice ISA.AV

Perfective

ISA.UV

in-

ACH ISA.AV

ni-

ISA.UV

ACH

i-

ISA.AV

i-

ISA.UV

ISA.AV

i-

ISA.UV

ACL

ACL

ACL

ACY ACY



ACH

e

i-

The two rules in (30) and (33) interact with a set of phonological processes to produce the perfective markers for the induced states of affairs in Table 7 (i.e., rows a, b, d, e, g, h, i, and j). Rules (30) and (33) do not produce the perfective markers for other verb classes since they only apply to verbs with the feature [Akclass:ISA]. Other rules, such as the one for achievement verbs in (43), are required to produce the perfective forms for other verb classes.

28

Towards a Realizational Approach to Morphology in RRG

(43) RRII,{Akclass:ACH, Asp:PFV}, V ()

=

Rule (43) interacts with a set of phonological processes to produce the perfective prefixes for the achievement verbs in rows c and f of Table 7. Perfective achievement verbs are marked by [n] if the root begins with an alveolar, otherwise they are marked by [i]. Rule (43) does not apply to the ablaut form meti “died” in row p because an ablaut rule which belongs to the same morphological block is more narrowly applicable than rule (43). Ablaut overrides rule (43) in accordance with the PƗϿini principle. “Choices among rules belonging to the same block are determined by a single universal principle (PƗϿini’s principle), according to which the narrowest applicable rule always overrides other applicable members of the same block” (Stump 2001:33). A single affix frequently serves as a cumulative exponent. For example, the infix in row e of Table 7 serves simultaneously as an exponent of the morphosyntactic features [Akclass:ISA], [Voice:UV], and [Asp:PFV]. Although the perfective allomorphs in Table 7 are dependent upon the verb class, verbs that belong to the same class are not necessarily inflected the same. Specifically, induced states of affairs are treated differently depending on whether they are actor voice or undergoer voice. Furthermore, as pointed out in §6, the three undergoer voice forms in (24), (25), and (26) do not share a derived stem. This is not a problem in a paradigm-based approach.

8.2.

Linking from Syntax to Semantics

Because linking from syntax to semantics involves interpreting overt morphosyntactic forms, it is more difficult than linking from semantics to syntax (Van Valin 2005). However, Bonggi speakers can predict much of the semantics from the morphological shape of the predicate. For example, given a hypothetical root whose shape is /root/, listeners can usually predict the following from the surface morphology of the predicate: Aktionsart class, voice, aspect, and whether or not the illocutionary force is imperative as seen in Table 8.

Michael Boutin

29

Table 8: Predictions given the hypothetical root /root/ Morphological Shape mo-root i-root root root root ƾo-root i-ƾo-root po-ƾo-root root root-on root-a٧ root-an

Aktionsart class ATTR.ST

Voice

Aspect

ISA ISA ISA ISA ISA ISA

ACT ACT ACT UND UND UND

imperfective perfective imperfective perfective imperfective imperfective perfective imperfective perfective imperfective imperfective

ISA

marked

imperfective

ACH ACH ACL/ACY ACL/ACY

activity

IF

imperative

imperative

imperative

UND

root-ei

ISA

marked

imperfective

imperative

UND

Although accomplishment verbs and activity verbs usually cannot be distinguished on the basis of morphological shape alone, if the subject is an undergoer then it is an accomplishment, if the subject is an actor it is an activity or an active accomplishment. Active accomplishments are activities with an endpoint added (cf. Pavey 2010:114). Although Aktionsart classes are encoded in the morphology of Bonggi, Aktionsart classes are not strictly determined by verb morphology. Whether a verbal predicate or sentence is telic is determined by the lexical semantics of the verb, obligatory and optional arguments, adjuncts, the discourse-level linguistic context, and general world knowledge associated with the meaning of sentences (Filip 1999:122).

9. Conclusion This paper has argued for a realizational approach to inflectional morphology within RRG in which inflectional morphemes are replaced by rules which relate the form of an inflected word to its morphosyntactic representation. Previous work on a realizational approach to morphology in RRG includes Everett (2002) and Martín Arista (2008).

30

Towards a Realizational Approach to Morphology in RRG

Van Valin and LaPolla (1997), Van Valin (2005:158), Cortés Rodríguez’s (2006), and Martín Arista (2008:122) claim that derivational affixes which change syntactic category occur in the lexicon. For example, both Cortés Rodríguez (2006:43) and Van Valin and LaPolla (1997:188) explain English agent nominalization by means of the lexical rule in (44). (44)

verb + er Æ [N verb + er] “xi which verbs” ([LS…(xi,…)…]), where “x” is the actor argument in the logical structure.

The rule in (44) is a word formation rule which applies to a verb base to produce a noun. The lexical material in (44) includes both lexemes and affixes. Cortés Rodríguez (2006) argues that derivational affixes are lexical units. According to him, derivational affixes should have a logical structure like lexemes.24 Some of the derivational processes which are described in this paper involve a change in syntactic category. For example, when the logical operator BECOME is added to the stative predicate be' (x, [dry']), the accomplishment verb koriƾ “dry”, meaning BECOME be' (x, [dry']), is derived from an adjective root. This type of derivational process is extremely productive in Bonggi as when the attributive state in (5), the accomplishment in (6), and the induced states of affairs in (7), (21), (22), (23), (24), (25), and (26) are all derived from the adjective root koriƾ “dry”. Van Valin (2005:47) has argued that the relationship between activity verbs and active accomplishments can be derived by lexical rules, and has suggested that other verb classes might also be derived by lexical rule. In a morpheme-based theory, changes in verb classes are described as a combination of a base with a derivational morpheme which expresses the meaning of the derived class. In a process-based approach like the one described here, changes in verb classes are explained in terms of changes in features. In the realizational approach to morphology described in this paper, verb class is a key morphosyntactic feature of inflectional rules (cf. Table 6). 25 The fundamental insight of processual approaches to morphology is that morphology is a set of relationships rather than a set of morphemes. 24

Being a lexicalist theory, traditionally RRG has not made an issue of the inflectional versus derivational distinction. Instead, RRG has presumed some version of the Lexicalist Hypothesis in which inflectional affixes are accounted for in the lexicon and are not sensitive to syntax. 25 RRG can inform Stump’s theory of Paradigm-Function Morphology via careful attention paid to Aktionsart classes in RRG.

Michael Boutin

31

In a morpheme-based approach, morphological rules/operations are defined in terms of morphemes. In a realizational approach, morphological rules/operations are defined in terms of features. From either perspective, the morphological operations involved in verb class changes are lexical; i.e., they occur in the lexicon. Furthermore, in both approaches, information about verb classes is available in the logical structure of verbs and semantic representations of clauses. The analysis of perfective and imperfective aspect in §6 and §8 provides evidence that a realizational approach is superior to a morphemebased approach. In a morpheme-based approach, one would expect to inflect an invariant stem with perfective aspect [Asp:PFV] or imperfective aspect [Asp:IPFV]. However, undergoer voice [Voice:UV] induced state of affairs [Akclass:ISA] do not share an invariant stem. As shown in (32) and (33), -on is a cumulative exponent of the features [Akclass:ISA], [Voice:UV], and [Asp:IPFV], while the infix is a cumulative exponent of the features [Akclass:ISA], [Voice:UV], and [Asp:PFV]. Verb class is a lexical category, aspect is an inflectional category, and the choice of voice takes place during the linking from semantics to syntax. In other words, different morphosyntactic features can be added throughout the linking process. Work by Marial Usón, Faber, and Guest on a semantic metalanguage for RRG is compatible with a realizational approach to morphology (e.g., Marial Usón and Faber (2005), and Marial Usón and Guest (2005)), as is Nolan’s work on a feature-based computational lexicon for RRG (e.g., Nolan 2004).26 This paper has taken a conservative approach to morpheme eradication by not trying to expunge all morphemes from the lexicon. I have simply argued that neither fully inflected words nor inflectional affixes should be included in an RRG lexicon.

26

Nolan’s lexicon includes both lexemes and affixes. Although his feature system accounts for the operator projection suggesting inflectional affixes, it appears to be compatible with a realizational approach to morphology.

32

Towards a Realizational Approach to Morphology in RRG

List of Abbreviations ACL ACT ACY AKCLASS ASP ATTR AV CL IF INGR ISA

accomplishment actor activity Aktionsart class aspect attributive actor voice noun class illocutionary force ingressive induced states of affairs

LS NIMP PSA PFV RR SR ST UND UV

logical structure non-imperative privileged syntactic argument perfective realization rule semantic representation state undergoer undergoer voice.

References Anderson, Stephen R. 1984. On representations in morphology: case marking, agreement and inversion in Georgian. Natural Language and Linguistic Theory 2:157-218. —. 1992. A-Morphous Morphology. Cambridge: Cambridge University Press. Aronoff, Mark, and Kirsten Fudeman. 2005. What is Morphology? Malden, MA: Blackwell. Boutin, Michael E. 2007. Lexical decomposition and locative predicates in Bonggi. In Mark Alves, Paul Sidwell, and David Gil, eds., SEALS VIII Papers from the 8th Annual Meeting of the Southeast Asian Linguistics Society 1998, 25-43. Canberra: Pacific Linguistics, Research School of Pacific and Asian Studies, The Australian National University. Online: http://www.jseals.org/. —. 2009. Influence of lexical semantics on reflexes and allomorphs of * and * in Bonggi. Journal of the Southeast Asian Linguistics Society 1.23-47. Online: http://www.jseals.org/. Cortés Rodríguez, Francisco José. 2006. Derivational morphology in Role and Reference Grammar: A new proposal. Revista Española de Lingüística Aplicada 19:41-66. Everett, Daniel. 2002. Towards an RRG theory of morphology. Lecture delivered at The 2002 International Conference on Role and Reference Grammar, held at the University of La Rioja. Unpublished manuscript. Filip, Hana. 1999. Aspect, Eventuality Types and Nominal Reference. Outstanding Dissertations in Linguistics. New York: Garland.

Michael Boutin

33

Halle, Morris, and Alec Marantz. 1993. Distributed morphology and the pieces in inflection. In Kenneth Hale and Samuel J. Keyser, eds., The View from Building 20, 111-176. Cambridge, MA: The MIT Press. Hockett, Charles F. 1954. Two models of grammatical description. Word 10:210-231. Kroeger, Paul R. 2005. Analyzing Grammar: An Introduction. Cambridge: Cambridge University Press. Levin, Beth, and Malka Rappaport Hovav. 1998. Morphology and lexical semantics. In Andrew Spencer and Arnold M. Zwicky, eds., The Handbook of Morphology, 248-271. Oxford: Blackwell. Marial Usón, Ricardo, and Pamela Faber. 2005. Decomposing semantic decomposition: towards a semantic metalanguage in RRG. Proceedings of the RRG ’05 Conference. Available online: http://linguistics.buffalo.edu/people/faculty/vanvalin/rrg/MairalRRG05 .pdf. Marial Usón, Ricardo, and Elizabeth Guest. 2005. Lexical representation based on a universal semantic metalanguage. RAEL Revista Electrónica de Lingüística Aplicada 5:125-173. Martín Arista, Javier. 2008. Unification and separation in a functional theory of morphology. In Robert D. Van Valin, Jr., ed., Investigations of the Syntax-Semantics-Pragmatics Interface, 119-146. Amsterdam and Philadelphia: John Benjamins. Mathiassen, Terje. 1996. A Short Grammar of Lithuanian. Columbus, OH: Slavica Publishers. Matthews, P.H. 1974. Morphology. 2nd edition. Cambridge: Cambridge University Press. McCarthy, John, and Alan Prince. 1998. Prosodic morphology. In Andrew Spencer and Arnold M. Zwicky, eds., The Handbook of Morphology, 283-305. Oxford: Blackwell. Ndjerareou, Mekoulnodji, Christy Melick, and Sarah Moeller. 2010. A brief grammatical sketch of Ngambay. GIALens: Electronic Notes Series 4.2. Available online: http://www.gial.edu/GIALens/vol4-2/ MelickMoeller Mekoulnodj-Ngambay.pdf. Nolan, Brian. 2004. First steps toward a computational RRG. In Brian Nolan, ed., RRG2004 Book of Proceedings, 196-223. Available online: http://linguistics.buffalo.edu/people/faculty/vanvalin/rrg/RRG2004%2 0Book%20of%20Proceedings.pdf. Pavey, Emma L. 2010. The Structure of Language: An Introduction to Grammatical Analysis. Cambridge: Cambridge University Press. Pustejovsky, James. 1995. The Generative Lexicon. Cambridge, MA: The MIT Press.

34

Towards a Realizational Approach to Morphology in RRG

Spencer, Andrew. 1998. Morphophonological operations. In Andrew Spencer and Arnold M. Zwicky, eds., The Handbook of Morphology, 123-143. Oxford: Blackwell. —. 2004. Morphology: an overview of central concepts. In Louisa Sadler and Andrew Spence, eds., Projecting Morphology, 67-109. Stanford, CA: CSLI Publications. Stump, Gregory T. 1998. Inflection. In Andrew Spencer and Arnold M. Zwicky, eds., The Handbook of Morphology, 13-43. Oxford: Blackwell. —. 2001. Inflectional Morphology: A Theory of Paradigm Structure. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 1993. A synopsis of Role and Reference Grammar. In Robert D. Van Valin, Jr., ed., Advances in Role and Reference Grammar, 1-164. Amsterdam and Philadelphia: John Benjamins. —. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press. —. 2009. An overview of role and reference grammar. Available online: http://linguistics.buffalo.edu/people/faculty/vanvalin/rrg/RRG_overvie w.pdf Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca, NY: Cornell University Press. Walton, Charles. 1986. Sama Verbal Semantics: Classification, Derivation and Inflection. Manila: Linguistic Society of the Philippines.

CASE SYNCRETISM IN TYPOLOGICAL PERSPECTIVE: AN RRG-OT ACCOUNT WATARU NAKAMURA TOHOKU UNIVERSITY

Abstract This paper provides an OT account of syncretism among core case morphemes in terms of their markedness relations revealed through the case hierarchy [CH] (Silverstein 1993) and a Role and Reference Grammar [RRG] case theory as outlined in Nakamura (1997, 1999a). Specifically, it derives the typological variation of case syncretism from the interactions among markedness constraints derived from the CH, faithfulness constraints, and an independently-needed morphosemantic constraint (Croft 1991).

Keywords Case, syncretism, markedness, Optimality Theory

1. Introduction Case marking displays a wide range of typological variation and has been a target of intense study within and outside of RRG (Van Valin 1993, 2005; Van Valin and LaPolla 1997). What adds to the complexity of the case-marking system is that it is often subject to syncretism, which is subdivided into absolute and contextual case syncretism (Calabrese 2008). The former involves a multiple correspondence between a case morpheme and more than one case feature value (however it is defined) that holds across the morphology of language, while the latter involves replacing a case morpheme with another one in certain nominal classes. It has been a common practice since Jakobson (1936/1983) (see also Bierwisch 1967, Neidle 1988, Blevins 1995, Wiese 1996, Baerman et al. 2005, and Calabrese 2008, among many others) to decompose cases into a

Case Syncretism in Typological Perspective: An RRG-OT Account

36

set of binary features and to derive their syncretism from a shared feature or features. The problem is that many examples of the case syncretism resist such a decomposition-based approach; it is often impossible to identify what is shared by more than one case. The aim of this paper is to propose an RRG-OT account of the typological variation of syncretism among the core case morphemes (i.e. nominative, accusative, ergative, dative, and genitive) with no use of feature decomposition. As an initial illustration of case syncretism, let us consider examples (1a)-(1f), taken from Kabardian (Northwest Caucasian) (Colarusso 1992: 167; Smith 1996: 108, 111, 113): (1) a.

b.

c.

d.

e.

f.

ॅ’ԥ-m šԥ-r fԥਸzԥ-m man-OBL horse-NOM woman-OBL jԥrԥjtáhsપ. (NOM.3)-IO-ACT-gave “The man gave the horse to the woman.” ॅ’ԥ-m šԥ-r jԥ-wԥh’áhsપ. man-OBL horse-NOM (NOM:3)-ACT-killed “The man killed the horse.” ъa-r žásપԥ-m mabáhna. dog-NOM night-OBL (NOM.3)-bark “The dog barks at night.” ॅ’ԥ-m fԥਸzԥ-m náxra nax‘‘ԥਸॳ߅. man-NOM woman-OBL older (NOM.3)-is “The man is older than the woman.” mázԥ-m jahъ. forest-OBL ACT-(NOM.3)carry “They carry it to the forest.” ъa-m Ø-yԥ-pa-r dog-OBL 3-POSS-nose-NOM “the dog’s nose”

Kabardian is a double-marking language with a two-way case-marking system. Unlike the nominative case suffix -r, which may appear once per clause, the oblique case suffix -m may appear multiple times. What is striking about Kabardian is that the oblique suffix marks transitive subjects (ERG), all oblique NPs (DAT), and adnominal possessors (GEN):

Wataru Nakamura

37

(2) Case Syncretism in Kabardian Case Case Morpheme NOM -r DAT -m ERG GEN An analogous “across-the-board” neutralization among oblique cases (e.g. instrumental, goal, comitative) is observed in Halkomelem (Salish), in which all oblique lexical NPs (and proper-noun adnominal possessors) are marked by the same case morpheme (Gerdts 1988).1 It is clear that it is impossible to identify what is shared by these three cases and that the multiple correspondence in Kabardian is not amenable to the feature decomposition-based approach to case syncretism, since it is a case of syncretism in which the syncretic forms clearly share no common (semantic or syntactic) feature. Before going into the main part of the present paper, it is necessary to make three assumptions. First, I assume a set of case assignment constraints in (3) (cf. Nakamura 1997, 1999a, 1999b): (3) Case Assignment Constraints a. At least one argument takes NOMINATIVE case. b. Non-macroroles arguments/adjuncts take DATIVE case. c. Undergoers take ACCUSATIVE case. d. Actor arguments take ERGATIVE case. e. At least one nominal argument takes GENITIVE case.2 (3a) accommodates the traditional observation (e.g. Jakobson 1936/1983) that nominative is the default case. (3b) defines dative as the default case for non-macrorole NPs (cf. Van Valin 1993). (3c,d) are derived from the observation that accusative and ergative case normally mark undergoers and actors, respectively. Finally, (3e) defines genitive as the default case for nominal arguments. Re-ranking of (3a)-(3d) yields the typological variation of case systems in (4) (Nakamura 1999a, 1999b):

1

The Halkomelem syncretism may be taken as a piece of evidence against a version of the case hierarchy (e.g. Blake 1994) that views the genitive and dative as being more marked than the accusative/ergative. 2 Many languages allow a genitive case morpheme to appear only once within an NP, but Japanese and Finnish (Hakulinen and Karlsson 1979: 395) permit the genitive morphemes to appear more than once within an NP.

38

Case Syncretism in Typological Perspective: An RRG-OT Account

(4) Typological Variation of Case Systems a. Accusative Case System: b. Ergative Case System: c. Accusative-Active Case System: d. Ergative-Active Case System:

(3b) >> (3a) >> (3c) >> (3d) (3b) >> (3a) >> (3d) >> (3c) (3b) >> (3c) >> (3a) >> (3d) (3b) >> (3d) >> (3a) >> (3c)

Second, given (3) and (4), I propose that the constraint rankings in (4) receive an input comprising a macrorole value or values of a verb’s argument(s) and output a case frame that specifies a SYNTACTIC case feature value of each argument. Table 1 shows how (4a) yields a “nom-acc” case frame when it receives a pair of an actor and undergoer argument as input: Table 1: Accusative Case System (Transitive Clauses) Input: Actor-Undergoer

(3b)

(3a)

 NOM-NOM

(3c)

(3d)

*!

* *

‫ڡ‬NOM-ACC  ERG-NOM  ERG-ACC

*! *!

The mapping between the syntactic case feature values output by (4a)-(4d) and their morphological realizations by case morphemes is the subject of the present paper. Third, I adopt the case hierarchy [CH] in (5), a combination of two markedness hierarchies of case morphemes (Silverstein 1976, 1993). These implicational hierarchies control the mapping between the syntactic case feature values and their morphological counterparts: (5) Case Hierarchy [CH] (adapted from Silverstein 1993: 485)   Gen Adnominal   щ a. Nom Њ Dat1 㧨 {Acc, Erg} b. Dat2 㧨 Loc, Instr ...

Propositional Adverbial/Propositional

(5a) states that the existence of a distinct genitive case morpheme implies that of a distinct accusative or ergative case morpheme, which, in turn,

Wataru Nakamura

39

implies that of a distinct dative morpheme. (5b) states that the existence of a distinct locative/instrumental case morpheme implies that of a distinct dative morpheme.3 I assume, contrary to Silverstein (1976, 1993), that no inherent markedness relation obtains between the accusative and ergative.4 An important point to note about the CH is that both the accusative/ ergative and genitive are elaborations of the fundamental contrast between the nominative and dative. The dative (as well as the nominative) is underspecified with respect to where it appears and is diverged into two domain-specific case morphemes, the accusative/ergative (clausal domain) and genitive (nominal domain). This means that while the genitive is more marked than the accusative/ergative, all the three occupy the same structural slot relative to the dative and that the accusative, ergative, and genitive are one degree more marked than the dative: (6) Case Hierarchy (Adnominal/Propositional: Revised) Marked 1 Marked 2 Marked 0 Gen Nom Dat щ Њ 㧨 Acc, Erg

[+N] [--N]

(6) shows that the genitive, accusative, and ergative case morpheme form a natural category (as opposed to the dative and nominative) in terms of degree of markedness.5 The essential idea behind (6) is that the CH is a nested markedness hierarchy that first ranks the case morphemes into three categories (“Nom”, “Dat”, “Acc/Erg/Gen”) and further hierarchizes the most marked category into two subcategories (“Acc/Erg”, “Gen”). The rest of the paper is organized as follows. Section 2 introduces a set of markedness and faithfulness constraints that allows us to describe the case syncretism. Section 3 illustrates the typological variation of the syncretism. Section 4 shows how the proposed set of constraints (plus an independently-motivated morphosemantic constraint) accommodates the 3

(5b) accommodates the syncretism between dative and other oblique case morphemes and is supplemented by Blansitt (1988), who establishes implicational relations among the dative, allative, and locative crosslinguistically (see Croft 1991 and Malchukov and Narrog 2009 for an alternative semantic account). 4 This means that the CH in (5a) requires the accusative and ergative to be more marked than the dative and less marked than the genitive, but it leaves it open whether the ergative is more marked than the accusative or not. 5 I follow Béjar and Hall (1999) in assuming that a shared degree of markedness can be a motivation for syncretism.

Case Syncretism in Typological Perspective: An RRG-OT Account

40

typological variation of the case syncretism. Section 5 is a conclusion.

2. Markedness and Faithfulness Constraints In this section, I will turn the CH into a set of markedness and faithfulness constraints that allows us to predict and describe the typological variety of case syncretism. First, I follow de Lacy (2006) in formalizing scale-based markedness relations in terms of a stringency hierarchy in (7a) rather than their universally fixed ranking (Prince and Smolensky 2004) in (7b). Under de Lacy’s proposal, a fixed ranking of markedness constraints is replaced by a subset structure within the constraint family, as shown in (7a): (7) a. 

b. 

Stringency Hierarchy Theory (de Lacy 2006) *{X}, *{X, Y}, *{X, Y, Z} Fixed Ranking Theory (Prince and Smolensky 2004) *X >> *Y >> *Z

Each of the markedness constraints in (7a) has to include the most marked member (“X”). It becomes progressively stringent and bans a larger, adjacent portion of the markedness hierarchy. These stringency-based constraints are freely rankable (like the other constraints in OT); no matter how they may be ranked, the more marked member involves more violations than the less marked member and loses to the less marked one (unless other constraints intervene). The notion of stringency hierarchy allows us to derive from the CH three markedness constraints in (8) under the assumption that nominative is underspecified with respect to its morphological value as well as its argument role (cf. Jakobson 1936/1983): (8) Markedness Constraints a. *{Gen} b. *{Gen, Acc/Erg} 1. *{Gen, Acc} 2. *{Gen, Erg} c. *{Gen, Acc/Erg, Dat} 1. *{Gen, Acc, Dat} 2. *{Gen, Erg, Dat} “*{Gen, Acc/Erg}” in (8b) is a shorthand for (8b1) and (8b2), while “*{Gen, Acc/Erg, Dat}” in (8c) is a shorthand for (8c1) and (8c2).

Wataru Nakamura

41

Second, I propose two types of faithfulness constraints in (9), which militate against the markedness constraints in (8a)-(8c): “MAX [Case]” in (9a) requires that each syntactic case value is realized by some case morpheme, while “IDENT [Case]” in (9b) dictates that case values are the same in the input and output:6 (9) Faithfulness Constraints a. MAX [Case (=Gen, Acc/Erg, Dat)] b. IDENT [Case (=Gen, Acc/Erg, Dat)]  c. IDENT [Gen, Acc, Dat] & IDENT [Gen, Erg, Dat] Given the CH, we may see that “IDENT [Case]” is a shorthand for (9c), a complex constraint that locally conjoins the two IDENT constraints and is violated when both of them are violated (Smolensky 1995).7 (9c) allows either the accusative or ergative case value (but not both) to be realized unfaithfully. Specifically, (9c) allows the accusative-dative syncretism when a distinct ergative case morpheme is available (since this case-marking system violates “IDENT [Gen, Acc, Dat]”, but satisfies “IDENT [Gen, Erg, Dat]”). Likewise, (9c) also allows the ergative-dative syncretism when a distinct accusative case morpheme is available. It is important to recall at this juncture that all of the accusative, ergative, and genitive are more marked than the dative by one degree according to (6). Given (6), I propose to add to (9c) a complex IDENT constraint in (9d) that requires the degree of markedness of a case morpheme to be kept constant in the input and output: (9) d.

IDENT [Gen/Acc, Dat] & IDENT [Gen/Erg, Dat] Shorthand: IDENT [DoM]

(9d) conjoins two subconstraints (“IDENT [Gen/Acc, Dat]” and “IDENT [Gen/Erg, Dat]”) and is violated when both of them are violated. The thrust of (9d) is to maintain a formal contrast between the fundamental opposition between the nominative and dative and its elaborations. What distinguishes (9d) from (9c) is that (9d) allows the genitive case value to be realized by either the accusative, ergative, or genitive case morpheme (since they all involve two degrees of markedness in (6)), while (9c) 6

The difference between MAX and IDENT constraints is that the former allow no feature deletion but feature changing, while the latter allow no feature changing but feature deletion. 7 The CH states that the existence of a distinct genitive case morpheme implies that of a distinct accusative and/or ergative case morpheme.

42

Case Syncretism in Typological Perspective: An RRG-OT Account

requires the genitive case value to be realized faithfully. (10) is a list of the markedness and faithfulness constraints proposed in this section: (10) Markedness Constraints a. 1. *{Gen} 2. *{Gen, Acc), {Gen, Erg} Shorthand: *{Gen, Acc/Erg} 3. *{Gen, Acc, Dat}, *{Gen, Acc, Dat} Shorthand: *{Gen, Acc/Erg, Dat} Faithfulness Constraints b. 1. MAX [Case] 2. IDENT [Gen, Acc, Dat] & IDENT [Gen, Erg, Dat] Shorthand: IDENT [Case] 3. IDENT [Gen/Acc, Dat] & IDENT [Gen/Erg, Dat] Shorthand: IDENT [DoM]

3. Examples of Case Syncretism I will illustrate the typological variety of the case syncretism in (11). They exhaust the possible patterns of syncretism (among the propositional and adnominal case morphemes) predicted by the CH in (5a): (11) Typological Variation of Case Syncretism8 a. Dat=Erg=Gen, Nom (Acc) (e.g. Kabardian) b. Dat=Acc=Gen, Nom (Erg) (e.g. Palauan) c. Dat=Acc=Erg=Gen, Nom (e.g. Yagnob) d. Dat=Gen, Nom (e.g. Halkomelem) e. Dat=Gen, Nom, Acc (e.g. Bengali) f. Dat=Gen, Nom, Erg (e.g. Djaru) g. Acc=Gen, Nom, Dat (Erg) (e.g. Estonian, Finnish) h. Erg=Gen, Nom, Dat (Acc) (e.g. Inuktitut) i. Acc=Gen, Nom, Dat=Erg No known example j. Erg=Gen, Nom, Dat=Acc (e.g. Tagalog) k. Dat=Acc, Nom, Erg, Gen (e.g. Hindi) l. Dat=Erg, Nom, Acc, Gen No known example 8 The round brackets in (11a,b,g,h) indicate that the case morphemes within the brackets may or may not be available. For example, Nyungar (Pama-Nyungan) uses the same case suffix to represent the ergative, genitive, and dative case value, while retaining a distinct accusative case suffix (Blake 1977; cf. Dixon 2002: 313).

Wataru Nakamura

43

First, the CH licenses the genitive, the genitive and accusative, the genitive and ergative, and the genitive, accusative, and ergative to be syncretized into the dative. This brings about the two-way case-marking systems in (11a)-(11d). Second, the CH allows the genitive to be syncretized into the dative, accusative, or ergative: (11e,f), (11g), and (11i) illustrate three- way systems with the genitive-dative, genitive-accusative, and genitiveergative syncretism, respectively. Third, the CH predicts three-way systems with both the genitive-ergative and accusative-dative syncretism in (11j) and those with the genitive-accusative and ergative-dative syncretism in (11i), but the latter don’t seem to be attested. Finally, the CH allows four-way systems with the accusative/ergative-dative syncretism in the presence of a distinct genitive case morpheme when a distinct ergative/ accusative case morpheme is available. (11l) doesn’t seem to be attested, however. I will come back to (11i,l) at the end of Section 4. (11a) is derived when the ergative and genitive case morpheme are syncretized into the dative. We saw in Section 1 that Kabardian uses the same case morpheme to represent all oblique NPs (DAT), transitive actors (ERG), and (proper-noun) adnominal possessors (GEN). (11b) is illustrated by the two-way case-marking system in Palauan (Austronesian). Palauan uses the oblique preposition er to represent the dative, accusative, and genitive case value (Josephs 1975: 207; Georgopoulos 1991: 26, 27, 29): (12) a. b. c. d. e. f.

ak-mo er a katsudo. R.1SG-go OBL  movies “I am going to the movies.” ng-kiltmekl-ii a ulaol a Peter. R.3SG-clean-3SG floor Peter “Peter cleaned the floor.” ng-diak ku-nguiu er a hong. NEG IR.1SG-read OBL book “I am not reading the book.” ng-mo er a ngebard er a klukuk. R.3SG-go OBL west OBL tomorrow “She is going to America tomorrow.” ak-uleldanges er a resensei er ngak R.1SG-IMPERF.honor OBL teachers OBL me “I respected my teachers.” A Romana a omeka er a rengalek er a kukau. Romana feed OBL children OBL taro “Romana is feeding the children the taro.”

Case Syncretism in Typological Perspective: An RRG-OT Account

44

(12a)-(12f) shows that the preposition er marks non-macrorole arguments and adjuncts (DAT), those human and/or specific-and-singular transitive undergoers (ACC) which occur in imperfective clauses, and adnominal possessors (GEN). (11c) is illustrated by Yagnob (Northeast Iranian), a double-marking language that displays a broader range of case syncretism than Kabardian and Palauan (Comrie 1981: 169-170). Specifically, it allows the same case morpheme to represent a variety of oblique NPs (DAT), transitive actors (ERG) in ergative constructions, definite transitive undergoers (ACC) in accusative constructions, and adnominal possessors (GEN). (11d) is illustrated by Halkomelem (Salish), a double-marking language that allows pronouns to display a four-way case distinction (i.e. nominative, dative, accusative, and genitive), while leaving lexical NPs with two case morphemes, nominative and dative (Gerdts 1988): (13) a.

b.

c.

d.

ni q’ࣶԥl-ԥt-ԥs șԥ s‫ࣶז‬éniপ tșԥ sce:‫ז‬tԥn AUX bake-TR-3.ERG DET woman DET salmon “The woman baked the salmon.” প ԥ kࣶșԥ ni cԥn q’ࣶáqࣶ-ԥt AUX 1.NOM club-TR DAT DET প ԥnপ-sápԥl-পԥ‫ז‬. 2.GEN-shovel-PAST “I hit him with your shovel.” প ԥ kࣶșԥ nԥ-télԥ. ni cԥn q’ࣶál AUX 1.NOM speak DAT DET 1.GEN-money “I spoke about my money.” প ԥ-ॅ’ John kࣶșԥ púkࣶ DET book-3.GEN DAT-DET John “John’s book”

Halkomelem marks proper-noun adnominal possessors with the dative case morpheme as in (13d), while leaving common-noun adnominal possessors nominative-marked. (11e,f) involve a genitive-dative syncretism illustrated by Bengali (Indo-Aryan) and Djaru (Pama-Nyungan). Examples (14a)-(14c) come from Bengali (Klaiman 1981: 3, 4, 20): (14) a.

aami-Ø caakor-Ø/tomaake I-NOM servant-NOM/you.ACC “I am looking for a servant/you.”

khNJjchi. am.seeking

Wataru Nakamura

b.

c.

d.

45

raastaay ektӏ meye-Ø/meye-ke dekhlaam. road.LOC a girl-NOM/girl-ACC saw “I saw a girl on the street.” aamaar tomaake mon-e po‫ގ‬be. me.DAT you.ACC mind-LOC will.fall “I shall remember you.” ha߮haat aamaar maathaay buddhi-Ø elo. suddenly me.GEN head.LOC idea-NOM came “I suddenly got an idea.”

(14a,b) show that Bengali has a split-accusative case-marking system in which humanness and definiteness play an important role. (14c,d) show that the dative and genitive case value are realized by the same case morpheme in Bengali. Furthermore, Djaru involves an ergative casemarking system and marks adnominal alienable and a few inalienable possessors with the same case morpheme as used to represent the dative case value (while leaving most of adnominal inalienable possessors nominative-marked) (Tsunoda 1981: 194-197).9 (11g) involves a genitive-accusative syncretism illustrated by two Finno-Ugric languages, Estonian and Finnish. Examples (15)-(16) come from Estonian (Matsumura 2001): (15) a. b. 

c.

(16) a.

9

Üliõpilased ehita-vad ise ühiselamu. students build-3PL themselves hall-of-residence.GEN “Students build a hall of residence for themselves.” (complete) Üliõpilased ehita-vad ühiselamu-t. students build-3PL hall-of-residence-PART “Students are building a hall of residence.” (incomplete) noore tüdruk-u kleit young girl-GEN dress “young girl’s dress” Arvo ost-is-Ø jalgratta. Arvo buy-PAST-3SG bicycle.GEN “Arvo bought a bicycle.”

I analyze this split marking as an instance of diagrammatic iconicity (Shapiro 1983; cf. Bybee 1985; Battistella 1990; Aristar 1997), in which the more marked a semantic content is, the more likely it is to take the more marked form. Under the assumption that inalienable possessors are semantically less marked than alienable ones, we may explain why adnominal alienable possessors are dative-marked (the more marked case morpheme), while most of adnominal inalienable possessors are nominative-marked (the less marked case morpheme) in Djaru.

Case Syncretism in Typological Perspective: An RRG-OT Account

46



b.

(17) a.



b.

Arvo ei ost-nud jalgratas-t. Arvo NEG buy-PAST.PART bicycle-PART “Arvo didn’t buy a bicycle (Lit. Arvo hasn’t bought a bicycle).” Ma telli-n tee ja koogi. I order-1SG tea.GEN and cake.GEN “I order tea and cake (for one person).” Ma telli-n tee-d ja kook-i. I order-1SG tea-PART and cake-PART “I order (an unspecified amount of) tea and cake.”

The accusative/genitive case morpheme encodes a nominal or aspectual boundedness (as illustrated by the alternations in (15a,b)-(17a,b)) in addition to encoding adnominal possessors as in (15c). This genitiveaccusative syncretism holds across the morphology of Estonian. Examples (18a,b) come from Finnish (Finno-Ugric: Vainikka 1993: 130, 142, 143): (18) a.



b.

c.

d.

Riita-n auto on ulkona. Riita-GEN car.NOM is outside “Riita’s car is outside.” Riita luki kirja-n Riita.NOM read.PAST.3SG book-GEN “Riita read the book.” Riita luki kirja-a Riita.NOM read.PAST.3SG book-PART “Riita was reading a/the book.” Riita ei luki kirja-a/*kirja-n Riita.NOM NEG read.PAST.3SG book-PART/*book-GEN “Riita didn’t read the book.”

Like Estonian, Finnish displays an alternation between accusative/genitive and partitive that is sensitive to the nominal/aspectual boundedness. What distinguishes the Finnish syncretism from the Estonian counterpart is that the genitive-accusative syncretism holds in lexical nouns alone in Finnish (Karlsson 2008). (11h) involves a genitive-ergative syncretism illustrated by Inuktitut. Let us consider examples (19a,b) (Eskimo-Aleut: Johns 1987: 12, 15):

Wataru Nakamura

(19) a. b.

47

Jaani-up natsiq kapi-ja-a. John-REL seal stab-DEC.TR-3SG.3SG “John is stabbing/stabbed the seal.” Jaani-up aggaa-nga John-REL glove-POSS.3SG “John’s glove”

These examples illustrate that ergative and genitive case value are realized by the same case morpheme (traditionally termed the “relative case” in the literature) in Inuktitut. It is important to note in this connection that the genitive-ergative syncretism in (11h) may be accompanied by the accusative-dative syncretism as in Tagalog (Austronesian), since the CH allows an accusative-dative syncretism when a distinct ergative case morpheme is available as in (11j). It makes no difference whether or not an ergative morpheme represents the genitive as well as the ergative case value, since we may term “ergative” a case morpheme that represents both the ergative and genitive case value, under the assumption that syncretism is a process by which the more marked form is assimilated into the less marked form. Examples (20a)-(20f) come from Kroeger (1993: 13, 14, 23, 32, 49, 131): (20) a. b. c.

d. e. f.

Pinutol ng-magsasaka ang-sungay ng-kalabaw. PERF-OV-cut ERG-farmer NOM-horn GEN-buffalo “The farmer cut off the buffalo's horn.” Bago ang-mga-kotse-ng iyon. new NOM-PL-car-LINKER that “Those cars are new.” Ibinigay lahat ng-mga-guro sa-mga-bata  IV-PERF-give all ERG-PL-teacher DAT-PL-child ang-pera. NOM-money “The teachers gave all the money to the children.” B-um-ili ang-lalake ng-isda sa-tindahan. PERF-AV-buy NOM-man GEN-fish DAT-store “The man bought fish at the store.” I-b-in-ili ng-lalake ng-isda ang-bata. BV-PERF-buy ERG-man GEN-fish NOM-child “The man bought fish for the child.” Binigyan lahat ng-mga-guro ng-pera DV-PERF-give all ERG-PL-teacher GEN-money

Case Syncretism in Typological Perspective: An RRG-OT Account

48

ang-mga-bata. NOM-PL-child “The teachers gave money to all the children.” Sa represents recipients (20c), locations (20d), and definite objects, while ng represents transitive actors (20a,c,f), adnominal possessors (20a), and indefinite objects (20d), while ang may mark transitive actors, undergoers, recipients/locations, benefactives, and instrumentals, depending on the voice of the verb.10 (21) is a summary of the Tagalog case syncretism:11 (21) Case Syncretism in Tagalog Case Feature Case Morpheme NOM ang/si (Nom) DAT sa/kay (Dat) ACC ERG ng/ni (Erg) GEN (11k) is illustrated by Hindi, a language that exhibits an accusativedative syncretism, while retaining a distinct ergative and genitive case morpheme (Mohanan 1994: 59, 60, 80):12 (22) a.

b.



10

c.

bacce-ne kitaab paʼфii. child-ERG book.NOM read.PERF “The child read a book.” raam-ne bacce-kaa naam pukaaraa. Ram-ERG child-GEN name.NOM call.PERF “Ram called the child’s name.” ilaa-ne haar-ko u߮фaayaa. Ila-ERG necklace-ACC lift.PERF “Ila lifted the/*a necklace.”

Tagalog has the actor, benefactive, instrumental, dative/locative, and objective voice. The argument role of a nominative-marked argument is indicated by the voice marker on the verb. 11 Si, kay, and ni in (21) are used to mark personal names. As illustrated by (21), case morphemes may involve allomorphic alternations correlated with semantic and/or phonological factors. In order to distinguish between case morphemes and their phonological realizations, I term the latter case forms. 12 Hindi has an aspect-based split-ergative system in which the ergative case clitic -ne marks transitive actor arguments in perfective clauses.

Wataru Nakamura

d.

e.

ilaa-ne bacce-ko/*baccaa Ila-ERG child-ACC/child.NOM “Ila lifted a/the child.” niinaa-ne bacce-ko kitaab Nina-ERG child-DAT book.NOM “Nina gave the child a book.”

49

u߮фaayaa. lift.PERF dii. give.PERF

Hindi assigns the case clitic -ko to recipients in ditransitive constructions and animate and/or specific undergoers in transitive constructions (see Mohanan 1994: 79-90 for more examples and discussion of the interaction between animacy and specificity/definiteness).13 These Hindi data provide another illustration that when both the ergative and accusative case value are somehow case-marked in a language that has a distinct genitive case morpheme, either the ergative or accusative may be realized by the less marked (i.e. dative) case morpheme. Finally, there seems to be no language that involves an ergative-dative and genitive-accusative syncretism as in (11i) or an ergative-dative syncretism alone as in (11l). I will come back to these two unattested patterns of case syncretism in Section 4.5: (23) Two Unattested Patterns of Syncretism (11i) (ERG=DAT, GEN=ACC) (11l) (ERG=DAT) Case Case Case Case Feature Morpheme Feature Morpheme NOM NOM DAT DAT ERG ERG ACC ACC GEN GEN

4. An OT Account of Case Syncretism This section demonstrates that the patterns of case syncretism in (11a)-(11i) predicted by the CH are derived from a competitive interaction between the set of markedness and faithfulness constraints in (10a,b).

13

Hindi has an inflectional case-marking system with “direct case” (nominative), “oblique case” (genitive/ergative/accusative/dative), and “vocative case” as well as the clitic-based case-marking system . See Section 4.4 for further discussion.

50

Case Syncretism in Typological Perspective: An RRG-OT Account

4.1. Kabardian, Palauan, and Yagnob Let us begin with the Kabardian case-marking system. Since Kabardian uses the same suffix to mark transitive undergoers and intransitive subjects to the exclusion of transitive actors, we may assume that Kabardian involves an ergative case system output by the constraint hierarchy in (4b). The problem here is that (4b) alone cannot assign the same case morpheme to transitive actors, non-macroroles, and adnominal possessors. We need to bridge the gap between the full-fledged ergative case system output by the constraint hierarchy in (4b) and the two-way case-marking system, illustrated in (1a)-(1f): (24) Case/Case-Marking System in Kabardian Case-Marking System Constraint Case System (Case Morpheme) Hierarchy (Case Feature) (4b) (Ergative) NOM -r (Nom) (3b) >> (3a) >> DAT (3d) >> (3c) -m (Dat) ERG (3e) GEN In order to identify the case morpheme that represents more than one case feature value, I propose (25) under the uncontroversial assumption that syncretism is a process by which the more marked form is assimilated into the less marked form: (25) Case Morpheme Principle [CMP] (cf. Nakamura 1997) When a single case morpheme represents more than one case feature, adopt the most unmarked morpheme as its name Given the CMP, I propose that the constraint hierarchy in (26) is responsible for the mapping between the set of case feature values output by the constraint hierarchy in (4b) and the impoverished two-way casemarking system. Table 2 shows how the genitive case feature value (GEN) is correctly realized by the dative case morpheme: (26) Mapping between Case Features and Case Morphemes in Kabardian MAX [Case] >> *{Gen, Acc/Erg} >> IDENT [Case], IDENT [DoM], *{Gen}, *{Gen, Acc/Erg, Dat}

Wataru Nakamura

51

Table 2: GENITIVE mapped to Dative in Kabardian14 Input: GEN Nom

MAX [Case]

*{G, A/E}

ID [Case]

*!

ID [DoM]

*{G}

*{G, A/E, D}

* *

‫ڡ‬Dat Erg

*!

Gen

*!

*

*

*

* *

*

Let us see how the evaluation proceeds in Table 2. The top constraint “MAX [Case]” requires the input has a correspondent in the output and rules out the first candidate (nominative). The third and fourth candidate violate the second-highest ranking constraint (“*{Gen, Acc/Erg}”) that militates against appearance of a genitive, accusative, or ergative case morpheme. Likewise, the dative morpheme emerges as the winner when the ergative case feature value (ERG) is given as input. The constraint hierarchy in (26) also accommodates the mapping between the Palauan case and case-marking system, summarized in (27), and extends further to the Yagnob case-marking system, which assimilates the genitive, accusative, and ergative to the dative: (27) Case/Case-Marking System in Palauan Constraint Case System (26) Hierarchy (Case Feature) (4a) (Accusative) NOM (3b) >> (3a) >> DAT (3d) >> (3c) ACC (3e) GEN

Case-Marking System (Case Morpheme) ø (Nom) er (Dat)

4.2. Bengali and Djaru The Bengali and Djaru case-marking system mark non-macroroles and adnominal possessors with the same case morpheme. (28) is the constraint 14

“ID” in Table 2 is an abbreviation for “IDENT”. The four lower-ranking constraints (“ID [Case]”, “ID [DoM]”, “*{G}”, and “*{G, A/E, D}”) constitute a block in which they are not crucially ranked with respect to each other.

Case Syncretism in Typological Perspective: An RRG-OT Account

52

hierarchy for Bengali and Djaru: (28) Mapping between Case Features and Case Morphemes in Bengali and Djaru MAX [Case] >> *{Gen} >> IDENT [Case] >> *{Gen, Acc/Erg} >> IDENT [DoM], *{Gen, Acc/Erg, Dat} Ranking “*{Gen}” between “MAX [Case]” and “IDENT [Case]” ensures that adnominal possessors receive a non-nominative case morpheme other than the genitive in Bengali and Djaru. Furthermore, ranking “*{Gen, Acc/Erg}” above “IDENT [DoM]” requires the genitive case value to be represented by the dative case morpheme. Tables 3(a,b) show that the genitive case value is correctly mapped to the dative case morpheme in Bengali and Djaru, respectively: Table 3(a): GENITIVE mapped to Dative in Bengali Input: GEN

MAX [Case]

Nom

*!

*{G}

ID [Case]

ID [DoM]

*{G, A/E, D}

*

‫ڡ‬Dat

*

Acc

*

Gen

*{G, A/E}

* *!

*!

* *

*

Table 3(b): GENITIVE mapped to Dative in Djaru Input: GEN

MAX [Case]

Nom

*!

*{G}

ID [Case]

ID [DoM]

*{G, A/E, D}

*

‫ڡ‬Dat

*

Erg

*

Gen

*{G, A/E}

*!

* *!

* *

*

(29) is a summary of the correspondence between the Bengali case and case-marking system. An analogous relation holds between the Djaru case

Wataru Nakamura

53

and case-marking system: (29) Case/Case-Marking System in Bengali Constraint Case System (28) Hierarchy (Case Feature) (4a) (Accusative) NOM (3b) >> (3a) >> ACC (3c) >> (3d) DAT (3e) GEN

Case-Marking System (Case Morpheme) Nom Acc Dat

4.3. Inuktitut and Estonian/Finnish I propose that (30) is the constraint hierarchy responsible for the genitiveergative syncretism in Inuktitut and the genitive-accusative syncretism in Estonian and Finnish: (30) Mapping between Case Features and Case Morphemes in Inuktitut and Estonian/Finnish MAX [Case] >> *{Gen} >> IDENT [DoM] >> IDENT [Case], *{Gen, Acc/Erg}, *{Gen, Acc/Erg, Dat} It is important to recall from Section 2 that “IDENT [DoM]” allows the genitive case value to be represented by the accusative or ergative case morpheme as well as the genitive case morpheme. This means that ranking “*{Gen}” and “IDENT [DoM]” above “IDENT [Case]” as in (30) leads the genitive case value to be represented by the accusative or ergative case morpheme (depending on their availability within each language). (31) provides a summary of the Inuktitut case and case-marking system, while Table 4(a) illustrates how the former is realized by the latter: (31) Case/Case-Marking System in Inuktitut Constraint Case System (30) Hierarchy (Case Feature) (4a) (Ergative) NOM (3b) >> (3a) >> DAT (3d) >> (3c) ERG (3e) GEN

Case-Marking System (Case Morpheme) Nom Dat Erg

54

Case Syncretism in Typological Perspective: An RRG-OT Account

Table 4(a): GENITIVE mapped to Ergative in Inuktitut Input: GEN Nom

MAX [Case]

*{G}

ID [DoM]

ID [Case]

*!

*

*{G, A/E}

*{G, A/E, D}

*!

Dat

*

‫ڡ‬Erg Gen

*!

* *

*

*

*

Likewise, (32) summarizes the Estonian/Finnish case and case-marking system, while Table 4(b) illustrates the mapping between them: (32) Case/Case-Marking System in Estonian/Finnish Constraint Case System Case-Marking System (30) Hierarchy (Case Feature) (Case Morpheme) (4a) (Accusative) NOM Nom (3b) >> (3a) >> DAT Dat (3c) >> (3d) ACC Acc (3e) GEN Table 4(b): GENITIVE mapped to Accusative in Estonian/Finnish Input: GEN Nom

MAX [Case]

*{G}

ID [Case]

*!

*

*{G, A/E}

*{G, A/E, D}

*!

Dat

*

‫ڡ‬Acc Gen

ID [DoM]

*!

* *

*

*

*

4.4. Hindi What is peculiar about Hindi is that it displays a clitic-based case-marking system (“Layer II”), composed of a nominative clitic -Ø, a genitive clitic -kaa, a dative/accusative clitic -ko, and an ergative clitic -ne, on top of its

Wataru Nakamura

55

three-way inflectional case-marking system (“Layer I”), summarized in (33) and illustrated in Table 5 (Masica 1991): (33) Two-Tiered Case-Marking System in Hindi Case Morpheme Case Feature Inflection (Layer I) Phrasal Clitic (Layer II) NOM Direct (=Nom) -Ø (Nom) DAT/ACC -ko (Dat) Oblique (=Dat) ERG -ne (Erg) GEN -kaa (Gen) Table 5: Hindi Inflectional Declension (Mohanan 1994; Kachru 2006) Phrasal Clitic

Word-level Inflection

Direct (=Nom) Oblique (=Dat) Vocative

Masculine Singular Plural baccaa bacce (“child”) bacce baccõ

Feminine Singular Plural kΩnya kΩnya-ͅ (“girl”) kΩnya kΩnya-õ

bacce

kΩnya

bacco

-Ø -ko, -ne, -kaa, ...

kΩnya-õ

(22a)-(22e) show that the oblique forms of nouns (e.g. bacce “child”) are used when the whole NPs are marked by a non-nominative clitic.15 This allows us to view the case clitics as an elaboration on the minimal case-marking system that comprises the nominative and dative.16 In order to derive the dative-accusative and the genitive-ergative-dative syncretism in (33), I propose (34a,b) as a constraint hierarchy for the Hindi inflectional and clitic-based case-marking system, respectively: 15

The oblique forms of nouns in Hindi (and other Indo-Aryan languages) usually cannot appear alone; they co-occur with one of the case clitics (Masica 1991: 239). 16 There has been a controversy over whether the case clitics reviewed above may be analyzed as case markers and whether the genitive clitic -kaa, which (unlike the other case clitics) agrees with the possesum in number, gender, and inflectional marking, may be analyzed as a case marker (or case morpheme) (see Mohanan 1994, Payne 1995, and Spencer 2007 for discussion). My view (irrespective of whether to analyze the genitive clitic as a genuine case marker or not) is that the nominal declension illustrated in Table 5 and the clitics in (33) combine to yield a complex case-marking system (see Spencer 2007 for an alternative view).

56

Case Syncretism in Typological Perspective: An RRG-OT Account

(34) Mapping between Case Features and Case Morphemes in Hindi a. Inflectional Case-Marking System (=26) MAX [Case] >> *{Gen, Acc/Erg} >> IDENT [Case], IDENT [DoM], *{Gen}, *{Gen, Acc/Erg, Dat} b. Clitic-based Case-Marking System MAX [Case] >> IDENT [Case] >> *{Gen, Acc} >> *{Gen, Erg} >> IDENT [DoM], *{Gen}, *{Gen, Acc/Erg, Dat} Two comments are in order about (34b).17 First, we need to evaluate the mapping between the accusative and ergative case value and their morphological counterparts simultaneously as in Table 6, since one’s realization as the dative case morpheme is dependent on the other’s retaining its distinct case morpheme. Second, “IDENT [Case]” requires the genitive and dative case value to be realized faithfully, while allowing either the accusative or ergative case value to be realized unfaithfully. In fact, ranking “*{Gen, Acc}” above “*{Gen, Erg}” as in (34b) requires the accusative case value to be realized unfaithfully. The lower-ranking markedness constraint “*{Gen}” ensures that the accusative case value is realized by the dative case morpheme:18 Table 6: ACC/ERG mapped to Dative/Ergative in Hindi Input1=ACC, Input2=ERG

MAX [Case]

ID [Case]

ACC->Acc ERG->Erg

*{G, A}

*{G, E}

*!

*

‫ڡ‬ACC->Dat

*

ERG->Erg ACC->Acc

*!

ERG->Dat ACC->Dat ERG->Dat

17

*!

ID [DoM]

*{G}

* * **

(34a) requires no comments here, since it is the same as (26), the constraint hierarchy proposed for the two-way case-marking systems in Kabardian, Palauan, and Yagnob. 18 For reasons of space, “*{Gen, Acc/Erg, Dat}” is omitted from Table 6.

Wataru Nakamura

57

4.5. Unattested Patterns of Syncretism As we saw at the end of Section 3, there are two unattested patterns of syncretism, summarized in (23). It is easy to see that they both involve an ergative-dative syncretism in the presence of a distinct accusative morpheme. A natural question that arises is why these two patterns don’t seem to be attested, even if they are licensed and predicted by the CH: (23) Two Unattested Patterns of Syncretism (11i) (ERG=DAT, GEN=ACC) (11l) (ERG=DAT) Case Case Case Case Feature Morpheme Feature Morpheme NOM NOM DAT DAT ERG ERG ACC ACC GEN GEN My proposal is that some other motivation than the markedness considerations plays a critical role in excluding these two patterns. An important clue comes from the force-dynamic causal chain, proposed by Talmy (1976) and extended by Croft (1991) and Langacker (1991), which visualizes the inherent causal asymmetries among co-arguments of a verb. We may follow the spirit of the causal chain hypothesis (Talmy 1976) in decomposing verbal events into a sequence of causal chains, as illustrated by (35a.b): (35) Causal Chain Analysis of Transitive and Ditransitive Clauses a. John broke the vase with a hammer. x ------------------------> y --------------------> z Agent/A[ctor] Instr[umental] Patient/U[ndergoer] b.

John gave a book to Tom. x -------------> (y) ------------> z --------------------> w Agent/A INSTR Theme/U Recipient/Non-MR

What is notable about the above analyses is that a recipient/non-macrorole argument of a ditransitive verb is the farthest away from an agent/actor argument on the causal chain in (35b). Another point to note is that the ergative case morpheme marks transitive actor arguments, while the dative case morpheme typically marks recipient arguments of ditransitive verbs.

58

Case Syncretism in Typological Perspective: An RRG-OT Account

Under the assumption that the nearer a co-argument is to another one on the asymmetrical causal chain, the more likely they are to receive the same morphological marking, we may assume that an agent/actor and a recipient/non-macrorole argument of a ditransitive verb are the last to share the same case morpheme. I propose to generalize this observation into a constraint in (36): (36) Constraint on Semantic Nearness (cf. Croft 1991: 187)19 The farther a co-argument of a verb is away from another one on the causal chain, the less likely they are to share the same morphological marking (e.g. case marking). This constraint applies to the interface between morphology and semantics. Apart from the question of how to reformulate (36) as an OT constraint, it works in tandem with the markedness and faithfulness constraints in (10a,b) to constrain the mapping between syntactic case feature values and their morphological counterparts. A potential exception to (36) is the two-way case-marking systems in Kabardian and Yagnob, which involve the ergative-dative syncretism. However, it is important to note at this juncture that they involve the “across-the-board” syncretism that includes the ergative-dative syncretism as its part. I assume that the broad range of case syncretism erases the semantic contrast among co-arguments of a verb (and its adjunct(s)) and thereby nullifies the force of (36) (cf. Croft 1991: 188-190).20

19

In his discussion of syncretism among thematic roles, Croft (1991: 187) divides them into antecedent thematic roles (those participants which precede the object in the causal chain (e.g. instrumental, manner, comitative, ergative)) and subsequent thematic roles (those participants which follow the object (e.g. result, recipient, benefactive)) and claims that no pattern of syncretism may subsume the antecedent and subsequent thematic roles. Croft’s primary interest is in predicting possible patterns of syncretism among oblique case morphemes in light of the causal chain model, while my interest is in deriving possible patterns of syncretism among the core case morphemes from the CH. It is interesting to explore the connection between these two seemingly complementary accounts of case syncretism, but space limitation precludes further discussion here. 20 It is not difficult to see that the distinction between the antecedent thematic roles (including what Croft terms “ergative”; see footnote 19) and subsequent thematic roles (including recipients) is wiped out by the across-the-board syncretism that subsumes the antecedent and subsequent thematic roles.

Wataru Nakamura

59

5. Conclusion I have proposed the set of markedness and faithfulness constraints in (10) and the constraint on semantic nearness in (36) (which comes from Croft (1991)) and have argued that these three types of constraints ranked in a particular way receive as input a set of syntactic case feature values (“case frame”) output by (4a)-(4d) and output their morphological realizations by case morphemes. (37) presents a whole picture of the two-stage case theory proposed in this paper: (37) Two-Stage Case Theory a. Generation of Syntactic Case Feature Values (“Case Frame”) Syntactic Case Feature Values candidate candidate candidate candidate candidate

Input (e.g. A-U)

Gen

Output (e.g. “NOM-ACC”)

Optimization through (4a)-(4d)

b. Syntax-to-Morphology Mapping Syntactic Case Feature Values

Morphological Case Feature Values candidate candidate candidate candidate candidate

Input

Gen

Output

Optimization through (26), (28), (30), and (34a,b)

First, the constraint rankings in (4a)-(4d) take as input a set of macrorole

60

Case Syncretism in Typological Perspective: An RRG-OT Account

(and thematic) values of a verb’s co-arguments and output the set of syntactic case feature values. For example, (4a) outputs a “nom.-acc.” case frame when it receives a pair of an actor and undergoer argument as input. Second, (26), (28), (30), and (34a,b) receive as input the set of syntactic case feature values (output by (4a)-(4d)) and output their morphological counterparts, which may further involve allomorphic alternations. Finally, I will leave it to another occasion to assess the validity of (10a,b) and (36) against a richer set of typologically diverse languages.

List of Abbreviations 1 2 3 AV BV DEC DET DV IMPERF IR

first person second person third person actor voice benefactive voice declarative determiner dative voice imperfective irrealis

IV OBL OV PART PERF POSS R REL TR

instrumental voice oblique objective voice partitive perfective possessor realis relative transitive

References Aristar, Anthony R. 1997. Marking and hierarchy types and the grammaticalization of case-markers. Studies in Language 21.2: 313-368. Baerman, Matthew, Dunstan Brown, and Greville G. Corbett. 2005. The Syntax-Morphology Interface: A Study of Syncretism. Cambridge: Cambridge University Press. Battistella, Edwin L. 1990. The Logic of Markedness. Oxford: Oxford University Press. Béjar, Susana and Daniel Currie Hall. 1999. Marking markedness: the underlying order of diagonal syncretisms. Proceedings of the 15th Eastern States Conference on Linguistics, 1-12. Bierwisch, Manfred. 1967. Syntactic features in morphology: general problems of so-called pronominal inflection in German. In Roman Jakobson, ed., To Honor Roman Jakobson: Essays on the Occasion of His Seventieth Birthday, 239-270. The Hague: Mouton. Blake, Barry J. 1977. Case Marking in Australian Languages. Canberra: Australian Institute of Aboriginal Studies.

Wataru Nakamura

61

—. 1994. Case. Cambridge: Cambridge University Press. Blansitt, Edward L., Jr. 1988. Datives and allatives. In Michael Hammond, Edith A. Moravcsik, and Jessica R Wirth, eds., Studies in Syntactic Typology, 173-191. Amsterdam and Philadelphia: John Benjamins. Blevins, James P. 1995. Syncretism and paradigmatic opposition. Linguistics and Philosophy 18.2: 113–152. Bybee, Joan L. 1985. Morphology: A Study of the Relation Between Meaning and Form. Amsterdam: John Benjamins. Calabrese, Andrea. 2008. On absolute and contextual syncretism: remarks on the structure of case paradigms and on how to derive them. In Asaf Bachrach and Andrew Nevins, eds., Inflectional Identity, 156-205. Oxford: Oxford University Press. Colarusso, John. 1992. A Grammar of the Kabardian Language. Calgary: University of Calgary Press. Comrie, Bernard. 1981. The Languages of the Soviet Union. Cambridge: Cambridge University Press. Croft, William. 1991. Syntactic Categories and Grammatical Relations. Chicago: University of Chicago Press. de Lacy, Paul. 2006. Markedness: Reduction and Preservation in Phonology. Cambridge: Cambridge University Press. Dixon, R.M.W. 2002. Australian Languages. Cambridge: Cambridge University Press. Georgopoulos, Carol. 1991. Syntactic Variables: Resumptive Pronouns and A’ Binding in Palauan. Dordrecht: Kluwer Academic Publishers. Gerdts, Donna B. 1988. Object and Absolutive in Halkomelem Salish. New York: Garland. Hakulinen, Auli and Fred Karlsson. 1979. Nykysuomen Lauseoppia. Helsinki: Suomalaisen Kirjallisuuden Seura. Jakobson, Roman. 1936. Beiträge zur allgemeinen Kasuslehre. Travaux du Cercle Linguistique de Prague 6: 240-288. Reprinted in Linda R. Waugh and Morris Halle, eds., 1983. Russian and Slavic Grammar: Studies 1931-1981, 59-103. Berlin: Mouton. Johns, Alana. 1987. Transitivity and Grammatical Relations in Inuktitut. Unpublished Ph.D. Dissertation, University of Ottawa. Josephs, Lewis S. 1975. Palauan Reference Grammar. Honolulu: University Press of Hawaii. Kachru, Yamuna. 2006. Hindi. Amsterdam and Philadelphia: John Benjamins. Karlsson, Fred. 2008. Finnish: An Essential Grammar. London and New York: Routledge. Klaiman, M.H. 1981. Volitionality and Subject in Bengali. Bloomington,

62

Case Syncretism in Typological Perspective: An RRG-OT Account

IN: Indiana University Linguistics Club. Kroeger, Paul. 1993. Phrase Structure and Grammatical Relations in Tagalog. Stanford, CA: CSLI Publications. Langacker, Ronald W. 1991. Foundations of Cognitive Grammar, vol. 2: Descriptive Applications. Stanford, CA: Stanford University Press. Malchukov, Andrej and Heiko Narrog. 2009. Case polysemy. In Andrej Malchukov and Andrew Spencer, eds., The Oxford Handbook of Case, 518-535. Oxford: Oxford University Press. Masica, Colin P. 1991. The Indo-Aryan Languages. Cambridge: Cambridge University Press. Matsumura, Kazuto. 2001. Esutoniago Bunpoo. [Estonian Grammar] Available online: http://www.kmatsum.info/eesti/opik/index.html. Mohanan, Tara. 1994. Argument Structure in Hindi. Cambridge: Cambridge University Press. Nakamura, Wataru. 1997. A Constraint-based Approach to Case Systems. Unpublished Ph.D. Dissertation, University at Buffalo. —. 1999a. Functional Optimality Theory. In Michael Darnell et al. (eds.), Functionalism and Formalism, vol. 2: Case Studies, 253-276. Amsterdam and Philadelphia: John Benjamins. —. 1999b. An Optimality-theoretic account of the Japanese case system. Studies in Language 23.3: 607-660. Neidle, Carol. 1988. The Role of Case in Russian Syntax. Dordrecht: Kluwer Academic Publishers. Payne, John R. 1995. Inflecting postpositions in Indic and Kashmiri. In Frans Plank, ed., Double Case, 283-298. Oxford: Oxford University Press. Prince, Alan and Paul Smolensky. 2004. Optimality Theory: Constraint Interaction in Generative Grammar. Oxford: Basil Blackwell. Shapiro, Michael. 1983. The Sense of Grammar: Language as Semeiotic. Bloomington, IN: Indiana University Press. Silverstein, Michael. 1976. Hierarchy of features and ergativity. In R. M. W. Dixon, ed., Grammatical Categories in Australian Languages, 112-171. New Jersey: Humanities Press. —. 1993. Of nominatives and datives. In Robert D. Van Valin, Jr., ed., Advances in Role and Reference Grammar, 465-498. Amsterdam and Philadelphia: John Benjamins. Smith, Henry. 1996. Restrictiveness in Case Theory. Cambridge: Cambridge University Press. Smolensky, Paul. 1995. On the internal structure of the constraint component Con of UG. Handout from talk at University of Arizona. Available on Rutgers Optimality Archive.

Wataru Nakamura

63

Spencer, Andrew. 2007. The possessum-agreement construction or “Does Albanian have a genitive case?”. Unpublished manuscript, University of Essex. Available online: http://privatewww.essex.ac.uk/~spena/ PossMod/poss_agr_constr1.pdf. Talmy, Leonard. 1976. Semantic causative types. In Masayoshi Shibatani, ed., The Grammar of Causative Constructions, 43–116. Syntax and Semantics, 6. New York: Academic Press. Tsunoda, Tasaku. 1981. The Djaru Language of Kimberley, Western Australia. Canberra: Pacific Linguistics. Vainikka, Anne. 1993. The three structural cases in Finnish. In Anders Holmberg and Urpo Nikanne, eds., Case and Other Functional Categories in Finnish Syntax, 129-159. Berlin: Mouton de Gruyter. Van Valin, Robert D., Jr. ed. 1993. Advances in Role and Reference Grammar. Amsterdam and Philadelphia: John Benjamins. —. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Wiese, Bernd. 1996. Iconicity and syncretism: on pronominal inflection in Modern German. In Robin Sackmann and Monika Budd, eds., Theoretical Linguistics and Grammatical Description, 323-344. Amsterdam and Philadelphia: John Benjamins.

MEANING CONSTRUCTION AND GRAMMATICAL INFLECTION IN THE LAYERED STRUCTURE OF THE IRISH WORD: AN RRG ACCOUNT OF MORPHOLOGICAL CONSTRUCTIONS BRIAN NOLAN INSTITUTE OF TECHNOLOGY BLANCHARDSTOWN, DUBLIN IRELAND

Abstract This paper examines meaning construction in Modern Irish within a layered structure of the word [LSW] in a way analogous to the layered structure of the noun phrase and the layered structure of the clause in Role and Reference Grammar [RRG]. In addition to characterising elements of morphologically motivated word-level meaning construction and grammatically related inflectional morphology of Irish, we examine the role of the lexicon and the need for a lexeme store in the lexicon and a morpheme inventory. We consider meaning construction and the formation of nominal and other lexical categories. Under morphologically motivated word-level meaning construction, we discuss how a new member of a lexical category is formed, and how lexical meaning may be modified within the LSW in RRG. We discuss the need for an inventory of morphological construction schemata for lexemes, similar to the syntactic inventory, such that morphemes that carry semantic or conceptual meaning might be suitably represented in a lexeme store within the lexicon as morphological constructions schematically represented. We argue for a repository of constructions within the grammar and a construction level workspace based on the idea of a workshop from Van Valin (2005). Our account considers both meaning construction and inflection.

Brian Nolan

65

Keywords Meaning construction, grammatical inflection, lexicon, construction repository, workspace, LSW, lexeme store, morpheme inventory

1. Introduction This paper has a number of goals. We propose a layered structure of the word within a RRG perspective and motivate this by reference to the word in Modern Irish, and the processes that operate on the word. We characterise elements of the morphology of Irish with respect to morphological meaning construction and inflection. In doing this we will consider several important issues related to the distinction between what goes in the lexicon and the status of morphemes with grammatical purpose. Specifically, while characterising elements of the morphology of Irish we also motivate that part of RRG concerned with morphology and its relationship to the lexicon and the morpheme inventory. We extend the RRG theory by proposing an RRG compatible theory of morphology that is functionally oriented through the use of morphological constructions which may be represented with appropriate constructional templates (schemata), and that is intended to connect in a compatible way with the main RRG model. While examining the Irish data for morphological meaning construction, we explore the general question: What should be included in an RRG theory of morphology? The basic theory of RRG is to be found in Van Valin and LaPolla (1997) and Van Valin (2005), including accounts of the LSC, the layered structure of the noun phrase [LSNP] and qualia representation of nominals (Van Valin 2005: 51ff), We intend that our account of the layered structure of the word be compatible with the LSNP and the LSC. An RRG theory of morphology must account for a wide variety of morphological processes and phenomena including those listed in (1). (1) a. Meaning construction and morphological category changing devices b. The role of the RRG lexicon within a morphological perspective c. A means of representing the semantic meaning of lexemes in the lexicon d. Inflection processes in support of grammatical phenomena e. A language specific morpheme inventory within the grammar

66

Meaning Construction and Grammatical Inflection

f. Compounding and incorporation in grammar g. The phonological word and clitics h. Word formation in Semitic languages In this paper, however, we limit our discussion to briefly dealing with (1a) to (1f) in the above list within an RRG account of morphological meaning construction and inflection as it is found with the Irish data. We will also make an important distinction between a lexeme, which may be of various kinds, in the lexicon as against morphemes with grammatical purpose in a morpheme inventory. We provide our discussion within an RRG perspective.

2. Constructions as Objects of Importance to the Grammar In this account we will recognise that the idea of morphological constructions are objects of significance understood within an RRG perspective (Nolan 2010a,b,c,d, 2011a,b,c,d, in press a,b). That is, for example, word level meaning construction involves the construction of lexemes of certain types (representing the traditional lexical categories and “parts of speech”) and the marking of these lexemes as grammatical words in syntax for grammatical purposes under inflection as a component part of their syntactic realisation. Specifically, the RRG characterisation of constructions is succinctly characterised by Van Valin (2005: 131ff) as: RRG recognises the importance of grammatical constructions, and they are represented in terms of constructional schemas. Cross-constructional and cross-linguistic generalisations are captured in terms of the general principles and constraints that constitute the linking algorithm, e.g. the actor-undergoer hierarchy, the layered structure of the clause, the privileged syntactic argument selection hierarchy. Only the idiosyncratic, language features of constructions are represented in constructional schemas. Hence constructional schemas, by virtue of their reference to general principles, permit the capturing of cross linguistic generalisations, while at the same time expressing language-particular properties of grammars.

A construction in RRG is presently identified within a constructional schema having the following typical (construction specific) format (2) as a generalised internal construction structure. A detailed description of these elements can be found in Van Valin (2005). This format, while useful, is not elaborate enough to sustain a morphological perspective of constructions

Brian Nolan

67

in RRG as an object of importance to the grammar due to its heretofore clausal and syntactic bias. (2)

RRG generalised construction schema Syntax: PSA Semantics: Linking Morphology: Pragmatics:

The present view of constructions in RRG pertains generally to a syntactic level that recognises a morphological component. Here we are concerned with the nature of the word and its formation, and with the lexeme component of the grammar within a layered structure of the word. Therefore, we present an elaborated view of RRG constructions of relevance to clausal and syntactic levels but also useable at a morphological level.

2.1. Construction Repository Importantly, we suggest that a construction has significance as an object of the grammar with relevance to the explication of RRG morphology within a layered structure of the word and, as such, constructions can be found at the morphosyntactic level within the grammar. Therefore, we consider a construction to be an important object within the grammar that is identified by a construction signature, which uniquely identifies the correct construction and its schema in the construction repository. The construction can be directly retrieved from the repository using the construction signature as its identifying key. Additionally, this paper posits that there is an isomorphic (1:1) relationship between the constructional signature and the stored schema that facilitates immediate discovery of the relevant and correct schema (if one exists). Once the schema is retrieved from the construction repository, an instance of it is activated to process the morphosyntactic input string (the input clause in the syntaxÆsemantics direction, for example, will produce an output logical structure). The internal linking within the construction is then activated as an executable process to map between lexicon-morphology-syntax-pragmatics-semantics according to the language specific internal specification of the particular construction. Our position is that the constructions reside in a structured construction

68

Meaning Construction and Grammatical Inflection

repository rather than a simple inventory, and that these constructions inherit from the lexicon as part of the interface between morphology and lexicon. That is, we suggest that this construction repository has an internal structure whereby constructions within it draw on information stored in the lexicon. Within this kind of a perspective, the construction repository is a database of constructions with associated schemas stored, not as a simple list, but with a means to enable an appropriate construction and its schema to be retrieved, as required, according to certain criteria, and activated. (3)

Construction repository Signature-1 Construction-1 Input Workspace Constructionbody Output

Signature-2 Construction-2 Input Workspace Constructionbody Output

Signature-n Construction-n Input Workspace Constructionbody Output

We suggest that each construction in the constructional repository is stored with some means to enable it to be identified and found. Because this key must be based on some morphosyntactic information typically, a pattern of occurrence, we call this key the construction signature. This signature may have constraints associated with its signature tokens. Each construction is posited to have a unique construction signature associated with the schema body (cx-body) to identify it. Therefore, any appropriate input that matches this signature pattern of occurrence and thereby satisfies this construction signature, will select the relevant construction from the construction repository.

2.2. Construction Signature We have therefore added the constructional signature to the construction schema as a necessary requirement. We must also recognise that a construction has an input. For example, from syntax, a clause is received for processing when the construction schema is activated following a schema retrieval based on the uniquely identifying signature match. Once the construction is activated and the various criteria at the syntaxsemantics-pragmatics interfaces are applied within the construction in the tokens within the syntactic string, an output is generated. This will deliver,

Brian Nolan

69

assuming a construction executing in the syntax-semantics direction, a rich populated logical structure. Specifically then, we can consider the constructional schema as having a) a unique signature, b) an input string of tokens of particular types that are processed in the activated schema within c) the construction workspace, and d) an output of a particular kind (where the exact nature of these depends on the direction of execution of the activated schema from for example, syntaxÆ semantics or semanticsÆ syntax), e) a construction body [cx-body] that appropriately encodes the relationship between lexicon, morphosyntax, semantics and pragmatics as mediated via the RRG linking system. Within this perspective, we can usefully consider the construction as a type of grammatical object that can be uniquely identified, has internal structure, accepts an input and produces an output. The activation of the construction instance is sensitive to the direction of application, as we mentioned, with the bi-directional linking system of RRG. We can therefore construe a construction within RRG schematically in some more detail as follows (4), as a structured grammatical object.

2.3. Construction Workspace RRG posits the idea of a workshop for processing (Van Valin 2005: 161). We locate this notion as a motivated processing space, which we will instead call a workspace, within each construction construed as a grammatical object with a robust computational capability. In a consideration of the relationship between the lexicon, the inventory of constructions and how the various rules are applied within the RRG workshop where some particular linguistically significant output is produced by application of these rules, one can construe this workshop, in our account this is called a workspace, as a generalised way for RRG to differentiate between what is stored in the lexicon (in a rich lexical representation) versus what is computed online in real-time production. In this account, each construction comes with its own workspace in which the internal linking processes of the construction store directly within the construction, the workspace, the (linguistic/morphological/syntactic) items input to the construction. Then, the construction linking processes manipulate this workspace in a principled way. Crucially, in this view, the workspace is local to the particular construction, and an intrinsic part of the internal structure of the construction.

70

(4)

Meaning Construction and Grammatical Inflection

Construction as an object in RRG SIGNATURE: some pattern of […x y z….] INPUT: (token 1, token 2, token n) WORKSPACE: token 1, token 2, token n, [LS] CONSTRUCTION BODY Syntax PSA Semantics Linking Morphology Prosody Pragmatics OUTPUT: [LS] or ij

In this paper, we will express these morphological constructions schematically to better facilitate our discussion of morphological meaning construction and grammatical inflection within the layered structure of the word of Irish.

3. The Role of Morphology in Grammar Morphology is the part of linguistics that deals with the internal structure of words, word formation, and the identification and study of morphemes. As such, it has interfaces to the lexicon, semantics, syntax and phonology. A morpheme is a grammatically relevant piece of a word that cannot be divided into smaller meaningful parts. More specifically, a lexeme is a type of morpheme having lexical meaning rather than simple grammatical purpose. Morphemes that are not lexemes (words) (i.e. those that are bound) are called affixes. A lexeme is considered to have a schematic representation that represents it as a construction. Depending on their position, an affix can be a prefix, suffix, infix (5), and circumfix. Additionally, an affix can be sensitive to the category of the lexeme to which it is morphologically fusing (6). Circumfixes come in two parts where one part attaches to the front of the word and the other to the back such that they may be analysed as a prefix and a suffix that apply to a stem simultaneously (7). Affixes attach to stems and the most embedded stem in a complex word is the root (i.e. it is a simple stem). Roots belong to lexical categories i.e. nouns, verbs, adjectives, prepositions (8). One way inflection is realised is through affixes (5), while derivation involves the creation of one lexeme from another (6), (7) and (8).

Brian Nolan

(5)

gluaisteán “car” (N.SG)

(6) a. b. c. d. e.

banc “bank” (N.SG) baincéir “banker” (N.SG) bácáil “bake” (V.PST) scríobh “write” (V.PST) cáiréis “care” (N)

(7) a. b. c. d.

easaontas “disagreement”: aontaigh “agree” easaontas “dis+agree+ment” easaontaigh “dis-agree” aontú “agree-ment”

(8)

71

– gluaisteáin “cars” (N.PL) – – – – –

baincéir “banker” (N.SG) baincéireacht “banking” (VN) bácéir “baker” (N.SG) scríbhneoir “writer” (N.SG) cáiréiseach “careful” (ADJ, STEM) V.PST N V.PST N

Examples of words and affixes for meaning construction. fish Æ fishery NÆN iasc Æ iascaireacht answer Æ answerable freagra Æ freagrach island Æislander oileán Æ oileánach VÆV íoc Æ reamhíoc pay Æ prepay chéimí Æ fóchéimí graduate Æ undergraduate ADJ Æ ADJ gné Æ fóghne species Æ subspecies N Æ ADJ tarbh Æ tarbhach bull Æ bullish anam Æ anamach soul Æ soulful VÆN damhsaigh Æ damsa dance Æ dance ADJ Æ ADV brón Æ brónach sad Æ sadly

Word formation and lexeme formation both refer to meaning construction. Grammatical inflection is usually distinguished from morphological meaning construction by application of the following general criteria. (9)

Criteria for distinguishing inflection from morphological meaning construction a. The syntactic context is usually the determining factor for the application of inflectional morphology (i.e., is the subject of a verb SG or PL, M or F, 1/2/3rd?). b. The application of the processes of morphological meaning construction does not depend on the syntactic context but must relate to meaning construction in some way.

72

Meaning Construction and Grammatical Inflection

c.

Word level morphological meaning construction generally results in i) a change in lexical meaning, or ii) the lexical category is changed for a particular word.

We examine morphological meaning construction first, is § 4 following, and propose a model of the layered structure of the word (LSW) to support the discussion. Then, later in § 5, we discuss inflection of Irish within the model.

4. Morphological Meaning Construction 4.1. Word-level Morphological Meaning Construction In morphological meaning construction, a category may be regarded as a morphological construction, a schematically represented constructional device, that provides the “part of speech” category type of the word that it creates. That is, it has an input and an output. Lexemes, and complex newly created lexemes, have semantics, are recorded in the lexicon, and work with the linking system of RRG. These constructions have a structure where the input lexeme is morphologically fused with the category affix to produce a new lexeme as output. Morphological meaning construction creates new lexemes and usually changes the lexeme class whereas inflection creates different forms of the same lexeme for grammatical purposes, for example, case, gender agreement etc. The input to morphological meaning construction is a lexeme. We make some key assumptions (10) while motivating our account. (10)

Assumptions regarding morphological constructions in RRG a.

b. c.

The construction that includes a category affix contains skeletal structure (a “slot”) for an input lexeme whereby the “part of speech” of that lexeme is usually changed from one category type to another. Morphological meaning construction operates over one argument slot per meaning construction instance, while allowing for multiple iterations in a continuous meaning construction series. Compounds can be treated as equivalent to morphological meaning construction (including both endocentric and exocentric compounds)

Brian Nolan

d.

73

A general working assumption is that affixes that are not inflectional and therefore syntactically related must be in service of morphological meaning construction.

In our discussion, we use the phi symbol ij to denote some morphological constructional function that changes the type of the lexeme input to it to a specific output type. This morphological construction takes another lexeme as an input lexeme and typically produces a lexeme of a new category type. A lexeme is a morpheme that is semantically meaningful and is therefore encoded as a lexical entry using a morphologically relevant version of a logical structure and which may be represented appropriately with qualia. We generalise this morphological construction with the schema below, its morphological logical structure within a lexeme construction schema, as: (11)

[[ĮInput_Lexeme] ⊕ [ȕCategory_Lexeme]]ijtype

The input lexeme ([ĮInput_Lexeme]) may occur in a pre- or –post position, indeed any affix position, according to the language under study. In this paper we are concerned with Modern Irish and will elaborate on this language following. The ⊕ operator in the morphological construction schema represents morphological fusion of the (alpha) Į input argument lexeme with the (beta) ȕ category lexeme yielding a composite output lexeme with (phi) ij type. Later, we will see some examples of this in (16) and following. Morphological meaning construction is word formation that either results in a new word class or a new instance of the same word class. We use the term word in a general sense here, of course, realising that it is, strictly speaking, a lexeme that will be realised within the syntax as a word and additionally, this lexeme is likely to undergo inflectional marking for grammatical purposes, the nature of which depends on the syntactic context. Affixes attach to roots or stems to form new words (in the general sense indicated above). Sometimes we may not see an overt affix, and this is traditionally called zero-derivation (12). This is a word-formation process, a morphological process of meaning construction that changes the lexical category of a word without changing its phonological shape.

74

Meaning Construction and Grammatical Inflection

(12) a b. c.

Adj crua “hard” (ADJ) dona “bad” (ADJ) tapaidh “quick” (ADJ)

Æ Æ Æ Æ

Adv go crua “hard” (ADV) an dona “very bad” (ADV) an tapaidh “very quick” (ADV)

It is well understood that, cross-linguistically, morphological meaning construction and inflectional affixes are applied over a fixed order in which the attachment order of the various affixes is significant. We present a generalised representation in (10). (13) a. Morphological meaning construction: [prefix-lexeme1–[[[ROOT]–suffix-lexeme1]–suffix-lexeme2]]

b. Grammatical inflection: [prefix-morpheme1–[[[Lexeme]–suffix-morpheme 1]–suffix-morpheme2]]

In Figure 1, we suggest an initial conceptualisation of the structural representation of the layered structure of the word. The morphological constituent projection in the layered structure of the word is important for meaning construction and the constituents here include the input argument lexeme and the type-changing lexeme. One category denoting lexeme (represented by phi ij) will determine the word type, irrespective of the lexeme input to the construction argument slot in the morphological template. The output of one meaning construction may be input to another meaning construction. The fact that speakers of the languages of the world add phonological material to either end of a word can lead to complex words and complex structures. We present in Figure 2 a generalised view of the grammar and the lexicon, indicating the relationship between the lexeme store of morphemes with semantic meaning, and the repertoire of morphemes that encode grammatical information within morphological inventory. We view the morphological elements in the grammar, which are language specific, as having internal structure and divided into the areas of the lexeme store (if they are conceptually meaningful) and the morpheme store (for those morphemes that provide grammatical function). Additionally, the lexemes have structured entries in the lexicon.

Brian Nolan

75

Figure 1: The RRG layered structure of the word in meaning construction for V Æ VAdj

Figure 2: The morphological inventory and the lexicon in grammar

76

Meaning Construction and Grammatical Inflection

4.2. Meaning Construction Processes and Adjectives Languages have been reported to use different kinds of morphological strategies. Here, we illustrate, by way of example to support out account, the morphological strategies that Irish employs for creating adjectives from lexical items belonging to other categories like nouns and verbs, and also from adjectives themselves. Adverbs can also be created from adjectives in a type changing meaning construction. Irish meaning construction processes (Nolan 2009) are regular and productive. In Irish we can find meaning construction processes involving the following strategies (14). The lexeme template inventory for these is contained in the lexeme store is indicated in (15). The type determining lexeme template schematically has the format of [[Į] ⊕ [ȕ]]ijtype, as per (11) earlier where the input lexeme is denoted by Į. We indicate how these lexemes morphologically fuse in a type changing meaning construction and we provide examples in the subsections following. As before, the ⊕ operator represents the fusion within the meaning construction of the input lexeme representing some concept with a category type to yield a specific output type of lexeme, the newly created output lexeme. You will notice that we include compound formation in the scope of this (11cd) and (12cd) as a type of meaning construction. (14) a. b. c. d. e.

ADJ VA ADJ N ADV

Å Å Å Å Å

V ADJ +ADJ ADJ +N ADJ

: adjective plus intensifier prefix : verbal adjective : adjective + adjective compound : adjective + noun compound : adverb

(15) a. b. c. d. e.

ADJ VA ADJ N ADV

Å Å Å Å Å

intensifier+ADJ V ADJ +ADJ ADJ +N ADJ

Lexeme constructional template : [INTENSIFIER_ ] ⊕ [ADJ]]ĭADJ : [[V] ⊕ [ _SUFFIX]]ĭVA : [[Adj] ⊕ [ADJ]]ĭADJ : [[ADJ] ⊕ [N]]ijN : [[ADJ] ⊕ [[]]ĭADV

INTENSIFIER+ADJ

We show specific examples of each of these forms in (13) – (19), within the following subsections.

Brian Nolan

77

4.3. Construction of Adjectives from Intensifier + Adjective An intensifier can prefix an adjective with the resulting form being another adjective. That is, the sortal type of adjective is retained as an output from the construction. (14a) ADJ Å INTENSIFIER+ADJ : [INTENSIFIER_ ] ⊕ [ADJ]]ĭADJ (16)

a.

láidir strong: ADJ “Strong”

b.

ró-láidir too: INTENSFIER+strong: ADJ “Too strong”

4.4. Construction of Adjectives from Verbs Adjectives can be created from verbs in several languages by changing the verbs into participles or verbal adjectives. We can see an example (17) of the perfective passive, and the schema of its syntactic pattern. We see an additional example of this in (18) where the (verbal) adjective, created originally from a verb, is deployed in adjectival function. Table 1 shows some of the verbal and verbal adjective forms. (14b) VA Å V : [[V] ⊕ [ _SUFFIX]]ĭVA (17) a.

b.

Perfective Passive Bhí an nuachtáin leite agam. AUX-PST DET newspaper read at+1SG LIT: “Be the newspaper read at me” “The newspaper was read by me.” Syntactic schema for perfective passive [AUX NPundergoer VA ( agPrep NPactor ) … ]

(18) Thuit sneachta1 fríd an oidhche Fall-PST snow.M through DET night.F agus bhí sé1 curtha glan anois and AUX-PST 3SG.M1 spread clean now “Snow fell through the night and it was spread clean now.”

78

Meaning Construction and Grammatical Inflection

Table 1. Some verbal adjectives with lexeme construction schema: [[V] ⊕ [ _SUFFIX]]ĭVA Basic verb

Verb meaning

Verbal Adjective

VA Gloss

ól dún las croch stad bog ceap fág buail sín bris goid ite rith caith léim beir lig aitraigh ceannaigh coinnigh imigh

drink close light hang stop move catch leave hit stretch break steal eat run spend jump catch let change buy keep go

ólta dunta lasta crochta stadta boghta ceapta fágtha buailte sínte briste goidte ite rite caite léimthe beirthe lighte aithraite ceannaithe coinnithe imithe

drunk closed lit hung stopped moved caught left beaten stretched broken stolen eaten run spent jumped caught let changed bought kept gone

4.5. Construction of Adjective from [adjective + adjective] under Compounding Adjectives can compound with other adjectives where the resulting construction output form is still an adjective. (14c) ADJ Å ADJ +ADJ : [[ADJ] ⊕ [ADJ]]ĭADJ (19)

a. bán ‘white’ + dearg “red” Î bán-dearg “pink” b. Chuir sí cóiriughadh úr-nuaidh ar an dreisiúr Put-PST 3SG.F ornament.M fresh+new on DET dresser “She put a fresh ornament on the dresser.”

Brian Nolan

79

4.6. Construction of Nouns from [adjective + noun] under Compounding Adjectives can form compounds with nouns where the resulting form is a nominal (20) and (21). The function of qualifying a noun may be accomplished by either using an independent adjective with the noun or by forming an adjective-noun compound. When the adjective is compounded with a noun, the adjective always appears as a prefix on the noun, according to the lexeme constructional schema in for the construction of nouns from [adjective + noun] under compounding. (14d)

N Å ADJ +N : [[ADJ] ⊕ [N]]ijN

(20) (21)

Lexeme constructional schema: [[ADJ] ⊕ [N]]ijN Chuir eagna an tseanduine cúl mór orm. Put-PST prudence.M DET old+person.M back.M large on+1SG LIT: “The prudence of the old person put huge reserves on me.” “The old folk’s prudence made me very resourceful.”

4.7. Construction of Adverbs from Adjectives Adjectives lose some of their prototypical characteristics when used as adverbs. Adjectives in Irish may be used as adverbs without modification but their connotation changes in that adjectives indicate a permanent or normal state whereas, when used as adverbials in functions in relation to a verb, or to denote the manner of some action, they indicate a temporary or abnormal state. Some adverbs have the same form as adjectives when used in an adverbial function. We illustrate in (23) the composition of the lexeme construction for the adverbs in (22) from adjectives. (14e) ADV Å ADJ : [[ADJ] ⊕ [[]]ijADV (22)

a.

Tá

sé

ag obair go crua. at work-VN [to hard]ADV “He is working hard.” b. Bhí sí saidhbhir go deo. AUX-PST 3SG.F rich [to ever]ADV “She was always rich.” c. Chodháil mé cuiosach maith. Sleep-PST 1SG [fairly-ADV well-ADJ]ADV “I slept fairly well.” AUX-PRS 3SG.M

80

(23)

Meaning Construction and Grammatical Inflection

a. Lexeme construction schema: b. Lexeme construction schema: c. Lexeme construction schema:

[[PP] ⊕ [ADJ]]ĭADV [[PP] ⊕ [ADJ]]ĭADV [[ADV] ⊕ [ADJ]]ĭADV

4.8. Motivating the Semantics of Lexemes The semantics of both lexemes, and newly constructed lexemes of various category types, is an issue of interest to morphology and meaning construction and the nature of the lexicon architecture. One approach to dealing with complex issues (such as, for example, over-extension, underextension, polysemy) in the context of the morphology-semantics interface is given in Pustejovsky (1995) and Van Valin (2005). Pustejovsky (1995) proposes an account of nouns in the lexicon in terms of qualia theory. The headings under which Pustejovsky attempts to capture the attributes of nominals are: constitutive, formal, telic, and agentive. The manner in which they relate together is indicated in (24). A lexical entry would look like (25b) and some examples of the English “_er” suffix are given in (26). We will build on these in our discussion and we will see how Irish differs. (24)

Qualia theory a. Constitutive role QC: The relation between an object and its constituents, or proper parts. 1. Material 2. Weight 3. Parts and components b. Formal role QF: That which distinguishes the object within a larger domain 1. Orientation 2. Magnitude 3. Shape 4. Dimensionality 5. Colour 6. Position c. Telic role QT: Purpose and function of the object 1. Purpose that an agent has in performing an act 2. Built-in function or aim that specifies certain activities

Brian Nolan

81

d. Agentive role QA: Factors involved in the origin or “bringing about” of an object 1. Creator 2. Artefact 3. Natural kind 4. Causal chain We might remember at this point that a word is, strictly speaking, a lexeme with semantic meaning encoded in the lexicon, and which will be realised as a word in syntax and subject to inflectional marking according to the particular syntactic context. In contrast, an inflectional morpheme of grammatical consequence and relevance to syntax is a simple morpheme in the morpheme store in the morphological inventory. (25)

a. b

“The door opened.” BECOME be´ (open´ (the door(x), {QC, QF, QT, QA} ))

(26)

a. b. c. d. where:

Persons: baker, dancer, gambler, driver Animals: pointer, retriever Material objects: blotter, eraser, fertilizer, shutter Immaterial objects: reminder, thriller, eye-opener a: denotes a type of actor b: denotes a type of creature c: denotes a type of material entity in the lexicon d: denotes a type of immaterial entity in the lexicon

All of these have their respective underlying lexemes and descriptive qualia as represented in a lexicon entry. We can also note the different event types involved in the verb creation by the English agentive suffix “– er” (26). In RRG, we represent the event types using various aktionsarten types (27). (27)

Aktionsarten Types Verb Class State Activity Achievement Accomplishment

Logical Structure predicate´ (x) or (x, y) do´ (x, [predicate´ (x) or (x, y)]) INGR predicate´ (x) or (x, y) BECOME predicate´ (x) or (x, y)

What is interesting is that the Irish data is quite different (28) with respect to variation across the agentive suffix. To illustrate these differences,

82

Meaning Construction and Grammatical Inflection

we take similar categories of person, animal, and objects, where (a) denotes a type of actor, (b) denotes a type of creature, (c) denotes a type of material entity in the lexicon, and (d) denotes a type of immaterial entity in the lexicon. Immediately we can see the variation across the agentive suffices. We can intuitively state the behaviour of the English agentive suffix, “–er”, as a rule (29), before examine the corresponding rule for Irish (30). Indications of the rules and their important distinctions for Irish are provided in (30). The logical structure of a clause that uses the bacáil “bake” verb is given in (31), and báicéir “baker” then elaborates the x argument. The logical structures operate according to the Actor-Undergoer Hierarchy, indicated in Figure 4. The rule in (29), for English, can be contrasted with the more complex and detailed rule (30) for Irish, where different distinctions are made. (28) a.

Persons:

b.

Animals:

c.

Material objects

d.

Immaterial objects

(29) a. Rule:

bolscaire “announcer” spásaire “astronaut” fuascailteoir “liberator” léachtóir “lecturer” báicéir “baker” rinceoir “dancer” damhsóir “dancer” cearrbhach “gambler” tiománaí “driver” treoir “pointer” snáthaid “pointer” gadhar “retriever” páipear suite “blotter” scriosán “eraser” leasachán “fertilizer” comhla “shutter” cuimhneachán “reminder” scéinséir “thriller” oscailt súl “eye-opener”

__aire is suffix __aire __eoir __óir __éir __oir __óir __ach __aí __oir __aid __ar Phrase, no suffix __án __án __a __án __éir Phrase, no suffix

V of category type + English suffix “-er” Æ Agentive N to do with action of V Which represents ‘somebody or something whose function or characteristic is to perform a particular act’. b. Construction schema: [[V] ⊕ [ _er]]ijN

Brian Nolan

(30)

83

a. Rule: V of category type + Irish suffix set ‘{aire|eoir|óir|éir|oir|ach|aí|…}’ Æ Agentive N [+anim, +human] to do with action of V Which represents: ‘some PERSON whose function or characteristic is to perform a particular act’. b. Rule: V of category type + Irish suffix set ‘-{oir|id|ar|…}’ Æ Agentive N [+anim, -human] to do with action of V Which represents: ‘some CREATURE whose function or characteristic is to perform a particular act’. c. Rule: V of category type + Irish suffix set ‘-{án|a|éir|…}’ Æ Agentive N [-anim, -human] to do with action of V Which represents: ‘some THING whose function or characteristic is to perform a particular act’.

The inheritance hierarchy of relations can be seen within Figure 3. The construction is constrained in virtue of the ontological nature of the nominal entity and considers whether it is [+/-ANIM] and [+/-HUMAN]. Based on this constraint on the construction being resolved, an appropriate construction with the correct constructional signature is selected.

Figure 3: The inheritance relations across constructional schema

After this, the body of the construction processes the V lexeme and its logical structure, inherited from the lexicon as part of the operation of the RRG linking system, and determines the appropriate nominal to be selected within that logical structure. The steps are as follows: a. b.

Select the correct generalised construction from the construction repository with the appropriate construction signature. Identify the constraining attributes on the signature. Here, as Modern Irish is sensitive to these within the construction, these constraints will reflect whether the actor in the verbal act is within a configuration of [+/-ANIM] and [+/- HUMAN].

84

c. d. e.

Meaning Construction and Grammatical Inflection

Select via inheritance the appropriate V from the lexicon as the construction input lexeme. Identify the logical structure of the V and store in the construction workspace. In the body of the construction, mediated by the constraints, select the x in the logical structure such that x is in the LS: [do´(ACTOR: x, 0)] pred´ (UNDERGOER: y)].

For Irish, for example, the logical structure of bacáil “bake” will be as presented in (31) and the selection of the actor will follow the actor undergoer hierarchy (Figure 4). When lexemes combine within a wordlevel meaning construction, for example, involving type coercion then the lexemes combine according to the particular morphological construction schema. These morphological meaning constructions are in the lexeme inventory within the lexicon, with a morphologically relevant logical structure, which we schematically show in (32). As we have already illustrated, for an instance where an actor N is formed as one who does the action of a specific V, we have the following in (33), based on the constructional schema. We suggest that qualia theory might usefully be extended into this to provide a suitably motivated deep grain size of appropriate relevance to human language. The frame-based lexeme constructions, and their representation as schematic morphological logical structures, proposed within the lexeme store part of the lexicon for word level meaning construction is intended to be compatible with this use of a qualia ontology. An input lexeme, employed in meaning construction within a morphological construction with some associated schema, will have a fully generalised specification in qualia theory terms that will help to modulate the constraints on the construction and its signature. (31)

Semantic Logical Structure bacáil “bake” V-PST [do´(ACTOR: x, 0)] CAUSE [BECOME baked´ (UNDERGOER: y)]

(32)

Generalised lexeme construction [[ĭǹINPUT_LEXEME] ⊕ [ijȕ_LEXEME]]ĭTYPE

Brian Nolan

(33)

85

a.

Construction: [[Į:V] ⊕ [ _éir]]ijtype:N Construction output: Ȧ:ij, representing some agentive (omega:phi) word of type N meaning “doer of V” where V is inherited from the lexicon

b.

Construction: [[bácáil-V] ⊕ [ _éir]]ijtype:N Construction output: bácéir “baker” (N.SG)

c.

Construction: [[scríobh-V] ⊕ [ _éóir]]ijtype:N Construction output: scríbhneoir “writer” (N.SG)

Figure 4: The Actor-Undergoer Hierarchy (Van Valin 2005)

We have considered morphological meaning construction in the LSW in our discussion of the Irish word. Now, we consider grammatically relevant inflection within our model.

5. Grammatical Inflection 5.1. Expressing Morphosyntactic Information Through the use of inflection we can express morphosyntactic information including the abstract syntactic categories of tense, aspect, number, and case. In this section we look at some of these, in particular case, definiteness, gender, along with various eclipsis and lenition mutations on the Irish noun to motivate our account. We present in Figure 5 a schematic

86

Meaning Construction and Grammatical Inflection

of a structural representation of the layered structure of the word as it might occur for inflection with a variation for infixation in Figure 6.

Figure 5: The RRG layered structure of the word under syntactic inflection

The phi symbol ĭ denotes some lexeme of a particular type that is to be the host of the inflectional changes. It receives a morpheme in the inflectional modification, which may involve a prefix, postfix, infix or circumfix. For ease of reference, we encode the various morphemes as prefix and suffix respectively. We indicate a constituent projection and an operator projection. While the constituent projection is important for derivation, the operator projection indicated here is in the service of syntax and therefore related to inflectional morphology. Schematically, we represent this as follows in (34). (34)

[[Ȗ] ⊕ [LEXEMEĭTYPE] ⊕ [į]] Where: Ȗ and į are morphological affixes and ĭ is a lexeme of some category type from the lexicon. and ⊕ denotes a process of morphological fusion for inflectional marking

Brian Nolan

87

Inflection in morphology is related to encoding operators on the lexeme within the LSW and therefore, depending on the sort type of the lexeme, these operators will vary. For example, the operations on a noun in a noun phrase will include those operators indicated in (35), while the operators on a clause and verb will include those operations indicated in (36), as appropriate. (35)

NP, CoreN and NuclearN operators (From Van Valin 2005) NuclearN operator o Nominal aspect (count-mass distinction, classifiers in classifier languages) CoreN operators o Number o Qualification (quantifiers) o Negation NP operators o Definiteness o Deixis

(36)

Clause and verbal operators (From Van Valin 2005) NuclearV operator o Aspect o Negation o Directionals (predicate) CoreV operators o Directionals (participant) o Event quantification o Root modality o Negation (internal/narrow-scope) Clausal operators o Tense o Evidentials o Illocutionary force

The lexeme is the stem to which the inflectional processes apply. For nouns, we need to distinguish between inherent and assigned inflection. Nouns, and pronouns, for example, are marked for a particular gender in the speaker’s mental lexicon. Here, gender is inherent but, of course, gender considerations may vary across the ontologies within a cultural community as a categorisation or classification device. For any other lexical category that reflects the gender of nouns and pronouns, i.e.,

88

Meaning Construction and Grammatical Inflection

adjectives and verbs, gender cannot be inherent – it must be assigned. Number is not generally inherent; hence it is not marked in the lexicon and it has the status of a COREN operator in the Layered Structure of the Noun Phrase [LSNP] in RRG. An example of assigned inflection is case. Nouns and pronouns in the lexicon do not have case and this is assigned within the syntax as a consequence of the RRG linking system from semantics to syntax. One of the primary functions of inflection is to signal dependencies between syntactic units as, for instance, are found with the syntactic relationships over agreement. The dependencies expressed by inflection in one language may be expressed by other lexical and syntactic means, such as periphrasis, word order etc., in another language.

5.2. Inflectional Case Marking in Irish Irish has common case marking for nominative, accusative and dative nouns, with distinct marking for dative as the object of a preposition, genitive and vocative cases. The nominative, accusative and dative case is therefore known as common case in Irish. There is no marking on the nouns for nominative and accusative cases as they have the same morphosyntactic form. A noun is considered to be in the dative case if it is the object of any of the simple prepositions. (37)

D’imigh an cailín abhaile. PVP+go-PST DET girl home “The girl went home.”

(38)

Cheannaigh sé an teach. buy-PST 3SG.M.NOM DET house.ACC “He bought the house.”

Some representative examples are provided following in (39) and (40). We can note here the case encoding as a prefix on the N. The nature of the prefix depends on the particular N, and varies accordingly. A noun is usually in the genitive case if it is: a) the direct object of a verbal noun, b) proceeded by a compound preposition, c) proceeded by phrases that denote indefinite quantity, or d) governed by a possessive adjective such that possession or ownership is characterised. An example of each is provided in (41).

Brian Nolan

(39)

The set of simple prepositions ag “at” ar “on” as “out of, from” chuig “to, towards” de “from, off, of” do “to, for” faoi “under, beneath, about” gan “without” go “to i ‘in’” idir “between” le “with” ó “from” roimh “before, in front of” thar “over, across, past” trí “through” um “about, at”

(40)

a.

ag an ngeata at DET gate.DAT “at the gate”

b.

ar an gcathaoir on DET chair.DAT “on the chair”

c.

roimh an bhfear before DET man.DAT “before the man”

89

90

Meaning Construction and Grammatical Inflection

(41) Non-inflected form

a.

ag deanamh na hoibre at doing-VN DET work.GEN “doing the work”

obair

b.

i lár na seachtaine in middle DET week.GEN “in the middle of the week”

seachtain

c.

go leor codlata to plenty sleep.GEN “enough sleep”

codladh

d.

siopa Sheáin shop (of) Sean.GEN “Sean’s shop”

Seán

d’.

fear an tí man (of) DET house.GEN “man of the house”

teach

Here, in these examples, we can note that morphological shape of the word changes in genitive case with the addition of the genitive suffix. Additionally, the plural form of the determiner may occur with Ns in NP constructions that are feminine (42). Table 2 shows some examples of morphological markings on the adjective for agreement across common case (NOM, ACC) and genitive case. Table 2. Examples of morphological markings on the adjective for agreement (Christian Brothers 1997:61) Common SG

bán direach bacach leisciúil mall maith buíoch gnách

‘white’ ‘straight’ ‘lame’ ‘lazy’ ‘slow’ ‘good’ ‘grateful’ ‘usual’

Genitive SG.M báin dírigh bacaigh leisciúil mall maith buíoch gnách

Genitive SG.F

Common

báine dírí bacaí leisciúla maille maithe buíthí gnáthaí

bána díreacha bacacha leisciúla malla maithe buíocha gnácha

PL

Brian Nolan

91

5.3. Gender Determination of Irish Nouns Irish has two genders, masculine and feminine. Nouns that are neuter in other languages are encoded as masculine or feminine in Irish. Sometimes the gender may not be obvious (42). The gender class of a noun in Irish is generally determined by its morphological shape (43), in particular by the form of the ending on the word. (42)

a. b.

cailín ‘”girl” has masculine gender in Irish stail “stallion” has feminine gender in Irish

(43)

Gender determination of an Irish N by its word ending (M) N-word endings (F) N-word endings -áil, -úil, -aíl, -ilt, -int, -áint,-ádh, -án, -ch, -éad, -éal, éan, -eál, -éar, -eír, -eoir, - íocht, -aois, -is, -chan, -irt, ach,-acht, -úil, -úint, -cht, -éir, oir, -s, -in, -ún, -úr, -úir eoir, -óir, -úir, -lann, -eog, -óg

The morphological shape of the word will influence the gender assignment of the noun. In particular, nouns ending in a broad consonant tend to be masculine, e.g. gasúr “child”, while nouns ending in a slender consonant are feminine, for example, muintir “people”. As to be expected, a number of exceptions apply. Therefore, an important morphological characteristic of Irish is that Irish consonants can be either broad or slender. This quality is determined by the vowels that follow or precede it. A slender consonant or consonant group will have either an ‘e’ or ‘i’ vowel, from the vowel quadrangle, on both sides of it. Alternatively, this vowel will be after the consonant if it is the first consonant in a word, or before the consonant if it is the last in the word. Similarly, a broad consonant will always have an “a”, “o”, or “u” before and after it.

92

(44)

Meaning Construction and Grammatical Inflection

vbroad.consonant.vbroad

or

vslender.c.vslender.

Figure 6: The generalised RRG layered structure of the word for infixation

It is the presence of this pair of mirrored vowels occurring morphologically within the Irish word that determines whether the consonant between them is broad or slender. If a consonant or group of consonants is in the middle of a word, with vowels on either side of it, the two mirrored vowels must be of the same type i.e. either both broad or both slender, as in the VCV patterns. We indicate diagrammatically how this works in Figure 6, for infixation. For example, in bád “boat” [ba:d], the b and d consonants are both broad whereas in fear “man” [f'ar], the consonant f is slender but r is broad. In glúin “knee” [glu:n'], the consonant gl is broad but n is slender and in mín “smooth” [m'i:n'], the consonants m and n are both slender.

Brian Nolan

93

5.4. Definiteness and Indefiniteness of Nouns Irish has no indefinite determiner and the absence of a determiner associated with a noun is indicative that the reading of the noun is indefinite. There are two forms of the definite determiner in Irish. (45)

Determiner forms an “he”

ƒ ƒ ƒ

na “the”

• • •

(46)

Is used in the singular. With masculine and feminine nouns, it is used in nominative, accusative, dative and genitive cases. Is used the genitive singular with masculine nouns. Is used in the genitive singular with feminine nouns. Is used in the plural with feminine nouns. Is used in the plural with masculine nouns.

Determiner - example of forms across gender N.M N.F fear

a man

bean

a woman

an fear

the man

an bhean

the woman

na fir

the men

na mná

the women

The definite singular determiner can morphologically conflate (47) with certain prepositions and the question particle. When this occurs the vowel of the determiner is dropped in the morphological fusion. These behave in a somewhat similar manner to prepositional pronouns but do not conjugate, as do the prepositional pronouns. (47)

Determiner morphological conflation form : den “from the” : don “to the” : faoin “about the” : ón “from the” : cén “what the” :

rule de+an Æ den do+an Æ don faoi+an Æ faoin ó+anÆ ón cé+anÆ cén

[[de] ⊕ [an]] Æ [den]

94

Meaning Construction and Grammatical Inflection

5.5. Eclipsis and Lenition on Irish Nouns 5.5.1. Eclipsis of Nouns Eclipsis and lenition are two grammatically conditioned initial mutations of Irish. We give an illustrative example following (48) for Irish where the first letter of the noun is eclipsed under certain conditions. (48)

The noun is eclipsed after the definite determiner when it occurs: after the dative singular except : when the initial letter is “d” or “t” which remain unchanged

in the genitive plural

a). ag an bhfear at DET man “at the man” b). as an gcistín out.of DET kitchen “out of the kitchen” c). ar an mbus on DET bus “on the bus” d). ón teach from+DET house “from the house” e). ceol na n-éan music (of) DET.PL birds “song of the birds” f). rogha na mban choice (of) DET.PL women “the women’s choice” g). hataí na bhfear hats (of) DET.PL men “the men’s hats”

5.5.2. Lenition of Nouns The initial consonant of a noun of Irish undergoes lenition in certain circumstances. (49)

The noun receives lenition after the definite determiner (except with a noun having an initial d-, t-, s-), when the noun is SG.F.NOM or SG.F.ACC.

Brian Nolan

N.SG.F.NOM or N.SG.F.ACC

95

a). Feicim an bhean see-PRS+1SG.NOM DET woman.SG.F.ACC “I see the woman.” b). Tá

an fhairraige ag trá sea.SG.NOM at ebbing-VN “The sea is ebbing.”

AUX-PRS DET

(50)

The noun receives lenition after the definite determiner (except with initial d-, t-, s-) when the noun is SG.M.GEN N.SG.M.GEN

a). hata an fhir hat (of) DET man.SG.M.GEN “the man’s hat” b). ag moladh an bhuachalla at praising-VN DET boy.SG.M.GEN “praising the boy”

We can now summarise our discussion regarding eclipsis and lenition as it occurs in Modern Irish.

5.5.3. Summary of the Eclipsis and Lenition Mutations The mutations that occur for both eclipsis and lenition across all forms can be summarised as follows (51).

96

(51)

Meaning Construction and Grammatical Inflection

Summary of the morphological mutations (based on Carnie 2008) Sounds that undergo the changes

Lenition1 (L1)

Sounds that don’t undergo changes

Sounds that undergo exceptional changes

b, m, f, t, d, s, c, g

-

Lenition2 (L2) Eclipsis1 (E1)

b, m, f, c, g p, b, f, t, d, c, g, vowels p, b, f, c, g p, b, f, c, g vowels vowels

The letters d, t, s do not lenite after a word ending in n. The sequences sp, st, sc never lenite. t, d -

s Æ ts -

t, d, vowels t, d, vowels -

s Æ ts -

Mutation name

Eclipsis2 (E2) Eclipsis3 (E3) T-prefixation (T) H-prefixation (H)

5.6. Inflections on Synthetic and Analytical Verb Forms Irish is a strict VSOX language, in common with the other Celtic languages. As Irish uses synthetic as well as analytical verb forms, the PSA ‘subject’ pronoun is often conflated as a suffix inflection on the verb (52a’) and (52b’). Additionally, tense encoding can cause adjustments to occur to the initial and final parts of the verb (52c). (52)

Synthetic and analytical verb forms a. Tagann mé anois come-PRS 1SG.NOM now “I arrive now.” a’.

Tagaim anois come-PRS+1SG.NOM now “I arrive now.”

b.

Molann mé an cailín praise-PRS 1SG.NOM DET girl.ACC “I praise the girl.”

Brian Nolan

b’.

Molaim an cailín praise-PRS+1SG.NOM DET girl.ACC “I praise the girl.”

c.

Thainigh an cailín come-PST DET girl.NOM “The girl arrived/the girl came.”

97

6. Discussion We have outlined some considerations applicable to a characterisation of the morphology of the Irish word. We considered both elements of wordlevel morphological meaning construction and grammatical inflection, situated within our conceptualisation of the RRG layered structure of the word in a way that is compatible with the layered structure of the noun phrase and the layered structure of the clause. We considered morphological meaning construction within an account that discussed how a new lexical category was formed, and how lexical meaning was modified in some way. We claim that lexemes are to be found within the lexicon, in a lexeme store, and formulated as concepts and considered as constructions with internal structure and which are of significance to the grammar. The role that constructions as grammatical objects play in a lexicalist functionalist model of grammar is of considerable theoretical importance (Diedrichsen 2010, this volume) and, if one accepts that they do play a role, raises the questions as to the nature of the construction and it’s role and additionally, where do these reside coherently and cohesively within the model. We posit that these constructions reside within a constructional repository with structured links to the lexicon. Therefore, this paper is a contribution to that debate with respect to syntactic and morphological constructions and RRG. We have indicated in our conceptualisation of the layered structure of the word how it would cater for both morphological meaning construction and inflection. In contrast to lexemes, grammatical morphemes are stored as part of a language specific morpheme inventory, similar in some respects to the syntactic inventory. In relation to the morpheme part of the morphological inventory, we schematically represented the application of the inflectional process through the inflectional morpheme constructional schema.

98

(53)

Meaning Construction and Grammatical Inflection

Morphological meaning construction and associated schema: [[ijĮInput_Lexeme] ⊕ [ijȕ_Lexeme]]ijtype Where: ⊕ is an operator in the morphological constructional template that represents morphological fusion of the Į input lexeme with the ȕ category lexeme to yield a composite output lexeme with ij type.

(54)

Grammatical inflection and associated schema: [[Ȗ] ⊕ [Lexemeĭtype] ⊕ [į]] Where: Ȗ and į are grammatical affixes and ĭ is a lexeme of some type from the lexicon and and ⊕ is an operator in the morphological constructional template that represents morphological fusion.

We have motivated a lexeme construction with generalised schematic representation as morphological logical structures that will reside in a frame-based entry in the relevant part of the lexeme store in the lexicon, and we illustrated lexeme type changing processes via this generalised morphological constructional schema. We have shown, for the layered structure of the word, how morphological meaning construction relates a lexeme to another lexeme. We represented lexemes as frame-based morphological constructions with their own inherent logical form, classified by type, where the types are assumed to inherit generalised properties in the lexicon. A preliminary consideration of the architecture of the lexicon construed an inheritance network within RRG has been discussed by Gottschalk (2010) and a discussion of the nature of categories of importance in RRG and other linguistic models is to be found within Rauh (2010). In this discussion, we have provided some insight into the morphology of Irish and its related grammatically relevant inflectional processes where morphemes and morphological change are in the service of syntax and grammatical purpose. Not all inflection is to do with pre- or postfixes. We saw that sometimes that infix adjustment is necessary too. We suggested how the layered structure of the word might apply in these situations as an extension of the operation of these constructions that deal with pre- and postfix adjustments according to the licensed adjustment pattern. Again, we have looked at a lexeme store and a morphological inventory for morphemes, proposed a layered structure of the word for Irish, applicable to both morphological meaning construction and grammatical inflection. In doing this, we provided significant examples of authentic Irish data to support our discussion of the layered structure of the

Brian Nolan

99

word in respect of morphological meaning construction and grammatical inflection within morphology in an RRG perspective.

List of Abbreviations ACC AUX DAT DET F GEN M

accusative auxiliary dative determiner feminine genitive masculine

NOM PL PRS PST SG VA VN

nominative plural present past singular verbal adjective verbal noun

References Aronoff, Mark and Kirsten Fudeman. 2005. What is Morphology? Oxford: Blackwell. Booij. Geert. 2010. Construction Morphology. Oxford: Oxford University Press Carnie, Andrew. 2008. Irish Nouns. Oxford: Oxford University Press. Christian Brothers. 1997. New Irish Grammar. Dublin: C.J. Fallon-Mount Salus Press. Diedrichsen, Elke. 2010. Towards a reconsideration of Constructional Schemas in RRG: Are all constructions driven by “constructions”? In Nakamura, Wataru, ed., Proceedings of the 10th International RRG Conference (RRG 2009), 52-79. Available at: http://wings.buffalo. edu/linguistics//people/faculty/vanvalin/rrg/ProceedingsofRRG2009_0 2.pdf —. [This volume]. The theoretical importance of Constructional Schemas in RRG. Gottschalk, Judith. 2010a. Storage of linguistic knowledge in the mental lexicon: An approach within Role and Reference Grammar. In: ITB Journal Issue 19: 20-45. Available at: http://www.itb.ie/files/journal/ issue-19.pdf. —. 2010b. N+N compounds in German: an analysis within Role and Reference Grammar. In: ITB Journal, Issue 20: 76–101. Available at: http://www.itb.ie/files/journal/issue-20.pdf. Katamba, Francis. 1993. Morphology. London: Macmillan Mac Congáil, Nollaig. 2004. Irish Grammar Book. Indreabhán, Conamara. Galway, Ireland: Cló Iar Chonnachta

100

Meaning Construction and Grammatical Inflection

Nolan, Brian. 2009. The functions, semantics and syntax of the adjective in Irish. In Guerrero, Lilián, Sergio Ibáñez, Valeria A. Belloro, eds., Studies in Role and Reference Grammar. México: Universidad Nacional Autónoma de México. —. 2010a. The layered structure of the word in Modern Irish. In Nakamura, Wataru, ed., Proceedings of the 10th International RRG Conference (RRG 2009), 52-79. Accessible at: http://wings. buffalo.edu/linguistics/people/faculty/vanvalin/rrg/ProceedingsofRRG 2009_02.pdf —. 2010b. The layered structure of the Modern Irish word: an RRG account of derivational and inflectional morphology. Paper delivered at a colloquium at the Heinrich Heine University of Dusseldorf, Germany. —. 2010c. The GET construction of Modern Irish and HibernoEnglish: passive and recipient variations. Paper presented at ‘Construction Grammar: New Perspectives for the Study of German and English’ - International Conference at the Christian-Albrecht University Kiel, Germany. —. 2010d. Constructional polysemy and argument realisation with the Irish GET verb. Paper presented at Workshop on Variation and Change in Argument Realization, University of Naples in Capri and Naples, Italy. —. 2011a. Characterising the effects of the referential hierarchy on Modern Irish ditransitive constructions. Paper presented at the International Workshop on Referential Hierarchies in three-participant constructions. The University of Lancaster, England. —. 2011b. Constructions as grammatical objects: a new perspective on constructions in RRG. Paper presented at the International Conference on Role and Reference Grammar on Functional Linguistics: Grammar, Communication & Cognition. Facultad de Letras, at the Pontificia Universidad Católica de Chile, Santiago de Chile. —. 2011c: Constructional templates at the morphology-lexicon interface – meaning and the layered structure of the Irish word. Paper presented at the workshop on Meaning construction at the crossroads of grammar, cognition and communication. Societas Linguistica Europaea, University of La Rioja, Logroño, Spain. —. 2011d: Characterising the effects of the referential hierarchy on Modern Irish ditransitive constructions. To appear in Anna Siewierska and Eva van Lier, eds., Referential Hierarchies in Three-participant Constructions, Online-Journal “Linguistic Discovery”:

Brian Nolan

101

http://journals.dartmouth.edu/cgibin/WebObjects/Journals.woa/2/xmlp age/1/issue —. In press a. Resolving ambiguity in the realisation of the Irish GET construction. To appear in Johanna Barðdal, Michaela Cennamo, and Elly van Gelderen, eds., Argument Structures: The Naples/Capri Papers. Amsterdam and Philadelphia: John Benjamins. —. In press b. The Structure of Irish: A Functional Account. London: Equinox Publishing Co. Ó Siadhail. Mícheál. 1996. Learning Irish. New Haven and London: Yale University Press. —. 1989. Modern Irish. Cambridge: Cambridge University Press. Pustejovsky, James. 1995: The Generative Lexicon. Cambridge, MA: The MIT Press. Rauh, Gisa. 2010. Syntactic Categories: Their Identification and Description in Linguistic Theories. Oxford: Oxford University Press. Spencer, Andrew. 1991. Morphological Theory. Oxford: Blackwell. Spencer, Andrew and Arnold M. Zwicky (eds.) 1998. The Handbook of Morphology. Oxford: Blackwell. Van Valin, Robert D., Jr. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax. Cambridge: Cambridge University Press.

PART 2. LEXICON, ARGUMENT STRUCTURE, AND CONSTRUCTION

ROMANCE ANTICAUSATIVES: A CONSTRUCTIONIST RRG APPROACH ROLF KAILUWEIT UNIVERSITY OF FREIBURG

Abstract In this paper, I will account for anticausative constructions as being located in the centre of an active-passive continuum ranging from real reflexive constructions to reflexive passives. My claim will be that the continuum can be modelled using the activity-calculus elaborated in Kailuweit (2005) for different classes of verbs of emotion. It will be shown that RRG Constructional Schemas can be used to account for the language-specific particularities of anticausative constructions in different Romance languages.

Keywords Anticausative variation, constructional schemas

Romance

languages,

activity-calculus,

1. Introduction In the below examples which express the eventualities of “hurt” and “save”, Brazilian Portuguese [BP] 1 shows an interesting contrast where the presence or absence of the pronominal element in anticausative constructions is concerned. (1)

1

hurt (arg1, arg2) (e.g. John hurt Louis with a knife) BP: João machucou Luis. John hurt Louis

The BP data were presented by Ulrich Reich at the workshop Linking Romance, FU Berlin, February 2009.

Rolf Kailuweit

(2) (3) (4) (5)

105

hurt (arg1, arg2) (e.g. Louis fell from the roof of his house) BP: Luis machucou. Louis hurt hurt (arg1, arg2) (e.g. Louis hurt himself preparing sushi) BP: Luis se machucou. Louis himself hurt save (arg1, arg2) (e.g. Louis has saved the file) BP: Luis salvou o arquivo. Louis saved the file save (arg1, arg2) (e.g. somebody–it doesn’t matter who–has saved the file) BP: O arquivo salvou rápido. The file saved quickly

As we will see later in detail, all Romance languages allow for transitive/bare intransitive pairings as illustrated by BP salvar “save” although they occur with other verbs. Hence, they seem to permit a causative alternation as already described for English, 2 without a pronominal element stemming from Latin sibi “self”. In addition, there is evidence in French, Italian, Spanish, and also in German–to mention a non-Romance language–for verbs of the machucar (“hurt”)-type in BP, i.e. verbs allowing for both a pronominal and bare intransitive construction. However, as far as the other Romance languages and German are concerned, an alternating verb does not denote the eventuality of “hurt”. As for linking from semantics to syntax, the appearance of the pronominal construction in BP seems to be triggered by a semantic feature of the argument which I will informally call “responsibility”. While neither Luis in (2) nor o arquivo “the file” in (5) are responsible for the change of state, Luis in (3) unintentionally performs an action to himself that leads to an injury. Onomasiologically, this linking problem seems to be the reason why this evident semantic difference is not coded in other Romance languages. Although examples parallel to those with machucar “hurt” are not especially easy to detect in BP, formally identical pairings found in French and Italian do not seem to reflect the same straightforward semantic contrast stipulated for BP machucar. Section 2 of this paper gives an overview of the phenomenon of causative and anticausative alternation, focusing on the concept of internal versus external causation. In Section 3, it will be shown that in the Romance languages, three formal classes of alternating verbs can be 2

Cf. Levin and Rappaport Hovav (1995: Ch.3).

106

Romance Anticausatives: A Constructionist RRG Approach

distinguished: verbs marking the anticausative with a pronominal element, verbs with a formally identical causative and anticausative variant and finally verbs allowing for both a marked and unmarked anticausative construction. Section 4 focuses on the semantics of the two constructions. They will be placed on an activity continuum, ranging from the real reflexive construction to reflexive passives. My claim in Section 5 will be that the continuum can be modelled using the activity-calculus elaborated in Kailuweit (2005) for different classes of verbs of emotion. Finally, it will be shown in Section 6 that RRG Constructional Schemas can be used to account for the language-specific particularities of anticausative constructions in different Romance languages. Hence, this paper will contribute to the current discussion about to what extent RRG can be considered a variant of Construction Grammar.

2. The (Anti-)Causative Alternation The given BP data appear to be instances of what has been called the causative alternation (Smith 1970; Levin and Rappaport Hovav 1995). As Sánchez López (2002: 80) points out, the transitive variant should be interpreted as a two-place predicate with a cause argument filling the subject position and an undergoer argument realized as a direct object. The undergoer appears as the subject of the intransitive or pronominal construction, while the cause argument can be realized as an adjunct. The data from BP has been slightly modified to ensure that the transitive and the intransitive or pronominal variant denote a corresponding type of eventuality: (6)

a. b.

(7)

a. b.

(8)

a. b.

A queda machucou Luis. the fall hurt Louis Luis machucou por causa da queda Louis hurt because of the fall João machucou Luis sem querer. John hurt Louis by accident Luis se machucou por causa de uma Louis himself hurt because of a inadvertência preparando sushi. inattentively preparing sushi A tecla Control+S salvou o arquivo. the button Control+S saved the file O arquivo salvou com a tecla Control+S the file saved with the button Control-S

Rolf Kailuweit

107

While the transitive variant is generally labelled the causative construction, there is a lot of terminological inconsistency as far as the intransitive variants are concerned. In accordance with Schäfer (2008),3 I will call both the pronominal and the bare intransitive construction anticausative. As Labelle (1992: 397) and Sánchez López (2002: 86) point out, the widespread term inchoative for the intransitive variant(s) is misleading, because only a few of the intransitives refer to the beginning of an action or an action soon to take place. According to Sánchez López (ibid.), arrepentirse “regret” is inchoative, hundirse “sink” denotes the beginning and the completion of a process and secarse “dry” is not inchoative: (9)

a. Juan se arrepintió. (“John started to feel sorry.”) b. El barco se hundió. (“The ships started to sink/The ship sank.”) c. La ropa se secó. (“The clothes dried.”)

BP machucarse is clearly not inchoative either. Luis se machucou in (7b) does not mean that he started to hurt himself, but that the action is already completed. Haspelmath’s (1993) documenting of the whole range of morphological procedures to mark either the causative or the anticausative construction was the beginning of an extensive, but fruitless discussion on which of the two constructions should be considered the basis construction disseminated in the literature. Van Valin (2005: 46) emphasized that there is evidence for both derivation directions. It has been claimed (cf. Van Valin ibid.) that derivation of the anticausative construction by means of a pronominal element is the default case for Romance languages. Nonetheless, the bare intransitive construction has also been documented in all Romance languages.4 For BP, a tendency towards the loss of the pseudo-reflexive has been proven (cf. Nunes 1995; Carvalho 2006). However, the direction of the derivation will not be discussed in this paper. In order to determine the verbs that enter the causative alternation, Levin and Rappaport Hovav (1995: 90s) distinguish between internally and externally caused eventualities. The distinction is based on Smith 3

The term “anticausative” was used in Spanish by Moreno Cabrera (1984). See Rothemberg (1974), Centineo (1995), and Carvalho (2006) for the French, Italian, and European Portuguese data, respectively. Sánchez López (2002: 89) cites a study of Levy (1994), who attests 25 cases of bare intransitives out of 300 Spanish causative-anticausative pairs.

4

108

Romance Anticausatives: A Constructionist RRG Approach

(1970: 107), who describes eventualities denoted by verbs like open or break as controlled by an external cause, whereas for eventualities such as those denoted by verbs like laugh or play control cannot be relinquished. Levin and Rappaport Hovav (1995: 91) reject the term “control”, because the distinction can be extended to non-agentive verbs, e.g. verbs of emission. The semantics of internal and external causation can be captured as follows: With an intransitive verb describing an internally caused eventually, some property inherent to the argument of the verb is “responsible” for bringing about the eventuality […] Unlike internally caused verbs, externally caused verbs by their very nature imply the existence of an “external cause” with immediate control over bringing about the eventuality described by the verb: an agent, an instrument, a natural force, or a circumstance. (Levin and Rappaport Hovav 1995: 91s)

According to Levin and Rappaport Hovav (1995), the distinction between internal and external causation should lead to two clear-cut classes: externally-caused intransitives should behave as unaccusatives, but internally-caused intransitives as inergatives. While the former allow for a transitive causative variant and reject embedding with MAKE,5 the latter should only extend their argument structure by means of the MAKE construction.6 As Labelle (1992) claims for French and Sánchez López (2002) for Spanish, in Romance languages the difference between internal and external causation should be reflected by the absence or presence of the pseudo-reflexive clitic. Labelle’s argumentation is based on Rothemberg (1974), who developed a similar idea using the terms external and internal causation:7 … pour les verbes intransitifs […] l’élément lexical assumant la fonction de sujet est le siège de l’action, du processus qui est vu comme se développant organiquement à partir de lui et rien qu’à partir de lui 5

Sánchez López (2002: 90) illustrates this behaviour with the following examples: i. El viento cerró la ventana (“The wind closed the window”) / ii.*El viento hizo cerrar la ventana (“The wind made the window close”). 6 Levin and Rappaport Hovav (1995: 90) cite the following examples taken from Smith (1970: 107): i. The green monster made Mary shudder / ii. *The green moster shuddered Mary. 7 Levin and Rappaport Hovav (1995: 85, 294) only cite Rothemberg (1974) in an endnote as a general source of inspiration.

Rolf Kailuweit

109

grâce à ses qualités inhérentes […] Quant aux verbes en construction pronominale […] l’élément lexicale assumant la fonction du sujet est également le siège de l’action, du processus qui est vu comme se développant à partir de lui. Le se de la récession marque pourtant que ses qualités inhérentes […] sont insuffisantes à elles seules pour permettre la réalisation de l’action ou du processus. (Rothemberg 1974: 67) (“...for intransitive verbs [...] the lexical element assuming the function of the subject is the host of the action, i.e. of the process that is seen as developing organically out of it and only out of it because of its internal qualities [...] As far as the pronominal construction is concerned [...] the lexical element assuming the function of the subject is the host of the action, too, i.e. of the process that is seen as developing out of it. However, the se of the recession marks that its internal qualities [...] are insufficient on their own to bring about the realisation of the action or the process.” [Translation R.K.])

However, the data is more complex. There is no clear-cut line between the two groups of verbs. The syntactic behaviour of bare intransitives in the Romance languages and in German is not consistent. We have already seen that in several Romance languages, bare intransitive verbs also enter the causative alternation. There are a considerable number of verbs (such as Italian aumentare “increase” or German kochen “cook” that only allow for the bare intransitive construction and show a transitive variant.8 While Labelle (1992) assumes that alternating French bare intransitives are unergatives, the choice of BE for compound tenses that we observe for most alternating bare intransitives in Italian and German hints at their unaccusativity. For modern Spanish, which no longer displays auxiliary variation, Mendikoechea (1999) and Sánchez Lopez (2002) also tend to classify alternating bare intransitives as unaccusatives. Nonetheless, unaccusativity is neither a sufficient nor a necessary condition for entering the causative alternation. While Italian unaccusative aumentare “increase” allows for a causative variant, the corresponding German unaccusative verb steigen does not. On the contrary, the German unergative kochen “cook” permits a causative construction.

8

In fact, Rothemberg (1974) seems to be the first systematic approach to alternating verbs in French. She leaves no doubt that the alternation is possible with and without pronominal marking.

110

Romance Anticausatives: A Constructionist RRG Approach

(10)

a. I prezzi (*si) sono aumentati. (“The prices increased.”) b. Gianni ha aumentato i prezzi. (“John increased the prices.”) (Folli 2002: 117s) c. Die Preise sind gestiegen. (“The prices increased.”) d. *Hans hat die Preise gestiegen. (“John increased the prices.”)

(11)

a. Die Kartoffen haben (*sich) gekocht. (“The potatoes cooked”) b. Hans hat die Kartoffeln gekocht. (“John cooked the potatoes.”)

In addition, the fact that the same verb in different languages appears in different anticausative constructions casts doubt on a clear correspondence between pronominal marking and external causation: (12)

a. Se hundió el barco. (“The ship sank.”) (Menikoechea 1999: 1587) b. La barca è affondata. (“The ship sank.”) (Folli 2002: 73)

In the following example, the relationship between marking and external causation should be acknowledged as a tendency. This can be formalized by Hypothesis I: (13)

HYPOTHESIS I: If the difference between internal and external causation is coded, external causation will be coded with the pronominal construction and not with the bare intransitive construction.

3. Verb Classification While the ongoing change in BP makes it difficult to classify the alternating verbs into different subgroups according to the morphological marking of their anticausative variant, such verb lists have been compiled for French, Italian and Spanish. Based on a detailed study of Rothemberg (1974), Labelle distinguishes three subgroups for French:9

9 See Heidinger (2009, 2010) for a diacronic approach that focuses on the quantitative relationship between the two constructions in the history of French. According to Heidinger, the pronominal construction spread at the expense of the bare intransitive (“labile”) construction; however, about 300 “labile” verbs still exist in Modern French.

Rolf Kailuweit

111

Verbs which necessitate the pseudo-reflexive construction only (Class A): (14)

s’alléger “become lighter”, s’abêtir “turn into a morin”, s’agrandir “become bigger”, s’allourdir “become heavier”, s’amaigrir “become thinner”, s’améliorer “improve”, s’américaniser “become Americanised”, s’assécher “dry out”, s’engourdir “become numb”, s’enkyster “encyst”, s’humidifier “become humid”, se calcifier “calcify”, se civiliser “become civilised”, se couvrir “become covered”, se nuancer “nuance”, se poisser “become sticky”, se rabougrir “shrivel up”

Verbs which entail only or mainly the bare intransitive construction (Class B): (15)

cuire “cook”, durcir “harden”, éclater “burst”, fonder “melt”, grandir “grow”, grosser “grow bigger”, maigrir “grow thinner”, moisir “mould”, pourrir “rot”, sécher “dry”, vieillir “age”

Finally, verbs that are can be utilised in both constructions (Class C): (16)

caraméliser “carmelise”, élargir “widen”, enfler “swell”, épaisser “thicken”, gonfler “inflate”, noircir “blacken”, ramollir “soften”, refroidir “cool”, rétrécir “get narrower”, rougir “become red”

For Italian, Folli (2002) establishes the same classes as Labelle (1992) for French,10 but the verbs belonging to a class do not always denote the same eventualities. (17)

(18)

10

Class A: alterare “alter”, aprire “open”, arrotolare “roll up”, bagnare “wet”, capovolgere “turn up side down”, chiudere “close”, dividere “divide”, estendere “extend”, restringere “shrink”, rompere “break”, rovesciare “overturn”, sbriciolare “crumble”, sfilacciare “fray”, svegliare “wake up” Class B: affondare “sink”, allungare “lengthen”, aumentare “increase”, bollire “boil”, cambiare “change”, diminuire “decrease”, guarire “heal”, invecchiare “age”, maturare “mature”, migliorare “improve”

Cavalho (2006) affirms the existence of the same three classes for European Portuguese.

112

(19)

Romance Anticausatives: A Constructionist RRG Approach

Class C: asciugare “dry”, bruciare “burn”, congelare “freeze”, cuocere “cook”, fondere “melt”, gelare “freeze”, ingrandire “enlarge”, raffreddare “cool”, riscaldare “heat”, sgonfiare “deflate”

Schäfer (2008) proves the existence of the same three classes in German. Based on the Corpus of Written Language COSMAS-II at the IDS Mannheim, he accounts for 17 German verbs that allow for both the marked and unmarked construction (Class C).11 As far as I know, the existence of Class C in Spanish has not been explicitly highlighted in the literature. Mendikoetchea (1999: 1604s), who cites the classification of Labelle (1992), does not give a corresponding list of Spanish verbs alternating in their anticausative construction. She only hints at aclarar “clear up” and oscurecer “darken”, which alternate with different object classes. (20) (21)

a. El día ha aclarado. (“The day has cleared up.”) b. El día ha oscurecido de repente. (“Suddenly, the day has clouded over.”) a. El jersey se ha aclarecido por sí solo. (“The sweater has bleached itself.”) b. Su fama se ha oscurecido por sí solo. (“His fame has dimmed on its own.”) (Mendikoechea 1999: 1599)

The small group of verbs of ‘internal causation’ 12 compiled by Mendikochea (1999: 1597) could be a starting point to proving the existence of the three classes in Spanish: (22)

adelgazar “slim”, arder “burn”, aumentar “increase”, crecer “grow”, empeorar “worsen”, empezar “start”, empobrecer “impoverish”, encoger “shrink”, enfermar “sicken”, enfriar “cool”, engordar “fatten up”, ennegrecer “blacken”, enrojecer “flush”, ensanchar “widen”, ensordecer “deafen”, envejecer “age”, florecer “blossom”, germinar “sprout”, hervir “boil”, mejorar “ameliorate”, palidecer “pale”

11 Schäfer (2008: 24-28) also refers to Greek examples which make the same distinction between three classes through the presence or absence of non-active morphology. I will not discuss Greek data in this paper. 12 The term is taken from Levin and Rappaport Hovav (1995). I will come back to its semantic implications in the next section.

Rolf Kailuweit

113

Most of them seem to be Class B verbs, however some are clearly Class A verbs, which always take the pronominal element as their intransitive variant: (23)

enfriarse “cool”, enfermarse “sicken”

Sánchez López (2002: 89-93) includes some verbs of internal causation in this list: (24)

ascender “ascend”, menguar rejuvenecer “rejuvenate”

“wane”,

finalizar

“finish”,

Nonetheless, some of the verbs allow for both the pronominal and the bare intransitive construction. Hence, Class C verbs also exist in Spanish: (25)

adelgazar(se) “slim”, empeorar(se) “worsen”, empobrecer(se) “impoverish”, encoger(se) “shrink”, engordar(se) “fatten up”, ennegrecer(se) “blacken”, enrojecer(se) “flush”, ensanchar(se) “widen”, envejecer(se) “age”, hervir(se) “boil”, mejorar(se) “improve”, rejuvenecer(se) “rejuvenate”

More examples of verbs can be found in the Spanish online Corpora (CREA of the Real Academia Española and Corpus of Español by Mark Davies). Not all verbs are equally accepted as instances of Class C. The use of hervirse was not found in CREA and is not accepted by the Dictionary of the Real Academia, but has been confirmed by sources in Argentina and Venezuela. For them, the pronominal variant tends to have the reading of “boil over”: (26)

El agua se hirvió. (“The water boiled (over).”) [non standard]

Sánchez López (2002: 91) denies the existence of a transitive variant for enrojecer “flush”, but this variant is attested in CREA: (27)

a. *La vergüenza enrojeció a Juan. (“The shame flushed John.”) (Sánchez López 2002: 91) b. Se degustó una potente sangría, que enrojeció a más de uno los carrillos. (CREA) (“They tasted a strong sangria that made many of them rosy-cheeked.”)

114

Romance Anticausatives: A Constructionist RRG Approach

To summarise this section: The contrast illustrated by BP machucar “hurt” seems to correspond to a largely attested complexity concerning the anticausative variant of verbs entering the causative alternation. French, Italian, German and also Spanish exhibit verbs that allow for both the marked and the unmarked anticausative variant. In the next section, I will account for the differences between the two constructions in order to shed some light on the given semantic contrast for BP machucar.

4. Semantic Contrasts between the Two Anticausative Constructions 4.1. Participant orientated Differences The BP data hint at a semantic difference concerning the “responsibility” of the argument denoted by the subject of two anticausative constructions. First of all, it must be stated that there is little evidence for the existence of the contrast shown in the machucar/salvar-examples, both in BP and in the other (Romance) languages. As far as BP is concerned, my sources would not exclude the pronominal construction with inanimate subjects, as the given salvar-data would suggest: (28)

a. A roupa (se) rasgou. (“The clothes tore.”) b. A mesa (se) quebrou. (“The table broke.”) c. A comida (se) estragou. (“The food rotted.”)

It would be fruitless to cite counter-evidence in order to prove that the contrast does not exist. Based on the given data, a more interesting point could be made by Hypothesis II: (29)

HYPOTHESIS II: If a contrast of responsibility is coded, the subject of the pronominal construction of a single verb will denote an argument more responsible for the outcome of the action than the argument denoted by the subject of the bare intransitive construction of the same verb.

Besides the given data for BP, there is evidence in other languages to prove the accuracy of Hypothesis II. The contrast is lexicalised with French brunir “tan, brown” and with the corresponding German bräunen. (30)

a. Les baigneurs brunissent. (“The bathers got a tan.”) b. Les baigneurs se brunissent au soleil. (Rothemberg 1974: 160)

Rolf Kailuweit

115

(“The bathers got a tan in the sun.”) c. Le poulet brunit. (“The chicken browned.”) d. *Le poulet se brunit. The pronominal construction implicates an intentionally acting subject. On the contrary, the bare intransitive construction seems odd with such a subject. (31)

a. Paul s’est bruni pour paraître plus séduisant. (“Paul got a tan to look more attractive.”) b. ??Paul a bruni pour paraître plus séduisant.

For other verbs that naturally take a human subject, these differences are subtle, but still detectable for native speakers. The following example from CREA suggests a certain responsibility of the person denoted by the logical subject of the infinitive: (32)

… él prefirió empobrecerse poco a poco y continuar operando la cafetería, a venderla y quedarse rico. (CREA) (“He preferred getting poorer and poorer keeping the coffee shop open to becoming rich by selling it.”)

My sources found the sentence less acceptable without the pseudoreflexive. In the same sentence they preferred the pronominal construction in the following pair: (33)

a. ??Juan empobreció por su propia culpa. (“John impoverished by his own fault.”) b. Juan se empobreció por su propia culpa.

Another piece of evidence stems from embedding constructions with MAKE. Folli (2002: 105ss) discusses the embedding under causative fare “make” for Italian. In Italian, the embedded infinitive appears obligatorily without the clitic: (34)

a. La mamma fa pettinare Maria. (“Mum makes Maria comb her hair.”) b. *La mamma fa pettinarsi Maria. (Folli 2002: 104)

For Class A verbs, Folli (2002: 105) observes that only a “transitive” reading is available.

116

Romance Anticausatives: A Constructionist RRG Approach

(35)

Maria fece chiudere la finestra => Maria made someone close the window => *Maria closed the window (Folli 2002: 105)

On the contrary, for Class B verbs the test with embedding under fare shows ambiguity. (36)

Maria fece diminuire la temperatura. => a. Maria made the temperature decrease. => b. Maria had someone decrease the temperature. (Folli 2002: 122)

The same holds for Class C verbs: (37)

Maria fece fondere il cioccolato. => a. Maria made the chocolate melt. => b. Maria had someone melt the chocolate. (Folli 2002: 131s)

The Spanish Class C verb empobrecer “impoverish” seems to show the same contrast as the Italian fondere: hacer empobrecer in the a-example is ambiguous. The construction could denote a process caused directly by the government itself or indirectly towards another causer, e.g. the banking sector. The sentence in b is unambiguous: the government is interpreted as the direct causer. (38)

a. El gobierno hizo empobrecer a la clase media. => The government impoverished the middle class. => The government had someone impoverish the middle class. b. El gobierno empobreció a la clase media. => The government impoverished the middle class

Notice that Spanish, in contrast to Italian, allows for the embedding of real reflexives without clitic-deletion: (39)

La madre le hizo peinarse a María. (“Mum made Maria comb her hair.”)

The embedding of Class A verbs under hacer “make” is not very common, but well attested.

Rolf Kailuweit

(40)

117

a. Hizo buena carrera, pero en la última curva el patín se le fue mucho, lo hizo abrirse un poco y separarse de la baranda. (CREA) (“He did a good race, but in the final curve the skate broke away, made straighten out a little bit and separate from the boards.”)

This example is not ambiguous: indirect causation can be excluded. As far as empobrecerse “impoverish” is concerned, analogous examples read quite strangely in the eyes of my sources. (41)

#El gobierno hizo empobrecerse a la clase media. (“The government made the middle class impoverish.”)

The example seems to suggest the interpretation that the government indirectly caused the impoverishment of the middle class, and yet that the middle class is being held accountable for its own impoverishment the same time, an interpretation that does not make a lot of sense. Note that with empobrecerse intentional acting is not completely excluded. In this case, the pronominal construction could be considered a real reflexive construction and allows for embedding under hacer: (42)

a. Jesús se empobreció por amor a vosotros. (2 Cor 8,9) (“Jesus made himself poor out of love for you.”) b. Diós hizo empobrecerse a Jesús. (“God caused Jesus to make himself poor.”)

To sum up, the preference for the embedding of the bare infinitive with a Class C verb such as empobrecer(se) shows that a certain responsibility for her unintentional behaviour can be attributed to the human being denoted by subject of these verbs when they occur in the pronominal construction. This corroborates the findings for BP machucar “hurt”. Schäfer (2008) puts forward another complex piece of evidence for the same contrast. In German, only bare intransitive constructions allow for a dative causer. The contrast is subtle, but clear when we exclude the dativus incommodi reading by adding aus Versehen “by mistake”: (43)

a. Die The b. Die The

Flasche bottle Flasche bottle

leerte emptied lief leaked

sich itself mir to me

mir. (*aus Versehen) to me by mistake aus Versehen aus. by mistake out

118

(44)

Romance Anticausatives: A Constructionist RRG Approach

a. Dem Chemiker verflüssigte sich (*aus Versehen) das Präparat. To the chemist liquidified itself by mistake the compound b. Dem Chemiker schmolz aus Versehen das Präparat. To the chemist melted by mistake the compound (Schäfer 2008: 47)

As we have already seen, bare intransitives entering the causative alternation also permit a causative adjunct: (45)

a. Marie a rougi de honte. (Labelle 1992 : 401) (“Mary flushed with shame.”) b. El paro aumenta con la inflación. (Sánchez López 2002: 89) (“The unemployment increases with the inflation.”)

If we consider these examples in line with Labelle (1992) and Sánchez López (2002) as instances of denoting internally-caused eventualities, internal causation does not seem to be incompatible with, in the words of Labelle (1992: 401), “the existence of a triggering cause”, but rather the process that the subject-argument undergoes “is presented as happening autonomously”. The dative causer could be considered another syntactical coding of a triggering cause. Nonetheless, causative adjuncts can also be found with the pronominal construction: (46)

a. Die Tür öffnete sich durch einen Windstoß The door opened itself by a blast (Schäfer 2008: 127) b. Toda la flota se hundió con el temporal All the fleet SE sank with the storm (Sánchez López 2002: 80)

Hence, causation expressed by the dative causer has to be of another quality. Notice that this construction is possible neither in French nor in BP. On the contrary, in Spanish and Italian the dative causer construction can be attested for both marked and unmarked anticausative constructions. According to Schäfer (2008: 69), the following examples can not only have a dativus incommodi reading, but a dative causer as well: (47)

a. A Juan se le rompieron las gafas to John SE DAT-CL broke the glasses b. A Juan le hirvió la leche to John DAT-CL boiled-over the milk

Rolf Kailuweit

(48)

119

a. A Francesca le ruppe il vaso di cristallo per errore to Francesca DAT-CL broke the cut glass by mistake b. A Francesca è bollito fuori il latte per errore to Franesca is boiled over the milk by mistake (Schäfer 2008: 69)

Schäfer (2008: 66s) concludes that there is no semantic blocking at all, but a syntactic reason for the incompatibility of the dative causer with the pronominal construction in German: the dative causer and the pseudoreflexive pronoun would have to be projected to the same specifier position, while the Spanish and Italian pseudo-reflexives are clitics that are attached to the verbal head. In this paper, I will not discuss the different syntactic representations that Schäfer’s study focuses on. If we consider the German data only, the contrast could be semantically interpreted in light of the machucar-alternation. If the pseudo-reflexive could hint at a higher degree of responsibility of the argument expressed by the subject in the sense of Hypothesis II, then there would be a cognitive dissonance with the dative causer that denotes the same semantic function. Of course, the Spanish and Italian data do not confirm such an inference. My Spanish sources corroborate the data given by Schäfer for romperse “break”. As far as hervir “boil” is concerned, they clearly prefer the non-standard pronominal variant, indicating that only this variant expresses the unintended change of state of boiling over; the intransitive variant is not seen as meaningful. (49)

a. A Juan se le hirvió la leche. (“John involuntary caused the milk to boil over.”) b. #A Juan le hirvió la leche. (“John involuntary caused the milk to boil.”)

The anticausative alternation here is interpreted aspectually, a reading which has been put forward in the literature and will be discussed in greater detail in the next section.

4.2. Aspectual Differences Although the given data from BP does not suggest any aspectual difference, this topic will be briefly dealt with, since aspectuality has played a major role for the analysis of the causative alternation in Romance languages. Zibri-Hertz (1987) claims that the two constructions

120

Romance Anticausatives: A Constructionist RRG Approach

differ in telicity in French: marked anticausatives are supposed to be telic and unmarked anticausatives atelic. Sánchez López (2002: 86) adopts this analysis for Spanish. Following Fernández Lagunilla and de Miguel (1999), she highlights the fact that marked anticausatives denote a change of state of cumulative character and focus on a resulting state. This could be proved by the existence of a resultative construction with estar + participle that is excluded for at least some unmarked anticausatives: (50)

a. Está hundido. (Sánchez López 2002: 86) (“It is sunk.”) b. *Está ascendido. (ibid.: 92) (Lit. “It is ascended.”)

Folli (2002) bases her syntactic analysis of Italian anticausatives on aspectual grounds: (51)

(52)

a. La sedia si è rota in un secondo. (“The chair broke in a second.”) (Folli 2002: 97) b. ?La sedia si è rotta, ma non è rotta. (“The chair broke, but it is not broken.”) (ibid.: 98) a. La temperatura è diminuita per un’ora. (“The temperature decreased for an hour.”) (ibid.: 118) b. La temperatura è diminuita, ma non è diminuita completamente. (“The temperature decreased, but it has not totally decreased.”) (ibid.: 120)

This data is to some degree problematic. As “in un secondo” is not a span of time in which a cumulative change of state could occur, rompersi “break” is classified as an achievement rather than as an accomplishment, but is still a telic verb. Nonetheless, Schäfer (2008) cites a set of counterexamples to Folli’s claim: (53)

(54)

a. La sua sfera d’influenza si è estesa per molti anni. (“Her sphere of influence expanded for years.”) b. La sua sfera d’influenza si è estesa, ma non si è estesa completamente. (“Her sphere of influence expanded, but is not yet completely expanded.”) a. La caffettiera è scoppiata in un secondo. (“The coffee machine exploded in a second.”) b. ??La caffettiera è scoppiata, ma non è scoppiata completamente. (“The coffee machine exploded, but it has not completely exploded.”) (Schäfer 2008: 17s)

Rolf Kailuweit

121

Labelle also states: It is not the case that the intransitive construction expresses the imperfectivity or atelicity of the process, while the reflexive construction expresses the perfectivity of [sic] telicity of the process. (Labelle 1992: 195)

But otherwise she assumes that: The aspectual flavour of the distinction between the intransitive construction and the reflexive construction is an indirect consequence of the semantics of the constructions. (ibid.)

My sources clearly prefer the marked construction of Class C verbs in telic contexts as we have already seen for hervir “boil” versus hervirse “boil over”: (55)

a. Con el gobierno peronista la clase media se ha empobrecido en pocos años. (“With the peronist government, the middle class impoverished in a few years.”) b. ?Con el gobierno peronista la clase media ha empobrecido en pocos años. c. La clase media ha empobrecido pero todavía no es pobre. (“The middle class has impoverished, but is not yet poor.”) d. ?La clase media se ha empobrecido pero todavía no es pobre.

Their intuitions can be captured by Hypothesis III: (56)

HYPOTHESIS III: If an aspectual contrast is coded with the same verb, the pronominal construction will be interpreted as telic, and the bare intransitive construction as atelic.

4.3. Interim Conclusion Starting from the BP data, slight but consistent semantic tendencies between the two anticausative constructions have been observed. The three hypotheses which have been put forward throughout this paper are interconnected in an organic way. The pronominal construction is– paradoxically–the more causative one of the two anticausative constructions. The different degrees of causativity of the subject argument of intransitive constructions can be organized into a hierarchy:

122

(57)

Romance Anticausatives: A Constructionist RRG Approach

a. REAL REFLEXIVES WITH INTENTIONAL DIRECT CAUSER Example: Jesus se empobreció por amor a vosotros. (“Jesus made himself poor out of love for you.”) > b. MARKED ANTICAUSATIVES WITH INTENTIONAL INDIRECT CAUSER

Example: Paul s’est bruni pour paraître plus séduisant. (“Paul got a tan to look more attractive.”) > c. MARKED ANTICAUSATIVES WITH UNINTENTIONAL RESPONSIBLE CAUSER

Examples: Luis se machucou por causa de uma imprudência preparando sushi. (Lit. “Louis hurt himself because of carelessness preparing sushi.”)/Juan se empobreció por su propia culpa. (“John impoverished by his own fault.”) > d. INTERNAL CAUSATION Examples: Luis machucou por causa da queda. (Lit. “Louis hurt because of the tumble.”)/La voiture ralentit. (“The car slowed down.”) (Rothemberg 1974: 192) > e. GENERIC EXTERNAL CAUSATION Example: La production se ralentit. (“The production slowed down.”) (Rothemberg 1974:192) > f. REFLEXIVE PASSIVES Example: Se firmaron los convenios. (“The treaties were signed.”) I would like to make two comments. First, the hierarchy reflects contrasts as far as they are coded. That means that the BP example o arquivo salvou rapido (“the file saved quickly”) should not be

Rolf Kailuweit

123

interpreted as an instance of internal causation. BP shows a tendency towards loose pronominal constructions. Therefore, it may resemble English in that the difference between external and internal causation is (generally) not expressed by using morphological marking. Second, the preference found in the Spanish data for combining the marked construction with a dative causer (see above: A Juan se le hirvió la leche) obviously contradicts the German data if we take them as evidence for a higher degree of causativity in the marked construction. 13 The Spanish data should be seen, too, in the light of general tendency, a tendency that is exactly opposed to the one observed for BP. In Spanish, the pronominal construction seems to extend to almost all anticausatives. If the contrast of the two constructions can be used to code differences in causativity–as we have seen for empobrecer(se) “impoverish”–this coding seems to neutralise in contexts of a dative causer interpreted as an unintentional external non-generic causer. The pseudo-reflexive does not code any higher degree of causativity in theses contexts, but operates as an aspectual marker. The fact that higher perfectivity is in line with higher causativity (expressed by the dative causer) could be explained by the general affinity of causation and telicity. Most causative constructions are telic, although caused atelic states (the dogs frightened the boy) or activities (The girl bounced the ball around) have been attested (cf. Van Valin and LaPolla 1997: 97).

5. The Theta-System Revisited The modelling of the semantic findings of the last section within a linking theory will be effected in two steps. First, I would like to shed some light on the formal coding of semantic roles. My considerations are inspired by Rozwadowska (1988) and Reinhart (2002). The different degrees of causativity will be coded by means of the feature-value calculus developed for verbs of emotion in Kailuweit (2005). In the second step, the calculus will be integrated in a linking theory based on Role and Reference Grammar (Van Valin and LaPolla 1997; Van Valin 2005). In Kailuweit (2005), I suggested a feature-based graduation to determine the argument’s activity degree for verbs of emotion. The calculus is based on three features–cause [c], mental [m] and result [r]– which are denoted by the values +, í and ±. [c] is a strong proto-agent feature, [m] is a weak proto-agent feature and [r] is a strong proto-patient 13

As far as I can see, there is no syntactic explanation of theses facts in line with Schäfer’s (2008) approach, either.

124

Romance Anticausatives: A Constructionist RRG Approach

feature. The presence [+] of a strong feature will duplicate the value of the presence of a weak feature. If an argument is underspecified for one feature [±] the value will be half. +c

±c

íc

+m

±m

ím

ír

±r

+r

4

2

0

2

1

0

0

í2

í4

Figure 1: Theta-features and values

Reinhart (2002) assumes that all transitive verbs taking a [+c, ±m]argument as subject enter the causative alternation. She does not distinguish between different variants of anticausatives. Therefore, she proposes a simple rule for the reduction of the external argument. The only distinction she makes is in the nature of the object-argument of the transitive construction. (58)

a. openacc ([+c], [ícím] ---> Re(open)[ ícím] b. worryacc ([+c], [íc+m] ---> Re(worry)[íc+m]

It is beyond the focus of this paper to comment on verbs of emotion although some of them show interesting parallels to change of state verbs which demand causative alternation. In Kailuweit (2005), I argued that the Romance pseudo-reflexive constructions of the worry-type are instances of anti-passive constructions, an argument which will not be elaborated on in this paper. As far as the causative alternation is concerned, Reinhart’s analysis is by far too coarse-grained. Nonetheless, her analysis of real reflexives will serve as a starting point for a finer grained formalisation of the results already obtained in the last section. (59)

a. shaveacc([+c+m]1,[ícím]2: Lucie shaved him. b. Rs(shave)([+c+m]1): Max shaved. (Reinhart 2002)

For Reinhart (2002), real reflexives always reduce the object argument. Since the focus of this paper is not on real reflexives, I will concur with this analysis without further discussion, 14 adding just the missing third feature [r] which will be necessary for the distinction of finer linking differences: 14

This analysis is not without problems for Italian (cf. Müller 2005).

Rolf Kailuweit

(60)

125

a. machucaracc ([+c+mír]1,[ícím+r]2): João machucou Luis. b. Rs(machucar)([+c+mír]1): Luis se machucou intencionalmente.

The transitive construction shows the highest possible activity contrast. A prototypical controlling agent and causer ([+c+mír]) brings about a change of state in a prototypical causatively affected patient [íc-m+r]. The semantics of the agent is not changed by the reflexive construction. However, the semantics of the marked anticausative variant of brunir “tan” cannot be described correctly if we adopt Reinhart’s proposal to simply delete the argument realised as the subject of the transitive variant. The subject of the marked anticausative construction is obviously not a prototypical patient, but an intentional indirect causer of the change of state. Although he would not get a tan without the help of the sun, he is exposing his body intentionally. Thus the object argument of the transitive construction is promoted semantically. The person denoted by this object becomes an indirect causer of the change of state that his body undergoes when the argument is realised as the subject of the intransitive construction. Notice that the argument takes the value ± for the r-feature. This corresponds to the fact that the argument undergoes a change of state, but is not prototypically causatively affected. We are dealing with an anticausative construction. Hence, neither the c-feature nor the r-feature can acquire a +-value: (61)

a. bruniracc ([+c+mír]1,[ ícím+r]2): Le soleil a bruni Paul. b. brunir ([±c+m±r]1): Paul s’est bruni pour paraître plus séduisant.

The human being denoted by the subject of machucar-se “hurt” in BP does not intentionally bring about the change of state that he or she is the victim of. Nonetheless, his or her responsibility can be coded by the ±value for the m-feature: (62)

a. machucaracc ([+c+mír]1,[ícím+r]2): João machucou Luis. b. machucar ([±c±m±r]1): Luis se machucou por causa de uma inadvertência preparando sushi.

Unmarked anticausative constructions denoting internal causation do not code any responsibility of the human being or object denoted by the subject-argument. This is coded by a negative value for the m-feature. It could be argued that a human being has intrinsic properties that lead to

126

Romance Anticausatives: A Constructionist RRG Approach

injuries when falling down, especially from a roof as suggested by the given BP data: (63)

a. machucaracc ([+c+mír]1,[ícím+r]2): João machucou Luis. b. machucar ([±cím±r]1): Luis machucou por causa da queda.

A slightly more passive argument appears in the subject position of marked anticausatives. As these constructions typically denote external causation, the c-feature takes a negative value. Note that the telicity effect is not caused by a higher value for the r-feature. This is in line with the argumentation of Labelle (1992) and Schäfer (2008). Telicity is only an indirect effect of the higher patienthood of the argument due to its cfeature value: (64)

a. ralentiracc ([+c+mír]1,[ícím+r]2): La crise économique ralenti la production. b. ralentir ([ícím±r]1): La production se ralentit.

Leaving the semantic field of anticaustives, reflexive passives take a prototypical patient subject [ícím+r], an argument that ranks at the lower pole of the activity hierarchy. It is identical to the object argument of the corresponding active transitive construction: (65)

a. firmaracc ([+c+mír]1,[ícím+r]2): Los ministros firmaron los convenios. b. firmar ([ícím+r]1): Los convenios se firmaron.

The following summarises the causativity-degree hierarchy: [+c+mír]

6

Luis se machucou intencionalmente.

[±c+m±r]

2

Paul s’est bruni pour paraître plus séduisant.

[±c±m±r]

1

Luis se machucou por causa de uma inadvertência preparando sushi.

[±cím±r]

0

Luis machucou por causa da queda.

[ícím±r]

í2

La production se ralentit.

[ícímír]

í4

Los convenios se firmaron.

Figure 2: Anticausative constructions in the centre of an activity hierarchy

Rolf Kailuweit

127

Note that the same verb machucar “hurt” appears in three different constructions. This raises the question of whether the linking between the semantics of each sentence and the syntactic construction is determined by different lexical entries or whether the constructions interfere to convey a certain interpretation. This question will be dealt with in the final section of this paper.

6. A Constructionist RRG Approach A projectionist approach has been put forward by Levin and Rappaport Hovav (1995): The meaning of a sentence and the linking of the arguments of the predicate is determined by the meaning of the predicate. Hence, each verb should be classified according to whether it denotes externallycaused or internally-caused eventualities. On the contrary, a constructionist view of the anticausative alternation can already be found in Rothemberg (1974): Un même verbe peut également se trouver dans l’une ou l’autre construction avec le même sujet […] on peut rencontrer aussi bien l’effort ralentit que l’effort se ralentit, dans le premier cas, par l’emploi de l’intransitif, le processus est considéré comme un processus naturel se réalisant uniquement grâce aux qualités inhérentes du sujet […] alors que dans le second cas on tient pour indispensable l’intervention d’autres facteurs… (Rothemberg 1974 : 192) (“The same verb can equally enter one or the other construction with the same subject [...] We can find just as well l’effort ralentit as l’effort se ralentit, in the first case, the uses of the intransitive marks that the process is considered as naturally developing only because of the inherent qualities of the subject […] whereas in the second case we consider the intervention of other factors as indispensable.” [Translation R.K.]).

In this section, I will sketch a constructionist approach to linking based on the semantics-to-syntax algorithm of RRG (Van Valin 2005). The organisation of RRG stipulates the existence of constructional schemas that intervene during the linking process in order to spell out the languagespecific characterisation of a special construction. It always consists of a list of syntactic, morphologic, semantic and pragmatic particularities that characterise the specific construction in a specific language.

128

Romance Anticausatives: A Constructionist RRG Approach

Figure 3: Organisation of RRG (Van Valin 2005: 134)

My claim is that the two anticausative variants can be represented as constructional schemas that interact with the information stored in lexical entries. As one could infer from the lists given in Section 2, for many verbs there exist arbitrary, language-specific restrictions that determine whether the verb takes on the pronominal construction, the bare infinitive construction or both. These idiosyncrasies are stored in the lexicon. Verbs that can potentially enter both constructions are lexically underspecified. Their lexical entry does not have to be doubled, but the semantic differences will be conveyed by a specific constructional schema. Morphologically, syntactically, and semantically, 15 anti-causative constructions are language-specific phenomena, but not universal syntactic devices. They can, however, be considered voice constructions and imply a universal argument reduction rule: (66)

General characterisation of basic voice constructions: a. Privileged Syntactic Argument [PSA] modulation voice: permits an argument other than the default argument […] to function as PSA. b. Argument modulation voice: gives non-canonical realization to a macrorole argument. (cf. Van Valin 2005: 116)

The general linking algorithm from semantics to syntax stipulates the selection of an appropriate syntactic template for a semantic representation following these general rules:

15

Pragmatic differences will not be dealt with in this paper.

Rolf Kailuweit

(67)

129

a. Syntactic template selection principle: The number of syntactic slots for arguments and argumentadjuncts within the core is equal to the number of distinct, specified argument positions in the semantic representation of the core. b. Language-specific qualifications of the principle in (a): 1. All cores in the language have a minimum syntactic valence of 1. 2. Argument-modulation voice constructions reduce the number of core slots by 1. […] (Van Valin 2005: 130)

Therefore, for both constructions, the general rules stipulate the realisation of the non-Actor-argument as subject (=PSA) and the reduction of the core slots to one. The constructional schemas intervene in order to spell out the language-specific characterisations of a voice construction. In the following outline, I will restrict myself to anticausative constructional schemas for Spanish, French and BP: (68)

(69)

(70)

CONSTRUCTION: Spanish marked anticausatives SEMANTICS: Actor: deleted; Undergoer [ícím+r]: promoted => {[ícím±r] v [±cím±r] v [±c±m±r] v [±c+m±r]}; telecity: (+) SYNTAX: PSA: Undergoer; Intransitivity: unaccusative; Dative causer: + MORPHOLOGY: Marking: reflexive PRAGMATICS: … CONSTRUCTION: Spanish unmarked anticausatives SEMANTICS: Actor: deleted; Undergoer [ícím+r]: promoted => [±cím±r]; telecity: (í) SYNTAX: PSA: Undergoer; Intransitivity: unaccusative; Dative causer: (+) MORPHOLOGY: Marking: í PRAGMATICS: … CONSTRUCTION: French marked anticausatives SEMANTICS: Actor: deleted; Undergoer [ícím+r]: promoted => {[ícím±r] v

130

(71)

(72)

(73)

16

Romance Anticausatives: A Constructionist RRG Approach

[±c±m±r] v [±c+m±r]}; telecity: (+) SYNTAX: PSA: Undergoer; Intransitivity: unaccusative; Dative causer: í MORPHOLOGY: Marking: reflexive PRAGMATICS: … CONSTRUCTION: French unmarked anticausatives SEMANTICS: Actor: deleted; Undergoer [ícím+r]: promoted => [±cím±r]; telecity: (í) SYNTAX: PSA: Undergoer; Intransitivity: unergative;16 Dative causer: í MORPHOLOGY: Marking: í PRAGMATICS: … CONSTRUCTION: BP marked anticausatives SEMANTICS: Actor: deleted; Undergoer [ícím+r]: promoted => {[ícím±r] v [±c±m±r] v [±c+m±r]}; telecity: (+) SYNTAX: PSA: Undergoer; Intransitivity: unaccusative;17 Dative causer: í MORPHOLOGY: Marking: reflexive PRAGMATICS: … CONSTRUCTION: BP unmarked anticausatives SEMANTICS: Actor: deleted; Undergoer [ícím+r]: promoted => {[ícím±r] v [±cím±r]}; telecity: (í) SYNTAX: PSA: Undergoer; Intransitivity: unaccusative;18 Dative causer: í MORPHOLOGY: Marking: í PRAGMATICS: …

According to Labelle (1992). According to Carvalho (2006). 18 According to Carvalho (2006). 17

Rolf Kailuweit

131

As far as language-specific particularities are concerned, the constructional schema accounts for syntactical differences, such as the lack of a dative causer construction in French and BP or the unergativity of French unmarked anticausatives, if we follow Labelle’s argumentation. However, this paper is mainly concerned with language-specific semantics. The different accessible semantic interpretations of the two constructions in each language can be inferred from the schemas. The difference between internal vs. external causation can be captured by the different kinds of undergoer promotion. While the unmarked anticausatives in Spanish and French convey a type of undergoer promotion [±cím±r] that corresponds to the semantics of internal causation, marked anticausatives in French are restricted to the other three types of undergoer promotion. On the contrary, Spanish marked anticausative constructions allow for expressing eventualities of internal causation, too. In BP it is the other way round: unmarked anticausative constructions denote both internal causation and external causation, as illustrated by the data given for saltar.

7. Conclusion (Anti-)causative variation is more complex than the well-known distinction of internal vs. external causation suggests. Anticausatives differ from passives by the fact that their argument is semantically promoted in comparison with the object argument of the transitive construction, denoting neither a prototypical patient nor a prototypical causer. In many languages, anticausatives can be divided into three classes, which is the case for Spanish. The semantics of anti-causativity are an effect of the constructions that an individual verb can enter. Marked and unmarked constructions vary according to the language in the degree of causativitypromotion they convey to the undergoer-argument of the transitive construction. It has been proven that RRG’s constructional schemas are powerful tools for dealing with a fine-grained semantic and syntactic analysis of anticausative constructions.

References Carter, Richard. 1988. On Linking: Papers by Richard Carter. Lexicon Project Working Papers 24. Cambridge, MA: Center for Cognitive Science, MIT. Carvalho, Guilhermina Mendes de. 2006. Inacusatividade e ergatividade na fala rural do PB. In Revista Inventário. 5. ed., Avaialble at:

132

Romance Anticausatives: A Constructionist RRG Approach

http://www.inventario.ufba.br/05/05gcarvalho.htm. Centineo, Giulia 1995. The distribution of si in Italian transitive/inchoative pairs. Proceedings of Semantics and Linguistic Theory 5: 54-71. Fernández Lagunilla, Marina and Elena de Miguel. 1999. Relaciones entre el léxico y la sintaxis: adverbios de foco y delimitadores aspectuales. Verba 26: 97-128. Folli, Raffaela. 2002. Constructing Telicity in English and Italian. Unpublished Ph.D. Dissertation, University of Oxford. Haspelmath, Martin. 1993. More on typology of inchoative/causative verb alternations. In Bernard Comrie and Maria Polinsky, eds., Causatives and Transitivity, 87-120. Amsterdam and Philadelphia: John Benjamins. Heidinger, Steffen. 2009. The persistence of labile verbs in the French causative-anticausative alternation. Paper presented at the 19th International Symposium on Theoretical and Applied Linguistics, Workshop “Typology of Labile Verbs: Focus on Diachrony”, Thessaloniki, 3-5 April, 2009. Heidinger, Steffen. 2010. French Anticausatives: A Diachronic Perspective. Berlin and New York: Walter de Gruyter. Kailuweit, Rolf. 2005. Linking: Syntax und Semantik französischer und italienischer Gefühlsverben. Tübingen: Niemeyer. Kliffer, Michael D. 1982. The case of the missing reflexive. Hispania 65: 424-427. Labelle, Marie. 1992. Change of state and valency. Linguistics 28: 375414. Levin, Beth and Malka Rappaort Hovav. 1995. Unaccusativity: At the Syntax-Lexical Semantics Interface. Cambridge, MA: The MIT Press. Mendikoetxea, Amaya. 1999. Construcciones inacusativas y pasivas. In Ignacio Bosque and Violeta Demonte, eds., Gramática Descriptiva de la Lengua Española, 1577-1629. Madrid: Espasa. Moreno Cabrera, Juan Carlos. 1984. La diátesis anticausativa: Ensayo de sintaxis general. Revista EspaĔola de Lingüística 12: 21-43. Müller, Natascha. 2004. Argumentreduktion im Französischen und Italienischen. In Rolf Kailuweit and Martin Hummel, eds., Semantische Rollen, 248-266. Tübingen: Narr. Nunes, Jairo. 1995. Ainda o famigerado SE. D.E.L.T.A, São Paulo 11.2: 201-240. Reinhart, Tanya. 2002. The theta system: an overview. Theoretical Linguistics 28: 229-290.

Rolf Kailuweit

133

Rozwadowska, BoĪena. 1988. Thematic restrictions on derived nominals. In Wendy Wilkins, ed., Syntax and Semantics 21: Thematic Relations, 147-165. New York: Academic Press. Rothemberg, Mira. 1974. Les Verbes à la Fois Transitifs et Intransitifs en Français Contemporain. The Hague and Paris: Mouton. Sánchez López, Cristina, ed. 2002. Las Construcciones con SE. Madrid: Visor Libros. —. 2002. Las construcciones con se: Estado de la cuestión. In Cristina Sánchez López, ed., Las Construcciones con SE, 13-163. Madrid: Visor Libros. Schäfer, Florian. 2008. The Syntax of (Anti-)Causatives: External Arguments in Change-of-State Contexts. Amsterdam and Philadelphia: John Benjamins. Smith, Carlota S. 1970. Jespersen’s ‘Move and Change’ class and causative verbs in English. In M. Ali Jazayery and Edgar C. Polomé, eds., Linguistic and Literary Studies in Honor of Archibald A. Hill, vol. 2: Descriptive Linguistics, 101-109. The Hague: Mouton. Van Valin, Robert D., Jr. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Zribi-Hertz, Anne. 1987. La réflexivité ergative en français moderne. Le Français Moderne 55: 23-54.

SE INCOMPATIBLE PREDICATES IN SPANISH: AN RRG EXPLANATION CARLOS GONZÁLEZ VERGARA PONTIFICIA UNIVERSIDAD CATÓLICA DE CHILE

Abstract This paper deals with the question of why some Spanish verbs (e.g. llover, morir, haber, gustar) are incompatible with the reflexive clitic se in the so-called “impersonal construction”. Based on the proposal that se involves a lexical rule that privileges an undergoer argument (González Vergara 2006, 2009), I propose that this behavior can be explained straightforwardly within the framework of Role and Reference Grammar [RRG] (Van Valin and LaPolla 1997): none of these verbs can privilege the undergoer, either because they have no macrorole argument at all (the llover type) or because their undergoer argument is already privileged (the morir, haber, and gustar type).

Keywords Reflexive, clitic, macrorole, Spanish

1. Introduction In Spanish, some verbal predicates seem to be incompatible with the reflexive clitic se, as illustrated by (1a)-(1d): (1) a.

*Se llueve REFL rain.3SG LS: do’ (rain’) b. *Se murió REFL die.PAST.3SG LS: INGR dead’ (Ø)

Carlos González Vergara

c.

*Se hay REFL there-is.3SG LS: be-in’ (cielo, nubes) d. *Se gusta REFL like.3SG LS: like’ (Ø, cine)

nubes en clouds in el the

135

el the

cielo sky

cine movies

(1a), (1b), and (1c,d), respectively, involve activity, achievement, and state. The verb in (1a) selects no argument, the verb in (1b) selects one, while the verbs in (1c,d) select two. These examples prompt us to ask what these predicates have in common and why they are incompatible with se. These questions lead to another question of what the meaning/function of se is. I will propose below that these data can be explained in terms of the properties of the clitic se.

2. The Nature of Spanish se In Spanish, the clitic se appears as an essential component of a wide range of syntactic constructions. One of the most widely known classifications of the uses of se divides them into the following six subtypes (adapted from Hernández 1966): (2) a. b. c. d. e.

Reflexive and reciprocal se mis amigos se golpearon. “My friends hit (themselves/each other).” Passive-reflexive se se construyeron muchas escuelas. “Many schools were built.” Impersonal-reflexive se se acusó a Pedro. “Pedro was accused.” “Interest se” Pedro se bebió una cerveza. “Pedro drank up a beer.” “Intrinsic se” 1. “Psychological” intrinsic-se Pedro se enojó. “Pedro got angry.” 2. “Physical” intrinsic-se with an animate/inanimate argument Pedro se levantó.

136

Se Incompatible Predicates in Spanish: An RRG Explanation

f.

“Pedro got up.” La puerta se cerró. “The door closed.” Middle se Esta puerta se cierra fácilmente. “This door closes easily.”

In González Vergara (2006, 2009), I proposed that the uses of se illustrated by (2b,c,e,f) involve a lexical rule that modifies the logical structure of the sentence in such a way as to diminish the actor’s role and to privilege the undergoer when it is present (cf. Centineo 1995; Van Valin and LaPolla 1997; Bentley 2004) and makes the undergoer argument serve as the privileged syntactic argument [PSA]. Specifically, I propose the lexical rule in (3), which may apply to predicates of any Aktionsart type: (3) Given any kind of logical structure, make the argument x of the predicate unspecified. a. State: pred’ (x, y) ļ pred’ (Ø, y) b. Activity: do’ (x, [pred’ (x, (y))]) ļ do’ (Ø, [pred’ (Ø, (y))]) c. Active accomplishment: 1. do’ (x, [pred1’ (x, y)]) & BECOME pred2’ (y) ļ do’ (Ø, [pred1’ (Ø, y)]) & BECOME pred2’ (y) 2. do’ (x, [pred’ (x)]) & BECOME be-LOC’ (y, x) ļ do’ (Ø, [pred’ (Ø)]) & BECOME be-LOC’ (y, Ø) d. Accomplishment/achievement: BECOME/INGR pred’ (x, y) ļ BECOME/INGR pred’ (Ø, y) e. Causative: 1. [do’ (x, Ø)] CAUSE [(BECOME/INGR) pred’ (y)] ļ [do’ (Ø, Ø)] CAUSE [(BECOME/INGR) pred’ (y)] 2. [do’ (x, Ø)] CAUSE [do’ (y, [pred’ (y)])] ļ [do’ (Ø, Ø)] CAUSE [do’ (y, [pred’ (y)])] Application of the lexical rule in (3) have a variety of consequences in (2a)-(2f), but they have in common that the actor argument that would otherwise be realized as the PSA is not available for the PSA selection after the application of the above lexical rule. As an illustration of how introduction of the reflexive clitic se changes a logical structure, let us consider the pair in (4): (4) a.

Pedro Pedro

ensució stained.3SG

la the

camisa. shirt

Carlos González Vergara

b.

137

“Pedro stained the shirt.” LS: [do’ (Pedro, Ø)] CAUSE [BECOME dirty’ (camisa)] LS: [do’ (Ø, Ø)] CAUSE [BECOME dirty’ (camisa)]

(4a) is a transitive sentence without the clitic se and involves a causative logical structure, while (4b) is the result of the application of the lexical rule to (4a). An analogous logical structure is observed in “intrinsic se” sentences as in (5a), passive-reflexive sentences as in (5b), impersonal-reflexive sentences as in (5c) (in order to obtain this meaning, it is necessary to change the inanimate argument camisa “shirt” with a human argument Juan) and middle se sentences as in (5d): (5) a.

b.

c.

d.

[do’ (Ø, Ø)] CAUSE [BECOME dirty’ (camisa)] “Intrinsic se” La camisa se ensució. the shirt REFL stained.3SG “The shirt got dirty.” Passive-reflexive se Se ensució la camisa. REFL stained.3SG the shirt “The shirt was stained.” Impersonal-reflexive se Se ensució a Juan. REFL stained.3SG to Juan “Juan was stained.” Middle se La camisa se ensucia fácilmente. the shirt REFL stained.3SG easily “The shirt gets dirty easily.”

3. Some Apparent Exceptions Let’s now return to the set of verbs in (1), which I have argue cannot occur with se (e.g. llover “rain”, morir “die”, haber “have”, gustar “like”). An immediate problem that comes to mind is that some of these verbs turn out to be compatible with se, as illustrated by (6a)-(6c): (6) a.

El techo se the roof REFL “The roof was rained.”

llovió rain.PAST.3SG

138

Se Incompatible Predicates in Spanish: An RRG Explanation

b. c.

Pedro se murió Pedro REFL die.PAST.3SG “Pedro died.” María se gusta María REFL like.PRES.3SG “María likes herself.”

It is important note that these verbs become compatible with se only when they involve the PSA. (6a,b) belong to the undergoer PSA construction, while (6c) is a reflexive construction. Nevertheless, these predicates cannot occur with se in the non-PSA (i.e. impersonal) construction. We are now in a position to refine our initial question and to ask why verbs as listed in (1) are not compatible with se in the non-PSA construction. (7a)-(7e) are an expanded set of those verbs: (7) a. b. c. d.

e.

llover “rain”, nevar “snow”, garúar “dizzle”, temblar “tremble” morir “die”, aparecer “appear”, crecer “grow”, envejecer “get old”, adelgazar “slim” hay “there is” alcanzar “have enough”, apenar “cause sorrow”, convenir “suit”, costar “take effort”, doler “feel hurt”, extrañar “feel the lack”, faltar “lack”, gustar “like”, importar “matter”, interesar “be interested”, molestar “be bothered”, preocupar “be worried”, quedar “have left”, sobrar “spare” dar pena/miedo/gusto “feel pity/fear/delight”

4. Proposal If se is the morphological expression of the lexical rule that privileges the undergoer argument, we can explain why se is not compatible with the predicates in (7). The reason is that their logical structures cannot privilege an undergoer argument, either because the undergoer is already the privileged argument or because their logical structures don’t have any argument in the first place. I propose four subtypes of se incompatible verbs: the llover type, morir type, haber type, and gustar type. Let’s examine all of these types below.

4.1. The llover type Verbs as llover “rain”, nevar “snow”, and temblar “tremble (the earth)”, usually known as “weather verbs”, involve an activity predicate do’ with no argument in its logical structure, as illustrated in (8):

Carlos González Vergara

139

(8) Ayer llovió/nevó /tembló yesterday rain.PAST/snow.PAST/tremble.PAST “Yesterday, it rained/snowed/trembled.” LS: yesterday’ (do’ (rain’/snow’/tremble’)) The consequence of having no argument is that these verbs have no undergoer argument. If there is no undergoer argument available, there is no way to privilege it. Therefore, these verbs are not compatible with se. It has to be noted again that apparent exceptions such as (9) are not really “weather verbs” as defined above, but predicates that have at least one semantic argument that is able to function as the PSA, as demonstrated by the agreement marker on the verb: (9) a. b.

Las carpas se llovieron the tents REFL rain.PAST.3PL “The tents were soaked in rain.” Los patios se nevaron the backyards REFL snow.PAST.3PL “The backyards got covered in snow.”

4.2. The morir type Verbs as morir “die”, aparecer “appear”, crecer “grow”, envejecer “get old”, and adelgazar “slim” are accomplishment or achievement verbs that have a state predicate with one argument in their logical structures. Since the base predicate is a state, the only arguments of these verbs are undergoers in accordance with the default macrorole assignment principles (Van Valin and LaPolla 1997). These verbs have no choice but to choose the undergoer arguments as their PSAs. This explains why these verbs cannot co-occur with se, because the undergoer already serves as the PSA: (10) a. *Se murió. LS: INGR dead’ (Ø) b. *Se creció. LS: PROC grown’ (Ø) c. *Se envejeció. LS: PROC old’ (Ø) d. *Se adelgazó LS: PROC thin’ (Ø)

140

Se Incompatible Predicates in Spanish: An RRG Explanation

It is worthy of notice that the morir type verbs cannot occur with se when they are part of the non-PSA construction, but that they can do so when they are part of the PSA construction as in (11) (see González Vergara 2006 for further discussion): (11) a. Tus mascotas se murieron your pets REFL die.PAST.3PL “Your pets died.” b. El equipo se creció the team REFL grow.PAST.3SG “The team grew stronger.” c. La actriz se envejeció The actress REFL get.old.PAST.3SG “The actress got old.” d. El niño se adelgazó The boy REFL slim.PAST.3SG “The boy slimmed.”

4.3. The haber type Haber “have” is a two-place state predicate one of whose arguments is a locative (Fernández Soriano and Táboas 1999). However, the locative argument fails to receive an actor status and is realized by a prepositional phrase. In contrast, the second argument of haber “have” (theme) serves as the PSA. In order to accommodate this linking, it is necessary to stipulate the number of macroroles as in (12). (12)

Hay nubes en be.PRES.3SG clouds in “There are clouds in the sky.” LS: be-in’ (sky, clouds) [MR1]

el the

cielo sky

Given that haber “have” involves only one macrorole (undergoer) because of [MR1], the undergoer argument serves as the PSA. Once again, we can see that haber “have” is not compatible with se, because the undergoer is already privileged as the PSA.

4.4. The gustar type Verbs such as gustar “like”, faltar “lack”, importer “matter” and sobrar “spare” exhibit an unexpected syntactic behavior. Gustar, for instance, is

Carlos González Vergara

141

semantically similar to the English verb like, but their syntactic behaviors are different. (13) a. I like apples. b. (A mí) me (to me) 1SG.DAT “I like the apples.”

gustan like.PRES.3PL

las the

manzanas apples

My proposal is that unlike the English verb like, gustar receives only one macrorole ([MR1]). Since gustar “like” is a state predicate, the only macrorole should be an undergoer. The actor-undergoer hierarchy requires us to choose the second argument as the undergoer. The first argument of the state predicate has no choice but to become a non-macrorole, which is optionally realized by a PP. The gustar type is not compatible with the clitic se, because it already privileges the undergoer argument as the PSA.

5. Summary As we have seen, all of those verbs that are incompatible with se in the non-PSA construction have one thing in common: they cannot privilege the undergoer argument, either because they have no macrorole argument (the llover type) or because their undergoer argument is already privileged (the morir, haber and gustar type). The above data and discussion support the proposal that Spanish se involves the lexical rule that modifies the logical structure of the predicate by diminishing the actor’s role and privileging the undergoer’s when it is present (González Vergara 2006, 2009).1

1

Verbs of the morir type sometimes co-occur with se in the non-PSA construction when they are in the imperfective aspect: Se muere/moría bien cuando se muere/moría REFL die.PRES/IMPERF well when REFL die.PRES/IMPERF por un ideal. for an ideal “One dies/used to die well when one dies/used to die for an ideal.” The meaning of the above example is similar to that of an attributive sentence (e.g. La muerte puede ser buena cuando es por un ideal “Dying can be good when it happens for an ideal”) (see Felíu Arquiola 2008 and González Vergara 2006, 2009 for related discussion). This and analogous examples remain as a genuine counterexample to the generalization put forward in this paper.

142

Se Incompatible Predicates in Spanish: An RRG Explanation

List of Abbreviations DAT PAST PL

dative past tense plural

PRES REFL SG

present tense reflexive singular

References Bentley, Delia. 2004. Unexpressed arguments: si-constructions in Italian. In Brian Nolan, ed., RRG2004 Book of Proceedings: Linguistic Theory and Practice: Description, Implementation and Processing, 17-48. Dublin: Institute of Technology Blanchardstown. Centineo, Giulia. 1995. The distribution of si in Italian transitive/inchoative pairs. Proceedings of Semantics and Linguistic Theory 5: 54-71. Hernández, César. 1966. Del se reflexivo al impersonal. Archivum 16: 39-66. Felíu Arquiola, Elena. 2008. Spanish middle sentences: a Role and Reference Grammar approach. In Rolf Kailuweit, Bjorn Wiemer, Eva Staudinger, and Ranko Matasoviü, eds., New Applications of Role and Reference Grammar: Diachrony, Grammaticalization, Romance Languages, 356-388. Newcastle: Cambridge Scholars Publishing. Fernández Soriano, Olga and Susana Táboas. 1999. Construcciones impersonales no reflejas. In Ignacio Bosque and Violeta Demonte, eds., Gramática descriptiva de la lengua española, 1723-1778. Madrid: Espasa. González Vergara, Carlos. 2006. Las construcciones no reflexivas con se: una propuesta desde la Gramática del Papel y la Referencia. Unpublished Ph.D. dissertation, Universidad Complutense de Madrid. González Vergara, Carlos. 2009. One rule to rule them all: logical structures for Spanish non-reflexive se sentences. In Lilián Guerrero, Sergio Ibáñez, and Valeria Belloro, eds., Studies in Role and Reference Grammar, 361-379. México: UNAM. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press.

THREE-PARTICIPANT EVENTS IN SERIAL VERB CONSTRUCTIONS AND THE SYNTAXSEMANTICS INTERFACE ANNA RICCIO UNIVERSITY OF NAPLES “L’ORIENTALE”

Abstract Three-participant events are typologically encoded in a variety of ways. The encoding strategy examined in this study is the serial verb construction [SVC]. In contrast to the three-place predicate strategy used typically in non-serializing languages, the serial verb strategy involves a sequence of two or more verbs. They function together as a single predicate and share the three participants between them (Margetts and Austin 2007). The aim of this study is to show that the syntax-semantics interface developed by Role and Reference Grammar [RRG] (Van Valin and LaPolla 1997; Van Valin 2005) provides a useful descriptive and theoretical framework for a typological investigation of SVCs encoding three-participant events. The resulting analysis reveals the syntactic and semantic properties of SVCs that differentiate them from both three-place predicate constructions and clause chaining (multi-clausal) constructions.

Keywords Three-participant event, serial verb construction, syntactic structure, semantic representation, semantic-to-syntax linking

1. Introduction Three-participant constructions have attracted a lot of attention in the last years. Since the works by Newman (1996, 1997, 1998), several studies

144

Three-participant Events in Serial Verb Constructions

have focused on the structural and semantic properties of three-participant constructions within different theoretical perspectives (Synder 2003; Lee 2004; Haspelmath 2005a, 2005b; Margetts and Austin 2007; McCormick and Theakston 2008; Malchukov, Haspelmath, and Comrie 2010). Three-place predicates (which are meant here as simple predicates) and other morphosyntactic encodings of three-participant events (including SVCs) take an “agent-like A”, a “recipient-like R” (which includes a recipient, beneficiary, goal, addressee, source, and location as its subtype), and a “theme-like T” (some thing or information conveyed by “A” to “R”) (Margetts and Austin 2007). The prototypical event denoted by a threeplace predicate is a “transfer-of-possession” event, in which an agent causes a change of location of something (theme). For example, the threeplace predicate give denotes a transfer event with an actively-instigating agent, a transferred patient/theme, and an animate recipient. I also follow Newman (1996: 136) in assuming that verbs of communication (e.g. tell, speak, ask, explain, show, reveal, address) involve a metaphorical extension from verbs of transfer (e.g. give) and that both of the two verb classes represent the transfer-of-possession event. A number of competing definitions of SVCs are available in the literature, but the definition adopted here is that proposed by Aikhenvald (2006: 1), who investigates SVCs from a typological perspective:1 A serial verb construction (SVC) is a sequence of verbs which act together as a single predicate, without any overt marker of coordination, subordination, or syntactic dependency of any other sort. Serial verb constructions describe what is conceptualized as a single event. They are monoclausal; their intonational properties are the same as those of a monoverbal clause, and they have just one tense, aspect, and polarity value. SVCs may also share core and other arguments. Each component of an SVC must be able to occur on its own.

The serial verb encoding a three-participant event consists of a main verb, which determines the basic structure of the event, and a minor verb (or two), which elaborates on the argument structure of the main verb. 2

 1

One of the earliest definitions of SVCs dates back to Christaller (1875), who made a distinction between an “essential combination”, in which one verb depends on the other, and “accidental combination”, in which two or more verbs expressing successive actions, share the same subject, and are joined together without any conjunction marker. The term “serial verb construction” was proposed later by Balmer and Grant (1929) and was reintroduced by Stewart (1963). 2 Aikhenvald (2006: 55) distinguishes between the symmetrical and asymmetrical SVCs and SVCs encoding three-participant events belong to the latter category.

Anna Riccio

145

The main areas of diffusion of SVCs are West Africa (in particular, the Kwa group and Atlantic Creoles), South-East Asia (Hmong-Mien, TaiKadai, and Mon-Khmer groups), America (central, southern, and northern), and Oceania (Austronesian, Papuan, Melanesian Pidgins, and Australian languages). An example of the serial verb strategy encoding a three-participant event is the following: (1) Rópu ke-wé leng nì book 3SG:NF-take:F give 1SG “He gave me the book.” (Skou; Donohue 2006: 390) The transfer-of-possession event is expressed by a sequence of the two predicates, wé “take” and leng “give”: the first predicate introduces the theme argument rópu “book”, while the second one introduces the recipient argument nì “me”. The agent argument is the proclitic ke- “he”. Unlike the three-place predicate construction, in which a simple predicate encodes the whole event, the SVC in (1) consists of two predicates each of which encodes its subevent. The combination of the two predicates as in (1) leads the two subevents to merge into the three-participant event. The paper is structured as follows. Section 2 outlines the semantics of SVCs encoding three-participant events and argues that they are divided into two classes, according to which participant (a recipient-like or themelike argument) is introduced by the minor verb. Section 3 describes the syntax of SVCs by listing some of their defining features and distinguishes them from multi-clausal constructions involving clause chaining. Section 4 describes the syntactic and semantic features of SVCs within the RRG framework. Section 4.1 describes the layered structures of the SVCs and distinguishes them from three-place predicate constructions. Section 4.2 compares the structures of SVCs with those of multi-clausal constructions (clause chaining constructions). Section 4.3 shows how the macrorole assignment proceeds in SVCs and illustrates the semantics-to-syntax linking in SVCs. Section 5 concludes this paper.

2. The Semantics of SVCs: R-type Serialized P and T-type Serialized P The SVCs examined in this study represent a variety of three-participant events which can be classified into four semantic classes (Margetts and Austin 2007; cf. Croft 1985; Pinker 1989; Levin 1993; Goldberg 1995):

146

Three-participant Events in Serial Verb Constructions

(a) Agent causes recipient to receive theme Verbs of Change of Possession: Give verbs Verbs of Sending and Carrying: Send Verbs; Bring and Take Verbs Verbs of Throwing: Throw Verbs (b) Agent causes theme to move to location Verbs of Putting: Put Verbs (c) Agent intends to cause recipient to receive theme Verbs of Change of Possession: Verbs of Obtaining (Get Verbs) Verbs of Creation (d) Agent acts to communicate information to recipient Verbs of Communication: Verbs of Transfer a Message Languages differ with respect to how these semantic classes are lexicalized or grammaticalized. Margetts and Austin (2007) propose another useful classification of SVCs encoding three-place participant events. They divide those SVCs into two major classes, R-type serialized P and T-type serialized P. They are also known in the literature as “give-serialization” and “takeserialization” (Green 1974; cf. Aikhenvald 2006). In the first class, minor verbs such as “give”, “show”, “submit”, “help”, “come”, and “go” introduce a recipient, beneficiary, goal, addressee, location, or source argument (R-type participant). In the second class, minor verbs introduce a theme or instrumental argument (T-type participant). The most common minor verb that licenses an extra theme-like argument is “take”. Example (2) illustrates an R-type serialized P, in which the minor verb dá “give” causes a valency-increasing extension of the main verb mandá “send” and introduces the recipient argument hen “her”: (2) Mí mandá biífi dá hen 1SG send letter give 3SG:F “I have sent letters to her.” (Saramaccan; Veenstra 1996: 107) Example (3) illustrates a T-type serialized P, in which the minor verb fa “take” introduces the theme argument buluku “book”: (3) Kofì fa-li buluku hԥ-leli Kasi Kofi take-PST book show-PST Kasi “Kofi showed the book to Kasi.” (Anyin; Van Leynseele 1975: 197)

Anna Riccio

147

The minor verb “take” has become semantically bleached to the extent that it no longer denotes physical grasp (or even a separable action) and only assigns a transitive or causative value to the main verb “show”. Tables 1 and 2 list serial verbs (composed of two verbs [V1] and [V2] or more verbs [V(n)]) with their argument roles (e.g. recipient [R], theme [T], beneficiary [B], location [L]). The asterisk-marked several predicates in the left-most column indicate that they can be expressed by the giveand take-serialization: Table 1: R-type serialized P GIVE-serialization bring*

build buy give*

put return sell* send

tell

throw* write

V1 bring bring take take take take take build give buy give give take take put give back sell send send send say say say throw stab

Arg T T T T T T T T R T T T T T T T T T T T T T T T T

V2 give go come put on come go arrive give build give reach go go come give give give give go come show go come go go

V(n)

give give go

give reach

Arg B L L L R R L B T R/B R R R R R R/B R R L R R R R R R

148

Three-participant Events in Serial Verb Constructions

Table 2: T-type serialized P TAKE-serialization bring* give* sell* show throw*

V1 take take take take take take

Arg T T T T T T

V2 bring give take sell show throw

V(n)

give

Arg R R R R R L

For illustration, consider the predicate “send” in Table 1. Example (2) combines the main verb “send” with the minor verb “give”. Examples (16) and (17) in the Appendix, illustrating SVCs in Thai (Tai-Kadai; Thailand) and Paamese (Austronesian; Vanuatu), show that the main verb “send” can also be combined with minor verbs such as “go (motion away from X)” and “come (motion toward X)”, each of them being a deictic verb. Unlike “give” and “come”, which introduce a recipient, the minor verb “go” licenses a locative argument. A further observation concerns the number of serial verbs in the construction. Several of the examined languages allow SVCs with more than one minor verb. In such SVCs, the additional minor verb typically adds no argument role of its own. This is the case with verbs such as “bring”, “throw”, and “write” in Table 1 (see examples (4), (5) and (9) below and examples (11), (12), (19), and (24) in the Appendix) and “give” in Table 2 (see example (4) in the Appendix).

3. The Syntax of SVCs SVCs consist of two or more morphologically independent verbs with no conjunction or any overt marker that separates them. This section outlines the word order, argument sharing, and syntactic linkage of SVCs. Word order is an important parameter of variation in SVCs. In SVCs of most serializing languages, an argument tends to occur adjacent to a verb that licenses it. Examples (1)-(3) (and the data in the Appendix) support this claim. Another observation to make is about the contiguity between verbs that constitute a serial verb. In examples (4) and (5), the serial verb “bring” is expressed by two contiguous verbs, but (4) and (5) are in contrast, in that each of the two verbs is syntactically independent in (4), while the two verbs in (5) are part of a syntactic word:

Anna Riccio

149

(4) Rri ve mê arròò 3PL take come water “They brought water.” (Tinrin; Osumi 1995: 213) (5) Me-ke-lam 3SG-take-come “He brought it.” (Sakao; Durie 1988: 10) Another syntactic feature of SVCs that is worthy of mention here is argument sharing. Verbs that form a serial verb share at least one semantic argument within the SVC. Examples (4) and (5) above illustrate a “samesubject” SVC, in which the subjects of the two verbs are coreferential. In contrast, example (6) illustrates a “switch-subject” SVC, in which the object argument of one verb is coreferential with the subject of the other: (6) Aê ka-kêng mo gê-dêng ngoc ngalalê 1S 1SR-give taro 3SR-reach GEN1S child “I gave taro to my child.” (Jabêm; Brashaw 1993: 151) The third structural feature of SVCs is the lack of any clause linkage marker. Such markers characterize multi-clausal constructions such as clause chaining constructions.3Consider example (7) below: (7) Hama mi-me da mo-mi! hammer take-SS 1SG give:2SG:IMPER-SG:OR “Give me the hammer.” (Koiari; Dutton 1996: 19) The three-participant event is encoded by the sequence of the verbs, mi “take” [V1] and mo “give” [V2], as in example (1) above. However, (7) is not a SVC, but a multi-clausal construction. The same-subject marker [SS] on the first verb mi “take”, which is absent in typical SVCs, indicates that the subjects of the two verbs coincide and that they refer to distinct events. This, in turn, suggests that the two verbs in (7) are not two constituents of

 3

The distinction between SVCs and clause chaining constructions may be more continuous than discrete. Crowley (2002: 18) proposes the following continuum between them: verbal compounds > nuclear serial verbs > core serial verb > clause chains > subordination clauses > coordination clauses.

150

Three-participant Events in Serial Verb Constructions

the serial verb, but they are two distinct predicates each of which has its own argument structure (see Section 4.2.).4

4. An RRG Analysis of SVCs This section provides a semantic and syntactic analysis of SVCs within the RRG framework.

4.1. Syntactic Representation of SVCs Both SVCs and three-place predicate constructions are monoclausal structures. Figure 1 diagrams the constituent projection of the simple three-place predicate “give” in English (Van Valin 2007):

Figure 1: The constituent structure of a three-place predicate construction

The verb give is preceded by one argument (“Pat”) and followed by two arguments (“the book” and “Kim”) (Van Valin 2007). The nuclear layer plus the three arguments of the verb give form a single core. A closer look at SVCs suggests that they involve two types of juncture, nuclear juncture and core juncture; serial verbs in SVCs may form either a nuclear-level or core-level serialization. Figure 2 shows that the two verbs that constitute the serial verb in example (4) are joined together to form a complex nucleus:

 4

For further investigations of the main differences between SVCs and clause chaining constructions, see Lynch (1983), Longacre (1985), Lehmann (1988), Haiman and Thompson (1998), Crowley (1998), and Kroeger (1999, 2004).

Anna Riccio

151

(4) Rri ve mê arròò 3PL take come water “They brought water.” (Tinrin; Osumi 1995: 213)

Figure 2: Example of a SVC involving a nuclear juncture

The verb mê “come” is an intransitive verb that adds no argument role other than the agent argument, which is coreferential with the agent argument of the verb ve “take”. These two verbs form a complex nucleus associated with the two arguments. It is instructive to compare example (4) with (8) with respect to the level of serialization: (8) Emil pan liv la bay Mari Emil take book DET give Mari ‘Emil gave the book to Mary.’ (Haitian Creole language; Andrews and Manning 1999: 105) Unlike the three-place predicate construction in Figure 1, the SVC in (8) consists of two independent nuclei, pan “take” [V1] and bay “give” [V2], each with its own argument role(s) and core. Figure 3 represents the constituent structure of the SVC in (8):

152

Three-participant Events in Serial Verb Constructions

Figure 3: Example of a SVC involving a core juncture

Examples (4) and (8) are in contrast, in that (4) involves a nuclear-level serialization, while (8) involves a core-level serialization. It is interesting to note in this connection that there are SVCs with three verbs that constitute a serial verb, as in example (9): (9) Malí ma dyana ba da pedéli Mary take banana go give father “Mary brought banana to her father.” (Fa d’Ambu; Post 1994: 201) The SVC in (9) presents a hybrid structure involving both a nuclear-level and core-level serialization, as represented in Figure 4:

Anna Riccio

153

Figure 4: Example of a SVC involving a nuclear-level and core-level juncture

The main verb ma “take” and its arguments form a core. This core is combined with another core (comprising a complex nucleus composed of ba “go” and da “give” and their shared argument pedéli “father”) to yield a complex core diagrammed in Figure 4.

4.2. Syntactic Representation of Non-SVCs In Section 3, SVCs were compared with and distinguished from clause chaining constructions. Example (7) illustrates the switch reference marker (in this case, the same-subject marker), one of the defining features of clause chaining constructions that serve to distinguish them from SVCs. Figure 5 below diagrams the constituent structure of (7): (7) Hama mi-me da mo-mi! hammer take-SS 1SG give:2SG:IMPER-SG:OR “Give me the hammer.” (Koiari; Dutton 1996: 19)

154

Three-participant Events in Serial Verb Constructions

Figure 5: Constituent structure of the clause chaining construction in (7)

What is notable about Figure 5 is that it involves no nuclear-level or corelevel serialization; what is conjoined in Figure 5 is the clausal units. The clause chaining consists of two separate clauses joined together by the clause-linkage marker (or complementizer) (Van Valin 2005: 205). The same-subject marker -ma indicates a temporal relation between the two verbs, mi- “take” and mo “give”. This suggests that the two verbs refer to two separate events and that the clause chaining construction in (7) is a multi-clausal (rather than multi-verb) construction.

4.3. Linking Semantics and Syntax in SVCs Under the RRG system of lexical decomposition (Van Valin and LaPolla 1997), simple three-place predicate encoding three-participant events and their related arguments have a logical structure like [do’ (x, Ø) CAUSE [BECOME predicate’ (y, z)]. The decompositional representations of verbs are intended to capture those aspects of verbal semantics that determine the morphosyntactic realization of their arguments. The mapping between the decompositional semantic representations of verbs and the morphosyntactic expressions of their arguments is mediated by macroroles [MRs] (actor and undergoer). RRG assigns MRs to verbal arguments on the basis of the actor-undergoer hierarchy [AUH] (a version of thematic hierarchy) and the default macrorole assignment principle, which ensures that single-argument and two argument verbs, respectively, receive one and two macroroles when no lexical prespecification of the number of MRs applies (see Van Valin and LaPolla 1997).

Anna Riccio

155

Given that the AUH leads the highest-ranking argument (agent) and lowest-ranking argument (theme) of a typical three-place verb to receive an actor and undergoer status, the recipient argument has no choice but to be become a non-macrorole core argument. Figure 1 above represents an indirect object construction, in which the recipient is expressed as an oblique argument [NMR].5 Figure 6(a) diagrams the constituent structure of the indirect object construction, which realizes the recipient argument as a non-macrorole core argument. In a SVC with a core-level serialization illustrated by example (8) (Figure 6(b)), however, the macrorole assignment proceeds in two cores. The consequence of this is that “Emil” and “book”, respectively, receive an actor and undergoer status from the main verb pan “take”, while the recipient argument “Mary” receives another undergoer status from the minor verb bay “give”:

Figure 6: MR assignments in three-participant constructions

 5

Three-participant events can also be expressed by double object constructions, as in Pat gave Kim a book. However, in this case, the theme argument book is analyzed as a non-macrorole argument. For further discussion of three-place predicates within the RRG framework, see Van Valin (2007).

156

Three-participant Events in Serial Verb Constructions

(8) Emil pan liv la bay Mari Emil take book DET give Mari ‘Emil gave the book to Mary.’ (Haitian Creole) Figure 7 presents a detailed view of how the semantics-to-syntax linking proceeds in example (8):

Figure 7: The semantic-to-syntax linking in example (8)

The broken line indicates that the theme argument liv “book” in LS2 is co-indexed with the theme argument in LS1 (cf. Baker 1989; Collins 1997). The linking in (8) proceeds as follows. First, the logical structures of pan “take” and bay “give” are [do’ (x, Ø)] CAUSE [BECOME have’ (x, y)] (LS1) and [do’ (x, Ø)] CAUSE [BECOME have’ (y, z)] (LS2), respectively. They form two distinct cores, but the one formed by bay has one less core argument than is in its LS. The effector-locative argument Emil in LS1 is assigned an actor status and serves as the privileged syntactic argument [PSA], which controls the verb agreement on the

Anna Riccio

157

matrix verb pan “take” and determines the interpretation of the missing argument (Emil) in the linked core (LS1) (Van Valin 2005). It is also important to note that the theme argument liv “book” in LS2 is co-indexed with the theme argument in LS1 and may not be expressed overtly. This suggests that the theme argument in LS2 is not available for macrorole assignment and that the recipient-possessor argument Mari has no choice but to receive an undergoer status from bay “give”.

5. Conclusion This paper has provided an overview of characteristic features of SVCs encoding three-participant events and has shown that the application of the RRG framework to the analysis of SVCs reveals their semantic and syntactic properties that serve to establish the unique status of SVCs. The serial verbs have been divided into two classes, the R-type serialization and T-type serialization (Margetts and Austin 2007; cf. Green 1974): the former is realized by minor verbs such as “give”, “show”, “submit”, “help”, “come”, and “go” which introduce a recipient, beneficiary, goal, addressee, location, or source argument as an additional argument role, while the latter is realized by minor verbs such as “take” which introduces a theme or instrumental argument (Section 2). The syntactic status of SVCs has been revealed by the RRG theory of clause linkage, according to which there are nuclear, core, and clausal junctures. RRG has allowed us to distinguish SVCs from clausal chaining constructions, on the grounds that the former involve a nuclear or core juncture, while the latter involve a clausal juncture (Sections 3 and 4). Finally, an RRG account of the semantics-to-syntax linking in SVCs has been described on the basis of the semantic and syntactic properties of SVCs reviewed above. It has been shown that SVCs with a core-level juncture/serialization involve more than one distinct core and that the macrorole assignments in those SVCs involve complications not observed in the corresponding simple predicate constructions.

158

Three-participant Events in Serial Verb Constructions

Appendix The data are based on the survey of descriptive grammars and other descriptive material. They are grouped according to the semantic classification presented in Section 2. (a) Agent causes recipient to receive theme ™Verbs of Change of Possession: Give Verbs ‘to give’

(1)

Khmer (Austro-Asiatic, Mon-khmer; Cambodia) KhЄom ٧aoy siԥphֽۨu tֽۨu nΉۨԥk 1SG take book go 2SG “I give you a book.” (Jacob 1968: 78)

(2)

Khmer (Austro-Asiatic, Mon-khmer; Cambodia) NΉۨԥk ٧aoy siԥphֽۨu mƥۨ:k khЄom 2SG take book come 1SG “You give me a book.” (Jacob 1968: 78)

(3)

Twi (Niger-Congo, Akan; Ghana) ƥ-de siká nó maa me 3SG-take money he give me “He gave me the money.” (Stewart 1963: 146, cited in Foley and Olson 1985: 54)

(4)

Anyi (Niger-Congo, Kwa; Ghana) Kòfi fà bùlúkú-ԥ‫ ܗ‬fá-mƗ Kàsí Kofi take book-DET take-give Kasi “Kofi gave the book to Kasi.” (Van Leynseele 1975: 201, cited in Foley and Olson 1985: 54)

(5)

Jabêm (Austronesian, Malayo-Polynesian; Papua New Guinea) A-kêng òbo gê-dêng napalê gê-jà 1PL:INCL-give cloth 3SG-to boys 3SG-go “We give cloth(s) to the boy.” (Durie 1988: 12)

Anna Riccio

159



‘to sell’

‘to return’

(6)

Yoruba (Niger-Congo, Benue-Congo, Defoid; Nigeria) ó tà-á fún mi 3SG sell-3SG give 1SG:O “He sold it to me.” (Lord 1993: 35)

(7)

Sranan (Suriname Creole English, Suriname) Mi teki fisi seri 1SG take fish sell “I sold the fish.” (Muysken and Veenstra 1995: 191)

(8)

Thai (Tai-Kadai, Kam-Tai) Dèk khƱ̣n naƾsƱ̣ hây khruu boy give-back book give teacher “The boy gave back the book to the teacher.” (Nichols and Woodbury 1985: 55)

™Verbs of Sending and Carrying: Send verbs; Bring and Take ‘to bring’

(9)

Sranan (Suriname Creole English; Suriname) Roy e tyari a pikin go na oso Roy PST bring DET child go LOC house “Roy brought the child to house.” (Jansen et al. 1978, cited in Dechert and Raupach 1989: 19)

(10)

Kana (Niger-Congo, Benue-Congo; Nigeria) Bàrilè èsúƗ lõ kpá núƗ Barile take SPEC:SG book bring:INSTRUM “Barile has brought the book.” (Ikoro 1995: 316)

(11)

Bamileke (Niger-Congo, Bantu; Camerun) á ká láh càk usá٧ ha a 3SG PST take pot come give 2SG “He brought me the pot.” (Hyman 1971, cited in Kroeger 1999: 235)

160

Three-participant Events in Serial Verb Constructions



‘to send’

(12)

Paicî (Austronesian, Malayo-Polynesian; New Caledonia) Ɲ pá tΉ‫ܮ‬Ήpà-ri pֺ‫ ۛܕ‬Ư Ɨ‫ܕ‬nyƝ‫ ܕ‬dari Iki‫̞ܕ‬ 3SG take arrive-TR go DET fire house Ikii “She brings the fire to Ikii’s house.” (Rivierre 1983, cited in Ozanne-Rivierre 2004: 335)

(13)

Igede (Niger-Congo, Benue-Congo; Nigeria) Ahi hu۱ olo chu 1PL take load put.on.head “We carried the load (on our heads).” (Bamgbose 1974, cited in Awoyale 1988: 29)

(14)

Yoruba (Niger-Congo, Benue-Congo, Defoid; Nigeria) Mo mú ìwé wá ilé I take book come home “I brought a book home.” (Stahke 1970: 61, cited in Nichols and Woodbury 1985: 25)

(15)

Kristang (Portuguese Creole; Malaysia, Singapore) Yo ja tizé isti floris da ku eli 1S PF bring this flower give R 3SG “I brought this flower for her.” (Forman 1993: 175)

(16)

Thai (Tai-Kadai, Kam-Tai) Song cotmaay pay Krungthep send letter go Bangkok “(S)he sends a letter to Bangkok.” (Natchanan Yaowapat [p.c.], cited in Malchukov et al. 2010)

(17)

Paamese (Austronesian, Malayo-Polynesian; Vanuatu) Kisa-n taselu-sien 2SG:FUT:DIST-send INDEF-message he-mai venau 3SG:FUT:DIST-come DAT:1SG “You will send a message to me.” (Crowley 2002: 178)

Anna Riccio

161



™Verbs of Throwing: Throw verbs ‘to throw’

(18)

Jamaican (English Creole) Im tek stuon fling ina di waata 3SG take stone throw in:LOC water “He threw the stone in the water.” (Paul and Veenstra 1992: 191)

(19)

Thai (Tai-Kadai, Kam-Tai) Khwaang luukbԥn pay hay khaw throw ball go give 3SG “(S)he threw the ball to him.” (Malchukov et al. 2010)

(b) Agent causes theme to move to location ™Verbs of Putting: Put Verbs (20)

‘to put’

Yoruba (Niger-Congo, Benue-Congo, Defoid; Nigeria) Mo dé filà fún 1SG put hat give 3SG “I put the hat on his head.” (Lord 1993:35)

(c) Agent intends to cause recipient to receive theme ™Verbs of Change of Possession: Verbs of Obtaining (Get Verbs) ‘to buy’

(21)

Gă (Niger-Congo, Kwa; Ghana) Mi he nook mi hă le 1SG buy something 1SG give 3SG:F “I bought something for her.” (McWhorter 1992: 11)

162

Three-participant Events in Serial Verb Constructions



™Verbs of Creation ‘to build’

‘to write’

(22)

Vietnamese (Austro-Asiatic, Mon-khmer, VietMuong) TƗ gƟi wǂ zào -le yì dǀng fángzi 3SG give 1SG:O build -ASP DET CL house “(S)he built an house for me.” (Liem 1979: 57)

(23)

Keo (Austronesian, Malayo-Polynesian; East Nusantara) Ja’o kéma dapu ti’i ‘ine 1SG build kitchen give mum “I built a kitchen for mum.” (Baird 2008: 60)

(24)

Jabêm (Austronesian, Malayo-Polynesian; Papua New Guinea) Na-so pepa ni-wesa ni-ndenga nanggi IR1S-stab paper IR3S-go IR3S-reach GEN1S lunggewe sister “I’ll write a letter to my sister (far away).” (Bradshaw 1993: 148)

(d) Agent acts to communicate information to recipient ™Verbs of Communication: Verbs of Transfer a Message ‘to tell’

(25)

Twi (Niger-Congo, Kwa; Ghana) Mi-ka DVҽm mi-N\ҽr no 1SG-speak word 1SG-show 3SG “I said him something.” (Riis 1854: 30, cited in Nichols and Woodbury 1985)

(26)

Loniu (Austronesian, Malayo-Polynesian; Papua New Guinea) Iy ipwey ime ete yó 3S R3S-say-it R3S-come AG 1S “He told me.” (Hamel 1993: 117)

Anna Riccio

(27)

163

Jabêm (Austronesian, Malayo-Polynesian; Papua New Guinea) Ja-sôm bing ê-ndêng lau 1S-speak word 3SG-go people “I address word(s) to the people.” (Durie 1988: 12)

List of Abbreviations 1 2 3 AG ARG ASP B DET CL DIST F FUT GEN IMPER INDEF IR L LOC

first person second person third person agent argument aspect beneficiary determiner classifier distal feminine future genitive imperative indefinite irrealis location locative

M NF O OR PST PERF PL R SG S SPEC SS INSTRUM T TR V1 V2

masculine non-feminine object object reference past tense perfective plural recipient singular subject specifier same subject instrumental theme transitive first verb second verb

References Aikhenvald, Alexandra Y. 2006. Serial verb constructions in typological perspective. In Alexandra Y. Aikhenvald and R. M. W. Dixon, eds., Serial Verb Constructions: A Cross-Linguistic Typology, 1-68. Oxford: Oxford University Press. Andrew, Avery D. and Christopher D. Manning. 1999. Complex Predicates and Information Spreading in LFG. Stanford, CA: CSLI Publications. Awoyale, Yiwola. 1988. Complex Predicates and Verb Serialization. Lexicon Project Working Paper 28. Cambridge, MA: Center for Cognitive Science, MIT. Baird, Louise. 2008. Motion serialization in Kéo. In Gunter Senft, ed., Serial Verb Constructions in Austronesian and Papuan Languages, 5574. Canberra: Pacific Linguistics.

164

Three-participant Events in Serial Verb Constructions

Baker, Mark C. 1989. Object sharing and projection in serial verb constructions. Linguistic Inquiry 20: 513-553. Balmer, William Turnbull and F.C.F. Grant. 1929. A Grammar of the Fante-Akan Language. London: Altantis Press. Bamgbose, Ayo. 1974. On serial verbs and verbal status. Journal of West African Languages 9.1: 17-48. Bisang, Walter. 1986. Die Verb-Serialisierung im Jabêm. Lingua 70: 13162. Bowden, John. 2008. Verb serialization in Taba. In Gunter Senft, ed., Serial Verb Constructions in Austronesian and Papuan Languages, 7598. Canberra: Pacific Linguistics. Bradshaw, Joel. 1993. Subject relationships within serial verb construction in Numbami and Jabêm. Oceanic Linguistics 32.1: 133-162. Bril, Isabelle. 2007. Nexus and juncture types of complex predicates in Oceanic Languages: functions and semantics. Language and Linguistics (Academia Sinica) 8.1: 267-310. Bril, Isabelle and Françoise Ozanne-Rivierre, eds. 2004. Complex Predicates in Oceanic Languages, Studies in the Dynamics of Binding and Boundedness. Berlin and New York: Mouton de Gruyter. Christaller, Johann Gottlieb. 1875. A Grammar of the Asante and Fante Language Called Tshi. Gregg Press. Collins, Chris. 1997. Argument sharing in serial verb constructions. Linguistic Inquiry 28: 461-497. Croft, William. 1985. Indirect object “lowering”. Proceedings of the Eleventh Annual Meeting of the Berkeley Linguistics Society, 39-51. Crowley, Terry. 1998. An Erromangan (Sye) Grammar. Honolulu: University of Hawaii Press. —. 2002. Serial Verbs in Oceanic: A Descriptive Typology. Oxford: Oxford University Press. Dechert, Hans-Wilhelm and Manfred Raupach, (eds.) 1989. Interlingual Processes. Tübingen: Gunter Narr Verlag. Donohue, Mark. 2006. Negation and grammatical functions in Skou. Language 82.2: 383-398. Durie, Mark. 1988. Verb serialization and “verbal prepositions” in Oceanic languages. Oceanic Linguistics 27: 1-23. Dutton, Tom E. 1996. Koiari. München and Newcastle: Lincom Europa. Foley, William A. and Mike Olson. 1985. Clausehood and verb serialization. In Johanna Nichols and Anthony Woodbury, eds., Grammar Inside and Outside the Clause, 17-60. Cambridge: Cambridge University Press. Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax

Anna Riccio

165

and Universal Grammar. Cambridge: Cambridge University Press. Foley, William A. and Robert D. Van Valin, Jr. 1985. Information packaging in the clause. In Timothy Shopen, ed., Language Typology and Syntactic Description: Clause Structure, 282-364. Cambridge: Cambridge University Press. Forman, Michael L. 1993. Verb serialization, word order typology, and Zamboangueño: a comparative approach. Oceanic linguistics 32: 163182. Goldberg, Adele E. 1995. Construction: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. Hamel, Patricia J. 1993. Serial verbs in Loniu and an evolving preposition. Oceanic linguistics 32.1: 111-132. Haspelmath, Martin. 2005a. Ditransitive constructions: the verb ‘give’. In Martin Haspelmath, Matthew S. Dryer, David Gil, and Bernard Comrie, eds., The World Atlas of Language Structures, 426-431. Oxford: Oxford University Press. —. 2005b. Argument marking in ditransitive alignment types. Linguistic Discovery 3.1: 1-21. Hyman, Larry and Erhard Voeltz. 1971. The linguistic status of Bamileke. In Chin-Wu Kim and Herbert Stahlke, eds., Papers in African Linguistics, 55-70. Carbondale and Edmonton: Linguistic Research, Inc. Ikoro, Suanu. 1995. The Kana Language. Leiden: Research School CNWS. Jackendoff, Ray. 1990. Semantic Structures. Cambridge, MA: The MIT Press. Jacob, Judith M. 1968. Introduction to Cambodian. Oxford: Oxford University Press. Jansen, Bert, Hilda Koopman, and Pieter Muysken. 1978. Serial verbs in the Creole Languages. Publikaties van het Instituut voor Algemene Taalwetenschap 20: 125-159. Lee, Ju-eun. 2004. Ditransitive Structures and (Anti-)Locality. Unpublished Ph.D Dissertation, Harvard University. Kroeger, Paul R. 1999. Review of “A theory of predicates” by Farrell Ackerman and Gert Webelhuth.Notes on Linguistics 2: 220-27. —. 2004. Analyzing Syntax: a Lexical-Functional Approach. Cambridge: Cambridge University Press. Lehmann, Christian. 1988. Towards a typology of clause linkage. In John Haiman and Sandra A. Thompson, eds., Clause Combining in Grammar and Discourse, 181-225. Amsterdam and Philadelphia: John Benjamins.

166

Three-participant Events in Serial Verb Constructions

Law, Paul and Tonjes Veenstra. 1992. On the structure of serial verb constructions. Linguistic Analysis 22: 185-217. Levin, Beth. 1993. English Verb Classes and Alternations: A Preliminary Investigation. Chicago: University of Chicago Press. Liem, Nguyen Dang. 1979. Cases and verbs in Pidgin French (Tay Boi) in Vietnam. Pacific Linguistics. Series A:Occasional Papers Canberra 57: 217-246. Longacre, Robert E. 1985. Sentences as combinations of clauses. In Timothy Shopen, ed., Language Typology and Syntactic Description: Complex Construction, 235-286. Cambridge: Cambridge University Press. Lord, Carol. 1973. Serial verbs in transition. Studies in African Linguistics 4.3: 269-296. —. 1993. Historical Change in Serial Verb Constructions. Amsterdam and Philadelphia: John Benjamins. Lynch, John. 1983. Switch-reference in Lenakel. In John Haiman and Pamela Munro, eds., Switch Reference and Universal Grammar, 209221. Amsterdam and Philadelphia: John Benjamins. Malchukov, Andrej L., Martin Haspelmath, and Bernard Comrie, (eds.). 2010. Ditransitive construction: a typological overview. Studies in Ditransitive Constructions: a Comparative Handbook. Berlin and New York: Mouton de Gruyter. McCormick, Anna and Anna Theakston. 2008. Investigating the Effects of Verb Transitivity, Verb Frequency and Construction Knowledge on the Ditransitive Construction. Manchester: University of Manchester Press. McWhorter, John. 1992. Substratal influence in Saamaccan serial verb constructions, Journal of Pidgin and Creole Languages 7.1: 1-53. Muysken, Pieter and Tonjes Veenstra. 1995. Serial verbs. In Jacques Arends, Pieter Muysken, and Norval Smith, eds., Pidgins and Creoles: An Introduction, 289-301. Amsterdam and Philadelphia: John Benjamins. Newman, John. 1996. Give: A Cognitive Linguistic Study. Berlin and New York: Mouton de Gruyter. —. 1997. The linguistics of Giving. Amsterdam and Philadelphia: John Benjamins. —. 1998. Recipients and ‘give’ constructions. In William van Belle and Willy van Langendonck, eds., The Dative, vol. 2: Theoretical and Contrastive Studies, 1-28. Amsterdam and Philadelphia: John Benjamins Nichols, Johanna and Anthony C. Woodbury, eds. 1985. Grammar Inside and Outside the Clause. Cambridge: Cambridge University Press. Osumi, Midori. 1995. Tinrin grammar, Oceanic Linguistics Special

Anna Riccio

167

Publications. Honolulu: University of Hawaii Press. Pinker, Steven. 1989. Learnability and Cognition: The Acquisition of Argument Structure. Cambridge, MA: The MIT Press. Post, Marike. 1994. Fa d’Ambu. In Jacques Arends, Pieter Muysken, and Norval Smith, eds., Pidgins and Creoles: An Introduction, 191-204. Amsterdam and Philadelphia: John Benjamins. Riis, H. N. 1854. Grammatical Outline and Vocabulary of the Ojilanguage, with Special Reference to the Akwapim-Dialect, Together with a Collection of Proverbs of the Natives. Basel: Bahnmaier. Rivierre, Jean-Claude. 1983. Dictionnaire paicî - français, suivi d'un lexique français - paicî. Paris: Société d’Etudes linguistiques et anthropologiques de France. Snyder, Kieran Margaret. 2003. The Relationship between Form and Function in Ditransitive Constructions. Unpublished Ph.D. Dissertation, University of Pennsylvania. Stewart, John. 1963. Some restrictions on objects in Twi. Journal of African Languages 2.2: 145-149. Van Leynseele, Hélène. 1975. Restrictions on serial verbs in Anyi. Journal of West African Languages 10.2: 189-218. Van Valin, Jr., Robert D. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press. —. 2007. The Role and Reference Grammar analysis of three-place predicates. Accessed June, 2009. http://linguistics.buffalo.edu/people/faculty/vanvalin/rrg.html. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Veenstra, Tonjes. 1996. Serial Verbs in Saramaccan: Predication and Creole Genesis. Dordrecht: ICG Printing.

THE THEORETICAL IMPORTANCE OF CONSTRUCTIONAL SCHEMAS IN RRG1 ELKE DIEDRICHSEN UNIVERSITY OF VECHTA, GERMANY

Abstract Constructional Schemas are a very important part of the RRG approach to the syntax-semantics-interface. They are suggested in order to capture the constructional knowledge that is available to speakers, but only in cases where “idiosyncratic, language-specific features of constructions” (Van Valin 2005: 132) have to be considered in the linking. This paper will reconsider the role of constructions for RRG. It will be argued that the construction is a “grammatical object” (after Nolan submitted a, b, this volume) that is acquired as a linguistic convention and stored as part of a repository of constructions that defines the linguistic knowledge of an individual. It has its own workspace, where the functional and grammatical specifications and processing rules are applied. Taking the constructional schema as a representative of the “construction as a grammatical object”, it is proposed to use constructional schemas in RRG as descriptive tools for constructions in general and argument structure constructions in particular. It will be demonstrated that constructional schemas provide a useful format for the description of the relevant characteristics of a construction and its role in the semantics-to-syntax linking. In this respect, constructional schemas are utilised to allow for the description of productivity and novelty in language use.

 1

I would like to thank Rolf Kailuweit and Brian Nolan for valuable comments and discussion. Any errors are mine.

Elke Diedrichsen

169

Keywords RRG, Construction Grammar, argument structure, constructional schemas, German

1. Introduction: Argument Realization in RRG Lexical theories of argument realization generally seek to explain the syntax of sentences on the basis of the meaning of the verb. There has to be a “mapping” of the lexical-semantic properties of the verb with the syntax of argument structure. It is posited that certain aspects of the verb meaning determine the number and the grammatical status of the arguments that will appear in a linguistic construction. Thus, verbs are classified according to the semantic features which in effect constitute the argument structure realization. Broadly, the following semantic factors are at issue: Causal notions, aspectual notions, event complexity and factors like sentience and volitionality (Levin and Rappaport Hovav 2005: 128). These are considered to be “grammatically relevant facets of meaning” (Levin and Rappaport Hovav 2005: 9) of the verb. The first step in the mapping is to decompose the relevant verb meanings. In Role and Reference Grammar [RRG], the verb meanings are decomposed with respect to a combination of causal and aspectual features. The aspectual features are assigned on the basis of the Aktionsarten-classification carried out by Vendler (1967), according to which features like telicity, punctuality, stativity and dynamicity are distinguished. The basic semantic predicates are connected by the use of do’ and operators like CAUSE and BECOME. RRG distinguishes six Aktionsart classes, all of which have additional causative counterparts. The assignment of verbs to the classes is to be done by tests which are called “independent criteria” (Van Valin 2005: 59), while it is admitted that some of the tests are either language-specific, do not apply in certain circumstances, or are misleading if different readings of verbs are involved. These tests are supposed to be the basis of the RRG approach to argument realization and syntactic organization.

170

The Theoretical Importance of Constructional Schemas in RRG

Table 1: Lexical representation for Aktionsart classes (Table 2.3 in Van Valin 2005: 45) Verb class

Logical Structure

State Activity Achievement Semelfactive

predicate’ (x) or (x, y) do’ (x, [predicate’ (x) or (x, y)]) INGR predicate’ (x) or (x, y) SEML predicate’ (x) or (x, y) SEML do’ (x, [predicate’ (x) or (x, y)]) BECOME predicate’ (x) or (x, y) BECOME do’ (x, [predicate’ (x) or (x, y)])

Accomplishment

Active Accomplishment do’ (x, [predicate1’ (x, (y))]) & INGR predicate2’ (z, x) or (y) Causative α CAUSE β; α and β are logical structures of any type. In many other theories of argument realization, a case frame (Fillmore 1968) or a theta-role-list as used in generative approaches is associated with a verb in its lexical entry. These lists determine the thematic relation that the verb will be associated with. In RRG, however, thematic relations are not supposed to play a paramount role. It is important to emphasize that in the system presented here, thematic relations play no direct role in lexical representation; the relevant semantic properties of the verbs are expressed by the decompositional logical structure representations, not by thematic relations. Thus even though a large number of role labels like agent, cognizer, theme and patient have been used in this discussion, they are merely mnemonics for argument positions in logical structure. They have no independent status. (Van Valin 2005: 60)

The thematic relations are defined according to the argument positions in the decomposed logical structure representation (after Jackendoff 1976). Thus, the argument positions are given for each verb separately. At a later stage in the semantics-to-syntax-linking, these verb specific roles are generalized to semantic macroroles. The paper will proceed as follows. In Sections 2 and 3, some more basic assumptions of lexical theories in general and RRG in particular will be discussed. It will be argued that the generalizations that are part of every theory of lexical decomposition, like the logical structures and the Generalized Semantic Roles, may not fully account for the general idea of

Elke Diedrichsen

171

the semantic motivation of argument structure constructions. Furthermore, the verb-centred algorithm of linking would have difficulties explaining valence-construction mismatches which occur in everyday language. In Section 4, the usability of a “construction”-based account, as the one pursued in Construction Grammar (e.g. Goldberg 1995, 2006), will be evaluated. In Construction Grammar, it is generally assumed that constructions are linguistic signs, which means that they are conventionalized pairings of form and meaning (Goldberg 1995, 2006). The notion of a strong “form/function”-correlation, however, cannot account for constructional ambiguity and the flexibility of constructions with respect to variable elements (Jackendoff 2002). Therefore, it will be argued that an account is needed that allows some variability in the architecture of linguistic constructions. Section 5 will introduce a linking algorithm for argument structure constructions in German, where the main influence on the argument structure is not seen in the verb semantics, but in the requirements of the construction, which is defined here as a “grammatical object” (Nolan submitted a, b, this volume) and represented as a constructional schema. It will be argued that the constructional schema that is traditionally used in RRG for specialised, language-specific constructions, is a very useful and underestimated tool for the description of argument structure constructions, in particular in those cases where the construction would not be predictable by the semantics of the verb.

2. Mapping Basic Logical Structures and Possible Thematic Relations in RRG The thematic relations continuum (Fig. 2.3 in Van Valin 2005: 58) lists the thematic relations that are possibly placed in the logical structure argument positions of the basic predicates. Agent and patient are presented as the endpoints of the continuum. As thematic relations are posited according to logical structure positions of single verbs, it may seem that there are a lot of thematic relations. In fact, there are only five “relevant distinctions”. These are the distinctions that fall out of the five possible argument positions in the logical structures of “activity” and “state” predicates, which are assumed to be basic. “Agent” is added as one of the thematic relations. It is ascribed only to verbs that lexicalize agency; for example murder as opposed to kill. DO signals agency in the logical structure.

172

The Theoretical Importance of Constructional Schemas in RRG

Arg of DO: AGENT 1st arg of do’ (x, …: EFFECTOR, MOVER, ST-MOVER, L-EMITTER, SEMITTER, PERFORMER, CONSUMER, CREATOR, SPEAKER, OBSERVER, USER 1st arg of pred’ (x, y): LOCATION, PERCEIVER, COGNIZER, WANTER, JUDGER, POSSESSOR, EXPERIENCER, EMOTER, ATTRIBUTANT 2nd arg of pred’ (x, y): THEME, STIMULUS, CONTENT, DESIRE, JUDGMENT, POSSESSED, SENSATION, TARGET, ATTRIBUTE, PERFORMANCE, CONSUMED, CREATION, LOCUS, IMPLEMENT Arg of state pred’ (x): PATIENT, ENTITY Figure 1: Thematic relations continuum in terms of logical structure argument positions; after Van Valin (2005: 58; Fig. 2.3)

Patient is listed as the single argument of a state predicate. It would be the argument of predicates like crushed, killed and smashed. The single argument of a state predicate, then, is something that displays the result of an action. The action that would be required to lead to this result is not listed, however, as it is not a basic logical structure. The second argument of a result-implying verb like smash occurs as y, not as x, cf. (1)

logical structure of smash (Van Valin 2005: 66) [do’ (x, Ø)] CAUSE [BECOME [smashed’ (y)]

The single argument in a verb like die is not a patient in a basic logical structure, either. The logical structure for die is more complex, as it involves an operator: BECOME dead’ (x) (Van Valin 2005: 45f.). The only occurrence of pred’(x) as a single argument in an active construction is in a result state predication, as in dead’ (x) or shattered’ (x) (cf. Van Valin 2005: 46ff. for discussion). Van Valin (2005: 48) makes a strong

Elke Diedrichsen

173

claim for the formal distinction of attributive/identificational and resultative predicates. The nature of this distinction is not clear, however. The pred’ (x) arguments mentioned in Van Valin (2005: 46) all involve destruction or death, and thus, their characterisation as “result states” is intuitively plausible. But what about a house that is in the state of being painted or a person that has been awarded a prize? It seems that the feature “result state” is not a feature of a predicate in itself. It is assigned on the basis of the preceding context information and coded by the logical structure construction predicate’ (x). As for agents, neither of the verb classes in Table 1 requires an agentive DO-predicate in its logical structure, not even a complex active accomplishment structure as the one in (2) (after Van Valin 2005: 47) (2)

Carl ate the pizza. do’ (Carl, [eat’ (Carl, pizza)]) & INGR consumed’ (pizza)

Thus, neither AGENT nor PATIENT appear naturally (i.e. without stipulation) as parts of a simple logical structure. The question is why they are listed among the “relevant distinctions” with thematic relations here, while RECIPIENT is not.

3. The Account of Macroroles In RRG, two macroroles, the two “Generalised Semantic Roles” are introduced. They are named “Actor” and “Undergoer”. By definition, Actor and Undergoer are “the two primary arguments in a transitive predication, either one of which can be the single argument of an intransitive verb” (Van Valin 2005: 60).

Figure 2: The Actor-Undergoer Hierarchy (after Van Valin 2005: 61)

Actor and Undergoer are supposed to be based on the logical structure representations of verbs and their arguments. The Actor-Undergoer

174

The Theoretical Importance of Constructional Schemas in RRG

Hierarchy determines the selection of the macroroles Actor and Undergoer from the logical structure of a verb as follows: This double hierarchy says simply that given the logical structure of a transitive verb, the leftmost argument will be the actor and that the rightmost argument will be the undergoer. (Van Valin 2005: 61)

3.1. Generalised Verb Meanings and Macroroles: Some Discussion The concept of macroroles resembles the concepts of “logical subject” and “logical object” semantically. The actor is the semantic counterpart of the traditional notion of “subject”, as it is the most agent-like argument. The undergoer is the semantic counterpart of the direct object. It is the most patient-like argument. While the traditional labels for grammatical relations, subject and object, are not used in RRG, the theory establishes the macroroles, which refer to semantic relations. Note that here with the description of macroroles, the thematic relations lists come back into play, which have been rejected before, in favour of the logical structures. Macroroles are generalizations across thematic relations. Actor is the subject of active transitive constructions, and Undergoer is the subject of passive constructions. Thus, the macroroles are not merely semantic; rather, they bridge the gap between semantic and grammatical relations. Macroroles are motivated by the fact that in grammatical constructions groups of thematic relations are treated alike. (Van Valin 2005: 60)

Macroroles can therefore be considered to constitute the link from semantics to syntax in the syntax/semantics interface. This way of proceeding has the advantage that it is basically functional, as the arguments in a syntactic construction can be given a characterisation based on the semantics of the verb. Furthermore, it is applicable crosslinguistically, as in all languages the arguments in a transitive predication can be distinguished in terms of an “agent-like“ and a “patient-like“ argument. The traditional syntactically based subject vs. object distinction is not cross-linguistically applicable, as languages with an ergative syntax, for example, lack a “subject” argument according to the traditional definition (Dixon 1994; Van Valin and LaPolla 1997; Diedrichsen 2006). However, in the formulation of the mapping between semantics and syntax some generalizations are carried out, which call into question the semantic motivation of syntactic facts, including argument structure relations.

Elke Diedrichsen

1.

175

The facets of meaning that are extracted from verb meanings are selected in virtue of their contribution to argument structure properties. Thus, the argument structure is decisive for the classification of verbs, not the verb semantics itself. The definition of argument structure properties “on the basis of verb classes” therefore becomes circular. Cf.: (…) advances in the understanding of argument realization regularities require isolating those semantic components which ultimately determine them. (Levin and Rappaport Hovav 2005: 16) In order to identify the grammatically relevant facets of verb meaning, it is crucial to recognize that verb meanings represent construals of events rather than the events themselves. (Levin and Rappaport Hovav 2005: 19)

2.

The argument structure positions do not fall out of the semantics of the verb itself. The positions in the logical structure representation are the argument positions in transitive and intransitive constructions. In the Actor-Undergoer Hierarchy the logical structures are designed to form a continuum, whose endpoints are agent and patient, respectively.

3.

The semantics-to-syntax linking is based on the requirements of the transitive construction. This becomes clear with Van Valin’s statement given above, where the basic definition of Actor and Undergoer does not give a semantic motivation, but rather the construction-based characterization that they are “the two primary arguments in a transitive predication” (Van Valin 2005: 60).

Thus, the definition of the “Generalized semantic roles” and also the classification of verbs with respect to “grammatically relevant facets of meaning” are based on the features of argument structure constructions. The positions in the logical structures are argument structure positions. Accordingly, it is questionable whether a theory with an elaborated account of logical structures, that explicitly denies the theoretical importance of thematic relations (see above), would necessarily need the concept of macroroles.

176

The Theoretical Importance of Constructional Schemas in RRG

3.2. Some Arguments against the Theoretical Importance of Macroroles It has been pointed out above that the argument positions found in argument structure constructions are basically characterized with respect to the role they play in the constructions. The Generalized Semantic Roles are subject to a generalization that abstracts away from the particular verb meaning in a way that makes them applicable for the argument structure pattern that is found in a transitive construction. The postulation of two macroroles instead of three makes it obvious that the basic constructional pattern is actually seen in the transitive, not in the ditransitive construction. This leads to problems with the assignment of a Macrorole to the third argument in a ditransitive construction (see Haspelmath 2008 and below for criticism). If the construction were to be considered to be responsible for the argument realization, this could lead to the abandonment of the problematic concept of macroroles. At least, these would not have to be considered to be basic elements of the theory. Abandoning the concept of macroroles would be a considerable change to the theory’s principles, but, in my opinion, it would be adequate in the following respects: ƒ

Many of the syntactic principles can be described without the help of macroroles. RRG is based on logical structures. Thematic relations are secondary, and thus the theoretical status of their generalization should not matter too much (cf. the statement from Van Valin given in Section 1).

ƒ

The definition of the macroroles is based on argument positions in logical structures and their position with respect to each other. The correlation of argument positions and semantic relations is carried out on the basis of the thematic relations continuum. Thus, “1st arg and 2nd arg of” and “leftmost” and “rightmost” suffice to identify the arguments in the logical structure. The thematic relations continuum is necessary to give a semantic reference to the argument positions in the logical structure (see also Michaelis and Ruppenhofer 2001). The number of arguments and their syntactic realizations are provided by the construction. Macroroles are not necessary.

ƒ

The signification of argument hierarchies is to map the thematic relations with the syntactic relations that appear in a sentence. These

Elke Diedrichsen

177

syntactic relations, however, are defined as argument positions in monotransitive constructions. ƒ

Recipients do not even appear in the Actor-Undergoer Hierarchy, even though they play a significant role in the syntax; for example, in recipient passives and secondative constructions (Diedrichsen 2008a; Haspelmath 2008). Still, the notion of a third macrorole is not accepted in RRG (cf. Van Valin 2004, 2005).

ƒ

Haspelmath (2008) shows that many syntactic processes like the omission of arguments can be described without referring to macroroles.

ƒ

Nolan (2010) shows that Machine translation Arabic-English on an RRG account works perfectly fine without macroroles as well.

While Haspelmath (2008) suggests the postulation of four macroroles for RRG, an extension of the idea of “constructional schemas” seems to me to be more promising, for the following reasons: 1.

The constructional schemas are there already, they do not have to be introduced into the theory. They are a very important part of RRG (Van Valin 2005: 131-135). Butler (2009: 28) points out that RRG is a “constructionist model” to a certain degree. What would be necessary, though, is to formulate constructional schemas for argument structure constructions.

2.

With constructions as main contributors of argument structure, it would be possible to describe the privileged syntactic argument [PSA], for example, with respect to the construction.

3.

The macroroles have been one source for the identification of the PSA. With a constructional account, the macroroles would be dispensable. As the previous discussion has shown, macroroles are in deficit for many reasons. They don’t suffice to describe syntactic processes and phenomena, in particular with respect to ditransitive constructions.

4.

It would be possible to treat constructions equally. Emerging constructions or spontaneous formations could be treated as constructions, not as mistakes or irregularities. This is especially

178

The Theoretical Importance of Constructional Schemas in RRG

important for the description of language change and variation. The fact that some constructions are more frequent than others would not be principally relevant for this description.

3.3 . Mismatches with Verb Valence and Argument Structure The discussion above involves theory-internal considerations that concern the definition and the prediction of the syntactic functions of argument positions with regular verbs. What about the following examples? (3)

(Frankfurter Allgemeine Sonntagszeitung, national weekly news magazine, 11-9-2008) denn es sei ungerecht, dass Frau as 3NsgNOM be.KONJ unfair that Mrs Ypsilanti “gescheitert wurde“. Y fail.PSTP be.PAST3SG “It was supposed to be unfair that Mrs Y “was failed.”

(4)

http://www.spiegel.de/netzwelt/web/0,1518,596618,00.html (found 09.05.2009) Was twittert mir Spiegel Online? whatNSGACC twitter.PRES3SG 1sgDAT SpiegelOnline (Name of the online magazine) “What is Spiegel Online twittering (to) me?”

(5)

(Common formulation among students at JLU Giessen): Ich bin eingeflext. 1sgNOM be.PRES1SG flex.inPSTP Lit.: “I am flexed in.” The past participle eingeflext means: registered in the online grading system FlexNow “I am registered in FlexNow.”

In (3), a passive construction is used with the one-place verb scheitern “fail” in order to imply that the failure was not the responsibility of the person herself, but was probably caused by someone else. So, the first of these examples shows a construction that has more argument positions than the valence of the verb would predict. In the next two examples, there isn’t even a valence that speakers could resort to in order to pick the “correct” construction – the verb in (4) is a loan word (twittern), that is used in German only in connection with the online-service Twitter. It is

Elke Diedrichsen

179

used with a ditransitive construction here, as it could be used with other expressions of communication, like erzählen “tell”, (Brief) schreiben “write (letter)”, e-mailen “e-mail”, and faxen “fax” as well. With (5), the situation gets even more complicated, as here, the form + meaning of the verb is an invention based on the name of the grading system. Thus, there is neither a valence nor an analogy of usage that speakers could resort to. Still, they create this new use and perpetuate it at least among a group of speakers. Furthermore, they have no problems at all using the new verb in an intransitive state passive construction: Flex is used as a verb stem, and the appropriate past participle form of it is formed with the regular German participle affixes ge- and –t. The lexical approach to argument structure would have problems explaining the productive use of constructions, for example with loan and novel verbs, but also with well-known verbs used in unusual constructions. As such examples are no exceptions, neither are they generally considered to be ungrammatical, and as they are a common part of the spontaneous and creative use of everyday language, the lexical approach seems to posit a too stative and restricted idea of the interplay between verb and construction.

4. Constructions as Conventions RRG considers certain generalisations, like the macroroles and the logical structures, to apply across constructions and across languages. They are captured in general principles. The language- and construction-specific features, however, are recorded in constructional schemas, which are related to the general principles, but may override them. In RRG, thus, linguistic descriptions based on the construction will only be used if the basic principles outlined in Sections 1-3 do not apply, because a language specific form-function correlation is on hand. In constructionist approaches, it is assumed that both the lexical semantics of the verb and the “construction” generally contribute to the actual argument structure expressed in a sentence. While the lexicon contributes a rich verb meaning with a “minimal valence” (Goldberg 1995; Fillmore and Kay 1997; Michaelis and Ruppenhofer 2001), the “linking construction” is the force that forms the actual sentence with its argument positions. In the RRG-oriented Lexical Constructional Model [LCM] developed by Ruiz de Mendoza and Mairal Usón (2006), it is assumed that there is a unification process between a lexical entry and a construction. “Meaning construction” (Ruiz de Mendoza and Mairal Usón 2006: 37) is achieved by

180

The Theoretical Importance of Constructional Schemas in RRG

the unification of a lexical template and a constructional template. The unification is ruled by internal constraints, which are the metalinguistic parts of the lexical template, and external constraints, which include mappings of templates based on metaphor and metonymy. The LCM gives a rich account of the semantic side of constructions by considering the contribution of the lexical template as an important component of constructional meaning. In this paper, the emphasis is on constructions as complex grammatical units. They may be associated with particular verb meanings, but should not be understood to be based on these. Both constructional idioms and argument structure constructions are considered to emerge in language use and get conventionalized as patterns whose recoverability is secured by gestalt perception, on the one hand, and their association with particular functions/meanings, on the other hand. The arguments in favour of constructionist approaches are manifold. As argued by Tomasello (2006), human learning generally works according to gestalt perception, rather than according to rules of combining single entities to meaningful complexes. The idea of constructions also makes sense with respect to the processing of syntactic structures. An incremental, “on-line” analysis of a syntactic structure requires an early projection of the emerging structure. The predictability of the emerging structure is achieved by the storage of recurring structural patterns (Auer 2006). One important argument in favour of constructions is the observation that idiomatic utterances like Cry me a river and He sneezed the napkin off the table (Goldberg 1995) exist, and that they are not exceptions, but very common in language use. These kinds of utterances are not decomposable into their parts. Their meaning cannot be explained with respect to the words that occur in the utterance. Construction Grammar describes these structures as having “sign value”, and calls them “constructions“. A “construction” is a ‘conventionalized pairing of form and function’ (Goldberg 2006: 3). As such, it should have meaning. The attribution of function/meaning on the one hand and recoverable formal properties on the other hand may turn out to be difficult for attested constructions, however. It seems that despite the intuitive appeal of positing a form/function correlation for constructions, this approach leads to new problems, as neither the function, nor the form of a construction is fully predictable, neither can they be mapped onto each other easily. Goldberg (1995) discusses the “polysemy” of ditransitive constructions assuming that there is a basic meaning of ditransitives from which tokens of the construction may deviate by implication or metaphor (Michaelis and Ruppenhofer 2001; Diedrichsen to appear). It is commonplace, however,

Elke Diedrichsen

181

that homonymy with words occurs as well, where there is more than one meaning associated with one form. Also, syntactic structures may exhibit ambiguity with respect to their constructional function, as discussed for the German bekommen-passive (Diedrichsen to appear) and the haben-passive (Diedrichsen submitted a). Syntactic reanalysis is a consequence of nondeterminate structures, and it may lead to the conventionalization of a different function for the same structure (Hopper and Traugott 2003; Nübling 2008). A theory of constructions has to account for the fact that the correlation of a form with a function is resolved by the speech act participants themselves in cases where there is no one-to-one correspondence. Moreover, as for the function/meaning part, there is hardly any specification as to what exactly is to be described here. Does the V2 sentence structure of German declarative sentences have “meaning”? Also, there is an obvious antagonism between the supposed memorability of form-/function complexes on the one hand, and their variability on the other hand. How is this to be resolved? According to Feilke (1996, 1998), both form and function of a linguistic sign originate through communication in a linguistic community. The sign is therefore “coined” in language use (Feilke 1998: 72). It gets established as a linguistic convention. Speakers’ knowledge of the convention is based on communicative experiences in their common culture. As for the variability, Jackendoff (2002) remarks that there is generally some gradience in the memorability of linguistic units, be they single words or large portions of text like in Hamlet, for example (Jackendoff 2002: 152f.) The extent to which a string of linguistic units can be memorized and recalled cannot be predicted, and thus, idioms in individual languages have varying complexity. Also, the extent to which parts of utterances are either stored or constructed online is not altogether predictable. Furthermore, the productivity of idiomatic constructions varies. For example, in English, there are expressions with semiproductive derivational morphology like denominal verbs. Not every noun has a corresponding denominal verb, and if it has one, the meaning is not predictable. (6) (examples after Jackendoff 2002: 158f.) a. butter a toast = put butter on a toast b. *mustard the sandwich= put mustard on the sandwich c. shelve the books = put the books on the shelf d. *table the books = put the books on the table

182

The Theoretical Importance of Constructional Schemas in RRG

In these examples, both the “put N on” and the “put on N” meanings are available, but not with every noun! Also, with many of the denominal verbs, the “put something in/on something else” interpretation is not sufficient to understand the meaning of the expression. To put one single book or a clock on the shelf would not mean “to shelve it”, and to put wine into a bottle would not suffice to bring about the state of affairs that is expressed in “to bottle the wine” (Jackendoff 2002: 158f.). There are, however, constructions in English which allow a “combination of specified constituents and free variables” (Jackendoff 2002: 175). Jackendoff gives examples of the one’s head off construction and the way construction, where some of the structural elements are fixed, while other elements are free variables. (7)

read, swim, google, twitter, email one’s head off=’do action to excess’ (after Jackendoff 2002: 173; some examples mine, ED)

(8)

swim your way to glory, drinks his way through the evening, homer their way into the hearts of America, google my way to wisdom = “traverse the path PP while/by doing V” (cf. Jackendoff 2002: 174; Goldberg 1995; some examples mine, ED)

The resultative construction even provides only the structure, which can be filled by free variables altogether. In these cases, the arguments are selected not by the verbs, but by the constructions (Jackendoff 2002: section 6.6). These examples show that “constructional idioms” have different degrees of variability. The slots for the variable elements are fixed in the structure and thus learnable. (…) when a pattern with a variable develops, the relation among the stored items “goes productive”: the pattern can participate in free combination with other lexical items that satisfy its typed variables, and new combinations need no longer be stored. (Jackendoff 2002: 189)

Regular patterns require a learning process that is comparable with the traditional conception of learning a rule. It accords the acquisition of a linguistic convention that takes place when a speaker is exposed to the use of the particular pattern in a language community. The variables are then learned from “instances” of the occurrence of the construction (ibid.), and the great variety of meanings and possibilities to fill the variables develop slowly in the course of the emergence of the construction (Jackendoff 2002: 190, after Tomasello 2000).

Elke Diedrichsen

183

The following section will suggest constructional schemas for selected constructions of German. It will be demonstrated that the features of argument structure constructions can be represented in constructional schemas and thus account for constructions where the argument structure is not predictable by the valence of the verb.

5. Constructional Schemas If the construction is not regarded as a solid form/function unit, but rather as a learnable pattern with a learnable variety of variables, the productivity and emergence of structures can be explained (see Jackendoff 2002: 180 for a similar argument). This approach is interesting for argument structure constructions as well, as here, the argument structure can be regarded as a productive structure, and the variables are the argument structure positions. RRG recognizes the importance of constructions by positing constructional schemas, but only for the “idiosyncratic, language-specific features of constructions” (Van Valin 2005: 132). Thus, there are constructional schemas for passives, antipassives, conjunction reduction and whquestions. These schemas are considered to be stored and applied in cases where the construction is not a direct consequence of the valence of the verb and the general argument realization principles applied in RRG, which include the logical structures, the macroroles and the ActorUndergoer Hierarchy. These are assumed to be the cross-linguistic basis of argument structure constructions. Accordingly, there are no constructional schemas for intransitive, transitive, and ditransitive constructions. The general idea of a constructional schema is that it informs about the features of a construction and thus reflects the knowledge that is stored with respect to a conventionalized construction in a language community. It includes syntactic, semantic and pragmatic properties of a construction. Van Valin (2005: 239) provides a representation and discussion of resultative constructions in English. He demonstrates an application of the constructional schema of the English resultative construction to Goldberg’s (1995: 9) famous “sneezed the napkin” example, where the unusual argument structure is resolved by the construction: The valency of the verb is not changed, but the “napkin” is presented as an argument of the full logical structure of the resultative construction (cf. Butler 2009: 29f. for discussion). My suggestion is to use the constructional schema as a general descriptive tool for constructions. It contains explicit information about the fixed structural elements that would help the speaker/hearer recover the

184

The Theoretical Importance of Constructional Schemas in RRG

construction in the online production of utterances. As speakers also know about the productivity and variability of constructions by convention, the constructional schema has to inform about variable elements of the construction as well. The advantages of such an extension of the theoretical impact of constructional schemas include a new compatibility of RRG for the description of syntactic change and variation. Thus, this approach facilitates the description of grammaticalization phenomena in RRG. The entire issue of spontaneity, change, novelty, and variation that has been missing in previous approaches to RRG, can be fitted easily into the theoretical framework without changing its major properties.

5.1. The New Role of Constructional Schemas in the Linking In this section, it will be demonstrated how a semantics-to-syntax linking for German sentences can be carried out by adding the benefits of a constructional perspective to constructional schemas, as discussed above, to the steps of the linking as suggested by Van Valin (2005), for example. One important aspect of this procedure is the decision as to the place the constructional schema will occupy in the linking. It has been shown for coordination and conjunction reduction (see Diedrichsen 2009: 163 ff.), that the constructional schema has to be the first position in the linking. It will be argued below that this holds for argument structure constructions as well. According to a new approach from Nolan (submitted a, b, this volume), the construction can be viewed as a “grammatical object” that is saved in a structured repository of constructions. This repository represents the linguistic knowledge of an individual in a linguistic community. A constructional schema is the representative of a grammatical object in linguistic theory. The construction as a grammatical object is highly complex and variable. It includes mechanisms for the processing of structures, which are considered to apply in a “workspace” that is part of the operational potential of the construction (Nolan submitted a, b, this volume; cf. also Van Valin 2005). Each grammatical object has a “signature” (term after Nolan submitted a, b, personal communication) that identifies the construction and activates the constructional schema, i.e. the functions, the language-specific properties and the processing mechanism. This “key” is based on morphosyntactic information; it is a morphosyntactic pattern. The “constructional signature” (Nolan submitted a, b) is unique to the

Elke Diedrichsen

185

construction and should therefore be distinctive. In an ideal case, there is a one-to-one correspondence between the morphosyntactic pattern and the function assigned to it. But, as argued in Section 4, in many cases there is no clear form-function correlation, and thus, there has to be another way of determining a given construction, when the signature is not univocal. As speakers are able to determine linguistic constructions despite their formal ambiguity, I suggest one needs to add a “constraint” to the “signature” that determines the construction in the particular situation of use (Diedrichsen submitted a). The constraint is a condition for the identification of a construction. It is effective on the basis of the immediate speech situation in terms of its linguistic and non-linguistic surroundings. By use of context-related knowledge, general knowledge and linguistic knowledge, the recipient of an utterance has to understand that the constraint is fulfilled, in order to process the construction in the way that is intended by the speaker. The determination of the meaning of signs is generally related to the situation of use, and therefore, such a “context constraint” for the determination of constructional functions is considered to be adequate. I will concentrate on the steps of the semantics-to-syntax-linking that are of interest with regard to argument realization. For the application of further details of the RRG-based semantics-to-syntax-linking, I refer the reader to Van Valin (2005), Van Valin and Diedrichsen (2006), and Diedrichsen (2009). The Macrorole assignment principles, the Case assignment rules, the PSA selection principle and verbal agreement can be neglected, as all of this is ruled by the construction. There would however, have to be some statement about the cases that are distinguished in a language and the distribution of overt vs. non-overt case marking. I leave this out here as well (see Van Valin and Diedrichsen 2006 and Diedrichsen 2009). In the following, I will suggest constructional schemas for the German sentence structures “declarative” and “W-Question”. Each of these will be combined with an argument structure constructional schema to illustrate the representation of actual sentence structures by use of constructional schemas (see Diedrichsen 2010 for an introduction and justification of constructional schemas for intransitive, transitive, and ditransitive constructions). Note that the RRG concept of syntactic templates with core slots is available, so the argument positions are to be understood as argument positions in the syntactic template, which involves a core, a periphery and a precore slot. They do not have to be derived from the semantics of the verb or the construction.

186

The Theoretical Importance of Constructional Schemas in RRG

5.2 . Linking Semantics to Syntax The linking from semantics to syntax involves five steps in the original version (e.g. Van Valin 2005: 129 ff.). It is a linking process from the semantic representation, which is the logical structure, of the verb, to the syntactic representation of the full sentence, where all of the syntactic features, like PSA selection, agreement, case marking, syntactic structure and word order are accounted for. Many of the steps in the linking process are considered to be universal, while the steps that involve syntactic features of a particular language, like case marking, are taken to be language-specific. In the semantics-to-syntax-linking presented here, many of the steps involve the reference to a language-particular constructional schema (cf. Table 2, for example). The benefit of this is, as argued before, that the system is able to account for varieties of uses of verb meanings, which is a considerable part of linguistic interaction and should not be neglected by a syntactic theory. While in the original version the arguments were assumed to be selected by the verb, it is now assumed that the construction selects the arguments. This also means that the constructional schema is the first position in any semantics-to-syntax-linking. It determines the semantic representation and also the selection of the template(s). Modified versions of the Completeness Constraint and the Core syntactic template selection principle are given below: (9) Completeness Constraint: All of the arguments explicitly specified in the semantic representation of a sentence, as determined by the construction, must be realized syntactically in the sentence, and all of the referring expressions in the syntactic representation of a sentence must be linked to an argument position in a logical structure in the semantic representation of the sentence. (10) Core template selection principles a. Core syntactic template selection principle: The number of syntactic slots for arguments within the core is equal to the number of distinct specified argument positions in the semantic representation of the core. b. Construction-based specifications: The number of distinct specified argument positions in the semantic representation of the core will be determined by the construction.

Elke Diedrichsen

187

Note here that in (10b), the language-specific qualifications are replaced by “construction-based specifications”. The particulars about the “minimum syntactic valence” and the fact that the passive reduces the number of core slots by one can be neglected, as these facts are all ruled by the construction.

5.3. Some Applications In the following, I will give two examples of the semantics-to-syntax linking. They involve combinations of constructional schemas (cf. Diedrichsen 2010 for more examples of the semantics-to-syntax linking). The first example is a simple transitive construction in a declarative sentence (cf. Van Valin and Diedrichsen 2006). (11)

David hat das D have.PRES3SG DETNsgACC gewaschen. wash.PSTP “David has washed the car.”

Auto car.SG

Table 2: Constructional schema for German declarative main clause construction CONSTRUCTION: German declarative main clause construction SIGNATURE: PrCS [COREVFIN............VINFIN etc. (see below)] PoCS CONSTRAINT: No rising intonation contour (would be mistaken as question, despite of the word order) WORKSPACE: Real-time processing according to the following construction-specific rules: SYNTAX: Template: 1-3 core arguments; V2-structure, where the finite verb is the first position in the core (PreCoreSlot position is obligatorily filled by one item: Argument/adjunct/non-finite verb/expletive element. Greater number of PrCS elements possible due to pragmatic requirements (Diedrichsen 2008b). “Bracket Structure“ (cf. Drach 1963): Last position in the core is filled by one of the following elements (if they appear): non-finite part of periphrastic verb form, separable prefix, part of collocation, predicative element (Hentschel and Weydt 2003; Nübling 2008).

188

The Theoretical Importance of Constructional Schemas in RRG

PSA [“subject”]: May be any core argument MORPHOLOGY: RPs: Case marking subject to noun type and declension class Auxiliary in the perfect: May be any perfect auxiliary PHONOLOGY: Intonation contour falls against the end of the core SEMANTICS: not specified, see PRAGMATICS for function PRAGMATICS: Illocutionary force: Declarative Focus structure: No restrictions; PSA = topic (default) Table 3: Constructional schema for German transitive construction CONSTRUCTION: German active transitive construction SIGNATURE: RP1...... VFIN.......RP2 (any order, any number of adjuncts, RP1 and RP2 may be distinguished by case marking) CONSTRAINT: VFIN is not a copula WORKSPACE: Real-time processing according to the following construction-specific rules: SYNTAX: Template: 2 core arguments: x, y PSA [‘subject’]: Highest ranking argument (default) Linking: Highest-ranking argument (agent) will be nominative, lowest-ranking Argument (patient) will be accusative (default) MORPHOLOGY: RPs: Case marking subject to noun type and declension class Auxiliary in the perfect: haben (default), sometimes sein with verbs of motion in a transitive use SEMANTICS: PSA is instigator of state of affairs (default), other core argument is affected or effected (there may be deviations) PRAGMATICS: Illocutionary force: Unspecified Focus structure: No restrictions; PSA = topic (default)

Elke Diedrichsen

189

Semantics-to-syntax-linking for (11): David hat das Auto gewaschen. Step 1. Combine the constructional schemas. a. Construct the semantic representation of the sentence, based on the construction (Table 3) and the predicator. do’ (x, [wash’ (x, y)] b. Determine the value of the operators to be expressed. (For the sake of ease of presentation, this will not be carried out here). c. Select the referring expressions to fill the variable positions in LS, according to the activation statuses of the referents (for this procedure, see Diedrichsen 2009). The following activation statuses are distinguished: Active=ACV: actively under consideration in the discourse by means of direct mention. Accessible=ACS: not actively under consideration, but readily recognized by the addressee due to world knowledge or occurrence in the situation. Inactive=INA: previously mentioned but not actively under consideration, not assumed by the speaker to be recognized by the addressee. Brand-new anchored=BNA: not previously mentioned but related to something already mentioned or accessible. Brand-new unanchored=BNU: not previously mentioned or related to anything previously mentioned (Van Valin 2005: 79 f.; Prince 1981; Chafe 1987). do’ (David, [wash’ (DavidACS, AutoACS)] In Step 2, the determination of the actor and undergoer assignments is carried out in the original model. As it is argued here that the argument realization is determined by the transitive construction, this step is obsolete here. Step 2. Determine the morphosyntactic coding of the arguments on the basis of the constructional schema in Table 3. a. PSA: Highest ranking argument b. Highest-ranking argument (agent) will be nominative, lowest-ranking argument (patient) will be accusative (default). The case marking is subject to RP type and declension class. c. Agreement marking: Finite verb agreement is always with the PSA.

190

The Theoretical Importance of Constructional Schemas in RRG

Step 3. Select the syntactic template(s) for the sentence, according to the general rules from Van Valin and Diedrichsen (2006) and Diedrichsen (2009). For the core template, follow the core template selection principles and the specifications in the Constructional Schemas. Step 4. Assign LS elements to positions in the syntactic representation, according to the constructional schema in Table 2. Semantics-to-syntax-linking for (4): (4)

Was twittert mir whatNSG twitter.PRES3SG 1sgDAT Spiegel Online? SpiegelOnline (Name of the online magazine) “What is Spiegel Online twittering (to) me?”

Here, a constructional schema for a German W-question, which is the functional equivalent of a WH-question in English (cf. Van Valin 2005: 133 for the respective constructional schema) is combined with a constructional schema for the German ditransitive construction. Table 4: Constructional schema for German W-question CONSTRUCTION: German W-question SIGNATURE: [PreCoreSlotW-word [COREVFIN............VINFIN etc. (see below)] PoCS] CONSTRAINT: No falling intonation contour (would otherwise be mistaken as a statement, in terms of a “rhetorical question“) WORKSPACE: Real-time processing according to the following construction-specific rules: SYNTAX: Template: 1-3 core arguments, V2-structure, where the finite verb is the first position in the core (PreCoreSlot position is obligatorily filled by W-word). “Bracket Structure“ (cf. Drach 1963): Last position in the core is filled by one of the following elements (if they appear): non-finite part of periphrastic verb form, separable prefix, part of collocation, predicative element (Hentschel and Weydt 2003, Nübling 2008). PSA [“subject”]: May be any core argument

Elke Diedrichsen

191



MORPHOLOGY: RPs: Case marking subject to noun type and declension class Auxiliary in the perfect: May be any perfect auxiliary PHONOLOGY: Intonation contour rises against the end of the core SEMANTICS: Contains an open proposition with a variable α, W-RP=α PRAGMATICS: Illocutionary force: interrogative Focus structure: narrow focus on W-RP (Default: PrCS) Table 5: Constructional schema for German ditransitive construction CONSTRUCTION: German ditransitive construction SIGNATURE: RP1...... VFIN.......RP2...RP3 (any order, any number of adjuncts, RPs may be distinguished by case marking) CONSTRAINT: VFIN is not a copula, Dative is “indirect object” (recipient, benefactive, malefactive) (cf. Diedrichsen in prep. for a discussion of Datives in German non-ditransitive constructions) WORKSPACE: Real-time processing according to the following construction-specific rules: SYNTAX: Template: 3 core arguments; x, y, z; one may appear in the Periphery (see below) PSA [“subject”]: Highest ranking argument Linking: Highest-ranking argument (agent) will be nominative, lowest-ranking Argument (patient/theme) will be accusative (default), second-highest ranking argument will be dative or in peripheral PP MORPHOLOGY: RPs: Case marking subject to noun type and declension class Auxiliary in the perfect: haben SEMANTICS: PSA is instigator of state of affairs (default), lowest-ranking core argument is affected or effected; third argument is recipient, benefactive or malefactive PRAGMATICS: Illocutionary force: Unspecified Focus structure: No restrictions; PSA = topic (default)

192

The Theoretical Importance of Constructional Schemas in RRG

Semantics-to-syntax-linking for (4): Was twittert mir Spiegel Online? Step 1. Combine the constructional schemas. a. Construct the semantic representation of the sentence, based on the construction in Table 5 and the predicator: do’ (x, twitter) CAUSE BECOME [know’ (y, z)] b. Determine the value of the operators to be expressed. (For the sake of ease of presentation, this will not be carried out here). c. Select the referring expressions to fill the variable positions in LS, according to the activation statuses of the referents do’ (Spiegel OnlineACS, twitter) CAUSE BECOME [know’ (ichACV, wasBNU)] Step 2. Determine the morphosyntactic coding of the arguments on the basis of the constructional schema for the ditransitive construction. a. PSA: Highest ranking argument b. Highest-ranking argument (agent) will be nominative, lowest-ranking argument (patient) will be accusative (default), second-highest ranking argument will be dative. The case marking is subject to RP type and declension class. c. Agreement marking: Finite verb agreement is always with the PSA. Step 3. Select the syntactic template(s) for the sentence, according to the general rules from Van Valin and Diedrichsen (2006) and Diedrichsen (2009). For the core template, follow the core template selection principles and the specifications in the constructional schemas. Step 4. Assign LS elements to positions in the syntactic representation, according to the constructional schema in Table 4.

6. Conclusion In this paper, some of the basic assumptions of RRG have been reorganized in order to account for a more flexible view of syntactic argument structure constructions. In particular, it has been argued that not a generalized notion of the semantics of the verb, in terms of “grammatically relevant facets of meaning” should be considered to be the functional basis of argument structure constructions, but rather the

Elke Diedrichsen

193

“construction” as gestalt that is stored and recovered in linguistic interaction. To this effect, it has been suggested to enhance the theoretical impact of the constructional schemas which are an important component of RRG-based linguistic descriptions anyway. It is proposed to use constructional schemas from RRG as descriptive tools for all kinds of constructions. The construction is viewed as an entity in the linguistic knowledge of a speaker: A grammatical object (Nolan submitted a, b, this volume). The constructional schemas serve as representatives of the constructions. They are supposed to reflect interlocutors’ knowledge of a construction. Thus, they can be filled with syntactic, semantic, and pragmatic information, but possibly also with other facets of knowledge that interact in the production and recovery of a construction in language use. With this account, the syntax/semantics interface is built on constructions, while the concept of macroroles has been found to be not sufficient for the description of many important phenomena, on the one hand, and gratuitous for the aims of the theory, on the other hand. The big advantage of a constructional account is seen in the possibility to describe structures with an unusual verb/construction combination. These are commonly found in everyday language, and they illustrate language creativity, spontaneity and variation. They can also be indicators of syntactic innovation in the sense of grammaticalization. For these reasons, they should not be ignored by a modern syntactic theory. Following Jackendoff’s (2002) approach, it is argued that the character of constructions as learnable items is to be taken seriously, but it has to allow for formal and semantic variability, which is not sufficiently considered in many Constructionist approaches that postulate a form/function correlation for constructions. The properties of the construction as given in a constructional schema are to be seen as based on conventions in language use. The recognition of constructions in language use is assured by formal properties called “signature” (Nolan submitted a, b, this volume). As there is hardly ever a direct correspondence between form and function, speakers’ ability to resolve ambiguities by context information is recognized as linguistic knowledge in the constructional schema as well, in terms of a “constraint” for the correlation of a morphosyntactic pattern with a particular construction. The paper has provided an illustration of the integration of constructional Schemas for argument structure constructions into the semantics-to-syntax linking for a number of attested examples from Standard German. Further investigations on the RRG/constructions framework will be necessary to give a more detailed account of the “construction as a

194

The Theoretical Importance of Constructional Schemas in RRG

grammatical object” and the principles behind the combinations of constructions. There has to be a thorough examination of the pragmatic factors behind the selection of constructions as well. As for the crosslinguistic aspects, a comparison of constructional schemas in different languages would be a very promising field of study.

List of Abbreviations ASP AUX COMP DEC DEIC DEM IF INDEF INF INGR IMP LOC MOD NEG NUC NUM

aspect auxiliary comparative degree declarative deictic demonstrative illocutionary force indefinite infinitive ingressive imperative locative modality negative nucleus number

PART PAST PERF POCS POSS PP PRCS PRED PRES PROP PSTP PURP REFL RP SUBJ TNS

particle past (tense) perfect postcore slot possessive prepositional phrase precore slot predicate present (tense) proper past participle purposive reflexive referential phrase subjunctive tense

References Auer, Peter. 2006. Construction Grammar meets Conversation: Einige Überlegungen am Beispiel von “so”-Konstruktionen. In S. Günthner and W. Imo, eds., Konstruktionen in der Interaktion, 291-314. Berlin and New York: De Gruyter. Butler, Christopher S. 2009. The Lexical Constructional Model: genesis, strengths and challenges. Available at: http://www.lexicom.es/drupal/ files/Butler_2009.pdf Chafe, Wallace. 1987. Cognitive constraints on information flow. In R. Tomlin, ed., Coherence and Grounding in Discourse, 21-51. Amsterdam and Philadelphia: John Benjamins. Diedrichsen, Elke. 2006. Ergativität und Diskurs. Berlin: LIT.

Elke Diedrichsen

195

—. 2008a. The grammaticalization of the bekommen-passive in an RRGperspective. In R. Kailuweit, B. Wiemer, E. Staudinger, and R. Matasovic, eds., New Applications of Role and Reference Grammar: Diachrony, Grammaticalization, Romance Languages, 87-145. Newcastle: Cambridge Scholars Publishing. —. 2008 b. Where is the precore slot? mapping the layered structure of the clause and German sentence topology. R. D. Van Valin, Jr., ed., Investigations of the Syntax-Semantics-Pragmatics Interface, 203-224. Amsterdam and Philadelphia: John Benjamins. —. 2009. Exploring the role of pragmatics in the interface: on the interrelations between activation status, choice of RP and syntax. In L. Guerrero, S. Ibáñez, V. Cerda, and A. Belloro, eds., Studies in Role and Reference Grammar, 143-177. México: The IIFL-UNAM Press. —. 2010. Towards a reconsideration of Constructional Schemas in RRG: are all constructions driven by “constructions”? In W. Nakamura, ed., Proceedings of the 10th International RRG Conference (RRG 2009), 52-79. Available at: http://linguistics.buffalo.edu/people/faculty/ vanvalin/rrg/ProceedingsofRRG2009_02.pdf —. to appear. What you give is what you GET? On reanalysis, semantic extension and functional motivation with the German bekommenpassive construction. In A. N. Lenz and G. Rawoens, eds., The Art of Getting: GET Verbs in European Languages from a Synchronic and Diachronic Point of View (Special edition of Linguistics) —. Submitted a. Zur “Inventarisierung” von idiomatischen und Argumentstruktur-Konstruktionen im Deutschen. To appear in A. Ziem and A. Lasch, eds., Grammatik als Inventar von Konstruktionen? Sprachliches Wissen im Fokus der Konstruktionsgrammatik, Series “Sprache und Wissen” Berlin: De Gruyter. —. Submitted b. The influence of person hierarchies on preferences with word order in German ditransitive constructions. In A. Siewierska and E. van Lier, eds., Referential Hierarchies in Three-participant Constructions, Online-Journal Linguistic Discovery. Available at: http://journals.dartmouth.edu/cgi-bin/WebObjects/Journals.woa/2/xml page/1/issue Dixon, R. M. W. 1994. Ergativity. Cambridge: Cambridge University Press. Drach, Erich. 1963. Grundgedanken der deutschen Satzlehre. 4th edition. Frankfurt am Main: Diesterweg. Feilke, Helmuth. 1996. Sprache als soziale Gestalt, Frankfurt am Main: Suhrkamp.

196

The Theoretical Importance of Constructional Schemas in RRG

—. 1998. Idiomatische Prägung. In I. Barz and Ö. Günther, eds., Zwischen Grammatik und Lexikon, 69-80. Tübingen: Niemeyer. Fillmore, Charles. 1968. The case for case. In E. Bach and R. Harms, eds., Universals in Linguistic Theory, 1-88. New York: Holt, Rinehart and Winston. Fillmore, Charles and Paul Kay. 1997. Construction Grammar. Available at: http://www.icsi.berkeley.edu/~kay/bcg/ConGram.html Goldberg, Adele E. 1995. Constructions. A Construction Grammar Approach to Argument Structure. London and Chicago: University of Chicago Press. —. 2006. Constructions at Work: The Nature of Generalization in Language. Oxford: Oxford University Press. Haspelmath, Martin. 2008. Ditransitive constructions: towards a new Role and Reference Grammar account? In R. D. Van Valin, Jr., ed., Investigations of the Syntax-Semantics-Pragmatics Interface, 75-100. Amsterdam and Philadelphia: John Benjamins. Hentschel, Elke and Weydt Harald. 2003. Handbuch der deutschen Grammatik. 3rd edition. Berlin and New York: De Gruyter. Hopper, Paul J. and Elizabeth Closs Traugott. 2003. Grammaticalization. 2nd edition. Cambridge: Cambridge University Press. Jackendoff, Ray. 1976. Toward an explanatory semantic representation. Linguistic Inquiry 7: 89-150. —. 2002. Foundations of Language. Oxford: Oxford University Press. Levin, Beth and Malka Rappaport Hovav. 2005. Argument Realization. Cambridge: Cambridge University Press. Michaelis, Laura A. and Josef Ruppenhofer. 2001. Beyond Alternations: A Constructional Model of the German Applicative Pattern. Stanford, CA: CSLI Publications. Nolan, Brian. 2010. UniArab: an RRG Arabic-to-English machine translation software. In W. Nakamura, ed., Proceedings of the 10th International RRG Conference (RRG 2009), 52-79. Available at: http://wings.buffalo.edu/linguistics//people/faculty/vanvalin/rrg/Procee dingsofRRG2009_02.pdf —. this volume. Meaning construction and grammatical inflection in the layered structure of the Irish word: an RRG account of morphological constructions. —. Submitted a. Constructional polysemy and argument realisation with the Irish GET verb. To appear In A. N. Lenz and G. Rawoens, eds., The Art of Getting: GET Verbs in European Languages from a Synchronic and Diachronic Point of View. (Special edition of Linguistics). Preliminary manuscript available at: http://itb-Dublin-

Elke Diedrichsen

197

Ireland.academia.edu/BrianNolan/Papers/415615/constructional/polyse my_and_argument_realisation_with_the_Irish_GET_verb —. Submitted b. Resolving ambiguity in the realisation of the Irish GET construction. To appear in B. Johanna, M. Cennamo, and E. van Gelderen, eds., Argument Structures: The Naples/Capri Papers. CSLI Publications/John Benjamins. Nübling, Damaris. 2008. Historische Sprachwissenschaft des Deutschen. 2nd edition. Tübingen: Narr. Prince, Ellen. 1981. Toward a taxonomy of given-new information. In P. Cole, ed., Radical pragmatics, 223-256. New York: Academic Press. Ruiz de Mendoza, Francisco J. and Ricardo Mairal Usón. 2006. Levels of semantic representation: where lexicon and grammar meet. Interlingüística 17: 26-47. Tomasello, Michael. 2000. Do young children have adult syntactic competence? Cognition 74: 209-253. —. 2006. Acquiring linguistic constructions. In W. Damon, R. M. Lerner, and R. S. Siegler, eds., Handbook of Child Psychology, Volume 2, Cognition, Perception and Language, 6th edition, 255-298. New York: Wiley. Van Valin, Robert D., Jr. 2004. Semantic Macroroles in Role and Reference Grammar. In R. Kailuweit and M. Hummel, eds., Semantische Rollen, 62-82. Tübingen: Narr. —. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press. —. 2007. Review of Goldberg 2006. Journal of Linguistics 43.1: 234-240. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Elke Diedrichsen. 2006. A Bonsai Grammar for German. Available at: http://linguistics.buffalo.edu/people/faculty/ vanvalin/rrg/BonsaiGrammarGerman.pdf

PART 3. SYNTAX-SEMANTICS INTERFACE

PP TYPES IN RRG: A TOP-DOWN APPROACH TO THEIR CLASSIFICATION SERGIO IBÁÑEZ CERDA UNIVERSIDAD NACIONAL AUTÓNOMA DE MÉXICO

Abstract Following Jolly’s work (1993) on preposition assignment, Role and Reference Grammar [RRG] (Van Valin and LaPolla 1997; Van Valin 2005) assumes three kinds of prepositional phrases [PPs] in terms of their semantic and syntactic status within the clauses: PPs which function as oblique core arguments; PPs which have the status of adjuncts; and PPs functioning as argument-adjuncts in the core. The aim of this paper is to build on this general schema and some additional proposals from Ibáñez (2009) and to propose a more fine-grained classification of PPs on the basis of three binary features ([±argument], [±core], and [±predicative]). The basic idea is that these binary features and their combinations provide a basis for a principled classification of PP types under the assumption that the three (i.e. semantic, syntactic, and categorical) dimensions are independent of each other. This assumption makes it possible to derive from the three binary features 8 logically possible PP types, which will be illustrated with data taken mainly from Spanish.

Keywords Syntax-semantics interface, argument-adjunct distinction

1. Introduction Following Jolly’s (1993) work on preposition assignment, RRG assumes three kinds of PPs in terms of their relations to predicates within the clauses in which they appear: 1) PPs introduced by a predicative

Sergio Ibáñez Cerda

201

preposition which code adjuncts and function as peripheral modifiers of the core (i.e. adjuncts in the periphery); 2) PPs introduced by a nonpredicative preposition which code verbal arguments (i.e. oblique core arguments); and 3) PPs which are introduced by a predicative preposition, but code verbal arguments (i.e. argument-adjuncts in the core). These three types of PPs are exemplified in (1a)-(1c): (1) a. b. c.

Adjuncts (in the periphery) John walks everyday in the park. Oblique core arguments Peter gave the book to Mary. Argument-adjuncts (in the core) Leslie put the book in /on/behind/under the box.

Building on this general schema and some additional proposals from Ibáñez (2009), this paper outlines a more fine-grained classification of PP functions in terms of three dimensions: a) the semantic nature of the participants coded (whether they are verbal arguments or not); b) their syntactic status (whether they belong to the core or periphery); and c) the nature of their prepositions (whether they are predicative or not). The main goal here is to develop an expanded system of PP types by considering possible combinations of the three features and to illustrate these PP types using data from Spanish. Before doing so, however, it is necessary to have a look at the RRG theory of the semantics-syntax interface.

2. The Semantics and Syntax of Argument Structure: Parallel Structures, Iconicity, and Mismatches 2.1. The RRG Conception of the Semantics-Syntax Interface The notions of argument (or actant) and adjunct (or circumstant) have gone through several changes, since they were first proposed by Tesniere (1959). This dichotomy has been used to refer to a syntactic or semantic distinction in the modern syntactic theories, but it is the case that most theories use the distinction in a syntactic way and that irrespective of their theoretical persuasion, they usually work with a very limited notion of what an argument is. Specifically, most of the syntactic theories assume that arguments are those clausal participants that are strictly required for producing a clause, i.e. those that require obligatory syntactic coding. In fact, in most cases, they take into account only those arguments that have a NP coding.

202

PP Types in RRG: A Top-down Approach to Their Classification

What distinguishes RRG from other theories is that as a functionally driven framework, it takes semantics as a starting point for the analysis of clause structure. The layered structure of the clause is motivated by two universal semantic distinctions: the one between the predicate and nonpredicating elements and the one between arguments and non-arguments among the non-predicating elements (Van Valin 2005):   Predicate + Arguments 



Non-arguments 

Figure 1: Universal semantic distinctions underlying the clause structure

The above distinctions are necessitated by the fact that human beings exchange conceptual representations of the world in communication. These representations are constituted by the states of affairs and the participants involved in the states of affairs. The states of affairs are denoted by predicates, while their participants correspond to the nonpredicating elements (arguments and non-arguments). Furthermore, the distinction between arguments and non-arguments is based on the idea that the states of affairs represented by the predicates inherently determine the number and type of the participants involved. In this view, the participants inherently required by the predicate are arguments, while those that are not are non-arguments. Based on these semantic distinctions, RRG proposes a layered structure of the clauses formed by three units: a) the nucleus; b) the core; and c) the periphery, as represented in Figure 2:

CLAUSE CORE NUCLEUS

PERIPHERY

Figure 2: The layered structure of the clause

RRG assumes that syntax and semantics are parallel structures in a similar way to Lexical Functional Grammar [LFG] (Bresnan 2001) and Functional Grammar (Dik 1997). This means that RRG has no room for any derivational representations and assumes a single level for each of the two parallel levels of representations, syntax and semantics.

Sergio Ibáñez Cerda

203

There are natural correspondences between the semantics and syntax. First, the predicate typically corresponds to the nucleus of the clause. Second, given the iconicity principle (Haiman 1980), it is expected that the predicate and its semantic argument(s) belong to the core. Finally, nonarguments are expected to occur in the periphery of the clause. RRG goes on to divide core arguments into direct core arguments and oblique core arguments on the basis of their morphosyntactic properties. Direct core arguments bear a direct morphological case 1, while oblique core arguments bear a non-direct case or preposition. Non-arguments, which are expected to occur in the periphery, have the status of adjuncts. In this general framework, a clause like “John gave a book to Mary at the library” can be represented as follows:

CLAUSE CORE

PERIPHERY

John gave a book to Mary

in the library

NUCLEUS Figure 3: Layered structure of “John gave a book to Mary in the library”

Both “John” and “the book” are direct core arguments, while “to Mary” is an oblique core argument. These three clausal participants are semantically required by the verb, while “in the library” serves as an adjunct. A fourth type of clausal complement originally proposed by Jolly (1993) is an argument-adjunct in the core. This corresponds to a PP that is introduced by a predicative postposition, but codes a verbal argument. It is important to emphasize at this juncture that the layered structure of the clause is not restricted by word order phenomena. It holds for any kind of grammatical system and allows us to identify semantic and syntactic correspondences that are common to all languages.

1

Direct cases are nominative and accusative in languages with an accusative case system. They are absolutive and ergative in languages with an ergative case system.

204

PP Types in RRG: A Top-down Approach to Their Classification

2.2. Mismatches between Semantics and Syntax Starting from the RRG conception of clause structure and incorporating the notion of parallel structure as developed in LFG, I claim that it is important to distinguish the semantic distinction between arguments of predicates and non-arguments (i.e. adjuncts) from the syntactic distinction between the core and periphery. Their independence makes room for mismatches between the semantic and syntactic level of representation, which will be illustrated in the next section. Before proceeding, a word is in order about how to define the core status with no reference to (semantic) argument status. Van Valin and LaPolla (1997) propose that one of the major criteria for determining the core status of the clausal participants is their possibility of serving as privileged syntactic arguments [PSA]. 2 The PSA is divided into the syntactic pivot (i.e. an NP that is the target of syntactic operations including relativization, coordinate reduction, equi-NP-deletion, raising, and reflexivization) and the syntactic controller (i.e. an NP which controls morphological expressions such as case marking and verb agreement). Given the RRG definition of the core status, we are ready to investigate possible combinations of the two distinctions referred to above.

3. A Classification of PP Types: A Top-down Approach In addition to the semantic distinction (arguments vs. non-arguments) and syntactic distinction (core vs. periphery), there is a third distinction that is needed to propose the classification of the PP types. Since the work of Jolly (1993), the categorical distinction between predicative and nonpredicative prepositions has been implemented in RRG. Predicative prepositions are used to code non-arguments and argument-adjuncts, while non-predicative prepositions are used to code oblique core arguments. It is convenient to represent the three distinctions in terms of three binary features, [±argument] (argument vs. non-argument), [±core] (core vs. periphery), and [±predicative] (predicative vs. non-predicative). Combining the distinction between predicative and non-predicative prepositions with the semantic and syntactic distinction outlined above yields 8 logically possible PP types: 3 2 The concept of PSA is proposed by RRG as an alternative to the standard view of grammatical relations such as the subject. 3 Almost all of the proposed labels for the PP types have been used in RRG studies. The contribution of the present paper is that it derives the variation of PPs from different combinations of the three features.

Sergio Ibáñez Cerda

1.

2.

3.

4.

5.

6.

7.

8.

205

Oblique Core Arguments [+argument] [+core] [-predicative] Peripheral Adjuncts [-argument] [-core] [+predicative] Argument-Adjuncts in the Core [+argument] [+core] [+predicative] Arguments in the Periphery [+argument] [-core] [-predicative] Argument-Adjuncts in the Periphery [+argument] [-core] [+predicative] Adjuncts in the Core [-argument] [+core] [+predicative] Adjunct-Arguments in the Core [-argument] [+core] [-predicative] Not labeled [-argument] [-core] [-predicative]

Of these eight PP types, the last one is ruled out due to functional considerations: it is impossible for a semantic adjunct with no core privilege to be introduced by a non-predicative preposition.

206

PP Types in RRG: A Top-down Approach to Their Classification

3.1. Oblique Core Arguments [+argument] [+core] [-predicative] The above combination of the features covers a class of verbal arguments that occur in the core, but are marked by non-predicative prepositions. Van Valin and LaPolla (1997) and Van Valin (2005) state that the most straightforward example of this type is the recipient arguments of transfer verbs in languages such as English: (2) Tony gave the book to Peter. The dative-marked arguments of transfer verbs in Spanish exhibit a different behavior from the English counterparts; they can be (and usually are) duplicated by a kind of agreement marker, namely the clitic le: (3) a. Luisa le dio el libro a Rogelio. “Luisa gave the book to Rogelio.” b. Luisa les dijo la noticia a sus hermanos. “Luisa told the notice to his brothers.” As pointed out by Belloro (2007), given this and other criteria, these dative-marked arguments in Spanish can be seen as some kind of direct core argument. Leaving aside this issue, in Spanish the goal and source arguments of intransitive verbs of motion and the goal arguments of change of place verbs are also oblique core arguments: (4) a. Juan fue al cine. “John went to the movies.” b. Tere salió de la ciudad. “Tere went out of the city.” c. Adriana puso el vaso en la mesa. “Adriana put the glass on the table.” As shown in Ibáñez (2005, 2009), the reason is four-fold. First, they are semantically required by the predicates they occur with. Second, although they seem to be optional in some contexts, they have a strong tendency to be coded overtly (70% to 90%). Third, they can serve as the PSA in coordinate clauses and in non-finite subordinate clauses. Finally, they are

Sergio Ibáñez Cerda

207

mostly coded with what are clearly canonical (at least in terms of frequency) prepositions, e.g. de (as occurs with source-oriented verbs such as salir “get out” and partir “leave”), a (as occurs with goal-oriented verbs such as ir “go” and venir “come”), en (as occurs with change of place verbs such as poner “put” and colocar “place”). Furthermore, what has been termed “suplementos” in the Hispanic Linguistics literature (Alarcos 1968) may also be analyzed as oblique core arguments. Examples are given in (5a,b): (5) a. Juan carece de oportunidades. “John lacks opportunities.” b. Juan aspira a un buen empleo. “John aspires to a good job.” c. *Juan carece. “John lacks.” d. *Juan aspira. “John aspires.” Besides being obligatory, these PPs are semantically required by verbs and are introduced by non-predicative prepositions. However, the prepositions are not predictable from the logical structure of the whole clause; they are idiosyncratically selected by verbs.

3.2. Peripheral Adjuncts [-argument] [-core] [+predicative] The above combination of features indicates that these PPs are not semantically required by the predicates they occur with, that they occur outside the core (i.e. in the periphery), and that the prepositions that head these PPs license the presence of a participant in the clause. This is the reason they are labeled “peripheral adjuncts”. The most straightforward example of this type of PP is the temporal and locative adjuncts that function as settings of the states of affair denoted by the predicates. (6a,b) are examples from Van Valin (2005): (6) a. John baked a cake after work. b. John baked a cake in the kitchen.

208

PP Types in RRG: A Top-down Approach to Their Classification

(7a,b) are the Spanish counterparts to (6a,b). (7c,d) show that this kind of PPs cannot function as controller and cannot occur in the core: (7) a. Marta corría en el parque para ejercitarse. “Marta ran in the park to exercise.” b. Julia cenó en la sala. “Julia had dinner in the living room.” c. Marta i corría en el parquej y _ i /*j se veía bien. “Marta ran in the park and (he/it) looked nice.” d. Julia i cenó en la sala nuevaj y_ i /*j se ve bien. “Julia had dinner in the new living room and (she/it) looks nice.” Van Valin and LaPolla (1997) propose that the prepositions that introduce these PPs are two-place predicates, one of their arguments being the whole core and the other being the participant introduced by the preposition. A standard RRG representation for (6a) is the LS in (8): (8) be-after' (work, [[do' (John, Ø)] CAUSE [BECOME baked' (cake)]]) Beyond this type of PPs, usually termed ‘setting’ PPs, there are other kinds of locative PPs in Spanish that cannot take the whole clauses within their scope, even though they are not verbal arguments and are headed by predicative prepositions (i.e. they serve as adjuncts in the periphery). This kind of PP is exemplified in (9): (9) Juan cortó el pastel sobre la tabla de madera. “John cut the cake on the wooden board.” Here the PP sobre la tabla de madera “on the wooden board” refers to a place on which the change of state takes place. More specifically, the wooden board is the place where the event of dividing the cake into pieces occurs. It is true that Juan is doing something that involves the wooden board, but he is clearly not located on the board itself. This means that the semantic scope of the PP is not over the entire core, but over only part of it. Thus, we can consider PPs such as the one in (9) as partial modifiers of the core. A possible LS for (9) is shown in (10): (10) do' (Juan, Ø) CAUSE [be-over' (tabla, [BECOME cut' (pastel)])] The adverbial predicate be over' modifies the sub-event of change of state and doesn’t include the activity sub-event within its scope.

Sergio Ibáñez Cerda

209

3.3. Argument-Adjuncts in the Core [+argument] [+core] [+predicative] The above combination of features indicates that: 1) the participant is an argument licensed by the predicate semantics; 2) it is in the core; and 3) its preposition is predicative. These properties distinguish this type of PP from the oblique core argument. These PPs can be labeled argument-adjuncts in the core. In Spanish this type is exemplified by the goal PPs of intransitive motion and change of place verbs, but it is important to note that they serve as argumentadjuncts only when they are introduced by non-canonical prepositions. The reason is that (as argued in Section 3.1.) they serve as oblique core arguments when they are introduced by the canonical prepositions. Noncanonical prepositions add a certain extra semantic specification and this explains why the non-canonical prepositions are predicative. 4 These PPs can serve as the PSA in conjunction reduction constructions, as demonstrated by (11a,b): (11) a. En la mañana, Juan i fue para la casa nueva j y i / j se veía bien. “In the morning, John went to the new house and (he/it) looked nice.” b. Juan i puso un pezh dentro de la pecera nueva j y *i / h / j se ve bien. “John put a fish inside the new fish bowl and (it) looks good.” This observation allows us to claim they are in the core and that they serve as argument-adjuncts in the core.

3.4. Arguments in the Periphery [+argument] [-core] [-predicative] As expected from the iconic principle operative in the semantic-syntactic correspondence, semantic arguments of the verb are canonically in the 4 There is some lexical variation and idiosyncrasy with respect to which locative prepositions may be taken by motion and change of state verbs.

210

PP Types in RRG: A Top-down Approach to Their Classification

core and enjoy syntactic privileges that distinguish them from canonical adjuncts. Nevertheless, there are cases in which the semantic argument is deprived of its syntactic privilege and ends up in the periphery. A well established example of this PP type is the agent PP of a passive construction, in which the agent loses all of its syntactic privileges and fails to be realized as a core argument. In fact, there are languages in which only direct core arguments may belong to the core, while all oblique arguments seem to have no access to the pivot or controller status and to belong to the periphery. This PP type with no voice involvement occurs with verbs of saying in Spanish. Examples are given in (12a,b): (12) a. Pedro habló con María (en francés). “Pedro talked to María in French.” b. Mauricio le declaró su amor a Tere (con un gesto). “Mauricio declared his love to Tere with a gesture.” c. Pedro habla francés perfectamente. “Pedro speaks French perfectly.” d. El gesto del hombre lo declaró todo. “The man’s gesture declared it all.” These PPs are optional and peripheral, but they code participants that are required lexically, as can be seen from the fact that they can be coded as a core argument as in (12c,d). Other examples of this PP type are illustrated by (13a,b): (13) a. El presidente reunió a los gobernadores con los alcaldes (en el salón principal). “The president gathered the governors with the mayors (in the main hall).” b. El presidente reunió a los gobernadores y a los alcaldes (en el salón principal). “The president gathered the governors and the mayors (in the main hall).” c. El salón principal reúne a los gobernadores y a los alcaldes. “The main hall gathers the governors and the mayors.” Verbs of gathering (e.g. reunir “gather”, agrupar “group”) require four arguments, an effector, a locative, and two themes and allow the two themes to be coded either as two syntactically independent phrases (one as

Sergio Ibáñez Cerda

211

a direct object NP and the other as a PP as in (13a)) or as a single NP formed by two coordinated NPs as in (13b). Another important point to note about these verbs is that their locative arguments can be coded as a PP introduced by en or as the subject of the clause, as in (13c) and that when the locative arguments are realized as PPs, they cannot function as a pivot, as shown by (14): (14) El presidente i reunió a los gobernadores con los alcaldes en el salón principal j y i / * j se veía bien. “The president gathered the governors with the mayors in the main hall and (he/it) looked nice.” (14) shows that el salon principal “the main hall” cannot be shared by the two conjoined clauses and that it is not part of the core.

3.5. Argument-Adjuncts in the Periphery [+argument] [-core] [+predicative] The above combination of features indicates that this type of PPs codes an argument participant introduced by a non-canonical preposition and that they don’t belong to the core. (15d,f) involve examples of argument-adjuncts in the periphery, but it is instructive to compare their goal PPs with the goal PPs in (15a,b,c,e): (15) a. Israel i llegó al edificio j y i / j estaba sucio. “Israel arrived at the building and (he/it) was dirty.” b. Julio i llegó hasta el edificio j y i / j estaba sucio. “Julio arrived at the building and (he/it) was dirty.” c. Lola i salió a la calle j y i / j estaba sucia. “Lola went out to the street and (she/it) was dirty.” d. Marisa i salió para la cabaña j y i /* j estaba sucia. “Marisa went out for the cabin and (she/it) was dirty.” e. Ramón i partió al bosque j y i / j estaba sucio. “Ramón left to the wood and (he/it) was dirty.” f. Toño i partió para el bosque j y i /* j estaba sucio. “Toño left for the wood and (he/it) was dirty.”

212

PP Types in RRG: A Top-down Approach to Their Classification

The predicates in (15a,b) inherently require and focalize a goal argument, which serves as a pivot in the coordinate conjunction constructions. This means that both of the goal arguments belong to the core and that they are argument-adjuncts in the core. In contrast, the goal PPs in (18c,e) are argument participants despite that the predicates they occur with focalize a source (see Ibáñez 2005 for related discussion). 5 The goal PPs that occur with those verbs that focalize a source (rather than a goal) argument may serve as a pivot in coordinate conjunction constructions, as illustrated by (15c,e), when they are coded by a canonical non-predicative preposition, while they cannot serve as the pivot, as illustrated by (15d,f), when they are coded by a non-canonical predicative preposition (in this case, para “for”). The contrast between (15a,b,c,e) and (15d,f) suggests that the goal PPs in (15a,b,c,e) are core arguments, while the goal PPs in (15d,f) occur in the periphery and serve as adjuncts.

3.6. Adjuncts in the Core [-argument] [+core] [+predicative] The above combination suggests that this PP type is not required by a predicate, but exhibits some important syntactic properties that give them a core status. This is the case with some manner, temporal and locative PPs illustrated in (16a)-(16c): (16) a. La mujer viste con elegancia. “The woman dresses with elegance.” b. María actuó en el momento adecuado. “María acted at the right moment.” c. El puente fue construido en el lado este de la ciudad. “The bridge was built in the east side of the city.” d. *La mujer viste. “The woman dresses.” e. *María actuó. “María acted.” 5

Ibáñez (2005) argues that each predicate in a given language focalizes different aspects of its frame (state of affairs) and that in Spanish, llegar “arrive” inherently focalizes a goal, while salir “go out” focalizes a source.

Sergio Ibáñez Cerda

213

f. *El puente fue construido. “The bridge was built.” The ungrammaticality of (16d)-(16f) suggests that the PPs in (16a)(16c) are obligatorily required. Furthermore, as Mora-Bustos (2009) shows, these PPs involve some word order restrictions. They may appear post-verbally (as in (16a)-(16c)) or pre-verbally, but they may occur preverbally only when the subject occurs post-verbally, as in (17a); otherwise, the sentences would become ungrammatical, as demonstrated by (17b,c): (17) a. Con elegancia viste la mujer. “The woman dresses with elegance.” b. ??Con elegancia la mujer viste. “The woman dresses with elegance.” c. ?? La mujer con elegancia viste. “The woman dresses with elegance.” It is important to note that free adverbial adjuncts have no such word order restriction, as demonstrated by (18a)-(18c) (Mora 2009): 6 (18) a. Pedro se afeita cuidadosamente. “Peter shaves carefully.” b. Cuidadosamente Pedro se afeita. “Peter shaves carefully.” c. Pedro cuidadosamente se afeita. “Peter shaves carefully.” To go back to the obligatory status of the PPs in (16a)-(16c), it remains to be explained why they are obligatorily required even though they are not required by the predicates. There are two options for explaining their obligatoriness. We may assume that they have to appear because of some kind of condition about the use of syntactic templates or that these PPs are required by pragmatic conditions about the relative communicative relevance (Goldberg and Ackerman 2001). I will leave it to another occasion to investigate this question for future.

6

Other diagnostics that Mora-Bustos (2009) proposes for establishing the core status of these PPs are negation scope and extraction possibility. For reasons of space, I will leave the pertinent examples out of this work.

214

PP Types in RRG: A Top-down Approach to Their Classification

3.7. Adjunct-Arguments in the Core [-argument] [+core] [- predicative] This kind of PP can be illustrated in Spanish with the ones that appear in a type of dative construction. As has been pointed out before by Gutiérrez (1978) and Demonte (1994), among others, there are two types of dative constructions in Spanish. The first one involves a PP that codes a recipient required by the predicate, while the second one involves a PP that isn’t an inherently required argument of the predicate it occurs with. The first and second one are illustrated by (19a,b) and (19c,d), respectively: (19) a. Susana (le) dio un regalo a María. “Susana gave a gift to María.” b. Dulce (le) ofreció un vaso de vino a Julio. “Dulce offered a glass of wine to Julio.” c. Mario le pintó la casa a Rosa. “Mario painted Rosa’ house (for her).” d. Ramiro le trabajó dos años al Sr. Rodríguez. “Ramiro worked for Mr. Rodríguez during two years.” e. *Mario pintó la casa a Rosa. “Mario painted Rosa’ house (for her).” f. *Ramiro trabajó dos años al Sr. Rodríguez. “Ramiro worked for Mr. Rodríguez during two years.” As can be seen from the ungrammaticality of (19e,f), the presence of the PP in (19c,d) depends on the co-presence of the clitic le, while this is not the case with (19a,b). Nonetheless, both the dative PPs in (19a,b) and (19c,d) serve as a pivot in non-finite subordinate clauses: (20) a. Juani dio un libro a Pedroj para _ j leer. “John gave a book to Peter to read.” b. Juan i le hizo un traje a Pedroj para _ j usar en la fiesta. “John made Peter a suit to wear at the party.” The fact that the dative PPs in (19a)-(19d) may function as the pivots suggests that all of them belong to the core. It is important to note here that since the presence of the PPs in (19c,d) is licensed by that of the clitic le, (19c,d) may be analyzed as applicative constructions (Ibáñez 2003) and that since a in (19c,d) is non-predicative

Sergio Ibáñez Cerda

215

in nature, it is part of an argument manipulation constructional scheme (the applicative scheme). All this suggests that the clitic le licenses a PP as in (19c,d) that is a semantic adjunct to serve as a core argument and that the dative PPs in (19c,d) are adjunct-arguments in the core. 7

4. Conclusion I have built on Jolly’s (1993) three-way classification of PPs, i.e. oblique core arguments, argument-adjuncts, and adjuncts, and have tried to show with reference to the Spanish examples that combining the three binary features, [±argument] (semantic status of a clausal participant), [±core] (its syntactic status), and [±predicative] (the nature of the preposition), yields 8 logically possible PP types. Detailed typological work is needed to confirm that the PP types that arise from the combination of the proposed set of features are useful for description of PPs in other languages.

References Alarcos, E. 1968. Verbo transitivo, verbo intransitivo y estructura del predicado. Archivum XVI: 5-17. Belloro, V. 2007. Spanish Clitic Doubling: A Study of the SyntaxPragmatics Interface. Unpublished Ph.D. Dissertation, State University of New York at Buffalo. Bresnan, J. 2001. Lexical Functional Syntax. Oxford: Blackwell. Demonte, V. 1994. La ditransitividad en español; léxico y sintaxis. In V. Demonte, ed., Gramática del Español, 431-470. México: El Colegio de México.

7

One might object that the term “applicative constructions” applies only to constructions in which a semantic adjunct functions as a direct or primary object and might suggest that (19c,d) should be analyzed as benefactive constructions instead. One problem with this alternative proposal is that the Spanish clitic le can be used to realize semantic adjuncts as indirect objects in intransitive as well as transitive constructions. Given that benefactive constructions are usually formed from transitive verbs (Shibatani 1996), it may not be appropriate to use the label “benefactive constructions” to name constructions such as (19c,d) (see Flores and Melis 2004 and Belloro 2007 for related discussion about the discourse-pragmatic status of dative PPs as in (19c,d)). In contrast, applicative constructions may be formed from either intransitive or transitive verbs (Peterson 2007). All this suggests that it is better to name (19c,d) applicative constructions rather than benefactive constructions.

216

PP Types in RRG: A Top-down Approach to Their Classification

Dik, S. 1997. The Theory of Functional Grammar. Part 1: The Structure of the Clause. Berlin and New York: Mouton de Gruyter. Goldberg, A. and F. Ackerman. 2001. The pragmatics of obligatory adjuncts. Language 77.4: 798-815. Gutiérrez, S. 1999. Los dativos. In I. Bosque and Y. V. Demonte, eds., Gramática Descriptiva de la Lengua Española, 1855-1928. Madrid: Espasa-Calpe. Flores M. and Ch. Melis. 2004. La variación diatópica en el uso del objeto indirecto duplicado. Nueva Revista de Filología Hispánica, 52.2: 329354. Haiman, J. 1980. The iconicity of grammar: isomorphism and motivation. Language 56: 515-540. Ibáñez, S. 2003. Introduciendo participantes en la estructura argumental: el caso del clítico le del español. In V. Sánchez Corrales, ed., Actas del XIII Congreso Internacional de ALFAL. San José de Costa Rica: Universidad de Costa Rica. —. 2005. Los Verbos de Movimiento Intransitivos del Español: Una Aproximación Léxico-sintáctica. México: INAH-UNAM. —. 2008. Saying verbs: deepening the lexical semantics. In R. D. Van Valin, Jr., ed., Investigations of the Syntax-Semantics-Pragmatics Interface, 3-21. Amsterdam and Philadelphia: John Benjamins. —. 2009. Some functions of prepositional phrases in Spanish. In L. Guerrero, S. Ibáñez, and V. Belloro, eds., Studies in Role and Referente Grammar, 459-490. México: IIFL-UNAM. Jolly, J. A. 1993. Preposition assigment in English. In R. D. Van Valin, Jr., ed., Advances in Role and Reference Grammar, 275-310. Amsterdam and Philadelphia: John Benjamins. Mora-Bustos, A. 2009. Fixed adjuncts. Paper presented at the 2009 International Conference on RRG, August 2009, Berkely, CA. Peterson, D. A. 2007. Applicative Constructions. Oxford: Oxford University Press. Shibatani, M. 1996. Applicatives and benefactives: a cognitive account. In M. Shibatani and S. A. Thompson, eds., Grammatical Constructions: Their Form and Meaning, 157-194. Oxford: Oxford University Press. Tesnière, L. 1979. Elementos de sintaxis estructural. Madrid: Gredos. Van Valin, R. D., Jr. and R. J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Van Valin, R. D., Jr. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press.

CLAUSE LINKAGE AND PURPOSE CLAUSES IN SOUTHERN UTO-AZTECAN LANGUAGES LILIÁN GUERRERO IIFL-UNIVERSIDAD NACIONAL AUTÓNOMA DE MÉXICO

Abstract This paper explores the syntactic and semantic properties of purpose clauses following the Interclausal Relation Hierarchy as proposed by Role and Reference Grammar. Based on data from Southern Uto-Aztecan languages, I show that the way purpose relations are expressed make use of different syntactic linkage types even within the same language. The analysis shows that these linkage types entail different degrees of syntactic tightness, i.e., from nuclear cosubordination and core coordination, to ad-core subordination. A first attempt to establish the logical structure for these purposive linkage types is also proposed.

Keywords Purpose clause, clause linkage, RRG, Uto-Aztecan

1. Introduction1 This paper explores the syntactic and semantic properties of purpose clauses in Southern Uto-Aztecan languages, within the framework of Role and Reference Grammar [RRG] (Van Valin and LaPolla 1997; Van Valin 1

This work is a revised version of the paper presented in the 2009 Role and Reference Grammar International Conference. An extended analysis on purpose clauses in Yaqui can be found in Guerrero (in review); a comprehensive typological overview of control relations and purpose clauses within Role and Reference Grammar is present in Guerrero (in press, 2011). This study was possible thanks to funding from CONACyT (project No. 83529).

218

Clause Linkage in Southern Uto-Aztecan Languages

2005). Compared with other adverbial subordinate clauses, purpose clauses have received less attention, and most of the studies have been based on English (Thompson 1985; Jones 1991) except for the recent typological studies presented by Verstraete (2008) and Schmidtke-Bode (2009). In regards to the study of purpose clauses, one is confronted with two apparently conflicting assumptions. Syntactically, purpose clauses are considered as a type of adverbial subordination, i.e., an argument-adjunct (Jones 1991) or a peripheral unit (Van Valin and LaPolla 1997; Ernst 2001: 355). Semantically, purpose clauses share certain meanings with other adverbial clauses such as consequence (reason and cause clauses), and future-oriented (sequential clauses), but they also share the notion of intention with certain complement clauses (modal and causative clauses) (Givón 2001; Cristofaro 2003). In turn, purpose can be grouped with other semantic relations within the domain of adverbial subordination, such as cause and reason, or can be aligned with certain complement relations as in desideratives. In the former, the syntactic and semantic properties of the dependent unit are minimally restricted by the main clause, as long as the event is unrealized. In the latter, some properties can be determined by the whole clause similar to the way in which complement-taking predicates specify the morpho-syntactic properties of a dependent unit. Syntactically, a purpose clause can have a range of manifestations. In English, for instance, a purpose clause can be formally introduced by a toinfinitive clause, as in John went out to run, by the complex preposition in order to, as in “John went to the CEO office in order to apply for the new position”, or by the special conjunction so that as in “John left early so that he won’t be late for the job interview”. In Guarijio, a Southern UtoAztecan language, at least three clause linkage types can be identified as the most common and unambiguous clauses coding a purpose relation: motion-cum-purpose (1a), intention (1b), and finality (1c). In the examples, co-referential arguments are co-indexed; the “_” is for illustrative purposes only and indicates a missing syntactic argument which is co-referential with an argument of the main unit (i.e., equideletion). (1) Guarijio a. Motion-cum-purpose linkage type wa’á maní, _ i eci-po=sa e’ego there be, sow-MOV.PURP.PL=EMPH then “There is the corn, go to sow it then.” (Miller 1996: 110)

Lilián Guerrero

219

b. Intentional linkage type Neei ehturiáwa-ni [ _i merikó ini-nári=a] doctor be-DESID=EMPH 1S.SG study-PRS “I am studying in order to become a doctor.” (Félix 2005: 322) c. Finality linkage type Tiburcioi hená [ka’í amój __i tewi-mí ruhka] Tiburcio come.PFV NEG 2SG.NS see-CLM like.this “Tiburcio came so that you couldn’t see him.” (Miller 1993: 136) The aim of this paper is twofold: to provide an overview of the clause linkage patterns expressing a purpose relation in Southern Uto-Aztecan languages, and to show that purpose clauses trigger a high degree of syntactic and semantic integration in terms of argument coding, operator restrictions, and the occurrence of clause linkage markers. Moreover, purpose clauses commonly entail a referential dependency among an argument of the main unit and an argument of the linked unit. As for the Guarijio examples above, the first two types are limited to actor control relations, that is, the actor of the main unit and the actor of the dependent unit must be co-referential, whereas the finality type allows actor and undergoer control relations. The next two sections characterize motioncum-purpose, intentional and finality clause linkages in Uto-Aztecan languages. The degree of syntactic integration is analyzed in §2, while the juncture-nexus relations for each of these linkages are presented in §3. For the semantic dimension (§4), a new sub-hierarchy is proposed in order to distinguish the different patterns of control relations found in purpose clauses. Finally, §5 concludes this paper. Two mayor branches are recognized for Uto-Aztecan family, the Northern branch (spoken mainly in the United States) and the Southern branch (spoken in Mexico and El Salvador). This study focuses on languages from the Southern branch. Five major groups are generally recognized: i) Tepiman (Southern and Northern Tepehuan, Mountain and Lower Pima, O’odham; ii) Taracahitan (Tarahumara varieties, Guarijio, Yaqui, Mayo), iii) Tubar (extinct), iv) Corachol (Cora, Huichol), and v) Nahuatl varieties (including Classic Nahuatl and Pipil). As a family, Uto-Aztecan languages show synthetic/agglutinant, headfinal properties, and follow a nominative-accusative case system. Except for Yaqui, the other languages have lost the case markers on the nouns, but most of them keep track of the nominative vs. non-nominative distinction on the pronominal system. In addition, Nahuatl variants, Cora and Huichol behave as head-marking languages since pronominal forms are obligatorily attached to the verb. The data used in this paper come from

220

Clause Linkage in Southern Uto-Aztecan Languages

published reference grammars, grammatical sketches, articles, papers, as well as other linguistic materials. Most of the Yaqui data come from my own fieldwork on this language.

2. The Syntax of Purpose Clauses A purpose construction is a clause that encodes a particular conceptual relation between two events, such that one of the events (the main event) is performed with the goal of obtaining the realization of another one (the purposive event) (Cristofaro 2003: 157). In Southern Uto-Aztecan languages three major syntactic linkages have been identified as the most common and unambiguous linkage patterns expressing purpose: motioncum-purpose, intentional and finality. Each of these linkage types has features that distinguish it from the others, yet some properties are consistently present across all linkage types, i.e., the main participant’s volition, a future-oriented event, and a strong preference for a shared participant.

2.1. Motion-cum-Purpose Linkage Very commonly, purpose clauses concern the purpose of motion, for example, the main participant might go somewhere in order to obtain the realization of the intended event. This means that the main actor and the dependent actor must be identical. In these cases, then the unmarked strategy links together the motion event and the desired event, with the two units usually sharing the TAM operators, and the two are placed together without a CLM. In Pima (2a) the motion and the intended events are contiguous independent verbs, while in Yaqui (2b) and Southern Tepehuan (2c), the motion event immediately follows the desired event deriving some sort of co-lexicalization. (2) a. Huma hihim [_i va’igiti igai] together RED.go bring DEM “Together they went to bring them.” (Pima; Estrada 1998: 34) b. Joan-Ø aabo koko-se-k Joan-NOM here sleep-MOV.PURP.PL-PFV “John came to sleep here.” (Yaqui)

Lilián Guerrero

221

c. Maic-ach tu-’i’-po-’ EXH-1NOM.PL EXT-drink-MOV.PURP.PL-FUT gu atuhl mu-cucsiñ DET gruel there-kitchens “Let’s go drink gruel at the ceremonial kitchens.” (Southern Tepehuan; Willet 1991: 171) There have been claims that purpose clauses do not logically entail the participants of the dependent unit, and that the performer of the main action does not necessarily controls the realization of the dependent one (Cristofaro 2003: 157). Yet the most striking characteristic of purpose clauses is the occurrence of unexpressed participants in the linked unit. When the main and dependent actors are the same, there may be a missing syntactic argument in the dependent unit which must be identical to the matrix actor; this is the case for all the examples in (2). However, the dependent actor can be overtly expressed as a pronominal form attached to the verb in those languages showing head-marking properties. For instance, Pajapan Nahuatl has two alternative structures involving a motion event in the main unit. In (3a), the two events are expressed in a complex predicate, and so the actor is expressed only once; in (3b), the two verbs are independent of each other and each must express its core arguments pronominally. (3) a. Ti-cho:-ga-ti-wis 2S.SG-cry-LINKER-come “You came to cry.” (Pajapan Nahuatl; Peralta 2007) b. Nii-yawi [nii-k-tegi-ti tro:ha chi:hli] 1S.SG-go 1S.SG-3PO-cut-AND a.lot chile “I am going to cut a lot of chile.” (Pajapan Nahuatl; Peralta 2007) Because of its very semantic nature, the TAM information in purpose clauses is usually unmarked or is restricted to future, potential, irrealis, or subjunctive forms as long as it is posterior to the time expressed in the main event. Thus, in a motion-cum-purpose linkage, the two events share all TAM operators, and they are usually marked on the main activity; the only exceptions are co-lexicalized units, as in (2b) above, where the TAM operators are attached to the last verb. In fact, the situation regarding TAM operators within the family is to some extent complicated. On the one hand, there are no “pure” tense markers, but rather tense aspect markers. On the other, the tense and aspect suffixes expressing future, irrealis,

222

Clause Linkage in Southern Uto-Aztecan Languages

potential as well as purpose seem to be historically derived from two sources: either from the suppletive motion stems *-sV/-pV “go (sg/pl)” or the “die” stem *-mV/-kV (sg/pl). Thus, the morphological glosses of these suffixes in the literature go from pure future tense marker, to inceptive, potential, to motion and purpose (4). The Nahuatl “andative” suffix -ti/-to (sg/pl) in (3) can be also related to motion, which explains why in the Sierra variant the suffix -ti is glossed as “timeless directional” (4c).2 (4) a. Ne mi sú-n-ma sipúcha. 1SG.NOM 2SG.ACC sew-APPL-FUT.SG skirt “I am going to sew a skirt for you.” (Tarahumara; Caballero 2008: 245) b. Ne-pi-heu-wakaxi-nanai-mie 1SG-ASI-TRAV-cow-buy-INC.SG “I will go to buy more cows.” (Huichol; Iturrioz and Gómez 2006: 205) c. Ki-mach-ti-to 3SG.O-learn-CAUSE-DIR.PAST “He went to teach him.” (Sierra Nahuatl; Robinson 1970: 57) The example in (5a) shows that a deontic modal operator as -maachi “should” in Yaqui has scope over the two units in a motion-cum-purpose clause, i.e. what Lupe should do is both, go somewhere and buy the meat. Languages might also show a different purpose structure in a situation where the main action takes place in order to prevent another event from occurring. Data on negative purpose is significantly sparse, but some constraints are clear: motion-cum-purpose limits the scope of negation to the main action only (5b). (5) a. Lupe-Øi wakas-ta _i jinu-se-maachi Lupe-NOM meat-ACC buy-MOV.PURP.SG-SHOULD “Lupe should go to buy the meat.” b. Joan-Øi kaa aabo _i kochi-se-k John-NOM NEG here sleep-MOV.PURP.SG-PFV “John didn’t come here to sleep/ *came here to not sleep.” (Yaqui)

2

Whether these morphemes synchronically encode purpose, genuine motion or purely tense meanings is a question that awaits further studies. See Haspelmath’s (1989) proposal on the evolution of infinitive-like forms from purposive actions.

Lilián Guerrero

223

2.2. The Intentional Linkage The second strategy, i.e., the intentional type, is something between the motion-cum-purpose and the finality linkage types. On the one hand, the main activity may be a motion predicate, as in the former strategy, though it may involve another activity instead. Unlike in the motion-cum-purpose strategy, however, there is a special verbal marker on the dependent unit expressing some sort of volition and purpose. There are two sub-types of this construction, characterized primarily by the type of the verbal marker used. The first sub-type takes a desiderative verbal suffix attached to the intended event, as shown in Yaqui (6a) and Guarijio (6c) below. Notice that in (6b) and (6d), the verbal suffixes -bae and -nári function as modal complement-taking predicates, i.e. co-lexicalization. (6) a. U

o’ou-∅i bwite-k [maso-ta _i me’e-bae-kai] man-NOM run.SG-PFV deer-ACC kill.SG-DESID-CLM “The man ran in order to kill the deer.” (Yaqui) b. U maso-∅ bwite-bae-k DET man-NOM run.SG-DESID-PFV “The deer wanted to run.” (Yaqui) c. Neéi ehturiáwa-ni [_i merikó ini-nári=a] doctor be-DESID=EMPH 1S.SG study-PRS “I am studying to become a doctor.” (Guarijio; Félix 2005: 322) d. I’wa=ni tehíba-nare here=1SG.NOM stay-DESID “I want to stay here.” (Guarijio; Miller 1993: 96) DET

So far, this linkage type has been identified in Taracahita languages inside the Uto-Aztecan family. Outside the family, it is also attested in Lealao Chinantec (Otomangue), and Fijian (Austronesia) (See Guerrero 2011). The second sub-type may additionally take the purpose and motion verbal suffix inside the linked unit, like -se/-bo in Yaqui (7a), -mi ~ -mia in Guarijio (7b), and possibly, -mea in Tarahumara (7c). That is, in this linkage type, the linked unit itself may function as a desiderative complex unit in the examples above, or as a motion-cum-purpose clause below, but here, it follows a unit expressing the main action. Notice again that the dependent clause in Yaqui is marked by -kai, whereas in Guarijio it is unmarked.

224

Clause Linkage in Southern Uto-Aztecan Languages

(7) a. Bwite-k _i [au _i esso-se-ka] run-PFV 3SG.REFL hide-MOV.PURP.SG-CLM San Hose de Waimam-mewi San José de Guaymas-DIR “He ran to hide himself in San José de Guaymas.” (Yaqui; Johnson 1962, kahe’eeme: 34) b. ma’chihéna=boga i wa’a te’pé-chi [_i no’o tepóre-mi] go out.SG = DEM there house-LOC 1SG.ACC greet-PURP “He went out of the house to say hello to me.” (Guarijio; Miller 1996: 105) c. Pé ’yáti _i simí-le [_i yo-méa] just quick go-PAST cure-FUT.SG “He went quickly in order be cured.” (Tarahumara; Burgess 1984: 142) As with the motion-cum-purpose clause type, the intentional clause type demands the main and dependent actors to be identical. A further restriction of the intentional clause type is that the actor must have volition, i.e. human being. Notice that in (6) and (7) above, there is a missing syntactic argument inside the linked event, which is co-referential to the main actor. With respect to the coding of TAM information, in Yaqui and Guarijio the dependent event must be completely unmarked in both sub-types, as in (8a-c). In the example (8d), the modal operator -maachi “should” has scope over the two units, i.e., Lupe should buy the meat and should cook the wakabaki. (8) a. Wanitai werumá puusi-ta-re [ _i kawé nene-narí=a] Juana big eye-make-PFV well see-DESID-EMPH “Juana opened her eyes wanting to see better.” (Guariijio; Félix 2005: 321) b. Lili-Øi aabo yepsa-Ø Lili-NOM there arrive.SG-PRS [_i jiak-nok-ta ne-u taa’a-bae-kai] Yaqui-word-ACC 1SG-DIR know-DESID-CLM “Lili comes there wanting to learn Yaqui from me.” (Yaqui) c. asi-rú = mu [no’ó wewé-mia] arrive-1.PFV =2SG.S 1SG.NS hit-PURP “You came to hit me.” (Guarijio; Félix 2005: 148) d. Lupe-Ø wakas-ta jinu-maachi Lupe-NOM meat-ACC buy-SHOULD

Lilián Guerrero

225

[_i wakabak-ta ya’a-bae-kai] wakabaki-ACC make-DESID-CLM “Lupe should buy meat to cook the wakabaki.” (Yaqui) With respect to negation, there is a pair of examples in Yaqui where the dependent unit (9b) can be negated independently of the main unit. There are not examples of this type in Guarijio. yecha’a-bae-kai] tobokta-k [a _i kom rise-PFV 3SG.ACC down put.SG-DESID-CLM [kaa _i into a beje’e-bae-kai] NEG and 3SG.ACC pay-DESID-CLM “They rose up to dominate it and did not pay it.” (Yaqui; Johnson 1962, hiakim ’etehoi: 30)

(9) a. ’emo

REFL.PL

In Guarijio, this linkage type is unmarked in terms of CLM, whereas in Yaqui, the dependent unit is marked by the adverbial suffix -kai. The distribution and functions of -kai are hard to typify, but it is limited to same-subject clauses, it demands equi-deletion, and it disallows any TAM information in the verb, which results in a participial- or gerund-like dependent clause. Other Yaqui adverbial clauses marked by -kai include manner clauses and simultaneous temporal clauses.

2.3. The Finality Linkage While the first two clause types are restricted to identical actors, the finality clause linkage allows different agents. Additionally, in the finality clause type, the linked unit is introduced by a special CLM. For instance, in Yaqui the finality clause linkage marked by -betchi’ibo “for, in order to” may be used for identical actors, as in (10a), as well as different actors, as in (10b). When the actors are different, the dependent actor must be marked as accusative. In Guarijio, this structure type make use of extra morphology like the adverbs ruhka and olaga “like this” at the end of the linked unit marked by -mi ~ -mia. Notice again that in (10c) the dependent subject is non-nominative. (10) a. U

o’ou-Ø i juya-u siika man-NOM mountain-DIR go.SG.PFV [_i mas-ta me’e-betchi’ibo] deer-ACC kill.SG-CLM “The man went to the mountain to kill the deer.” (Yaqui) DET

226

Clause Linkage in Southern Uto-Aztecan Languages

b. U

maso-Øi bwite-k [u-ka o’ou-taj ka deer-NOM run.SG-PFV DET-ACC man-ACC NEG me’e-ne-betchi’ibo] ai 3SG.ACC kill.SG-POT-CLM “The deer ran quickly in order for the man not to kill it.” (Yaqui) c. Tiburcioi hená [ka’í amój __i tewi-mí ruhka] Tiburcio come.PFV NEG 2SG.NS see-PURP like.this “Tiburcio came so that you couldn’t see him.” (Guarijio; Miller 1993: 136) d. Oí-re [ihí-bo olága] invite-PAST drink-PURP.PL like.this “(Hei) invited them ___i+v to drink it.” (Guarijío; Miller 1993: 206) DET

When the actors are different, other instances of semantic correlation among different participants may take place. In contrast to the main actor controlling the identity of the dependent actor in (10a), the main actor controls the dependent undergoer in “the deer ran in order for the man not to kill it” in (10b), and in “Tiburcio came so that you couldn’t see him” in (10c). In (10d), the controller of the missing argument includes the main actor and someone else. A more complex situation is found when the main clause is transitive. In (11a) the main undergoer controls the identity of the missing actor in the linked unit; though the theme is also shared, there is a co-referential pronoun am “them” in the dependent unit. In (11b) the main undergoer controls the identity of an optional instrument phrase. (11) a. Min-Øi u-ka yoi-taj kaba’i-mk reuwa-bae Fermín-NOM DET-ACC foreigner-ACC horse-PL lend-DESID [_ j amk wiria-ne-betchi’ibo] 3PL.ACC feed-POT-CLM “Fermin wants to lend the foreigner the horses in order for him to feed them.” (Yaqui) b. Min-Øi u-ka tractor-taj jinu-k Fermin-NOM DET-ACC tractor-ACC buy-PFV [enchik a-ej bwia-ta tekipanoa-ne-betchi’ibo] 2SG.ACC 3SG.ACC-with land-ACC work-POT-CLM “Fermín bought the tractor in order for you to work the land with it.” (Yaqui) Traditionally, purpose is analyzed as belonging with other adverbial relations like reason and cause constructions (Dixon and Aikhenvald

Lilián Guerrero

227

2009). A feature that distinguishes purpose from other adverbial clauses is temporal anchoring: the linked state of affairs must be unrealized at the time of the former action for purpose, while for reason/causal relations it may be already realized at the time of the main event as long as it is posterior (Kortmann 1997: 86; Hengeveld 1998: 357; Ernst 2001: 327; Longacre 2007: 381). This contrast of finite vs. non-finite units can be seen in Huichol; the purpose clause in (12a) and the reason clause (12b) share the same linkage marker -ki, but only in the latter is the linked unit fully independent with respect to operators. (12) a. kareta nei-p-e-nanai cart 1SG.S-ASI-INV-buy.PFV [kiye-xi nei-’ikata-mi-ki] wood-PL 1SG.S-carry-IMPL-CLM “I bought the cart to carry out the wood.” (Huichol; Gómez 1988: 172) b. tei-pi-’u-ti-tsuana [tei-mi-heu-ta-hiwerie-tsie-ki] 1PL.S-ASI-EXP-MULT-cry 1PL.S-ASI-EXT-UNIT-concern-TNS-CLM “We are crying because we were concerned.” (Huichol; Iturrioz and Gómez 2006: 204) This does not mean, however, that the dependent verb must be unmarked, but rather that the TAM information is limited to future, potential or other irrealis-like markers. In Yaqui, the dependent verb inside a -betchi’ibo clause may be unmarked or may be marked by the ‘potential’ suffix -ne; contrast (10a) and (10b) above. Moreover, the modal operator has scope within the main action only, hence in (13a) “Lupe” should buy the meat, but she is not obligated to cook it. This follows from the fact that a deontic modal operator can modify the relation between one actor and a sequence of events performed by the same participant, whereas it cannot modify relationships between the actor and distinct states of affairs which may be realized, at least potentially, by a different participant. In addition, only this linkage type allows the occurrence of temporal adverbs independently modifying the dependent unit. (13) a. Lupe-Ø wakas-ta jinu-maachi Lupe-NOM meat-ACC buy-SHOULD ya’a-ne-betchi’ibo] [_i/k wakabak-ta wakabaki-ACC make-POT-CLM “Lupe should buy the meat in order to cook the wakabaki.”

228

Clause Linkage in Southern Uto-Aztecan Languages

b. Lupe-Øi wakas-taj jinu-k Lupe-NOM meat-ACC buy-PFV yooko ya’a-ne-betchi’ibo] [_i/k wakabak-ta wakabaki-ACC tomorrow make-POT-CLM “Lupe bought meat in order to cook the wakabaki tomorrow.” In a finality clause type, either the main unit (14a) or the linked unit (14b, c) can be negated. (14) a. Goyo-Øi kaa aabo siika Goyo-NOM NEG here go.SG.PFV [enchi _i bit-ne-betchi’ibo] 2SG.ACC see-POT-CLM “Goyo did not come here to see you (e.g. but to see me).” (Yaqui) b. Goyo-Øi yeu siika Goyo-NOM out go.SG.PFV [enchi _i kaa bit-ne-betchi’ibo] 2SG.ACC NEG see-POT-CLM “Goyo left so that he didn’t see you.” (Yaqui) c. Tiburcio hená [ka’í amó tewi-mí ruhka] Tiburcio come.PFV NEG 2SG.NS see-PURP like.this “Tiburcio came so that you couldn’t see him.” (Guarijio; Miller 1993: 136) Since purposes are essentially goal-oriented, they are usually introduced by CLMs commonly specific to benefactive and dative arguments, as well as recipients, allatives, and goals. In Yaqui and Pima, for instance, this clause type makes use of the postposition encoding beneficiaries; contrast the uses of vuika “for” in Pima marking a benefactive (15a) and purposive (15b) unit. Another possible marker seems to be the suffix -ki which marks instrumental (15c), cause and finality (15d) in Huichol; some cognates are found in Guarijío -ke signaling beneficiary, dative, possession (Félix 2005), and Comache -ki marking cause and beneficiary (Charney 1989). in taan a-daad-vuika (15) a. Higai timiti-m tortilla-PL 1SG.NS ask 3SG.NS-mother-BEN 3SG “She asked me for tortillas for her mother.” (Pima; Estrada 1988: 80)

Lilián Guerrero

229

b.

Aani sudag nukad [_i i’i-ag-vuika] 1SG water have.IMPFV drink-FUT-PURP “I don’t have water to drink.” (Pima; Estrada 1988: 59) c. Miikii yu-kiye-ki mei-pe-i-kuuwaazi 3PL REFL-stick-INSTR 3PL.S-AS-3SG.O-beat “They beat him with their stick.” (Huichol; Comrie 1982: 103) d. Kareta nei-p-e-nanai [kiye-xi nei-’ikata-mi-ki] cart 1SG.S-ASI-INV-buy.PFV wood-PL 1SG.S-carry-IMPL-INST “I bought the cart to carry out the wood.” (Huichol; Gómez 1988: 172)

The Spanish loanword para “for” is found in same-actor purpose clauses in some Nahuatl dialects (16a-b), regardless of the identity of the actors, while in the complex preposition para que ~ paque “in order to, so that” is used in Pima for different-actor purpose clauses (16c). nii-ya [pa ompá-i nii-pashsalo-(a) (16) a. Neli 1SG.NOM 1SG.S-go CLM there-around 1SG.S-walk-(PRS.S.SG) “I am going there in order to walk around.” (Mexicanero Nahuatl; Canger 2001: 54) b. Shi-kij-wahwana [para in ∅j-kisa-s] IMP-3SG.O-scratch CLM DET 3SG.S-get.out-FUT “(You) scratch it in order to get the spine out.” (Mexicanero Nahuatl; Canger 2001: 137) c. Huaan him [pake ap kova ap niidiam] Juan go.SG.PFV CLM 2SG NEG.EMPH 2SG see.PROB.CONT “John left so you don’t see him.” (Pima; Estrada 1988: 123) Finally, in Tarahumara, Tepehuan, and O’odham, finality clauses are introduced by a general subordinator which also introduced complement and other adverbial clauses. These general clause linkage markers appear in the middle of the two clauses; see the marker -na in Southern Tepehuan (17a) and by ma- in O’odham (17b). (17) a. Gu chi-chioñi bopa-mit [na-mit _i tusa-m gu tai] DET RED-man run-PFV CLM-PFV extinguish-O DET fire “The men run to extinguish the fire.” (Southern Tepehuan; García 2008) b. Am a-t hii-x g huani LOC MD-TNS go-PFV DET Juan

230

Clause Linkage in Southern Uto-Aztecan Languages

[ma-t _i wo ñii-x a husi] CLM-TNS FUT see-PFV DET José “Juan went to see José.” (O’odham; Saxton 1982: 261) Diessel (2001) claims that in languages in which adverbial clauses have a final subordinator, the dependent unit tends to precede the main clause, whereas in languages with an initial subordinator, the adverbial unit may occur in either sentence-initial or sentence-final position. For purpose, in particular, it has been observed that the purposive unit usually follows the main action, although it may also appear pre-posed with a different, more general meaning (Thompson 1985: 61). In the family, the purposive unit prefers to follow the main action regardless whether the CLM appears at the beginning, as in (16) and (17), or at the end of the linked unit, as in the rest of the examples. In sum, from a syntactic standpoint, a purpose clause can be expressed by means of different clause linkage types. The syntactic and semantic dimensions of these purpose clause types can be neatly captured by the theory of clause linkage as proposed by Role and Reference Grammar; the syntactic dimension is presented in (§3) and the semantic relations in (§4).

3. The Juncture-Nexus Relationships of Purpose Clauses In their seminal analysis, Foley and Van Valin (1984: 264) elaborated the concept of syntactic boundedness. The central idea was that the functionalsemantic notion of dependence is not equivalent to the purely syntactic notion of embeddedness and structural dependency. Further works within RRG offer an original model for the study of clause linkage. Three theoretical aspects of clause linkage are considered: (i) the theory of juncture, (ii) the theory of nexus, and (iii) the interclausal semantic relation among the units. The theory of juncture deals with the units which make up complex sentences: nucleus, core, clause and sentence. In a nuclear juncture, there is a single core containing two nuclei taking a single set of core arguments. In a core juncture, there is a single clause containing more than one core, each with its own set of arguments. In a clausal juncture, whole clauses are joined and each clause may be fully independent of the others. The sentence is the major syntactic unit. The theory of nexus concerns the syntactic relationship between the units in the juncture, subordination, coordination, and co-subordination, each type distinguished on the basis of structural or operator dependencies. In subordination, the linked unit functions either as an argument (as in complementation), or as a modifier

Lilián Guerrero

231

(as in adverbials). In coordination, the two units are added together in a sequence in a relationship of equivalence and independence at the level of the juncture. Cosubordination shows properties of both: there are two equivalent units joined together (as in coordination), but one unit depends on the other (as in subordination), in terms of operators. What distinguishes cosubordination from coordination is operator sharing: in the former, the linked verb must be dependent upon the matrix verb for expression of one or more operators at the level of juncture, whereas in the latter, the two verbs can, but do not need to be independently specified for the relevant operators. Thus, the term clause linkage denotes linking of units at any layer (nucleus, core, clause, and sentence), through any of these nexus relations, resulting up to eleven combinations organized into the Syntactic Relations Hierarchy ranked in terms of their syntactic tightness, as seen in the first column in Figure 1. The linkage types at the right edge are combinations of whole clauses constituting sentences (e.g. sentential or clausal coordination); moving to the left edge, the linked units lose more and more features of an independent clause until they are reduced to a bare nucleus (e.g. nuclear co-subordination). These syntactic combinations express a wide variety of semantic relations between the units in the juncture, which are organized in terms of their degree of semantic cohesion, presented the second column in Figure 1, next page. This Interclausal Relation Hierarchy reflects the basic iconic principle governing the interaction of the two hierarchies: the closer the semantic relation between the two propositions is, the stronger the syntactic link joining them. That is, the semantic relations at the top end should be realized by the strongest syntactic combination (e.g. nuclear, core junctures & cosubordination, subordination nexus), whereas perception, cognition and discourse predicates which are at the bottom of the scale, should be realized by the loosest syntactic linkage (i.e. core, clausal junctures & subordinate, coordinate nexus). 3 It is important to keep in mind that these juncture-nexus types are abstract linkage relations, not grammatical constructions, and languages rarely make use of all of them. This means that each of type may be realized by more than one grammatical construction in a particular language, and vice versa, the same grammatical construction may involve different linkage types.

3

As a manifestation of the stronger the semantic relation, the tighter the morphosyntactic bond between the units, it is very common for the function at the top end of the hierarchy to grammaticalize into complex predicates (i.e., nuclear cosubordination) and, lastly, into grammatical morphemes.

232

Clause Linkage in Southern Uto-Aztecan Languages

Subordination involves some further distinctions, since subordinate adjuncts may function either as arguments or adjunct modifiers (Van Valin 2005: 194). Generally, subordinate units within complementation trigger a “daughter” relationship with the main core, while adverbial subordination may occur in the periphery of the core it modifies, i.e. “peripheral” subordination introduced by predicative adpositions like English where, before or after. For instance, in “I saw Bartola after she entered the house”, the relationship of after she entered the house is the same as that of the PP “I saw Bartola inside the house”. Following Bickel’s proposal (1993, 2003), this peripheral modification is referred to as “ad-core subordination” in RRG, because the subordinate clause is a modifier of the matrix core and occurs in the peripheryCORE. In contrast, adverbial clauses introduced by English because, if, although are not in the peripheryCORE, but in the periphery of the clause unit, i.e., ad-clausal subordination (Van Valin 2005: 194-195). Unlike ad-core subordinate clauses, they do not express the spatial or temporal setting of the event expressed by the core; they express the reason or condition for the event expressed by the clause as a whole, e.g. “Bartola entered the house because it is raining”. It is also predicted that the ad-core subordinate linkage should be closer to the matrix core than the ad-clausal subordinate linkage, since the former is construed as modifying the core, rather than the whole clause. Although a more detailed study is needed in order to determine the juncture-nexus relations for each of the Uto-Aztecan languages completely, some generalizations can be made at this point. Without a doubt, the motion-cum-purpose linkage type is the most complex structure to characterize because it may vary from language to language. (i) The main actor and the dependent actor must be identical. Depending on the structural properties of the language in question, the controllee can be covert (i.e., Taracahita, Tepiman) or overt (Nahuatl, Corachol). That is, there is a dependency among the two units in terms of argument coding. (ii) The linked event is future-oriented but it is unmarked; this means that the two linked units must share the relevant TAM operators, including negation, which yields operator dependence as well. (iii) Most of the time, the linked unit lacks a CLM and (iv) the purposive unit follows the main action except in co-lexicalized structures. All these properties together yield nuclear cosubordination or core cosubordination, depending on the valence of the dependent verb. In the former juncture, there is a single set of core arguments (2b-c, 3a), while in the latter there is a single shared argument (i.e. the actor), as in (2a, 3c). The Yaqui motion-cum-purpose linkage in (2b) triggers a nuclear

Lilián Guerrero

233

cosubordination, as represented in Figure 2. In the Figures, both the constituent projection and the operator projection are shown. Strongest Nuclear cosubordination Nuclear subordination Nuclear coordination Core cosubordination Core subordination daughter peripheral Core coordination Clause cosubordination Clause subordination daughter peripheral Clause coordination Sentential subordination Sentential coordination Weakest Syntactic relations

Closest Causatives[1] Phase Manner Motion Position Means Psych-action Purposive Jussive Causative [2] Direct perception Indirect perception Propositional attitude Cognition Indirect discourse Direct discourse Circumstances Reason Conditionals concessive Simultaneous actions Sequential actions Situation-Situation:unspecified Loosest Semantic relations

Figure 1. Interclausal Relations Hierarchy (Van Valin 2005: 209)

234

Clause Linkage in Southern Uto-Aztecan Languages

SENTENCE CLAUSE PERIPHERY

CORE NP

NUCLEUS NUC1

NUC2

PRED

PRED

ADV V

V

Joan aabo kochi

-se

V

V

PRED

PRED

NUC 1

NUC2

NUCLEUS

-k

ASP

CORE CLAUSE

TNS

SENTENCE Figure 2. Nuclear cosubordination in Yaqui (2b)

In clauses like those in (4), there is a periphery modifying the nucleus of the clause, i.e. peripheryNUCLEUS, since the purposive marker may be also interpreted as a future or potential aspect marker, e.g. ad-nuclear junctures. If this is so, the representation for the Huichol ad-nuclear subordination clause in (4b) above can be the following:

Lilián Guerrero

235

SENTENCE CLAUSE CORE PRO

NP

NUCLEUS PRED V

PERIPHERY NUC PRED V

Ne-pi-heu-wakaxi

-nanai

-mie

V PRED NUCLEUS

ASP

CORE CLAUSE SENTENCE Figure 3. Ad-nuclear subordination in Huichol (4d)

As for the two sub-types of the intentional linkage, they take a complex predicate inside the linked unit, i.e., either the desiderative verbal suffix, or the motion and purpose suffix. In both cases, (i) the main actor and the dependent actor must be identical and there must be a syntactic missing argument in the linked unit. (ii) The linked event depends on the main unit in terms of the TAM operators, but it may be independently negated. (iii) The linked unit is marked by a same-subject CLM in Yaqui but it is unmarked in Guarijio, and (iv) the purposive unit tends to follow the main action. The fact that the dependent actor cannot be overtly expressed and that the TAM must be shared by the two cores, but not negation, suggests core coordination. In the representation of Yaqui

236

Clause Linkage in Southern Uto-Aztecan Languages

intentional purpose clause in (4a), the linked unit itself takes a nuclear cosubordination linkage type. SENTENCE CLAUSE CORE CORE2

CORE1 NP

NUCLEUS NP

NUCLEUS NUC1 NUC2

PRED V U o’ou

CLM

PRED

PRED

V

V

me’e

-bae

bwite-k maso-ta V

-kai

V

PRED

PRED

NUCLEUS

NUCLEUS

CORE1

CORE2 CORE

TNS

CLAUSE SENTENCE

Figure 4. Core coordination in Yaqui (4a)

The finality linkage type seems to be the most flexible structure, since it is the only one that allows both same-actor and different-actor constructions. (i) When the two actors are identical, the controllee can be covert (Taracahita, Tepiman) or can be overt (Nahuatl, Corachol); when it is overt, the actor commonly appears in a non-nominative form. When the

Lilián Guerrero

237

two actors are different, the dependent actor must be explicit; if it is realized as a pronominal form, it must be a non-nominative form. (ii) The linked event depends on the main unit for tense but not for modal or aspectual information, or even negation. (iii) The linked unit is explicitly introduced by a CLM; in some of the languages, the CLM corresponds to a postposition that also introduces nominal arguments (i.e., benefactives, instruments), while in others, it is a general subordinator marker. All these features result in a less tight linkage type in comparison with the other two, i.e. ad-core subordination. In the representation of the Yaqui purpose clause in (14b), the linked unit is a core unit modifying the matrix core, and so it occurs in the peripheryCORE. At this point, it is important to know what a typical adverbial construction looks like in these languages in order to understand how a purpose clause differs from these. The examples below illustrate typical cases of ad-clausal subordination in Yaqui. Reason clauses are introduced by bweituk “because” and they may be fully marked by tense, aspect and modal operators, in addition to negation. More importantly, the linked unit explicitly expresses its actor participant as a nominative noun (or full nominative pronoun), as in (18a), regardless it is identical or not to the main actor. In point of fact, reason clauses do not entail any reference dependency among the two units, as in (18b). This linkage type is, therefore, a good example of ad-clausal subordination where the linked unit is a subordinate clause. (18) a. Tei saja-k [bweituk itepoi ka enchi bicha-k] 1PL.NOM go.PL-PFV because 1PL.NOM NEG 2SG.ACC see-PFV “We left because we didn’t see you.” b. [bweituk Mary-Ø ka yepsa-k] te saja-k because Mary-NOM NEG arrives-PFV 1PL.NOM go.PL-PFV “We left because Mary never arrived.”

238

Clause Linkage in Southern Uto-Aztecan Languages

SENTENCE CLAUSE CORE NP

PERIPHERY

NUCLEUS

PP

PRED V Peo

NUCP

COREP NP

NP NUCLEUS

yeu=siika-k

PRED V enchi ka a bicha-ne betchi’ibo

V

V

PRED

PRED

NUCLEUS

NUCLEUS

CORE CLAUSE SENTENCE Figure 5. Ad-core subordination in Yaqui (14b)

NEG TNS

CORE

ASP

Lilián Guerrero

239

SENTENCE CLAUSE CORE NP

NUCLEUS

PERIPHERY CLM

CLAUSE

PRED

CORE

V Te

NP

NP NUCLEUS

saja-k

PRED bweituk

V

itepo ka enchi bicha- k V

V

PRED

PRED

NUCLEUS

NUCLEUS

CORE CLAUSE

NEG TNS

CORE CLAUSE -TNS

SENTENCE Figure 6. Ad-clausal subordination in Yaqui (18a)

4. The Semantics of Purpose Clauses Notionally, a purpose clause evokes intention and thoughts, future expectation, and the participant’s willingness for another state of affairs to take place, which alone can be a reason and motivation to do something. In RRG, the interclausal semantic relations are determined by the interaction of sub-hierarchies like those in (19):

240

Clause Linkage in Southern Uto-Aztecan Languages

(19) Semantic sub-hierarchies (Van Valin 2005; Guerrero 2006) a. Temporal hierarchy: Phase of a single event > simultaneous events > sequential events > unspecified b. Causal hierarchy: physical > verbal > underspecified[non-defeasible], inferred [defeasible] c. Participant’s mental disposition [PMD]: Intention > internal/direct experience > mental experience: commitment > mental experience: reasoning > non-mental experience: report d. Necessarily shared participant [NSP]: Yes > No Based on these semantic sub-hierarchies, the three major juncturenexus relationships are evaluated in Table 1. Notice that the only strategy that can be clearly differentiated is motion-cum-purpose which can be conceived as a macro-event where the motion and the intended actions are two phases of a single event (the first value of the temporal scale). Table 1. Degree of semantic cohesion of purpose clauses motion-cum-purpose intentional finality

temporal 1st value 3rd value 3rd value

causal 4th value 4th value 4th value

PMD 1st value 1st value 1st value

NSP 1st value 1st value 1st value

Within RRG, Cutrer (1993) and Van Valin (2009) have suggested that in English, purpose clauses optionally share the main actor, e.g. “John bought a book (for you) to read”, while the undergoer must be always shared by the two units. The main undergoer may control the dependent undergoer as in “John bought a book to read *it/to read *the magazine”, as well as the dependent actor as in “John gave her the book in order to sell it”. Moreover, Cutrer and Van Valin both identify two related but not identical “purpose” clauses based on syntactic grounds involving the pivot in the linked unit. The examples below come from Van Valin (2009): (20) a. b. c. d. e. f.

Pat brought the booki for her sister to read ___i *Pat brought the book for her sister to read it Patj brought the booki ___ j to read ___i *Patj brought the book ___ j to read it Pat brought the book in order (for her sister) to read it *Pat brought the book in order (for her sister) to read __

Lilián Guerrero

241

Thus, while sharing the actor is optional (20a) and (20c), sharing the undergoer is obligatory (20b) and (20d). This property distinguishes between purpose clauses in (20a, c) from rationale clauses in (20e), since in the latter there is no obligatory controller-controllee relationship of any kind (i.e. there is not a missing syntactic argument). See Guerrero (in press, in review, 2011) for an alternative analysis on this topic. If the NSP sub-hierarchy in (19d) refers to a semantically shared argument semantically, then all purpose linkage types, including rationale clauses, satisfy the first value. In contrast, if this scale is intended for syntactically shared arguments, then only ‘pure’ purpose clauses satisfy the first value; those clauses showing a co-referential pronoun will satisfy the second value. Indeed, purpose clauses strongly entail a referential dependency among one argument of the main unit and one argument of the linked unit. The controller may be the actor or undergoer, while the controllee may be any argument inside the main unit (Guerrero, in review). The semantic subhierarchy in (19e) seeks to capture these patterns of reference control over an argument inside the linked unit. (19) e. Referential control [RC] The identity of an argument in the linked unit is controlled by the main actor > by the main undergoer > by another main participant > there is no control relation. This scale is compatible with the NSP sub-hierarchy but it looks for a referential dependency among two arguments, regardless of the form of the controllee. Thus, the first value reflects such constructions demanding identical actors; the second value represents such structures requiring the matrix undergoer to control the dependent actor, while the third value encodes cases where a control relation holds between any other main argument and any dependent argument. Table 2. The semantic degree of purpose relations (revised) motion-cumpurpose intentional finality

temporal 1st value

causal PMD 4th value 1st value

NSP 1st value

RC 1st value

3rd value 3rd value

4th value 1st value 4th value 1st value

1st value 1st value 1st value 1st/2nd value

242

Clause Linkage in Southern Uto-Aztecan Languages

A first try to establish the logical structures for these purposive linkages is presented below based on the logical structures proposed in Van Valin (2005: 207). (21) Logical structures for purposive and modifying sub-events and reason relations (preliminary proposal) a. Motion-cum-purpose linkage want’ (xi, LS2) ∧ DO (xi, [motion’ (xi)] ¡ CAUSE [LS2… xi…]) b. Intentional linkage want’ (xi, [LS2… x…] ∧ DO (xi, [LS1… xi …] ¡ CAUSE [LS2… xi…]) c. Finality linkage want’ (xi, LS2) ∧ DO (xi, [[LS1… xi…]] ¡ CAUSE [LS2…..y….]] d. Reason linkage [LS1] because´ [LS2]

5. Conclusion The aim of this paper was to examine the syntactic and semantic properties of purpose clauses in Southern Uto-Aztecan languages. It was shown that purpose clauses make use of different clause linkage types which concerns different degrees of syntactic and semantic tightness. The Interclausal Relation Hierarchy as proposed by RRG can capture the syntactic and semantic correlations of these linkages, especially if the Referential Control semantic sub-hierarchy is taken into consideration.

List of Abbreviations ACC AND APPL AS(I) BEN CLM CONT DEM DESID DET DIR EMPH EXH

accusative andative applicative assertive benefactive clause linkage marker continuative demonstrative desiderative determiner directional emphatic exhortative

MOV NEG NOM NS O PAS PAST PFV PL PO POT PRO PROB

motion negation nominative non subject object passive past perfective plural primary object potential pronominal probable

Lilián Guerrero FUT IMP IMPFV INAN INC INSTR INV LOC

future imperative imperfective inanimate inceptive instrumental invisible locative

PRS PURP RED REFL S SG TNS TRAV

243

present purpose reduplication reflexive subject singular tense traversal

References Bickel, Balthasar. 2003. Clause linkage typology. Lecture series delivered at the 2003 International Role and Reference Grammar Conference, UNESP, São Jose do Rio Preto, Brazil —. 1993. Belhare subordination and the theory of topic. In K. Ebert, ed., Studies in Clause Linkage, 23–55. Arbeiten des Seminars für Allgemeine Sprachwissenschaft, no. 12. Zurich: University of Zürich. Burgess, Don. 1984. Western Tarahumara. In R.W. Langacker, ed., Studies in Uto-Aztecan Grammar 4. SIL & UTA. Canger, Una. 2001. Mexicanero de la Sierra Madre Occidental. Archivo de Lenguas Indígenas de México 24. México: El Colegio de México. Cristofaro, Sonia. 2003. Subordination. Oxford: Oxford University Press. Cutrer, Michelle. 1993. The semantic and syntactic factors in control. In R. D. Van Valin, Jr., ed., Advances in Role and Reference Grammar, 167196. Amsterdam and Philadelphia: John Benjamins. Diessel, Holger. 2001. The ordering distribution of main and adverbial clauses: a typological study. Language 77.3: 433-455 Dixon, R.M.W. and Alexandra Aikhenvald. 2009. The Semantics of Clause Linking. Oxford: Oxford University Press. Ernst, Thomas. 2001. The Syntax of Adjuncts. Cambridge: Cambridge University Press. Estrada, Zarina. 1998. Pima Bajo de Yapechi, Chihuahua. Archivo de Lenguas Indígenas de México 21. México: El Colegio de México. Félix, Rolando. 2005. A Grammar of River Warihio. Unpublished Ph. D. Dissertation, Rice University. Foley, William and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. García Salido, Gabriela. 2008. The encoding of serial verbs in Southeastern Tepehuan. Handout from talk at the Seminario de Complejidad Sintáctica, USON.

244

Clause Linkage in Southern Uto-Aztecan Languages

Gómez, Paula. 1999. Huichol de San Andrés Cohamiata, Jalisco. Archivo de Lenguas Indígenas de México 22. México: El Colegio de México. Guerrero, Lilián. 2006. The Structure and Function of Yaqui Complementation. Munich: Lincom. —. 2009. Extending the challenge of control phenomena. Paper presented at the 2009 International Conference on Role and Reference Grammar. University of California at Berkeley/LSA. California. —. 2011. Revising the controller-controllee relations. Paper delivered at the 2011 International Conference on Role and Reference Grammar. Chile. —. In press. Más sobre controladores y pivotes: el caso de las relaciones de propósito. In R. Mairal, L. Guerrero and C. González-Vergara, eds., Gramática del Papel y la Referencia. Introducción, Avances y Aplicaciones. Madrid: AKAL —. In review. Goal, desire and finality: different strategies for the same purpose. International Journal of American Linguistics. Givón, Talmy. 2001. Syntax. Amsterdam and Philadelphia: John Benjamins. Haspelmath, Martín. 1989. From purposive to infinitive: a universal path of grammaticalization. Folia Linguistica Historica X/1-2, 287-310. Societas Linguistica Europea. Hengeveld, Kee. 1998. Adverbial clauses in the languages of Europe. In J. Van der Auwera and D. O. Baoill, eds., Adverbial Constructions in the Languages of Europe, 335-419. Berlin and New York: Mouton de Gruyter. Iturrioz, José Luis and Paula Gómez. 2006. Gramática Wixarika. Munich: Lincom. Jones, Charles. 1991. Purpose Clauses: Syntax, Thematic and Semantics of English Purpose. Dordrecht: Springer. Johnson, Jean. 1962. El Idioma Yaqui. México: INAH. Kortmann, Bernd. 1997. Adverbial Subordination. A Typology and History of Adverbial Subordinators Based on European Languages. Berlin and New York: Mouton de Gruyter. Longacre, Robert. 2007. Sentences as combination of clauses. In T. Shopen, ed., Language Typology and Syntactic Description II, 372420. Cambridge: Cambridge University Press. Miller, Wick. 1996. Guarijío: Gramática, Textos y Vocabulario. México: UNAM. —. 1993. Guarijío de Arechuyvo, Chihuahua. Archivo de Lenguas Indígenas de México 16. México: El Colegio de México.

Lilián Guerrero

245

Peralta, Valentín. 2007. Construcciones de propósito y movimiento asociado en nawatl de Pajapan, Veracruz. Handout from talk at the Friends of the Uto-Aztecan languages. Saxton, Dan. 1982. Papago. In R.W. Langacker, ed., Studies in UtoAztecan Grammar 3. Arlington: SIL & UT. Schmidtke-Bode, Karsten. 2009. A Typology of Purpose Clauses. Amsterdam and Philadelphia: John Benjamins. Thompson, Sandra. 1985. Grammar and written discourse: initial vs. final purpose clauses in English. Text 5: 55-84. Van Valin, Robert D., Jr, and Randy J. LaPolla. 1997. Syntax. Structure, Meaning and Function. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2009. Privileged syntactic arguments, pivots and controllers. In L. Guerrero, S. Ibáñez, and V. Belloro, eds., Studies in Role and Reference Grammar, 45-68. México: UNAM. —. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press. Verstraete, Jean-Christophe. 2008. The status of purpose, reason, and intended endpoint in the typology of complex sentences: implications for layered models of clause structure. Linguistics 46.4: 757-788.

MODALITY IN TAIWAN SIGN LANGUAGE CHIEN-HUNG LIN AND JUNG-HSING CHANG NATIONAL CHUNG CHENG UNIVERSITY

Abstract This paper discusses modality in Taiwan Sign Language within the framework of Role and Reference Grammar (Van Valin and LaPolla 1997; Van Valin 2005), with the goal of accounting for the correlation between the semantic and syntactic representations of modals. First, it is observed that the modals with a wider semantic scope (e.g. epistemic modal) have a freer syntactic distribution. Second, the same modals in different syntactic positions exhibit different degrees of subjectivity. To convey stronger subjectivity, the modal in the sentence-final position tends to involve non-manual features. Third, the modal with a wider semantic scope (e.g., epistemic modal) always precedes the one with a narrow scope (e.g. deontic modal).

Keywords Taiwan Sign Language, modality, epistemic modals, deontic modals

1. Introduction Modality is a grammatical category on a par with tense and aspect. Generally speaking, tense is concerned with anchoring events in time, whereas aspect is related to the internal temporal structure of events (Comrie 1976, 1985). Modality differs from tense and aspect in that it does not refer to the properties of the event, but rather to the speaker’s attitude to the proposition of the sentence. Thus, modality is defined as the grammaticalized expression of the speakers’ attitudes and opinions toward the proposition of the sentence (Lyon 1977; Palmer 2001). In spoken languages, modality is expressed through various linguistic devices, such as morphological inflections, lexical items, syntactic patterns, or intonation

Chien-hung Lin and Jung-hsing Chang

247

(Bybee and Fleischman 1995). For example, English uses modals to distinguish a judgment about a proposition from a statement. The sentence in (1a) is a statement without expressing the speaker’s attitude, whereas the sentence in (1b) makes use of the modal “may” to convey the speaker’s lack of confidence to the proposition. The sentence in (1c) uses the modal such as “must” to convey the speaker’s confidence in his reporting of facts. (1) a. Kevin is sick today. b. Kevin may be sick today. c. Kevin must be sick today. Over forty years of researches on sign languages have shown fruitful results and evidenced that signed languages and spoken languages have many linguistic properties in common, such as diachronic language acquisition time (Newport and Meier 1985; Lillo-Martin 1999), the speed that the proposition is transmitted (Bellugi and Fisher 1972; Fischer et al. 1999), the complex grammatical structures that encode the relationship between form and meaning (Klima and Bellugi 1979; Sutton-Spence and Woll 1999; Liddell 2003; Tai 2005, 2008; Chang et al. 2005; Tai and Su 2006; Sandler and Lillo-Martin 2006; Chang 2008, 2009). The major difference between signed languages and spoken languages resides in how their messages are produced and perceived. The messages of spoken languages are articulated by vocal tract and perceived through audition, whereas those of signed languages are produced by physical action and perceived visually. That is, human language can be produced and perceived through two different channels, namely, the auditory-vocal channel in spoken languages and the visual-gestural channel in signed languages. It has been well-accepted that modality is a cross-language grammatical category. In signed languages, modality can be expressed through two manual ways or one non-manual way. The first manual way is to use lexical items, for instance, British Sign Language [BSL] uses modals such as “SHOULD”, “CAN” and “MUST” (Sutton-Spence and Woll 1999:126). The second manual way is to express scalar degree of tenseness, strength and size in the verb sign, for example, in BSL, the movement for “MUST^ASK” is larger, tenser and stronger than that of “COULD^ASK” (Sutton-Spence and Woll 1999:126). However, the non-manual way is to incorporate non-manual features into the sentence, such as head nodding, chin lifting and so forth (Wilcox and Wilcox 1995).

248

Modality in Taiwan Sign Language

Modality in signed languages has been studied in many different ways. For example, some studies discuss the expression of modality (Fisher and Gough 1978; Padden 1988; Ferreira-Brito 1990; Wilcox 1996; Sutton-Spence and Woll 1999). Others explore the grammaticalization of modals (Wilcox and Wilcox 1995; Wilcox and Wilcox 1995; Shaffer 2000; 2002; Janzen and Shaffer 2002; Meir 2003; Wilcox and Shaffer 2005). Still others investigate the syntactic distribution of modals (Aarons, Bajan, Kegl, and Neidle 1995; Shaffer 2004). However, not much attention is given to the correlation between the semantic and syntactic representations of modals This paper discusses modality in Taiwan Sign Language [TSL] within the framework of Role and Reference Grammar [RRG] (Van Valin and LaPolla 1997; Van Valin 2005), with the goal of accounting for the correlation between the semantic and syntactic representations of modals. This paper has centered on the following issues. The first issue discusses the syntactic distribution of modals in TSL. The second issue explores the correlation of syntactic position and the degree of subjectivity. The third issue concerns the correlation between the semantic scope of modals and their ordering, when there are double modals within a sentence. The methodology we use to elucidate the above issues is a discourse-based analysis, and the TSL data discussed in this paper were collected from deaf informants and from sign language news in Taiwan Public Television Service by the authors. This paper is organized as follows: Section 2 introduces the theoretical background, Section 3 discusses the syntactic distribution of modals, Section 4 explores the correlation between syntactic positions and degree of subjectivity, Section 5 investigates the word orders of modals, while Section 6 concludes this paper.

2. Theoretical Background RRG is a linguistic theory based on languages with diverse structures such as Lakhota, Tagalog and Dyirbal, with an aim to capture and explain the interaction of syntax, semantics and pragmatics (Van Valin and LaPolla 1997; Van Valin 2005). RRG postulates three representations: syntactic, semantic, and discourse-pragmatics representation. The syntactic and semantic representations are linked together by a set of linking algorithms, and pragmatics will be taken into consideration while operating linking algorithms. With regard to syntactic representation, it consists of two projections. One is a constituent projection which deals with verbs, arguments, and adjuncts, and the other is an operator projection which deals with functional categories such as modality, aspect, negation, tense and so forth.

Chien-hung Lin and Jung-hsing Chang

249

As for constituent projection, the representation of a clause structure is called the layered structure of clause. The predicating element is realized by a verb without taking any argument and it is defined as “nucleus [NUC]” in the layer structure of the clause. The next layer is the “core [CORE]” which contains a NUC and core arguments of the predicate. The non-argument of predicate is projected as “periphery [PERIPHERY]” which joints to the CORE to formulate a “clause [CLAUSE]”. Figure 1 shows the constituent projection. The verb “drive” is projected as a NUC, and then combines with the core arguments, “Sarah” and “the car”, to formulate a CORE. The temporal adverb “yesterday” is a non-core argument, so it is projected as a PERIPHERY and linked to the CORE. Finally, the CORE incorporates with PERIPHERY to formulate a CLAUSE. The CLAUSE is further projected as a SENTENCE. SENTENCE CLAUSE CORE ARG

NUC

PERIPHERY ARG

PRED NP Sarah was not able to

V

NP

drive

the car

ADV yesterday.

Figure 1: Constituent projection

As for operator projection, each layer may be modified by one or more operators. The scope of nucleus operators is over NUC; thus, they modify the action without reference to the core arguments. Core operators modify the relation between the action and the core arguments. Clausal operators modify the entire clause as a whole. Note that the operator projection mirrors the constituent projection in terms of layered structure. Three layers of operators are demonstrated as in Table 1.

250

Modality in Taiwan Sign Language

Table 1: Operators in layered structure of the clause (Van Valin 2005: 9) Nuclear operators: Aspect Negation Directionals Core operators: Directionals Event quantification Modality (root modals) Internal negation Clausal operators: Status (epistemic modals, external negation) Tense Evidentials Illocutionary force Figure 2 shows the operator projection. The modal “able” conveys the ability of Sarah, thus it is projected as a Modality which modifies the CORE. The negation “not” negates Sarah's ability to carry out an action, so it is projected as a Negation which modifies the CORE. The copula “was” conveys the time of the event and the type of speech act, thus it is identified as a clausal operator modifying the entire proposition. Sarah was not able to drive the car yesterday. V NUC Modality Negation Tense Illocutionary force Figure 2: Operator projection

CORE CORE CLAUSE CLAUSE

Chien-hung Lin and Jung-hsing Chang

251

RRG is characterized by its layered representation. Within the syntactic representation, the operator projection mirrors the constituent projection in terms of the layered structure. In this paper, the operator projection will be adopted to account for the syntactic distribution of modals in TSL. In Section 3, the syntactic distribution of modals and its correlation with the degree of subjectivity will be discussed.

3. Syntactic Representation of Modality 3.1. Syntactic Distribution of Modals Following Palmer (1979), modality is classified into deontic and epistemic modality. Epistemic modality is concerned with the speaker’s degree of commitment to the truth of the proposition (Payne 1997:246). Deontic modality is concerned with permission or obligation for the performance of actions (Hoye 1997:43). In simple terms, epistemic modality conveys necessity or possibility, while deontic modality denotes permission and obligation. The English sentences with epistemic and deontic modals are given in (2) and (3). The sentence in (2) has an epistemic interpretation, since the modal “may” conveys the speaker’s inference of the proposition. The sentence in (3) has a deontic interpretation, since the modal “can” is used to signify the speaker’s permission of the event. (2) Mark may be in his office.

[Epistemic modality]

(3) Kelly can take a ten-minute break.

[Deontic modality]

Like spoken languages, it has been noted that signed languages also have such two-way distinction in modality (Wilcox and Wilcox 1995; Shaffer 2000, 2002; Janzen and Shaffer 2002; Wilcox and Shaffer 2005). Further, it has been suggested by Shaffer (2000) that in American Sign Language there is an iconic correlation between the semantic and syntactic representations. She points out that the modals with lower subjectivity (e.g., deontic modals) tend to occur in the preverbal position, while the ones with higher subjectivity (e.g., epistemic modals) usually occur in the postverbal position. In TSL, the epistemic modals can occur in the preverbal position, the sentence-final position, and the sentence-initial position, as exemplified in (4)-(6). The TSL modal “MUST” in (4) conveys necessity of the proposition, while the modals in (5) and (6) convey possibility of the proposition.

252

Modality in Taiwan Sign Language

(4) In the preverbal position: NIGHT DRIVE MUSTE LIGHT. “Driving in the night, you must turn on the light.” (5) In the sentence-initial position: NOW AIRPLANE WRECKAGE PART FIND-OUT. MAYBEE AIRPLANE PEOPLE TOTAL 228 HEALTHY DIFFICULT. “Parts of the aircraft’s wreckage were found out now. The 228 people may have died in the crash.” (Sign language news in Taiwan 06/03/2009) (6) In the sentence-final position: MAY JUNE PLUM^RAIN TIME SHOULDE. “May and June should be the raining seasons.” (Sign language news in Taiwan 06/02/2009) In addition, it is found that the epistemic modality can also be realized via non-manual features. That is, the occurrence of epistemic modals is optional, since the epistemic interpretation can be expressed in a non-manual way, as shown in (7). The sentences in (7a) and (7b) have the same interpretation, but they differ in the syntactic representation. In (7a), the TSL modal “MUST” is used to convey the speaker’s strong commitment. In (7b), the non-manual features, such as upward-backward head tilt and strengthened movement, are employed to convey the same degree of commitment, as shown in Figure 4c. (7) a. HE DRIVE MUST UNABLE “I am sure that he is unable to drive.”

a. HE

b. DRIVE

Figure 3 (upward-backward head tilt)

c. MUST

d. UNABLE

Chien-hung Lin and Jung-hsing Chang

253

b. HE DRIVE UNABLE “I am sure that he is unable to drive.”

a. HE

b. DRIVE

c. UNABLE

Figure 4

Further, it is interesting to find that the signer can manipulate non-manual features to reflect the speaker’s degree of commitment. Comparing the sentence (8) with (7b), it shows that the non-manual features, such as head tilt and lip corners depressed, are used to signify the speaker's uncertainty. (8)

(head tilt and lip corners depressed) HE DRIVE UNABLE “I am uncertain that he is unable to drive.”

a. HE

b. DRIVE

c. UNABLE

Figure 5

Unlike epistemic modality, deontic modality can only be expressed in a manual way. In TSL, deontic modals can occur in the pre-verbal position and the sentence-final position. As shown in (9) and (10), the TSL modals such as “CAN” and “SURE” are employed to convey the speaker’s permission for the event.

254

Modality in Taiwan Sign Language

(9) In the preverbal position: FILM PEOPLE CAND ENTER LOOK-FOR FILM-EDITING SUPPORT AND-SO-FORTH. “The film staffs are allowed to enter (the center), and look for support such as film editing and so forth.” (Sign language news in Taiwan 06/02/2009) (10) In the sentence-final position: NOW SENIOR-HIGH-SCHOOL STUDENT NOON GO-OUT EAT SURED. “Senior high school students are allowed to eat out by noon now.” (Sign language news in Taiwan 06/03/2009) The discussion above demonstrates that epistemic modality can be expressed either manually or non-manually, whereas deontic modality can only be expressed manually. It also shows that epistemic modals have a much freer syntactic distribution than deontic modals. It is reasonable to assume that the semantic scope of modals plays a role in determining their syntactic distribution. RRG postulates two operators to account for modality, which are “modality [MOD]”and “status [STA]”. The modality conveying the sense of ability, permission, and obligation (e.g., deontic modals) is defined as the core operator MOD, while the modality conveying the sense of necessity and possibility (e.g., epistemic modals) is categorized as the clausal operator STA. That is to say, the modal with a wider semantic scope has a much freer syntactic distribution than the one with a narrow semantic scope.

3.2. Syntactic Positions and Speaker's Subjectivity It is well known that some languages have the corresponding tentative forms to express a weaker strength of modality. For instance, the modals in English, “will”, “can”, and “may”, have corresponding tentative forms, such as “would”, “could”, and “might”. An interesting parallelism observed in TSL shows that TSL uses different syntactic distribution to locate different degrees of subjectivity. Subjectivity is defined as devices whereby the speaker, in making an utterance, simultaneously comments upon that utterance and expresses his attitude to what he is saying (Lyons1977:739).

Consider the sentences in (11) and (12). These two sentences only differ in the syntactic position of the modal, but they express different degrees of

Chien-hung Lin and Jung-hsing Chang

255

subjectivity. To convey stronger speaker's subjectivity, the modal in the sentence-final position tends to involve non-manual features, such as upward-backward head tilt and chin lift, as shown in Figure 7. (11) In the preverbal position: ECONOMY SITUATION BAD, POOR^PEOPLE GOVERNMENT ABLED TAKE-CARE+POOR^PEOPLECL “Due to economic recession, the government will take care of the poor.” (12) In the sentence-final position: ECONOMY SITUATION BAD, POOR^PEOPLE GOVERNMENT TAKE-CARE+ POOR^PEOPLECL ABLED “Due to economic recession, it is obligatory for the government to take care of the poor.”

Figure 6: ABLE (from Ex.11)

Figure 7: ABLE (from Ex.12)

Likewise, epistemic modals can convey different degrees of subjectivity by manipulating their syntactic positions, as exemplified in (13) and (14). These two sentences only differ in the syntactic position of modal. To convey stronger subjectivity, the modal in the sentence-final position occurs with non-manual features, as shown in Figure 9. (13) In the preverbal position: SISTER STUDY DILIGENT. SHE MUSTE EXAM SUCCESS “My sister studies so hard. She can pass the exam.” (14) In the sentence-final position: SISTER STUDY DILIGENT. SHE EXAM SUCCESS MUSTE “My sister studies so hard. She surely can pass the exam.”

256

Modality in Taiwan Sign Language

Figure 8: MUST (from Ex.13)

Figure 9: MUST (from Ex.14)

Due to visual-gestural channel of signed languages, the signer can resort to non-manual features to vary the degree of subjectivity. TSL has shown that non-manual features can be used to express epistemic modality. Further, to convey stronger subjectivity, the modal in the sentence-final position tends to involve non-manual features.

4. Double Modals in TSL Some spoken languages allow single modal in a sentence, such as Standard English, while some spoken languages allow double modals in a sentence, such as Catalan, Icelandic, Chinese, and English dialects in the southern United States (DiPaolo 1989; Nagle 2003). With regard to TSL, a sentence containing two modals is allowed. However, it is noted that the word order of modals is closely related to the interpretation of modality, as in (15) and (16). These two sentences only differ in the word order of the modals, and they have different interpretations. In (15), the TSL modal “SURE” precedes the modal “CAN”, in which “SURE” is interpreted as in epistemic sense, while “CAN” is in deontic sense. In (16), the modals are in the inverse order. Thus, “SURE” is interpreted as in deontic sense, while CAN is in epistemic sense. 1 (15) HE GO TAIPEI SUREE CAND. “It is sure that he is allowed to go to Taipei.” (16) HE GO TAIPEI CANE SURED. “It is probable that he will go to Taipei.”

1 Due to the process of meaning extension, a single modal may convey different modal senses. In many languages, the different modal senses may share the same modal forms (Hoye 1997)

Chien-hung Lin and Jung-hsing Chang

257

Sometimes, the reverse order of modals may lead to unacceptability, as given in (17a) and (17b). In the sentence-final position, the TSL modal “ABLE” cannot precede the modal “SHOULD”. It seems that the order of modals have to comply with some semantic constrains. In the following discussion, RRG is employed to account for the correlation of the word order of modals and their interpretations in TSL. (17) a. ELDER-BROTHER GO DRIVE EXERCISE, HE DRIVE SHOULDE ABLED. “The elder brother ever took driving lessons. It’s sure that he is able to drive.” b. *ELDER-BROTHER GO DRIVE EXERCISE, HE DRIVE ABLE SHOULD. To capture the order of operators, RRG proposes the basis principle of scope assignment to govern operators. This principle indicates that the semantic scope of clausal operator is over the core operator, and the scope of core operator is over the nucleus operator. Those operators are syntactically ordered with respect to each other in terms of two rules, as given in (18). The first is the universal operator linear precedence rule which indicate that the operators are syntactically ordered and represented with respect to each other in terms of the scope principle. The second is language-specific linear precedence rules which indicates that the operators are simply line up according to their scopes on one side of the nucleus or the other. That is, the order of operators in non-verb-final languages is illocutionary force, tense, status, modality and aspect, whereas that in verb-final languages is in the inverse order. (18) a. Universal operator linear precedence rule CLAUSE CORE NUCLEUS b. Language-specific linear precedence rules OPs > NUC or NUC > OPs (Van Valin and LaPolla 1997:72) It has been mentioned that the deontic modal is categorized as the operator MOD, and the epistemic modal as STA. That is, the epistemic modal has to precede the deontic in the non-verb-final languages and the modals are in the reverse order in the verb-final languages. Take Chinese for example. As in (19), the epistemic modal “yiding” conveys the speaker’s confidence about what he is saying, and the deontic modal “hui” points out the participant’s ability. Thus, the epistemic modal “yiding”

258

Modality in Taiwan Sign Language

precedes the deontic modal “hui”. (19) Chinese lanqiu Zhangsan yidingE huiD da Zhangsan must able play basketball “Zhangsan must know how to play basketball.” We would like to illustrate the word order of modals, using verb-final languages such as Turkish, and dependant-marking languages such as Japanese. In Turkish, the deontic modal “emi” precedes the epistemic modal “yebil”, as given in (20). In Japanese, the deontic modality can be expressed by altering the form of verb, for instance “hanas-er-u” in (21a), or by an individual morpheme, for instance “dekiru” in (21b), and the epistemic modal “hazu” consistently follows the deontic modal in Japanese, as shown in (21a) and (21b). (20) Turkish Gel-emi D-yebil E-ir-im Come-ABLE.NEG-PSBL-AORIST-1SG 'I may be unable to come.' (Van Valin and LaPolla 1997:44) (21) Japanese a. ᓐߪਛ࿖⺆߇⹤ߖࠆߪߕߛ‫ޕ‬ Kare wa chuugokugo ga hanas-er-u D hazu E da. He TOP Chinese NOM speak can should COP “He should be able to speak Chinese.” b. ᓐߪਛ࿖⺆߇⹤ߔߎߣ߇ߢ߈ࠆߪߕߛ‫ޕ‬ Kare wa chuugokugo ga hanasu koto ga dekiru D “He TOP Chinese NOM speak NMZ NOM can hazu E da. should COP” “He should be able to speak Chinese.” Figure 10 summarizes how epistemic modals and deontic modals are represented in the operator projection and projected onto the syntactic representation in terms of universal operator linear precedence rule.

Chien-hung Lin and Jung-hsing Chang

Epistemic modal

Deontic modal

V Deontic modal

259

Epistemic modal

NUC MOD STA

CORE

MOD

CLAUSE

STA

SENTENCE Figure 10: The operator projection and the syntactic representation of modals

It has been mentioned that the modals in TSL can occur either in the preverbal position or in the sentence-final position. The combination of an epistemic modal and a deontic modal can occur in the sentence-final position, as in (22a), the preverbal position, as in (22c) or both positions, as in (22e). It shows that the TSL epistemic modal “SHOULD” has to precede the TSL deontic modal “ABLE” either in the preverbal position or in the sentence-final position. When two modals occur separately, the epistemic modal “SHOULD” has to occur in the preverbal position, while the deontic modal “ABLE” occurs in the sentence-final position. That is, the epistemic modal always precedes the deontic modal in a sentence. The word order of modals in TSL is summarized as in Table 2. (22) a. GRADUATE FINISH, HE GO FIND WORK SHOULDE ABLED “After graduation, it is probable that he will seek a job.” b. * GRADUATE FINISH, HE GO FIND WORK ABLED SHOULDE c. GRADUATE FINISH, HE SHOULDE ABLED GO FIND WORK d. * GRADUATE FINISH, HE ABLED SHOULDE GO FIND WORK e. GRADUATE FINISH, HE SHOULDE GO FIND WORK ABLED f. * GRADUATE FINISH, HE ABLED GO FIND WORK SHOULDE Table 2: The word order of modals in TSL The word order of modals Epistemic modal > Deontic modal V V Epistemic modal >Deontic modal Epistemic modal V Deontic modal

In general, deontic modals modify the relation between the participant and the action, and epistemic modals are concerned with speaker’s judgment. Following the linear precedence rules, the modals are ordered

260

Modality in Taiwan Sign Language

with respect to each other in terms of their semantic scopes, in which the deontic modals should occur more closely to the verb than the epistemic modals. It is worthy of noting that in TSL the combination of modals occurring in the sentence-final position contradicts this generalization. As shown in (22a), the epistemic modal occurs more closely to the verbal predicate than the deontic modals in the sentence-final position. In addition, the epistemic modal has to precede the deontic modal, but not vice versa, when these two modals occur separately, as exemplified in (22e) and (22f). That is to say, in TSL, the modal with a wider semantic scope (e.g., epistemic modal) always precedes the modal with a narrow semantic scope (e.g., deontic modal). The operator projection and the syntactic representation of the modals in TSL are summarized as in Figure 11. Epistemic modal

Deontic modal

V Epistemic modal

Deontic modal

NUC MOD STA

CORE CLAUSE

MOD STA

SENTENCE Figure 11: The operator projection and the syntactic representation of modals in TSL

5. Conclusion This paper has discussed modality in TSL within the framework of RRG. It has been found that modality in TSL can be expressed either by manual or by non-manual ways. By manual way, the modal with a wider semantic scope (e.g., epistemic modal) has a freer syntactic distribution than the one with a narrow semantic scope. In addition, the modal in the sentence-final position conveys stronger subjectivity than the same one in the preverbal position. The modal with a wider semantic scope (e.g., epistemic modal) always precedes the one with a narrow semantic scope (e.g., deontic modal), when a sentence contains two modals at the same time. However, by non-manual way, the epistemic modality can be expressed via non-manual features, and the signer can manipulate these non-manual features to convey different degrees of subjectivity.

Chien-hung Lin and Jung-hsing Chang

261

List of Abbreviations

^ +

1 A

B

A> B

the scope of non-manual markers compounds Two signs are produced currently with different hands first person The scope of A is over B A precedes B

AORIST CL COP D E NEG NMZ NOM PSBL SG TOP

aorist tense classifier predicate copular deontic modals epistemic modals negation nominalizer nominative possibility singular topic

References Aarons, Debra, B. Bahan, J. Kegl, and C. Neidle. 1995. Lexical tense markers in American Sign Language. In Karen Emmorey and Judy S. Reilly, eds., Language, Gesture, and Space, 225-253. Hillsdale, NJ: Erlbaum. Aarons, Debra, B. Bahan, J. Kegl, C. Neidle. 1995. Lexical tense markers in American Sign Language. In Karen Emmorey and Judy S. Reilly, eds., Language, Gesture, and Space, 225-253. Hillsdale, NJ: Erlbaum. Bybee, Joan, Revere Perkins, and William Pagliuca. 1994. The Evolution of Grammar: Tense, Aspect, and Modality in the Languages of the World. Chicago: University of Chicago Press. Comrie, Bernard. 1976. Aspect. Cambridge: Cambridge University Press. —. 1985. Tense. Cambridge: Cambridge University Press. DiPaolo, Marianna. 1989. Double modals as single lexical items. American Speech 64:195-244. Ferreira-Brito, Lucinda. 1990. Epistemic, alethic, and deontic modalities in a Brazilian Sign Language. In Susan Fisher and Patricia Siple, eds., Theoretical Issues in Sign Language Research. vol. 1: Linguistics, 229-260. Chicago: University of Chicago Press. Fischer, S. D., and B.Gough. 1978. Verbs in American Sign Language. Sign Language Studies 18:17-48. Hoye, Leo.1997. Adverbs and Modality in English. New York: Longman. Janzen, T., and B. Shaffer. 2002. Gesture as the substrate in the process of ASL grammaticalization. In Richard P. Meier, Kearsy Cormier, and David Quinto-Pozos, eds., Modality and Structure in Signed and

262

Modality in Taiwan Sign Language

Spoken Languages, 199-223. Cambridge: Cambridge University Press. Klima, Edward, and Ursula Bellugi. 1979. The Signs of Language. Cambridge, MA: Harvard University Press. Lillo-Martin, Diane. 1999. Modality effects and modularity in language acquisition: the acquisition of American Sign Language. In William C. Ritchie and Tej K. Bhatia, eds., Handbook of Child Language Acquisition, 531-567. London: Academic Press. —. 2002. Where are all the modality effects? In Richard P. Meier, Kearsy Cormier, and David Quinto-Pozos, eds., Modality and Structure in Signed and Spoken Languages, 241-262. Cambridge: Cambridge University Press. Lyons, John. 1977. Semantics. 2 vols. Cambridge: Cambridge University Press. Meier, Richard P. 2002. Why different, why the same? explaining effects and non-effects of modality upon linguistic structure in sign and speech. In Richard P. Meier, Kearsy Cormier, and David Quinto-Pozos, eds., Modality and Structure in Signed and Spoken Languages, 1-26. Cambridge: Cambridge University Press. Meir, Irit. 2003 Grammaticalization and modality: the emergence of a case-marked pronoun in Israeli Sign Language. Journal of Linguistics 39: 109-140. Nagle, Stephen J. 2003. Double modals in the southern United States: syntactic structure or syntactic structures? In Roberta Facchinetti, Manfred G. Krug, and Frank Palmer, eds., Modality in Contemporary English, 349-371. Berlin and New York: Mouton de Gruyter. Padden, Carol. 1988. Interaction of Morphology and Syntax in American Sign Language. New York: Garland Press. Palmer, F. R. 1979. Modality and the English Modals. London: Longman. —. 2001. Mood and Modality. Cambridge: Cambridge University Press. Payne, Thomas E. 1997. Describing Morphosyntax: A Guide for Field Linguists. Cambridge: Cambridge University Press. Romano, Christine. 1991. Mixed headedness in American Sign Language: evidence from functional categories. Papers from the Third Student Conference in Linguistics 1991 (MIT Working Papers in Linguistics), 241-254. Shaffer, Barbara. 2000. A Syntactic Pragmatic Analysis of the Expression of Necessity and Possibility in American Sign Language. Unpublished Ph.D. Dissertation, University of New Mexico. —. 2002. CAN'T: the negation of modal notions in ASL. Sign Language Studies 3: 34-53. —. 2004. Information ordering and speaker subjectivity: Modality in ASL.

Chien-hung Lin and Jung-hsing Chang

263

Cognitive Linguistics 15: 175–195. Shepherd, S.C. 1993. The acquisition of modality in Antigua creole. In Norvert Dittmar and Astrid Reich, eds., Modality in Language Acquisition, 171-184. New York: Walter de Gruyter. Stokoe, William. 1960. Sign language structure. Studies in Linguistics Occasional Papers 8. Buffalo, NY: University of Buffalo Press. Sutton-Spence, Rachel, and Bencie Woll. 1999. The Linguistics of British Sign Language: An Introduction. Cambridge: Cambridge University Press. Traugott, Elizabeth Closs. 1989. On the rise of epistemic meanings in English: an example of subjectification in semantic change. Language 65.1: 31-55. Van Valin, Robert D., Jr. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Wilcox, P. 1996. Deontic and epistemic modals in ASL: a discourse analysis. In Adele E. Goldberg, ed., Conceptual Structure, Discourse and Language, 481-492. Stanford, CA: CSLI. Wilcox, P., and S. Wilcox.1995. The gestural expression of modality in American Sign Language. In Joan Bybee and Suzanne Fleischman, eds., Modality in Grammar and Discourse, 135-162. Amsterdam and Philadelphia: John Benjamins. Wilcox, S., and B. Shaffer. 2005. Modality in American Sign Language. In William Frawley, ed., The Expression of Modality, 207-237. Berlin and New York: Mouton de Gruyter. Wilcox, S., and P. Wilcox. 1995. The gestural expression of modality in ASL. In Joan Bybee and Suzanne Fleischman, eds., Mood and Modality, 135-162. Amsterdam and Philadelphia: John Benjamins.

PART 4. SYNTAX-INFORMATION STRUCTURE INTERFACE

THE LEFT PERIPHERY AND FOCUS STRUCTURE IN JAPANESE MITSUAKI SHIMOJO UNIVERSITY AT BUFFALO, THE STATE UNIVERSITY OF NEW YORK

Abstract Given the elaborate morphosyntactic representations of topic and focus in Japanese, it has been a challenge to describe fully how the different types of topic and focus are differentiated in the language. This study integrates subordinate focus structure (Erteschik-Shir 1997, 2007) into Role and Reference Grammar [RRG] and examines various topic and focus elements which are linked with the so-called left periphery of a sentence in Japanese. For each topic/focus type, a default focus structure representation is proposed, and it is also extended to account for particular focusshifting effects which result from marked use of the topic/focus phrase.

Keywords Topic, focus, left periphery, Japanese

1. Introduction Despite the elaborate morphosyntactic representations of topic and focus in Japanese, it has not been described fully how different types of topic and focus are represented in the syntax of Japanese. In particular, the relationship between contrastive topic and contrastive (or narrow) focus has been unclear with respect to their common and discrete properties. Thus, this paper examines various topic and focus elements which are particularly linked with the so-called left periphery of a sentence in Japanese and attempts to present detailed structural and focus-structure

Mitsuaki Shimojo

267

representations in RRG (Van Valin and LaPolla 1997; Van Valin 2005). More specifically, the goals of the paper consist of the following: (1) to describe the structural and pragmatic properties of the left periphery in Japanese, with special reference to the so-called topic marker -wa and nominative marker –ga, (2) to define different types of topic and focus (topic, contrastive topic, contrastive focus, restrictive focus, etc.) by integrating the notion of subordinate focus-structure (Erteschik-Shir 1997, 2007) into focus structure projection in RRG, and (3) to capture “unexpected” uses of topic and focus, which bring out particular focusing or defocusing effects, by pinpointing the extended use of topic and focus. It is widely known in the so-called discourse-configurational languages that the left periphery of a sentence exhibits pragmatic prominence, which is commonly associated with the terms topic and focus. In these languages, word order is often restricted according to the position of topic and focus, hence, rigid focus structure as found in French, Toba Batak, Sesotho, and Italian (Van Valin 1999). However, as shown in (1), the pragmatically specialized nature of the left-periphery (e.g. “yesterday” and “what”) is observed even in English, which is considered as a language of rigid syntax in which topicalization and focus fronting are not required, except for wh-fronting. (1) Yesterday, what did John give to Mary in the library? The critical property of the left periphery is that the left edge of a sentence is commonly ambivalent between topic and focus, not only because the left periphery often hosts separate topic and focus phrases in the same sentence, but also because it may host a phrase that is both topic and focus like. In a number of languages, sentence initial topics represent so-called contrastive or switch topics, a specific type of topic that has some focus properties, and sentence initial foci represent restrictive or narrow foci, a specific type of focus that has some topic properties (Erteschik-Shir 2007).1 It should be noted, however, that not all languages require the left periphery; for example, topicalization is optional in Danish, but it is strongly preferred in languages such as Hungarian, Japanese and Korean. With respect to ordering within the left periphery, a topic typically precedes a focus. In Hungarian, for example, a topic is followed by a restrictive focus, and regular focus appears in situ post-verbally (Horvath 1 These languages include Catalan, Czech, Danish, German, Hungarian, Italian, Japanese, Korean, and American Sign Language.

268

The Left Periphery and Focus Structure in Japanese

2000). The topic-focus ordering in the left periphery applies to Japanese also, as will be discussed in Section 6. The properties of the left periphery outlined thus far are captured by the RRG layered structure of the clause, as shown in Figure 1. The leftdetached position [LDP] represents a sentence-initial topic, and this topic position is followed by a Pre-core slot [PrCS], a fronted focus position. The PrCS houses a specialized focus, which may be a narrow focus as in the English example, or a contrastive or restrictive focus as in the case of Japanese. A regular focus may appear in situ anywhere in the clause within the potential focus domain.2 SENTENCE LDP

CLAUSE PrCS

CORE RP

NUC

PERIPHERY RP

PRED ADV

RP

Yesterday,

what

V did

John

give

PP to Mary

in the library?

Figure 1: LDP and PrCS as the left periphery (Van Valin 2008)

The discussion for the present study will proceed as follows. In Section 2, I will give an overview of topic and focus NPs that are relevant to the left periphery in Japanese. Section 3 provides a summary of the previous proposals in RRG and lays out the remaining issues to be examined. In Section 4, I will discuss the representation of subordinate focus structure, and in Section 5, I will show how the representation is applied to the observed range of topic and focus NPs in Japanese. Section 6 integrates the focus structure representation into the RRG clause structure representation. Sections 7 and 8 describe how the irregular use of topic and focus-NPs is captured in RRG and how the various focus-shifting effects are predicted. Section 9 incorporates the current proposal into the 2

There is also a post-core slot, for a post-verbal focus position in a verb-final language such as Japanese, and a right detached position, for a post-verbal topic position. See Section 9 and also Shimojo (1995, 2009).

Mitsuaki Shimojo

269

linking algorithm and demonstrates the essential roles played by both the focus structure projection and the discourse representation structure.

2. Wa and Ga: Topic and Focus in Japanese First, an overview is in order for some morphosyntactic properties of Japanese that is relevant to the present study. The canonical word order in Japanese is SOV, though it allows some flexibility such as preposing and postposing (or left and right dislocation) of arguments and adjuncts (Shimojo 1995). In general, post-nominal markings are sensitive to the focus structure, especially the topic marker -wa and the nominative marker -ga. A topic marked argument and adjunct appear sentence-initially (e.g. StopOV, OtopSV). A non-sentence-initial topic marked argument or adjunct represents explicit contrastiveness. An argument that is part of the focus normally remains in situ, but a non-subject in narrow or contrastive focus may be preposed (e.g. Ofoc(S)V). Also, subjects of particular predicate types which denote permanent and generic states of affairs are topicmarked unless they represent narrow focus (and marked with -ga in such a case). On the other hand, subjects of “presentational” verbs, such as aru “to exist (for nonanimate)”, iru “to exist (for animate)”, kuru “to come”, favor the nominative marking, and topic marking for such subjects typically evokes an explicit contrastive sense. The alternation of post-nominal markings for subjects is illustrated in (2)-(4), which are taken from Lambrecht (1994: 223). (2) What happened to your car? (kuruma-wa) kosyoosita car-TOP broke.down “(The car) broke down.” (3) What happened? kuruma-ga kosyoosita car-NOM broke.down “The car broke down.”

[predicate-focus]

[sentence-focus]

(4) I heard your motorcycle broke down. KURUMA-GA kosyoosita [argument-focus] car-NOM broke.down “THE CAR broke down.”

270

The Left Periphery and Focus Structure in Japanese

The answer in (2) represents predicate-focus, in reply to the question “what happened to your car?”. Although kuruma “the car” in the answer does not need to be repeated, wa is clearly the choice over ga if it is overtly present. If the nominative argument is either part of the focus or the narrow focus, as shown in (3) and (4) respectively, it is marked with ga. Also, it has long been observed (Kuno 1973) that the topic marking is also used to represent contrastiveness, as shown in (5). (5) I heard Ken and Sayaka came yesterday. ken-wa kita Ken-TOP came “Ken came (but Sayaka didn’t).” The use of wa in (5) puts Ken and Sayaka in contrast with respect to the proposition in question; it implies Sayaka did not come. If the topic marker for Ken is replaced with the nominative marker, the sense of contrast is lost and the sentence becomes awkward in the given context. For the purpose of the present study, I use the term topic to refer to NP’s marked with wa because it is how the language marks the fronted non-focus part of the sentence. It should be noted, however, that in discourse, zero anaphora is the default choice for a continuing topic, and an overt wa-marked argument is typically used for contrastive topics (Shimojo 2005). For the definition of focus, I follow Lambrecht’s (1994) characterization that focus represents non-presupposed information. The characterization of topic and focus above serves as the basis for the present analysis, which I will extend later to capture some irregular use of wa and ga, where wa is used when ga is expected, and vise versa.

3. Previous Claims in RRG and Remaining Issues 3.1. Hasegawa (1992, 1996) The classic dichotomy of topic and focus with respect to wa and ga is reflected in Hasegawa’s analysis within RRG, as summarized in (6). A topic NP-wa is placed in a LDP and a narrow-focus NP-ga in a PrCS. A NP-ga which is part of a sentence-focus is within the CORE, under an ARG node (which is labeled referential phrase [RP] in the current RRG layered structure of the clause (Van Valin 2008)). Hasegawa (1996: 41) places a contrastive NP-wa in PrCS because “[c]ontrastive NP-wa’s and

Mitsuaki Shimojo

271

narrow focus NP-ga’s are cognitively similar: both convey the idea ‘THIS entity, but not something else.’” (6) Topic NP-wa: LDP Narrow-focus NP-ga, Contrastive NP-wa: PrCS Part of sentence-focus NP-ga: within CORE Besides the obvious association of a topic with LDP and a narrowfocus with PrCS, the merit of Hasegawa’s analysis is the recognition of the property common to contrastive wa and narrow-focus ga NP’s, as they both exhibit a “focus” property. At the same time, however, this proposal fails to capture properties that separate the two types of “focus” NPs in PrCS and properties shared by topic NP-wa and contrastive NP-wa. In other words, why are some NP’s in PrCS marked with wa and others marked with ga, and why are some NP-wa’s in LDP and other NP-wa’s in PrCS, despite the same post-nominal marking?

3.2. Shimojo (1995) A quantitative study of spoken Japanese in TV talk shows in Shimojo (1995) offered a different view with respect to topic and contrastive NPwa’s. Frequency counts of subject NPs in discourse pointed to a common ground that there was an overall tendency to associate NP-wa’s, whether contrastive or non-contrastive, with recent reference in the preceding discourse, hence, to exclude them from the focus domain of the sentence. The merit of the discourse-based study was the finding that contrastive wa, as well as topic wa, exhibited a tendency to represent previously given referents, which suggests that contrastiveness is in principle independent of the givenness of referents. As expected, NP-ga’s in the spoken Japanese data dominantly represented brand-new referents or inactive referents to be re-introduced after some absence. Given these findings, the following structural representations in RRG were proposed. (7) Topic NP-wa, Contrastive NP-wa: LDP Narrow-focus NP-ga: PrCS Part of sentence-focus NP-ga: within CORE

272

The Left Periphery and Focus Structure in Japanese

3.3. Remaining Issues Although each of the preceding studies outlined above points out critical properties of the topic and focus NPs, their mutually exclusive claims have fallen short of a satisfactory account. While Hasegawa (1992, 1996) fails to show how contrastive NP-wa’s are differentiated from narrow-focus NP-ga’s, the discourse generalizations presented in Shimojo (1995) do not tease apart the two types of NP-wa’s and fails to show the same focus property associated with contrastive NP-wa’s and narrow-focus NP-ga’s. Thus, these shortcomings collectively point to the need of a framework to properly capture the properties of topic and focus NPs, consisting of a common property for NP-wa’s (both contrastive and non-contrastive), a common property for NP-ga’s (both part of broad-focus and narrowfocus), and a common property for contrastive NP-wa and narrow-focus NP-ga. Furthermore, Shimojo’s (2005) recent extensive discourse-based study of argument forms in Japanese has revealed that, in informal Japanese conversations, the use of wa is ambivalent in terms of givenness of referents represented by the wa-marked NPs. Both subject NP-wa’s and direct object NP-wa’s represented given and new referents almost equally; hence, it has been suggested that the index of newness/givenness of referents per se does not predict the use of NP-wa.3 On the other hand, the NP-wa’s in the data were found to be predominantly contrastive (91% of the total 336 cases); therefore, it has been suggested that contrastiveness is the reliable index to predict the use of wa. Yet, the discourse-based findings do not dissociate NP-wa from the givenness of referents (i.e. the non-focus of a sentence) and they vividly point to the fact that NP-wa’s do represent either given or new information and a theory also needs to capture the “irregular” use of wa, where the “topic” marking is associated with new information.

3

Of the total 252 subject NP-wa’s, 145 (58%) were used for referents in the referential distance [RD] range of 1 through 10 clauses, and 107 (42%) were used in the RD range of 11 clauses and over, including “no previous reference”. Of the total 25 direct object NP-wa’s, 13 (52%) were used for RDs of 1 through 10 clauses and 12 (48%) were used for RDs of 10 clauses and over, including “no previous reference”.

Mitsuaki Shimojo

273

4. Subordinate Focus Structure In order to overcome the previous shortcomings discussed above, I adopt the representation of subordinate f(ocus)-structure (Erteschik-Shir 1997, 2007) and incorporate it into the RRG notions of clause and focus structure. A subordinate f-structure is a complex f-structure with a f-structure embedded in a matrix f-structure which is a main topic or focus. Using this representation, Erteschik-Shir illustrates how specific indefinite NPs are licensed as topic with the following examples.4 (8) A person I know is famous. (9) A personfoc [Itop know_ ] (10) Itop know [a person]foc In essence, it is the specificity of specific indefinites that allows it to be a topic. The subject of the sentence in (8), “a person I know” has a subordinate f-structure of its own, as shown in (9). In the f-structure represented by (10), the indefinite “a person” is the focus and “I” in the relative clause is the topic, hence there is specificity on “a person” (the 1st person subject is permanently available as topic). Thus, the f-structure of (9) is equivalent with the f-structure of the simple sentence given in (10). Furthermore, a subordinate f-structure can provide an explanation for the fact that indefinite NPs can be topics if contrastiveness is involved, as shown in (11). (11) a. #A dog is intelligent. b. A DOG is intelligent, a CAT is not. Contrastive elements function as both topics and foci because a contextually available set serves as topic and the element selected from the set serves as a focus. In the case of (11b), “a dog” has a subordinate fstructure that is included in the full f-structure given in (12).

4

Erteschik-Shir points out that, unlike non-specific indefinites (e.g. “a person”), specific indefinites can be topicalized in languages such as Danish, and they also can serve as the subject of individual-level predicates (e.g. “A student I know is intelligent”, “#A student is intelligent”). See Erteschik-Shir (2007: 19-22) for the discussion of other topic tests.

274

The Left Periphery and Focus Structure in Japanese

(12) [{dogfoc, cat}top]top [is intelligent]foc The fully spelled out sentence of (12) is provided in (13). However, the subordinate topic “cat” can be dropped, as in the first clause of (11b), or pronounced, as in (13). (13) A dog, not a cat, is intelligent. Erteschik-Shir (2007: 50) states that the bi-level representation of focus structure is particularly useful to tease out the complex focus structure associated with contrastive elements, which function as both topics and foci, because it accounts for the various contrastive readings that are derived by different ways of imposing one f-structure on the other.5 In what follows, I will discuss subordinate f-structure of NP-wa and NP-ga in Japanese and provide their classification in terms of contrastive topic and focus types.

5. NP-wa and NP-ga: Five Topic/Focus Types Using the subordinate f-structure representation, I propose the f-structure representations for the five topic and focus types listed in (14). Each type is discussed with examples below.6 (14) a. b. c. d. e.

5

Contrastive NP-wa: [{xfoc, y}top]-WAtop [predicate]foc Contrastive (narrow) focus NP-ga: [{xfoc, y}top]-GAfoc [predicate] Restrictive focus NP-ga: [{xfoc, …}top]-GAfoc [predicate] Non-contrastive (broad) focus NP-ga: x-GAfoc [predicate]foc Topic NP-wa: [{xfoc, …}top]-WAtop [predicate]foc

This hierarchical representation of focus structure has advantage over previous approaches that derive “contrastive focus”, such as Choi (1999), who combines cross-classifying features [+new] (i.e. focus) and [+prom] (i.e. contrastive). The present framework requires only the notions of “topic” and “focus” to capture contrastiveness and is capable to represent the f-structure of a complex NP as given in (8). 6 The f-structure given in (14e) represents what would be called “restrictive topic”, though this particular f-structure type is not discussed in Erteschik-Shir (2007).

Mitsuaki Shimojo

275

5.1. Contrastive NP-wa: [{xfoc, y}top]-WAtop [predicate]foc The contrastive NP-wa, or contrastive topic, requires a contextually provided set such as {x, y} and {x, y, z}. The set as a whole is a topic and, from the set, a particular element (e.g. x) is selected; hence, the selected element is a focus. The NP-wa is the topic of the matrix-level f-structure and the predicate is the focus. Overall, this topic type represents predicate-focus, despite the contrastiveness involved. In (15b), for example, the contrastive subject represents a complex f-structure [{anifoc, imooto}top]-WAtop, where “brother” (focus) is selected from the given set (topic), and the NP-wa as a whole represents a topic of the matrix fstructure. (15) a watashi-wa ani-to imooto-ga imasu I-TOP older.brother-and younger.sister-NOM exist “I have an older brother and a younger sister” b ani-wa tookyoo-ni sundeimasu older.brother-TOP Tokyo-in living “(My) older brother lives in Tokyo.” “Sister”, which is in the subordinate f-structure, can be overtly present as shown in (16); however, the NP-wa containing just the subordinate focus element is sufficient to express the contrastive topic in the given context. (16) imooto janakute ani-wa tookyoo-ni sundeimasu younger.sister not older.brother-TOP Tokyo-in living “(My) older brother, not younger sister, lives in Tokyo.”

5.2. Contrastive (Narrow) Focus NP-ga: [{xfoc, y}top]-GAfoc [predicate] Contrastive focus NP-ga and contrastive topic NP-wa share the same subordinate f-structure; in both cases, the element (focus) is selected from a contextually provided set (topic). However, for contrastive focus NP-ga, the NP-ga is the focus of the matrix f-structure. An example is provided in (17). (17) A1 gokyoodai-wa? siblings-TOP “(How about your) siblings?”

276

The Left Periphery and Focus Structure in Japanese

B1 ani-to imooto-ga imasu older.brother-and younger.sister-NOM exist “(I) have an older brother and a younger sister” A2 sorede donata-ga tookyoo-ni irassharu no? and who-NOM Tokyo-in exist Q “And who is in Tokyo?” B2 ani-ga tookyoo-ni imasu older.brother-NOM Tokyo-in exist “(My) older brother is in Tokyo.” In (B2), “(my) brother” singles out a member of the previously given set, hence, the contrastiveness, x, not y. In the matrix f-structure, the predicate is not focused and the NP-ga is the narrow focus, hence the complex f-structure [{anifoc, imooto}top]-GAfoc.

5.3. Restrictive Focus NP-ga: [{xfoc, …}top]-GAfoc [predicate] Restrictive focus is distinguished from contrastive focus because the elements of the set are not clearly defined, only restricted (Erteschik-Shir 2007). In Japanese, NP-ga is the unmarked form for this type of focus. The subordinate f-structure is identical to that of topic NP-wa; for both, elements are selected from a restricted but unspecified set (see 5.5 below). An example is given in (18). (18) A tomodachi-de dareka ryuugakushita? friend-among anyone studied.abroad “Did any of your friends study abroad?” B tanaka-ga ryuugakushita Tanaka-NOM studied.abroad “Tanaka studied abroad.” The element is selected from a set that is contextually restricted but not clearly defined, such as one’s friends; therefore, this is distinguished from contrastive focus (Erteschik-Shir 2007). In (18B), “Tanaka” is selected from a contextually restricted set of B’s friends, hence [{Tanakafoc, …}top]-GAfoc. It should be noted that, because the set is not clearly defined in this focus type, the complement of the selected set is not eliminated; therefore, there may be other friends who studied abroad, besides Tanaka. For example, the sentence in (18B) can be followed by a separate sentence “And Suzuki studied abroad also”.

Mitsuaki Shimojo

277

5.4. Non-Contrastive (Broad) Focus NP-ga: x-GAfoc [predicate]foc With broad-focus NP-ga, a sentence typically represents sentence-focus. In this focus type there is no contrastiveness involved; therefore, there is no bi-level f-structure for this type, as exemplified by the subject NP in (19), which represents tomodachi-GAfoc. (19) senshuu tomodachi-ga nihon-kara last.week friend-NOM Japan-from “A friend came from Japan last week.”

kita came

5.5. Topic NP-wa: [{xfoc, …}top]-WAtop [predicate]foc Lastly, topic NP-wa deserves a more detailed discussion. Contrary to the dichotomic distinction of two types of wa’s, topic and contrastive, I claim a common underlying property of wa, which is rooted in the selection of (a) particular element(s) of a given set. What separates the “topic” NP-wa from the “contrastive” one is the unspecified-set reading for the former. The elements of a set are not overtly specified for topic NP-wa’s; thus, the subordinate f-structure for the NP is identical with that for restrictive focus NP-ga. Both involve selection of particular elements from a restrictive, but unspecified, set. The integration of the two types of NP-wa’s is motivated by discourse-based observations (Suzuki 1995; Shimojo 2005) that wa is predominantly used for either an overt contrastive relationship or a shift topic, which involves a selection of a new topic to be presented in contrast with the previous one. In Japanese, zero anaphora is preferred for continuation of the same topic without contrastiveness involved. Furthermore, it has been observed in Shimojo (2005) that, when a speaker refers to a continuing topic overtly (rather than through zero anaphor) in the conversation, a bare NP with no post-nominal marking is the preferred form, unless there is contrastiveness involved in the reference. Overall, these discourse-based observations point to the particular property of NPwa’s; they represent a specialized type of topic, which is not captured by the notion of givenness or non-focus. An example of topic NP-wa is given in (20). The subject NP-wa represents the subordinate focus [{watashi-no anifoc, …}top]-WAtop, which is embedded within the topic. In other words, the topic wa is inherently

278

The Left Periphery and Focus Structure in Japanese

contrastive, with an implicit contrast with the other elements of a contextually unspecified set.7 (20) ima watashi-no ani-wa tookyoo-ni sundeimasu now I-LK older.brother-TOP Tokyo-in living “(My) older brother lives in Tokyo now.” As in the case of restrictive focus, the set is not clearly defined in this topic type; therefore, the complement of the selected set is not eliminated. For example, the sentence in (20) can be followed by a sentence “And my younger sister lives in Tokyo also”. On the other hand, this is not possible in the case of contrastive NP-wa, where the set is clearly defined. For example, it is very awkward to continue after (15B) with “And my younger sister lives in Tokyo also”.

6. NP-wa and NP-ga in RRG Clause Structure The focus structures proposed for the range of NP-wa and NP-ga thus far lead to the RRG representations shown in Figure 2. By default, NP-wa is outside the potential focus domain [PFD] and NP-ga is (part of) the actual focus of the sentence (NP-ga in PrCS is always the narrow-focus of the sentence). The common property of ga in all positions is the focus in the matrix f-structure. It is the narrow focus if there is a subordinate focus, i.e. if contrastiveness arises. On the other hand, the inherent property of wa is the topic in the matrix f-structure (hence, outside the PFD) and the required set representation, whether the contrastiveness is explicit or implicit. The association of contrastive NPwa’s with the LDP, which is outside the PFD, can be demonstrated by a truncation test as follows.

7

Note that in Lambrecht’s example given in (3) earlier, the subject NP-wa is given in parentheses, by which I assume Lambrecht implies the degraded acceptability for the overt NP-wa in the particular context. In this example, zero anaphora for the subject is clearly preferred since there is no contrastiveness involved.

Mitsuaki Shimojo

279

SENTENCE LDP

CLAUSE PrCS

CORE RP

NP-wa [{xfoc, y}top]-WAtop [{xfoc, …}top]-WAtop

NP-ga [{xfoc, y}top]-GAfoc [{xfoc, …}top]-GAfoc

x-ga x-GAfoc

NUC PRED

PFD Figure 2: Default assignment of NP-wa/ga in LSC

For a contrastive topic, the selection of an element per se (i.e. the contrastive topic NP alone) does not constitute the most informative part of the sentence with respect to the purpose of the utterance; therefore, the sentence results in unacceptability (due to incompleteness) in the given context if the NP alone remains by the truncation. This is shown in (21b), in contrast with the full sentence given earlier in (15b). (21) a watashi-wa ani-to imooto-ga imasu I-TOP older.brother-and younger.sister-NOM exist “I have an older brother and a younger sister” b # ani-wa older.brother-TOP “(My) older brother.” For a contrastive focus NP-ga, on the other hand, the same truncation is acceptable which shows that the remaining NP-ga is the most informative part of the sentence, as exemplified by (22B2), which is the truncated version of (17B2).8 8

Of course, the truncated sentences are incomplete without the predicate; therefore, they are informal (or sloppy) sounding. But these acceptable fragments still make sense in the given context, unlike the unacceptable contrastive topic ones.

280

The Left Periphery and Focus Structure in Japanese

(22) A1 gokyoodai-wa? siblings-TOP “(How about your) siblings?” B1 ani-to imooto-ga imasu older.brother-and younger.sister-NOM exist “(I) have an older brother and a younger sister” A2 sorede donata-ga tookyoo-ni irassharu no? and who-NOM Tokyo-in exist Q “And who is in Tokyo?” B2 ani-ga older.brother-NOM “(My) older brother (is in Tokyo).” This is also the case with a restrictive-focus NP-ga. The NP can remain by truncation, as shown in (23B), which is the truncated version of (18B). (23) A tomodachi-de dareka ryuugakushita? friend-among anyone studied.abroad “Did anyone of your friends study abroad?” B tanaka-ga Tanaka-NOM “Tanaka (studied abroad).” Thus, the observation above can be summarized as follows. (24) Contrastive topic: [{xfoc, y}top]-WAtop {xfoc, y}top alone is not informative for the purpose of the utterance, hence WAtop Contrastive/restrictive focus: [{xfoc, y}top]-GAfoc, [{xfoc,…}top]-GAfoc {xfoc, y}top alone is informative for the purpose of the utterance, hence GAfoc I should note that the overall structural representations which I presented in Figure 2, NP-wa in a LDP and NP-ga in a PrCS or the CORE, are not new since these structural assignments were proposed in Shimojo (1995) as discussed earlier. However, the current proposal captures the details that were missed in the previous work. First, the proposed bi-level f-structure captures the subtypes of NP-wa and NP-ga combined with the

Mitsuaki Shimojo

281

clause structure, as shown in Figure 2. More importantly, the proposed representations account for the topic-focus ambivalence of contrastive NPwa’s. While their topic property is captured by the matrix f-structure (top), their focus property is captured by the subordinate f-structure (foc). Also, the shared topic property is structurally reflected by being hosted by a LDP. Furthermore, the default structural and functional representations proposed in Figure 2 provide the means to account for the extended use of NP-wa and NP-ga, which achieve particular pragmatic effects. The marked use of these NPs include NP-wa used in a PrCS and a RP inside the CORE, and NP-ga used as a non-focus RP and a subject RP of certain types of predicates which normally take NP-wa for subject. These are discussed in the following sections.

7. Marked Use of NP-wa 7.1. Wa in PrCS Perhaps the most peculiar, but possible, use of wa is for a narrow-focus NP in a PrCS, as exemplified by (25B’). In the answer to the question, the NP-ga is the default form to represent the narrow-focus, as shown in (B); however, NP-wa in (B’) achieves a particular pragmatic effect as follows. (25) (At a party) A dare-ga baakuree-no gakusee? who-NOM Berkeley-LK student “Who are Berkeley students?” B karera-ga baakuree-no gakusee desu they-NOM Berkeley-LK student COP “They are Berkeley students.” B’ karera-wa baakuree-no gakusee desu they-TOP Berkeley-LK student COP “They are Berkeley students (I don’t know about others).” This peculiar use of wa has been described as anti-exhaustive listing by Kuroda (2005). According to Kuroda, if a NP-ga is used as expected for the contrastive focus, there is an “only x” (exhaustive listing) reading, which conforms to the Gricean maxim of quantity, i.e. “make your contribution as informative as is required (for the current purposes of the exchange)” (Grice 1975: 45). On the other hand, if a NP-wa is used, the speaker is not giving a description of the situation (a thetic judgment in Kuroda’s characterization) as expected for a NP-ga sentence. With the use

282

The Left Periphery and Focus Structure in Japanese

of wa (which represents a categorical judgment), the speaker has committed himself only to the proposition that “they are Berkeley students” only with respect to “they”. This speech act implicates that the speaker leaves the possibility of others “being Berkeley students” open, hence, antiexhaustive listing implicature. SENTENCE CLAUSE PrCS

CORE NUC

NP-wa [{xfoc, y}top]-WAtop [{xfoc, …}top]-WAtop

PRED

PFD Figure 3: NP-wa in PrCS Defocusing effect on PrCS due to WAtop (shift to predicate-focus)

In RRG with the f-structure representations proposed in the present study, the observed anti-exhaustive listing effect is explained as follows. The PrCS is inherently associated with a contrastive or restrictive subordinate f-structure due to the default contrastive or restrictive focus NP-ga; thus, the subordinate f-structure of NP-wa does not bring out the “contrastive” or “restrictive” effect for the PrCS (because it is already there). However, since a PrCS is a narrow focus element, the matrix fstructure of NP-wa achieves defocusing, shifting the f-structure of the sentence from the narrow-focus to the predicate-focus. In other words, in the speaker’s choice of NP-wa for the PrCS, the sentence is presented as if it has predicate-focus, i.e. predication for the given entity, despite the contextually determined actual focus on the PrCS. Therefore, the defocusing effect brings out an “at least” reading, as in (25B’) “As for them, they are Berkeley students” without any reference to entities which are not selected for the predication. Hence, in RRG, the anti-exhaustive listing reading as observed by Kuroda is captured by the interplay of the

Mitsuaki Shimojo

283

contextually determined focus structure of the sentence and the lexical property of wa, i.e. its inherent f-structure. A NP-wa in the PrCS is shown in Figure 3.

7.2. Wa for a Focus RP The use of NP-wa for a focus RP achieves an effect similar to NP-wa in a PrCS discussed above; however, the effect involves contrastiveness as well as a focus shift in this case. An example is given in (26B’). (26) A baffaroo doo? Buffalo how “How is Buffalo?” B yuki-ga huru yo snow-TOP fall FP “It snows.” B’ yuki-wa huru yo snow-TOP fall FP “It snows (at least/but it’s not that cold, etc.)” In response to the question in (26A), the answer represents sentencefocus, hence the NP-ga as default, as shown in (B). If a NP-wa is used, it brings out a marked contrastive and “at least” reading, as indicated by the translation for (B’). In this case, there are two separate effects imposed by the use of wa. First, we expect a similar defocusing effect as in the case of NP-wa in PrCS, but to a lesser degree. It is a shift from sentence-focus to predicate-focus, due to the switch in the matrix f-structure from GAfoc to WAtop. Because the predicate is in focus to begin with, the focus shift is not expected to be a major shift as in the use of wa in PrCS, i.e. a switch from narrow-focus to predicate-focus. Yet, due to the imposed predicatefocus (i.e. predication for a selected entity), the focus shift brings out the “at least” reading, such as “it snows at least”. What is striking in this case is the marked contrastive reading imposed by the use of wa, because a broad-focus NP-ga, the default form which would be used otherwise, is not associated with the subordinate f-structure. The RRG representation is given in Figure 4.

284

The Left Periphery and Focus Structure in Japanese

SENTENCE CLAUSE CORE RP

NUC

NP-wa [{xfoc, y}top]-WAtop [{xfoc, …}top]-WAtop

PRED

PFD Figure 4: NP-wa for focus RP Defocusing effect on RP due to WAtop (shift to predicate-focus) AND Contrastive effect on RP due to {xfoc, y/…}top

The same argument applies to the use of wa for a non-subject RP. Consider the examples given in (27)-(29). The sentence in (27) exemplifies the default post-nominal marking for the nominative and accusative arguments. In the sentence in (28), on the other hand, the object argument is marked with wa. It is well known that wa marking for a nonsubject brings out a marked contrastive reading. I claim that the imposed contrastiveness is due to the subordinate f-structure associated with the NP-wa. Also, there is an “at least” reading for the sentence, which is again explained by the defocusing on the NP-wa, and thus, the focus shift to the other elements of the sentence. These dramatic effects are partly suppressed if the NP-wa is preposed in the LDP, as shown in (29). The LDP is a non-focus position; therefore, there is no defocusing effect caused by the NP-wa in LDP. However, the contrastive reading is still there, which is the default for a NP-wa, whether explicit or implicit, depending on the context. (27) taroo-ga ringo-o Taro-NOM apple-ACC “Taro ate an apple.”

tabeta ate

Mitsuaki Shimojo

285

(28) taroo-ga [ringo-wa]RP tabeta Taro-NOM apple-TOP ate “Taro ate an apple (at least, but didn’t eat an orange, etc.)” tabeta (29) [sono ringo-wa]LDP taroo-ga the apple-TOP Taro-NOM ate “The apple, Taro ate.” I should note that so-called “presentational” sentences are not ideal hosts for NP-wa’s, even for their marked use. In (30), the predicate iru “there is/exist” requires the hearer’s attention to the entity that is introduced. Similarly, in (31), the indefinite subject NP “man” solicits a context in which it is the first-time introduction of the referent. In both cases, a focus shift away from the newly introduced referent is not desirable. Also, the preceding context appropriate for these sentences would not contain a set of entities out of which “bear” or “man” is singled out; therefore, contrastiveness (whether explicit or restrictive) associated with wa is not desired. (30) mite, mukooni kuma-ga/*wa iru! look over.there bear-NOM/*TOP exist “Look, there is a bear over there!” (31) kinoo otoko-ga/*wa taihosareta yesterday man-NOM/*TOP was.arrested “Yesterday, a man was arrested.”

8. Marked Use of NP-ga 8.1. Ga for a Non-Focus RP Just like a NP-wa may be used when it is not expected to be used per the given focus structure of the sentence, the marked use of a NP-ga is possible, but drawing focus on the NP. One obvious possibility is the use of ga for a non-focus RP, which is outside the actual focus domain, though it is within the CORE. The RRG representation is given in Figure 5.

286

The Left Periphery and Focus Structure in Japanese

SENTENCE CLAUSE CORE RP

NUC PRED

x-GAfoc

PFD Figure 5: Ga for a non-focus RP Focusing effect on the RP due to GAfoc

The marked use of NP-ga’s is best suited to the re-introduction of a referent that has already been given in the preceding discourse. Because the referent is previously given, it would represent non-focus of the sentence, hence, NP-wa, a NP without a post-nominal marker, or even zero anaphora if the overt reference is not needed. However, the reintroduction is more effectively achieved if the NP is presented as focused, hence, the ga marking. In fact, it has been observed in narrative discourse that NP-ga’s are used for previously introduced referents in order to achieve particular discourse effects. Maynard (1987) argues that such use of ga is part of the narrator’s staging strategy, in which a NP-wa is used for “thematized” main story characters and the avoidance of wa-marking (hence, the use of NP-ga if subject) is the strategy of non-thematizing. In essence, nonthematized referents are presented as if new, as they become the focus of attention. Consider the part of a story given in (32), which is a slightly modified example taken from Maynard (1987: 60). (32) a mukasi, mukasi, arutokoro-ni oziisan to obaasan-ga once.upon.a.time one.place-in old.man and old.woman-NOM orimasita existed “Once upon a time, there were an old man and an old woman in some place.”

Mitsuaki Shimojo

287

b sizukana yama-de hutari-wa siawaseni kurasiteorimasita quiet mountain-in two-TOP happily were.living “In a peaceful mountain, the two were living happily.” c aruhi oziisan-ga yama-e sibakari-ni one.day old.man-NOM mountain-to firewood.collecting-to ikimasita went “One day, the old man went to the mountain to collect firewoods.” The first two sentences of the discourse represent the usual referential progression from ga to wa; a NP-ga is used for the initial introduction of a referent and a NP-wa is used for a subsequent mention. In (32.c), “old man” is re-introduced afresh with the ga-marking, which goes well with the temporal transition in the story between (b) and (c).9

8.2. NP-ga for Particular Types of Predicates (Constant/Habitual state of affairs) One final point to be discussed is the use of NP-ga with a particular range of predicates. It has been pointed out by Kuno (1973) that the neutral description reading (i.e. the sentence-focus reading) is not allowed for a NP-ga used with a predicate that denotes a constant or habitual state of affairs. If a NP-ga is used with such a predicate, it imposes a narrow-focus reading (i.e. exhaustive-listing in Kuno’s term) on the NP, as exemplified by (33). In other words, a NP-wa is the default form for the subject of this type of predicate, as shown in (34), unless a narrow-focus is intended. (33) [taroo-ga]PrCS gakusee da Taro-NOM student COP “It is Taro that is a student.”

9

Watanabe (1990) observes a similar use of NP-ga in narrative discourse, in which story characters are re-introduced with the ga marking to signal a perspective shift, such as a switch of viewpoint from the narrator to a particular character in the story. Furthermore, Yamaguchi (2007) points out that selfcontained facts are often presented without topic NP’s, as observed in newspaper articles, such that every sentence is presented afresh, being disconnected from each other. Such discourse is likely to contain more NP-ga’s even for the same referents.

288

The Left Periphery and Focus Structure in Japanese

(34) [taroo-wa]LDP gakusee Taro-TOP student “Taro is a student.”

da COP

The contrast observed in (33) and (34) is described in RRG terms as follows. The type of verb discussed here is inherently associated with an argument NP which is the topic in the matrix f-structure (i.e. [{xfoc, y}top]top or [{xfoc, …}top]top); thus, the NP is assigned to a LDP by default. However, if a NP-ga replaces the NP-wa, the minimal change possible involves a switch from [{xfoc, y/...}top]top to [{xfoc, y/...}top]foc, ([{xfoc, y/...}top]foc is more equivalent with the f-structure of wa than x-GAfoc). I.e., the subordinate f-structure (i.e. {xfoc, y}topor {xfoc, …}top) remains since it is part of the property of contrastive/restrictive NP-ga also and the subordinate f-structure does not have to be eliminated by the wa-ga switch. The contrastive or restrictive NP-ga is assigned to a PrCS for the narrow-focus reading.

9. Linking Before concluding the present study, I incorporate the current proposal into the linking algorithm which I previously proposed to capture the discourse-based alternation of argument forms in Japanese. As I discussed previously (Shimojo 1995, 2009), Japanese uses the full range of COREexternal positions: LDP, PrCS, post core slot [PoCS], and right detached position [RDP]. On the basis of the findings in spoken Japanese (Shimojo 2005), the following form-function relationship has been proposed. Table 1: Form-function relationship: nominative and accusative arguments Functional Properties Givenness Identifiable Unidentifiable Contrastiveness Contrastive Absolutive Saliency Default Defocusing

Grammatical Means Morphology Zero anaphor -ga (nominative), -o (accusative) -wa Zero particle Syntax Pre-verbal argument Post-verbal argument

Mitsuaki Shimojo

289

What is relevant to the present study is the CORE-external positions at the left periphery, i.e. LDP and PrCS, and these are the default periphery positions, as indicated in Table 1 (see under “saliency”). The COREexternal positions at the right periphery, PoCS and RDP, are associated with arguments by which the referents are defocused in cataphoric discourse; therefore, the right periphery in Japanese is associated with a marked discourse function. In Shimojo (2009: 132-137), I proposed the linking algorithm to capture argument assignment to PoCS and RDP. Given the bi-level f-structure representations proposed in the present study, the linking algorithm from semantics to syntax is revised in (35). (35) Linking algorithm: semantics Æ syntax 1. Construct the semantic representation of the sentence, based on the logical structure of the predicator. 2. Determine the actor and undergoer assignments, following the actorundergoer hierarchy. 3. Determine the morphosyntactic coding of the arguments. a. Select the privileged syntactic argument, based on the privileged syntactic argument selection hierarchy and principles. b. Assign the arguments the appropriate case markers and/or postpositions. (i) If an argument has the f-structure [{xfoc, y}top]top or [{xfoc, …}top]top, or if it is a matrix focus but needs to be defocused, assign wa to the argument(s). If the referent(s) requires absolutive specification, assign no post-nominal marking. (ii) If an argument is a matrix focus, or if it is a matrix topic but needs to be focused, assign appropriate case markers, based on the case assignment rules for accusative constructions. (iii) If neither (i) nor (ii) above applies, use no morphosyntactic instantiation for the argument (i.e. zero anaphora). 4. Select the syntactic template(s) for the sentence, following the syntactic template selection principle (and language-specific qualifications). a. If an argument has no syntactic instantiation, use appropriate truncated syntactic templates. b. Use the LDP for a wa-marked argument, but place it in a RP for an argument of a sentence-focus. Use the PrCS for a ga or wa-marked narrow-focus argument. c. If the referent(s) of the argument(s) requires defocusing, use PoCS (default) or RDP (for a right detached topic). 5. Assign arguments to positions in the syntactic representation of the sentence. If there is no syntactic position to assign the argument(s) to,

290

The Left Periphery and Focus Structure in Japanese

link them directly with the corresponding referents in the discourse representation structure. The major points of revision are how the arguments are assigned the wa or ga-marking (35.3.b.i, ii) and how the left and right periphery positions are all incorporated (35.4.b, c). For the former, the f-structure representations for the different NP forms are specified in the algorithm. The linking also captures the marked use of wa and ga for the specified (de)focusing effects. The interplay of the left and right periphery is captured such that the left periphery is the default topic and focus position and the arguments are linked with the right periphery only for the specialized function, defocusing of referents. Likewise, the linking algorithm from syntax to semantics is revised in (36), which incorporates the f-structure specifications. (36) Linking algorithm: syntax Æ semantics 1. Obtain an appropriate clause structure upon parsing the sentence. 2. Determine the macrorole(s) and other core argument(s) in the clause. 3. Retrieve from the lexicon the logical structure of the predicate in the nucleus of the clause. If the clause structure contains no predicate, retrieve it from the presupposition discourse representation structure.10 4. Link the arguments determined in step 2 with the arguments determined in step 3 until all core arguments are linked. If an argument is marked with wa, assign either [{xfoc, y}top]top or [{xfoc, …}top]top, as determined by the discourse representation structure, to the argument in the focus structure projection, and establish a contrastive link in the discourse representation structure with a proper referent. If the argument is zeromarked, cancel existing contrastive links if any. If there is a ga-marked argument in the PrCS, assign either [{xfoc, y}top]foc or [{xfoc, …}top]foc, as determined by the discourse representation structure, to the argument in the focus structure projection. If there is an unlinked argument position(s) in the semantic representation, retrieve the corresponding referent(s) directly from the discourse representation structure (for zero anaphora). 5. If there is an element in the PoCS or RDP, assign it in the remaining unlinked argument position in the semantic representation of the clause. Defocus the corresponding referents in the discourse representation structures and if the argument is wa or zero marked, follow the procedure in step 4. 10

This is required for zero anaphora for verbs (Shimojo 2008).

Mitsuaki Shimojo

291

6. If the f-structure specification for an argument retrieved by the wa/gamarking does not match the actual focus structure projection determined by the discourse representation structure, the former overrides. Adjust the current discourse representation accordingly. The revised linking algorithm from syntax to semantics includes the retrieval of the f-structure specifications for the pragmatically specialized arguments, i.e. a wa-marked argument and a ga-marked narrow-focus argument in the PrCS (36.4). Thus, the linking is done in connection with the focus structure projection as well as the discourse representation structure. Furthermore, the linking accommodates the marked use of NPwa/ga so that the actual focus structure projection is adjusted accordingly to reflect the anticipated pragmatic effects (36.6).11

10. Conclusion I hope to have demonstrated that bi-level f-structure representations provide a useful means to lay out the functional properties associated with the observed range of topic and focus types in Japanese and shed light on the remaining issues in the previous proposals. In particular, the framework successfully pinpoints the functional ambivalence associated with contrastive topic and focus. On one hand, they are both topic as they need to be linked with the preceding context, but on the other hand, they are both focus due to the contrastive property. It is noteworthy that what is assigned to the left periphery is a topic-like entity, whether contrastive/restrictive topic or contrastive/restrictive focus, and this conforms to the cross-linguistic observation that the left periphery is essentially a topic position (Erteschik-Shir 2007). Furthermore, the default f-structure properties of the different topic and focus types serves as the basis to account for their extended use, which leads to focus shifting in the focus structure projection. These irregular uses of topic and focus NPs may be a reflection of the speaker’s “manipulation” of focus structure for particular pragmatic effects, yet, it needs to be captured in the grammar.

11

Although the exact mechanism for such adjustment in the discourse representation structure will need to be examined in a future study, File Change Semantics (Heim 1982; Vallduví 1990; Portner and Yabushita 1998) may serve as a possible framework. Focus shifts proposed in this study are essentially manipulations in the way information is “filed”. See Lee and Shimojo (2010) for an application of the framework to topic marking in Japanese and Korean.

292

The Left Periphery and Focus Structure in Japanese

List of Abbreviations COP FP LK

copula sentence final particle linker

Q TOP

question marker topic

References Choi, Hye-Won. 1999. Optimizing Structure in Context: Scrambling and Information Structure. Stanford, CA: CSLI. Erteschik-Shir, Nomi. 1997. The Dynamics of Focus Structure. Cambridge: Cambridge University Press. —. 2007. Information Structure: The Syntax-Discourse Interface. Oxford: Oxford University Press. Grice, H. Paul. 1975. Logic and conversation. In Peter Cole and Jerry L. Morgan, eds., Speech Acts: Syntax and Semantics 3, 41-58. New York: Academic Press. Hasegawa, Yoko. 1992. Syntax, Semantics, and Pragmatics of TE-Linkage in Japanese. Unpublished Ph.D. Dissertation, University of California, Berkeley. —. 1996. A Study of Clause Linkage: The Connective TE in Japanese. Stanford, CA: CSLI. Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. Unpublished Ph.D. Dissertation, University of Massachusetts, Amherst. Horvath, Julia. 2000. Interfaces vs. the computational system in the syntax of focus. In Hans Bennis, Marin Everaert, and Eric Reuland, eds., Interface Strategies, 183-207. Amsterdam: Royal Netherlands Academy of Arts and Sciences. Kuno, Susumu. 1973. The Structure of the Japanese Language. Cambridge, MA: The MIT Press. Kuroda, S.-Y. 2005. Focusing on the matter of topic: A study of wa and ga in Japanese. Journal of East Asian Linguistics 14: 1-58. Lambrecht, Knud. 1994. Information Structure and Sentence Form: Topic, Focus, and the Mental Representation of Discourse Referents. Cambridge: Cambridge University Press. Lee, EunHee and Mitsuaki Shimojo. 2010. Mismatch of topic between Japanese and Korean. ms. University at Buffalo, The State University of New York. Maynard, Senko K. 1987. Thematization as a staging device in the Japanese narrative. In John Hinds, Senko K. Maynard, and Shoichi

Mitsuaki Shimojo

293

Iwasaki, eds., Perspectives on Topicalization: The Case of Japanese WA, 57-82. Amsterdam and Philadelphia: John Benjamins. Portner, Paul and Katsuhiko Yabushita. 1998. The semantics and pragmatics of topic phrases. Linguistics and Philosophy 21: 117-157. Shimojo, Mitsuaki. 1995. Focus Structure and Morphosyntax in Japanese: Wa and Ga, and Word Order Flexibility. Unpublished Ph.D. Dissertation, University at Buffalo, The State University of New York. —. 2005. Argument Encoding in Japanese Conversation. Hampshire: Palgrave Macmillan. —. 2008. How missing is the missing verb? The verb-less numeral quantifier construction in Japanese. In Robert D. Van Valin, Jr., ed., Investigations of the Syntax-Semantics-Pragmatics Interface, 285-304. Amsterdam and Philadelphia: John Benjamins. —. 2009. Focus structure and beyond: Discourse-pragmatics in RRG. In Guerrero Valenzuela, Lilián, Sergio Ibáñez, and Valeria A. Belloro, eds., Studies in Role and Reference Grammar, 75-95. México: The IIFL-UNAM Press. Suzuki, Satoko. 1995. The functions of topic-encoding zero-marked phrases: a study of the interaction among topic-encoding expressions in Japanese. Journal of Pragmatics 23: 607-626. Vallduví, Enric. 1990. The Information Component. Unpublished Ph.D. Dissertation, University of Pennsylvania. Van Valin, Robert D., Jr. 1999. A typology of the interaction of focus structure and syntax. In E. Raxilina and J. Testelec, eds., Typology and Linguistic Theory: From Description to Explanation, 511-524. Moscow: Languages of Russian Culture. —. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press. —. 2008. An overview of Role and Reference Grammar. ms. HeinrichHeine-University Düsseldorf Universitätsstr. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Watanabe, Noriko. 1990. WA and GA: From the perspective of the deictic center in discourse. In Hajime Hoji, ed., Japanese/Korean Linguistics, 129-140. Stanford, CA: CSLI.

TOPIC, FOCUS, AND WORD ORDER IN THE ACQUISITION OF SPANISH ANTOINETTE HAWAYEK UNIVERSIDAD AUTÓNOMA METROPOLITANA-IZTAPALAPA

Abstract The aim of this paper is to explain Mexican children’s apparently error-free production of unaccusative and transitive sentences with the word order variation between SV and VS. All of the utterances with an NP subject in the UAMI corpus are analyzed in the contexts in which they were produced. The analysis shows that the subjects of unaccusative and transitive sentences occur preverbally when the children are talking about them, while the subjects occur postverbally when they are talking about the whole situation. The above analysis suggests that Mexican children up to the age of 4 are sensitive to topichood (rather than focushood) when they produce unaccusative and transitive sentences and that the notion of topichood enables the very young children to establish a contrast between the SV and VS order.

Keywords Focus, topic, information structure, discourse-pragmatics

1.

Introduction

One of the motivating questions for proposing and developing Role and Reference Grammar [RRG] (Van Valin and LaPolla 1997; Van Valin 2005) is how the interaction of syntax, semantics and pragmatics in different grammatical systems can best be captured and explained. In this paper, I will deal with manifestations of the interaction of syntax, semantics and pragmatics in the early stage of language acquisition. I will

Antoinette Hawayek

295

analyze the acquisition of information structure and, in particular, its interaction with syntax. Since the late 1980’s, RRG has addressed questions related to language acquisition. In the acquisition of the layered structure of the clause, one of the distinctions that has to be learned is the opposition between predicating and non-predicating elements (Van Valin and LaPolla 1997: 642). I will examine the word order in the early stage of language acquisition and will argue that the analysis of utterances in the contexts in which they were produced suggests that children place the subject before the verb when they topicalize it, while they place it after the verb when they topicalize the whole sentence. From the RRG perspective, one of the most important ways in which languages differ from each other is in terms of how discourse-pragmatics interacts with the linking between syntax and semantics (Van Valin 2005). In some languages, discourse-pragmatics can affect word order. Focusrelated word order variation is a well-attested phenomenon observed in a number of languages. For Mexican Spanish, it has been demonstrated that in the unmarked case (i.e. the word order used to felicitously answer questions such as “What happened?” and “What’s happening?”), the subject surfaces before the verb in sentences with agentive verbs, but it shows up post-verbally in sentences with unaccusative verbs (Gutiérrez Bravo 2006). The unmarked word order is shown in (1): (1) VS SV SVO

for sentences with unaccusative verbs for sentences with unergative verbs for sentences with transitive verbs

When the focus falls on the subject, the word order changes as in (2): (2) SV VS VSO, VOS

for sentences with unaccusative verbs for sentences with unergative verbs for sentences with transitive verbs

I report on the spontaneous production of 8 monolingual Mexican children (UAMI corpus). This longitudinal study was carried on at the Universidad Autónoma Metropolitana Day Care Center #2 (CENDI 2). At the beginning of the study, children were between 1.11 and 2.1 and they were between 3.7 and 3.10 at the end of the study. Children were recorded once a week from 8:00 A.M. to 12:30 P.M. while they carried out the regular CENDI program. The utterances recorded show that children produced all the orders attested in adult Spanish: SV, VS, SVO, VSO, and



296

Topic, Focus, and Word Order in the Acquisition of Spanish

VOS. The analysis of the context in which they were produced shows that all of them may be considered grammatically correct.

2. Early Production Data 2.1. Unaccusative Sentences First, verbs in (3a)-(3d) used by the children participating in this study are unaccusatives (with no subject or clitic required in adult Spanish).1 (3) a. (se) cayó “It fell.” b. (se) rompió “It broke.”

c. (se) acabó “It is finished.” d. cabe “It fits.”

2.1.1. VS Order Almost all intransitive verbs with an NP recorded in this longitudinal study are unaccusatives. Examples such as (4a)-(4k) were uttered to announce that an event was taking place or had just taken place. The NP subject surfaces postverbally, as in sentence-focus unaccusative constructions in adult Spanish. (4)

a. ya viene el perro. already comes the dog “The dog is coming.” b. se fue la luz. clitic went the light “The light went out.” c. se cae Arturo. clitic falls Arturo “Arturo is going to fall.” d. no vino Tonatzin. no came Tonatzin “Tonatzin did not come.” e. ya se murió la araña. already clitic died the spider “The spider died.”



(E. 2.3.27) 2 (F. 2.4.1) (D.A. 2.3.9) (M. 2.2.21) (Ed. 2.3.27)

This is confirmed by questionnaires applied to parents. Children are identified by first letter(s) of their names; numbers show the age (years, months, and days, in this order).

1 2



297

Antoinette Hawayek

f. ya se va Pedro. already clitic leaves Pedro “Pedro is leaving.” g. ya llegó el agua. already arrived the water “The water is here.” h. se cayó el tractor. clitic fell the tractor “The tractor fell.” i. cuidado si no, va a salir sangre. careful if not going to come out blood “Be careful or it is going to bleed.” j. sale sangre. comes out blood “It is bleeding.” k. me salió sangre me comes out blood I was bleeding

(E. 2.4.4) (M. 2.5.5) (F. 2.2.18) (T. 2.6.8) (A. 2.9.16) (S. 2.8.7)

2.1.2. SV Order Contrary to what is observed in (4a)-(4k), the subject surfaces before the unaccusative verb in (5). (5) a. Marita no vino. Marita no came “Marita did not come.” b. Dani no vino. Dani no came “Dani didn’t come.” c. ese tronco se movió. that trunk clitic moved “That trunk moved.” d. mi mamá se cayó de la cama. my mother clitic fell off the bed “My mother fell off the bed.” e. mi mama se fue. my mother clitic left “My mother left.” f. Fernando se cayó así. Fernando clitic fell like this “Fernando fell down like this.”



(T. 2.3.29) (Dal. 2.4.7) (F. 2.3.9) (F. 2.4.89) (M. 2.6.11) (E. 2.3.17)

298

Topic, Focus, and Word Order in the Acquisition of Spanish

The contexts in which these utterances were produced suggest that they did not answer a “what happened” type question; in (5a)-(5f), the subjects surface preverbally. This is exactly what happens in adult Spanish when the focus domain is restricted to the subject in unaccusative sentences. The VS order in (4) and the SV order in (5) apparently suggest that children have acquired how to assign focus in unaccusative sentences; the utterances in (4) exhibit the VS order used in adult Spanish to assign focus to the whole sentence, while the utterances in (5) exhibit the SV order used in adult Spanish to assign a narrow focus to the subject. The fact that Spanish children apparently follow the basic word order in adult Spanish should come as no surprise, because it has been observed that children’s early utterances generally observe the basic word order of their language whether it is SVO, OSV, VOS or other (Brown 1973; Bowerman 1973; Neeleman and Weerman 1997; Lust 2006).3 However, I will argue below that there is a reason to believe that the information structure system operative in (4) (VS) and (5) (SV) is different from the one used in adult Spanish and that the word order variation in (4) and (5) is not sensitive to where to place focus in unaccusative sentences.

2.2. Discussion The utterances in (4) and (5) apparently suggest that children under the age of 3 place the subject in preverbal or postverbal position depending on whether they assign a sentence focus to the whole sentence or assign a narrow focus to the subject. However, there is a reason to believe that children under the age of 4 are not sensitive to focus structure.

2.2.1. Focus Structure Foci have been defined in many ways, but their definitions usually center on the distinction between old and new information or the one between shared and non-shared information (Lambrecht 1994, 2000). It is important to note at this juncture that it has been shown that the pragmatic ability of children under the age of 3 is not developed to the extent that they can always distinguish between shared and non-shared information (Wimmer and Perner 1983; Schaeffer 2000; Tomasello et al. 2005; Surian et al. 2007). For example, Wimmer and Perner (1983) use a

 3

Evidence from brain research has shown that serial order mechanisms that serve the purpose of string detection exist (word sensitive sequence detectors are assumed to operate on pairs of elementary units) and that they may be part of the grammar machinery in the brain (Pulvermüller 2002, 2003a, 2003b).



Antoinette Hawayek

299

nonverbal technique to show that normal children, up to the age of 4, are unable to represent others’ beliefs when they differ from their own beliefs. Tomasello et al. (2005) also argue that the ability to understand others’ beliefs does not emerge until around the age of 4 in human ontogeny.4 Given that assigning focus to a whole sentence or a part of it presupposes understanding of others’ beliefs (i.e. being aware that some information is shared by the speaker and hearer), it seems unlikely that the children who uttered (4a)-(4k) and (5a)-(5f) chose the word order on the basis of where to assign focus within a sentence. The above consideration requires us to seek an alternative account of the word order variation in (4) and (5) by identifying the linguistic “cue” that leads very young children to choose the VS or SV order. 5

2.2.2. Looking for “Cues” RRG assumes that humans are endowed with the cognitive faculties to learn and use language. However, this begs a question of what kind of evidence is available to the child in the input from caregivers and from the situation in which utterances are used (Van Valin and LaPolla 1997: 641). This is one of the primary issues to be addressed in child language acquisition research. 6 It may be instructive to have a look at Schaeffer’s (2000) account of acquisition of direct object scrambling in Dutch, according to which the difficulty Dutch children have in their acquisition of object scrambling is attributed to their underdeveloped pragmatic ability. 7 In adult Dutch, direct objects can occur before or after elements such as a negation or adverbial, depending on whether they have a referential reading or not (Schaeffer 2000). Referential nouns refer to entities with distinct properties that make them unique members of their class, while non-referential nouns refer to any member of the class (or no member at all). A problem arises when Dutch children encounter an indefinite direct object, since it is ambiguous between a referential and a non-referential



4 A strong support for Wimmer and Perner’s (1983) and Tomasello el al.’s (2005) proposal comes from research on the evolution of language. Recent studies in this area assume that normal children up to the age of 4 are unable to represent others’ beliefs when they are different from their own beliefs (cf. Fitch 2010: 135-136). 5 A cue can simply be a linear pattern found in the observed data. 6 Signals are required to activate special purpose mechanisms (including language acquisition device), some of which originate outside the organism (Lewontin 2000). 7 Up to the age of 4.5, Dutch children fail to scramble direct objects in certain obligatory contexts (Schaeffer 2000: 47).



300

Topic, Focus, and Word Order in the Acquisition of Spanish

reading. Schaeffer (2000: 33-34) notes that a non-referential indefinite object may be placed before a negation, as in (6a), while a referential indefinite object may be placed after an adverbial, as in (6b). (6) a. das Els niet een problem goed kan uitleggen that Els not a problem well can explain “that Els cannot explain a problem well” b. das Marieke gisteren een (bespaald/zeker) boek gkocht heft that Marieke yesterday a certain book bought has “that Marieke bought a certain book yesterday” Schaeffer proposes to attribute the difficulty Dutch children have in acquiring the object scrambling to the well-established fact that 2-4 year old children cannot make a distinction between shared and non-shared information, which underlies the distinction between a referential and nonreferential object. 8 Comparing the Dutch data with the Spanish data in (4) and (5) suggests that it may not be easy for 2-4 year old children learning Spanish to assign focus (sentence focus or narrow focus), because doing so presupposes their ability to distinguish between shared and non-shared information. This observation prompts us to propose an alternative explanation for the word order variation in Spanish. An important clue comes from the fact that a topic is placed at the beginning of a sentence in Spanish and that the topicalized subject occurs preverbally (whether the verb is unaccusative, unergative, or transitive). My proposal is that the preverbal position serves as an unambiguous cue 9 for the topicalized subject: the children place the subject in preverbal position when they are talking about it, while they place the subject in a postverbal position when they are not talking about it. It is important to emphasize at this juncture that a topic is not synonymous with the “given” or “presupposed” part of the utterance (Van Valin and LaPolla 1997: 203-204). 10 This runs counter to much of the literature on information structure, but what is presupposed is not a topic or its referent, but the fact that the topic referent is expected to play a role in a given proposition, due to its status as a center of interest (Lambrecht

 8

Another observation to make is that the Dutch indefinite article een “a(n)” is not a reliable marker to be used as a cue. 9 Cues must be frequent enough and not too hard to detect.  10 If a topic were identified with the “given” or “presupposed” part of an utterance, it would require the speaker to be aware of what might be accessible information to the hearer and to share the information with the hearer. 



301

Antoinette Hawayek

1994: 151).11 All the speaker needs to do is to become aware of what a sentence s/he is about to utter is about. Given that the word order variation in (4) and (5) is not sensitive to the focus assignment, but rather to the topic assignment, we may summarize the correlation as in (7): (7) a. SV order, illustrated in (5) The subject appears preverbally when it serves as a topic b. VS order, illustrated in (4), The subject appears postverbally when the whole sentence serves as a topic The validity of (7a,b) is confirmed by a contextual analysis of (5a)-(5f), which suggests that the children were talking about the subjects when they uttered (5a)-(5f). (5) a. Marita no vino. Marita no came “Marita did not come.” b. Dani no vino. Dani no came “Dani didn’t come.” c. ese tronco se movió. that trunk clitic moved “That trunk moved.” d. mi mamá se cayó de la cama. my mother clitic fell off the bed “My mother fell off the bed.” e. mi mama se fue. my mother clitic left “My mother left.” f. Fernando se cayó así. Fernando clitic fell like this “Fernando fell down like this.”

(T. 2.3.29) (Dal. 2.4.7) (F. 2.3.9) (F. 2.4.89) (M. 2.6.11) (E. 2.3.17)

It is generally agreed that for the hearer to accept a referent as a topic, it must have been mentioned in the immediately preceding discourse. The subjects of (5a,b) are typical topics, because they were mentioned earlier

 11

See also Erteschik-Shir (2007: 19), who defines a topic as “what a sentence is about”. 



302

Topic, Focus, and Word Order in the Acquisition of Spanish

in the conversation. However, it is important to note that not all topics require previous mention. It is true that there is a correlation between the pragmatic state of the topic referent and its acceptability as a topic and that the more accessible the topic referent of an utterance is, the less processing effort will be required to interpret the utterance (Van Valin and LaPolla 1997: 204), but the presence of a tree trunk on the scene allows the child “E” to make “that trunk” serve as the topic of (5c). Furthermore, those subjects that have no previous mention in the preceding discourse or no situationally accessible referent may be considered as a topic if they constitute permanent features of the world (e.g. president, moon) (Erteschik-Shir 2007:18). This applies to (5d)-(5f). Specifically, mi mamá “my mother” in (5d) may be regarded as a permanent feature of the child’s world. The contextual analysis shows that the children including F are not talking about a situation, but about their mothers. F laughs and puts stress on mi mamá when uttering (5d). M is not announcing that his mother left when he utters (5e). M is protesting that his mother left in spite of her promise not to do so. Finally, when E utters (5f), E wants to tell the other children that Fernando is special. In this utterance, the subject may function as a temporarily available topic because Fernando just fell in front of E and other children.

3. Transitive Verbs To further validate the hypothesis that subjects in preverbal position serve as topics in the stages of acquisition under study, I will turn to transitive sentences (VSO, VOS, and clitic VS) in the UAMI corpus and will show that all the preverbal subjects are topics.

3.1. VSO Order The contextual analysis shows that the subjects of VSO sentences in (8) are not the most important issues for the children. They don’t serve as the topics and don’t occur preverbally. (8) a. no me compró mi mamá una panterita. not me bought my mother a panther diminutive “My mother did not buy me a little panther.” b. me compró mi mamá un vestido lindo. me bought my mother a dress pretty “My mother bought me a pretty dress.”



(A. 3.2.5) (M. 2.11.16)

Antoinette Hawayek

c. me ponió mi mamá cremita. me put my mother cream diminutive “My mother applied cream.” d. le quitó tu hija la pistol. her took away your daughter the gun “Your daughter took the gun away from her.” e. así hace el conejo la nariz. like this the rabbit make the nose “The rabbit moves the nose like this.”

303

(A. 3.9.14) (A. 2.10.28) (E. 2.3.27)

For example, when uttering (8a,b), the children (A and M) are talking about presents they received. In other words, they are not talking about their mothers, but about what they did or didn’t get. The fact that mi mamá “my mother” in (8a,b) doesn’t serve as a topic explains why mi mamá doesn’t occur preverbally. An analogous account holds for (8c)-(8e).

3.2. VOS Order VOS sentences are not very frequent in Spanish and it has been claimed that they sound awkward unless a very specific context is provided (cf. Gutiérrez Bravo 2006). However, all of (9a)-(9e) are grammatical and are not awkward. (9) a. ya me cortó la uña mi mama. already me cut the nail my mother “My mother has already cut my nail.” b. me cortó mi pelo mi papá. me cut my hair my father “My father cut my hair.” c. a mí, me cortó el pelo la señora. to me, me cut the hair the lady “My father cut my hair.” d. me quitó mi papel mi maestro. me took away my paper my teacher “My teacher took my paper away.” e. tiene tierra mi zapato. has sand my shoe “My shoe has sand.”

(M. 2.10.9) (M. 3.0.10) (S. 2.10.3) (Ed. 2.4.2) (Dal. 2.5.17)

In (9a)-(9c), children were talking about having had their hair or finger nails cut, but not about who did it. Another point to note is that pelo “hair”



304

Topic, Focus, and Word Order in the Acquisition of Spanish

and uña “nail” occur with the transitive verb cortar “cut” often enough to form a VO unit. These observations explain why the subject follows the VO unit in (9a)-(9c). Furthermore, (9d) was uttered as an answer to the question of why the child was not working. The reason is that his paper was taken away from him. What was topicalized in (9d) is the verb phrase. This explains why the subject mi maestro “my teacher” occurs after the verb phrase. Likewise, the child was talking about what she just found out (“have sand”) when she uttered (9e). The sand in her shoes made her uncomfortable. This explains why the subject occurs after the VO.

3.3. Clitic VS The subjects of all clitic VS sentences in (10) occur postverbally, because they don’t serve as the topics: (10) a. mira, lo hizo Karen. look it made Karen “Look, Karen made it.” b. ya la tiró mi maestro. already it threw my teacher “My teacher threw it away.” c. le pegó Marta. him hit Marta “Marta hit him.” d. me mordió Zaira. me bit Zaira “Zaira bit me.” e. me lo regaló el doctor. me it gave the doctor “The doctor gave it to me.” f. me lo limpió la maestro. me it cleaned the teacher “My teacher cleaed it.” g. para que no me lo robe mi maestro. so that not me it steal the teacher “so that my teacher does not steal it from me.”

(T. 2.2) (T. 2.6.29) (S. 2.4.2) (E. 2.4.11) (M. 2.6.18) (T. 2.6.29) (S. 2.8.7)

T was talking about the animal in (10a) and the garbage the teacher threw away in (10b), both of which are represented by the third person object clitic lo “it”. It is clearly these objects that are topicalized in (10a,b). Children were explaining why somebody was crying when they uttered



Antoinette Hawayek

305

(10c,d), but they were not talking about who made them cry. Finally, (10e)-(10g) are the answers the children (M, T, and S) gave when they were asked about the entity represented by the third person object clitic lo “it”. They were taking about the entity that functions as the direct object, but not about the subject. These observations explain why the subjects in (10a)-(10g) don’t occur preverbally.

3.4. SVO Order Transitive sentences with SVO order in adult Spanish involve either of two information structures: either the subject functions as a topic or the whole sentence is focused. It is interesting to note in this connection that all SVO utterances registered in the UAMI corpus were produced at a later stage of acquisition than the SV sentences with unaccusatives verbs in (4)-(5). 12 This means that there is a possibility that those children that uttered transitive sentences with SVO order just started to be aware that the subject may also occur before a verb when the whole sentence is focused. This subsection provides a contextual analysis of SVO sentences illustrated in (11a)-(11h) in order to investigate whether all of them involve a topicalized subject or may be analyzed as involving a sentence focus (as in adult Spanish). (11) a. esta Ariana me prestó su agua. (M. 3.6.1) this Ariana lent me her water “Ariana lent me her water.” b. este popote tiene grande el agujero. (F. 3.2.4) this straw has big the hole “This straw has a big hole.” c. Daniel no me prestó su material. (E. 2.9.2) Daniel not me lent his material “Daniel did not lend me his material.” d. mi mamá está haciendo la comida para comer con una cucharita. (E. 2.9.2) my mother is making the meal to eat with a spoon diminutive “My mother is cooking the meal to be eaten with a little spoon.” e. mi mama me compró un coche que funciona. (Ed. 3.6.29) my mother me bought a car that functions “My mother bought me a car that functions.”

 12

SV sentences with unaccusative verbs in (4) and (5) were registered at 2.3.9, first SVO sentence was registered at 2.9.



306

Topic, Focus, and Word Order in the Acquisition of Spanish

f. mi mama me va a comprar una calabaza grande y una chiquita. (M. 3.6.16) my mother me is going to buy a pumpkin big and one small “My mother is going to buy me a big and a small pumpkin.” g. mi abuelita me hace un sweater de borrego. (S. 3.1) my grandmother me makes a sweater of lamb (wool) “My grandmother is knitting a lambs-wool sweater for me h. mi tía me tejió un sweater. (A. 3.7) my aunt me knitted a sweater “My aunt knitted me a sweater.” The subjects in (11a)-(11c) qualify as topics, because Adriana, Daniel and the straw were on the scene of the conversation and were pointed at by the speakers, while the subjects of (11d)-(11h) are arguably the permanent features of the child’s world and thereby can function as the topics even with no previous mention of them. 13 However, a careful contextual analysis of (11d)-(11h) shows that their subjects are not topics, because the children who uttered (11d)-(11h) were not talking about their mother, grandmother, or aunt. First, the children who uttered (11d)-(11f) were not talking about their mothers, but they refer to an event that had happened, was happening, or was going to happen: (11d) was uttered when E saw a girl playing with a little spoon and pretending she was cooking, (11e) was uttered when another child refused to lend a toy to Ed, while (11f) was uttered by M, who wasn’t allowed to cut pumpkins in the way she liked. Second, those children who uttered (11g,h) were not talking about mi abuelita “my grandmother” or mi tía “my aunt”. (11g,h) were uttered when other children were talking about their sweaters. This observation suggests that S and A were also talking about the sweaters that had been or were being knitted for them and that neither mi abuelita “my grandmother” nor mi tía “my aunt” serves as the topic in (11g,h). If we adopt a pragmatic definition of focus as “that information which is relatively the most important or salient in the given communicative setting and considered by the speaker to be the most essential” (Dik 1997: 326; cf. Erteschik-Shir 2007: 38), we may, at least, argue that the whole sentences are focused in (11d)-(11h) and that those children who uttered (11d)-(11h) are learning that transitive sentences with SVO order may



13 In typical Mexican “extended” families including those of the children who uttered (11g,h), “my grandmother” and ”my aunt” as well as “my mother” constitute the permanent features.



Antoinette Hawayek

307

involve a sentence focus. Before closing this section, I would like to mention that we may assume that sentence-focus constructions do have a topic, an implicit topic that indicates the here and now of the discourse (Erteschik-Shir 2007: 1617).14 As (11d)-(11h) are arguably related to what was happening or what had just happened before they were uttered, we may analyze (11d)-(11h) as all-topic (i.e. sentence-topic) sentences. This means that it is possible to analyze (11d)-(11h) as involving a sentence topic (rather than a sentence focus) and to argue that they fall within the scope of (7), repeated below. (7) a. SV order, illustrated in (5) The subject appears preverbally when it serves as a topic b. VS order, illustrated in (4), The subject appears postverbally when the whole sentence serves as a topic

4. Conclusions In this paper, I have examined all the SV, VS, SVO, VSO, VOS, and clitic VS utterances registered in the longitudinal study of monolingual Mexican children (the UAMI corpus). The 8 subjects of this study produced all the word orders attested in adult Spanish with no apparent error. We have seen that children start producing unaccusative sentences with the word order variation that is subject to (7a,b). The basic idea is that the word order of unaccusative sentences is decided by where to place a topic within a sentence; if it is placed on the whole sentence, the VS order emerges, while if it is placed on the subject, the SV order emerges. To further confirm the validity of (7), I examined transitive sentences with VSO, VOS, and clitic VS order and showed that children follow (7) the same two rules; they only place the subject in preverbal position if they are talking about it, if they are not, they place it post verbally. Transitive sentences with SVO order do not usually appear at the very early stages of acquisition. When children start producing them, the preverbal subjects of some of their utterances do not seem to be a topic. I have taken this as an indication that before children turn the age of 4, they start understanding that transitive sentences with SVO order are used not only to topicalize their subjects, but to indicate that the most important information is expressed by the whole sentence. I have argued that if we

 14

Researchers who adopt a definition of a topic as a pivot for assessment argue that every sentence should have one topic.



308

Topic, Focus, and Word Order in the Acquisition of Spanish

may assume that sentences-focus constructions have a topic, all-topic (i.e. sentence-topic) constructions may function as a bridge to sentence-focus constructions, which are not acquired by children up to the age of 4. Finally, a word is in order about unergative sentences. Their word order variation was not investigated in this study, because the UAMI corpus contains only one pair of the SV and VS order. However, these two examples do not contradict (7). At the age of 2.3.27, one child (Ed) produced an unergative sentence with VS order when asked what was going on, while he produced the one with SV order (using the same verb). (12) a. están llorando los bebés. are crying the babies “The babies are crying.” b. Marita también lloraba. Marita also was crying “Marita was also crying.”

(Ed 2.3.27) (Ed. 2.3.27)

The child appealed to the VS order when he was talking about the whole event, while he used the SV order when he was taking about the subject.

References Bowerman, Melissa. 1973. Early Syntactic Development: A CrossLinguistic Study with Special Reference to Finnish. Cambridge: Cambridge University Press. Brown, Roger. 1973. A First Language: The Early Stages. Cambridge, MA: Harvard University Press. Dik, Simon C. 1997. The Theory of Functional Grammar, Part 1: The Structure of the Clause. Berlin and New York: Mouton de Gruyter. Erteschik-Shir, Nomi. 2007. Information Structure: The Syntax-Discourse Interface. Oxford: Oxford University Press. Fitch, W. Tecumseh. 2010. The Evolution of Language. Cambridge: Cambridge University Press. Gutiérrez-Bravo, Rodrigo. 2006. Structural Markedness and Syntactic Structure: A Study of Word Order and the Left Periphery in Mexican Spanish. New York: Routledge. Lambrecht, Knud. 1994. Information Structure and Sentence Form. Topic, Focus and the Mental Representations of Discourse Referents. Cambridge: Cambridge University Press. —. 2000. When subjects behave like objects: an analysis of the merging of S and O in sentence-focus constructions across languages. Studies in



Antoinette Hawayek

309

Language 24: 611-682. Lewontin, Richard C. 2000. The Triple Helix: Gene, Organism and Environment. Cambridge, MA: The MIT Press. Lust, Barbara. 2006. Child Language. Acquisition and Growth. Cambridge: Cambridge University Press. Neeleman, Ad and Fred Weerman. 1997. L1 and L2 word order acquisition. Language Acquisition 6: 125-170. Pulvermüller, Friedemann. 2002. A brain perspective on language mechanisms: from discreet neuronal ensembles to serial order. Progress in Neurobiology 67: 85-111. —. 2003a. Sequence detectors as a basis of grammar in the brain. Theory in Bioscience 122: 87-103. —. 2003b. The Neuroscience of Language: On Brain Circuits of Words and Serial Order. Cambridge: Cambridge University Press. Surian, Luca, Stefania Caldi, and Dan Sperber. 2007. Attribution of beliefs by 13-month-old infants. Psychological Science 18: 580-586. Van Valin, Robert D., Jr. 1999. A typology of the interaction of focus structure and syntax. In E. Raxilina and J. Testelec, eds., Typology and the Theory of Language: From Description to Explanation, 511-524. Moscow: Languages of Russian Culture. —. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Schaeffer, Jeanette. 2000. The Acquisition of Direct Object Scrambling and Clitic Placement: Syntax and Pragmatics. Amsterdam and Philadelphia: John Benjamins. Tomasello, Michael, Malinda Carpenter, Josep Call, Tanya Behne, and Henrike Moll. 2005. Understanding and sharing intentions: the origins of cultural cognition. Behavioral and Brain Science 28: 675-735. Wimmer, Heinz and Josef Perner. 1983. Beliefs about beliefs: representation and constraining function of wrong beliefs in children’s understanding. Cognition 13: 103-128.





PART 5. COMPUTATIONAL APPLICATIONS OF RRG





UNIARAB: RRG ARABIC-TO-ENGLISH MACHINE TRANSLATION BRIAN NOLAN AND YASSER SALEM INSTITUTE OF TECHNOLOGY BLANCHARDSTOWN, DUBLIN IRELAND

Abstract This paper presents a proof-of-concept machine translation system called UniArab that supports fundamental aspects of Arabic, including lexical category, agreement and tenses. UniArab is based on the linking algorithm of Role and Reference Grammar [RRG] (syntax to semantics and vice versa). UniArab takes MSA Arabic as input in the native orthography, parses the sentence(s) into a logical meta-representation based on the fully expanded RRG logical structures and generates perfectly grammatical English output with full agreement and morphological resolution. UniArab utilizes an XML-based implementation of elements of the RRG in software. To analyse Arabic, we extract the lexical properties of the Arabic words. From the parse, UniArab create a computer-based representation for the logical structure of the Arabic sentence(s). We use RRG to motivate the computational implementation of the architecture of the lexicon in software and implement in software the RRG bidirectional linking system to build the parse and generate functions between the syntax-semantic interfaces. Through seven input phases, including the morphological and syntactic unpacking, UniArab extracts the logical structure of an Arabic sentence. Using the XML-based metadata representing the RRG logical structure, UniArab then accurately generates an equivalent grammatical sentence in the target language through four output phases.



Brian Nolan and Yasser Salem

313

Keywords Arabic Machine Translation, RRG, computational linguistics, Java programming, XML

1. Introduction This paper reports on recent work the development of a rule-based semantically oriented Interlingua bridge framework for machine translation of Arabic language processing using the RRG linguistic model. Machine translation is a sub-field of computational linguistics that investigates the use of computer software to translate text (or speech) from one natural language to another. Our system has been developed and is able to analyse Arabic sentences in native orthography, and extract their logical structure. Through a detailed study of the Arabic language, we have been able to develop an analyser that can successfully process many of the unique features and challenges present in Arabic. This logical structure is then used in the generation phase, where the sentence(s) is translated into another language, in this case, English. The Arabic language is written from right to left and it has complex, language-specific grammar rules and a relatively free word order. These distinguishing features pose a major challenge in processing Arabic text for linguistic analysis. Our framework demonstrates that RRG is a feasible foundation for building multi-language machine translation [MT] systems. Arabic is a Semitic language originating in the area presently known as the Arabian Peninsula. The Arabic language is one of six major world languages, and one of the six official languages of the United Nations. The version of Arabic we consider in this work is Modern Standard Arabic [MSA]. When we mention Arabic throughout this paper, we mean MSA, which is a distinct, modernized form of Classical Arabic (Alosh 2005). MSA is the universal written language of the Arabic-speaking population, printed in most books, newspapers, magazines, official documents, and reading primers for children. Most of the oral Arabic spoken today is more divergent than the written Arabic language, because of dialectal interference. However, MSA is the literary and standard variety of Arabic used in writing and formal speeches today (Schulz 2005). In this paper we discuss the RRG UniArab MT research project and the Interlingua model of Arabic MT that we designed and built using Java and XML. With this we discuss the challenges inherent within Arabic MT and the part that RRG played in helping to overcome many of the challenges. The architecture of the lexicon and its design and implementation in XML



UniArab: RRG Arabic-to-English Machine Translation

314

is discussed, along with a presentation of the results produced by the UniArab software evaluation

2. The RRG Linguistic Model RRG is a model of grammar that posits a direct mapping between the semantic representation of a sentence and its syntactic representation (Van Valin 2005). We claim that RRG is very suitable for machine translation of Arabic via an Interlingua bridge implementation model. RRG is a mono strata-theory, positing only one level of syntactic representation, the actual form of the sentence and its linking algorithm can work in both directions from syntactic representation to semantic representation, or vice versa. In RRG, semantic decomposition of predicates and their semantic argument structures are represented as logical structures. The lexicon in RRG takes the position that lexical entries for verbs should contain unique information only, with as much information as possible derived from general lexical rules. The main features of RRG are the use of lexical decomposition, based upon predicate semantics, an analysis of clause structure and the use of a set of thematic roles organized into a hierarchy in which the highestranking roles are Actor (for the most active participant) and Undergoer (Van Valin 2005). RRG characterises the relationship between syntax and semantics and can account for how semantic representations are mapped into syntactic representations. RRG also accounts for the very different process of mapping syntactic representations to semantic representations. Of the two directions, syntactic representation to semantic representation is the more difficult since it involves interpreting the morphosyntactic form of a sentence and inferring the semantic functions of the sentence from it. Accordingly, we have chosen to implement Arabic to English as the translation direction and the basis of the parse and generate functions in this version of our software.

3. Interlingua Approach of Arabic MT The Interlingua approach is to develop a universal language-representation for text. In effect, in an Interlingua there is no transfer map, and the MT model thus has two main stages: input-PARSE-analysis and outputGENERATE.



Brian Nolan and Yasser Salem

315

Figure 1: MT – Transfer vs. Interlingua approach

Interlingua-based MT is done via an intermediate semantic representation, based on RRG logical structures, of the source language text. An Interlingua is designed to be a language independent representation from which translations can be generated to different target languages.



UniArab: RRG Arabic-to-English Machine Translation

316

Figure 2: MT – Our Interlingua approach

3.1. UniArab: Lexical representation in an Interlingua system Transfer oriented translation systems (Figure 1) do not scale up when additional languages are added beyond the initial source (SL1) and target (TL1) language pairs, and very quickly this leads to a translation complexity problem between languages. Additionally, of course, in simple transfer-based systems there are no problems if, for a particular language pair, there are morphosyntactic one-to-one equivalents; problems do arise, however, when there is more than one target word for a single source word. Implementation of an Interlingua bridge architecture solves (Figure 2) the translation complexity problem as automatic language translation is made from a source language into a kernel meta representation (the input PARSE phase) and generates to a target language from the meta representation (the GENERATION phase). Ambiguity problems for an Interlingua in a multilingual system are still likely if one of the languages involved has two or more potential forms for a single given word in one of the other languages. A semantically oriented approach to MT can potentially disambiguate more easily than other strategies. For an Interlingua to be completely language-neutral, it must represent not the words of one or another of the languages, but language-independent lexical units. Any distinction that can be expressed lexically in the



Brian Nolan and Yasser Salem

317

languages of the system must be represented explicitly in the Interlingua representation (Hutchins 2003). We use the RRG logical structures as the basis of our meta-representation in the Interlingua Bridge with a lexicon encoded in XML. The UniArab system can generate a target language through classifying every Arabic word in the input source text by creating a metarepresentation of the sentence(s) input as a text in a fully populated RRGstyle logical structure including its various nominals and their associated features of [def+, masc+], etc. There are six major parts of speech in Arabic. These are verbs, nouns, adjectives, proper nouns, demonstratives, adverbs and we create a seventh for purposes of our software, which we have simply called the `other' category for Arabic words that do not fit into any of previous six categories. The major parts of speech in the Arabic language have their own attributes, and we use these attributes within the UniArab system. For example, verbs in the Arabic language agree with their subjects in gender. Arabic words are masculine and feminine; there is no neutral gender. In the UniArab system we record the gender associated with a verb in the syntax for a particular subject NP. Adjectives and demonstratives also agree with the subject in gender too. In Arabic, words come into three categories with regards to number. They are: (1)

Singular, indicating one

(2)

Dual, indicating two

(3)

Plural, indicating three or more.

The UniArab system records these attributes of gender and number. It is important to understand that source language specific features may not be used, or may be significantly different, in the target language. For example, the Arabic number category of dual is not relevant in English. The UniArab system is directly based on RRG and uses logical structures for each verb in the lexicon.

3.2. Challenges of Arabic to English MT Arabic words can often be ambiguous due to the three-letter root system. Most words are derived from a three-letter root that is modified to create the different derivations. In some morphological derivations one or more of the root letters is dropped, resulting in possible ambiguity. Arabic has a large set of morphological features (Al-Sughaiyer and Al-Kharashi 2004).



UniArab: RRG Arabic-to-English Machine Translation

318

These features are normally in the form of prefixes or suffixes that can completely change the meaning of the word (see Figures 3 and 4). This means an MT may need to apply a thorough analysis in order to obtain the root or to deduce that in one word there is in fact a full sentential proposition.

Figure 3: The root and pattern characteristics of Arabic

Arabic has a relatively free word order (Figure 5) and this poses a significant challenge to MT due to the vast possibilities to express the same sentence in Arabic. For the elements of subject (S), verb (V) and object (O), Arabic's relatively free word order allows the combinations of SVO, VSO, VOS and OVS. For example, consider the following word orders: (1) V N N and (2) N V N. This means that we have a challenge to identify exactly which are the subject and the object. An example of the RRG layered structure of the Arabic clause is presented in Figure 6.



Brian Nolan and Yasser Salem

Figure 4: The tri-consonantal roots and word formation in Arabic

Figure 5: The challenges of Arabic for MT



319

UniArab: RRG Arabic-to-English Machine Translation

320

Figure 6: The layered structure of the Arabic clause

4. The UniArab System UniArab is a proof-of-concept system supporting the fundamental aspects of Arabic, such as the parts of speech, agreement and tenses. UniArab stands for Universal Arabic machine translator system. UniArab is based on the linking algorithm of RRG (syntax to semantics and vice versa). The conceptual structure of the UniArab system is shown in Figure 7. The system accepts Arabic as its source language. The morphology parser and word tokenizer have a connection to the lexicon, which holds all attributes of a word. UniArab was developed in the Java programming language with the lexicon encoded in XML. UniArab stores all data in XML format. This data can then be queried,



Brian Nolan and Yasser Salem

321

exported and serialized into any format the developer wishes. The system can understand the part of speech of a word, agreement features, number, gender and the word type. The syntactic parse unpacks the agreement features between elements of the Arabic sentence into a semantic representation (the logical structure) with the state of affairs of the sentence. In UniArab we have a strong analysis system that can extract all attributes from the words in a sentence. The structure of the UniArab system in Figure 7 breaks down into the several phases, which are described below.

Figure 7: The conceptual architecture of the UniArab system

Phase (1) Input of Arabic language sentence: The input to the system consists of one or more sentences in Arabic. Phase (2) Sentence Tokenizer: Tokenization is the process of demarcating and classifying sections of a string of input characters. In this phase the system splits the text into sentence tokens. The resulting tokens are then passed to the word tokenizer phase. Phase (3) Word Tokenizer: In this phase, sentences are split into tokens. For example, for the Arabic sentence (4a), read from right to left, the output (4b) of Phase 3 is as follows.



UniArab: RRG Arabic-to-English Machine Translation

322

(4)

a. b.

qr’a ‫ه‬Ɨld ƗlktƗb “Khalid read the book.”

qr’a ‫ه‬Ɨld ƗlktƗb

Phase (4) Lexicon XML Data-source: A set of XML documents for each component category of Arabic. More details will be in Sections 6 and 7. Phase (5) Morphology Parser: Directly works with both the Lexicon and Tokenizer to produce the word order. A connection is made to the datasource of Phase 4, which has been implemented as a set of XML documents. The use of XML has the added advantage of portability. UniArab will effectively work the same regardless of the operating system. To understand the morphology of each word, we first tokenize each sentence and determine the word relationships. Phase 5 of the system holds all attributes specific to each word of the source sentence. Phase (6) Syntactic Parser: Determines the precise phrasal structure and category of the Arabic sentence. At this point, the types and attributes of all words in the sentence are known. Phase (7) Syntactic linking (RRG): We must first develop the link from syntax to semantics out of the phrasal structure created in Phase 6, if we are to create a logical structure that will generate a target language and also act as the link in the opposite direction from semantics to syntax. The system should answer the main question in this phase, “who does what?” In this case the actor is “Khalid” and the undergoer is “the book”, as in (4) above. Phase (8) Logical Structure: The creation of logical structure is the most crucial phase. An accurate representation of the logical structure of an Arabic sentence is the primary strength of UniArab. The results of the parse can be seen in the following logical structure for the verb “read”. (5)

a. b.



Verb “read”: sg 3rd.m PAST qr’a where : the Proper Noun is: Khalid sg unspec.m: ‫ه‬Ɨld and the Noun is: the book sg def.m: ƗlktƗb.

Brian Nolan and Yasser Salem

323

We also have the challenge of inferring the indefinite article, from the information unpacked in Phase 5 and Phase 6, as this does not exist in Arabic. All of the unique information for each word can thus be taken from the lexicon to aid in the creation of a logical structure of the target language. Phase (9) Semantic to Syntax: Assuming we have an input and have produced a structured syntactic representation of it, the grammar can map this structure from a semantic representation. In this phase, the system uses a linking algorithm provided by RRG, to determine actor and undergoer assignments, assign the core arguments and assign the predicate in the nucleus. We determine the grammatical subject by analysing the agreement marking on the verb and the various nominals. The system uses the semantic arguments of logical structures. Phase (10) Syntax Generation: The generation phase from the Interlingua Bridge meta-representation to the morphosyntax of a particular target language will, of course, depend on the characteristics of the target language. In our proof-of-concept software, we generate to grammatically correct English (see also Phases 11 and 12 below). The generation phase implements the RRG semantics-to-syntax linking system. Phase (11) Generate English Morphology: The system generates English morphology in an innovative way, generating the tenses that are not existent in Arabic but which do exist in English as well as the copula verb of “to be” correctly, as appropriate. Our solution is to recognize the difference between morphological features and syntactic functional categories. The tense features must be determined analytically, and expressed correctly for the target language, in this instance, English. Phase (12) English Sentence Generation: The process of generating an English sentence can be as simple as keeping a list of rules and these rules can be extended through the life of the MT system. The system will apply some operations in English such as vowel change in the lexical item of English to denote SG vs. PL, for example, man vs. men. Sometimes this accompanies affixations: break/broke/broken (=broke+en) to denote various tense and aspect distinctions. Having described the various Interlingua phases, we now discuss in more detail, in the next section, the GENERATION from metarepresentation (i.e. the logical structure) to target language.



UniArab: RRG Arabic-to-English Machine Translation

324

5. UniArab - Generation The target language generation phases in the UniArab system follow the syntactic realization model. Generation takes as input, the universal logical structure of the input sentence(s) and produces, as output, the grammatically correct morphosyntax of the target language. The UniArab system is a universal machine translator, which means that it can translate Arabic into any other natural language. The UniArab system is evaluated using Arabic as source language into English as the target language. In the UniArab system, Phases 9, 10, 11, and 12 are for generation of the target languages, in our case this is English. First, the Semantic to Syntactic phase determines the actor and undergoer assignments, assigns the core arguments and assigns the predicate in the nucleus. The system uses semantic arguments of logical structure. In the UniArab system we keep all word attributes whether they are used in the target language or not. In this case, the gender of the noun “the book”, in Arabic is masculine, but in English “book” has neutral gender. In Phase 10, Syntax Generation, and Phase 11, Generate English Morphology, UniArab uses target language rules to generate the syntax. The verb’s logical structure indicates to UniArab how many arguments the verb takes. For example, the logical structure will be as in (6a), from the lexicon. Now the UniArab system replaces x with “Khalid”, and y with “the book”, after which it now holds the following (6b): (6)

a. b.

Read: Read:

[do'(x, [read'(x, (y))])] [do'(Khalid, [read'(Khalid, the book)])]

In the last phase, English Sentence Generation, the UniArab system builds the final shape of a sentence: “Khalid read the book”. Moreover, there are some special cases, like the UniArab system adding the copula verb “to be” into the English copula sentence, or changing the source language verb’s tense to an appropriate and grammatically correct tense in the target language, depending on the tense distinction in the target language. Also, the word order in the target language must be considered and applied correctly.

6. An XML-based Lexicon In order to build this system and represent the data sources, we use the Java with the XML language (Bray et al. 2008). XML has become the default standard for data exchange among heterogeneous data sources



Brian Nolan and Yasser Salem

325

(Arciniegas 2000). The UniArab system allows data to be stored in XML format. This data can then be queried, exported and serialized into any format the developer wishes. We choose to create our data source as XML, for optimum support on different platforms. It was also easier as we used Arabic letters, not Unicode, inside the data source, and XML fully supports Arabic. We created our search engine for the lexicon using Java. The lexicon is represented as an XML data object

6.1. Advantages of XML XML gives us a generalized way to store data, which is not married to any particular technology. This makes it easy to store information, and retrieve and manage it later, as required. Using XML to manage information offers a number of advantages, including the following: (7) 1.

Easily build: A well-formed data element must be enclosed between tags. The XML document can be parsed without prior knowledge of the tags. XML allows one the possibility of defining ones own application relevant tags, such as tags representing data description or data relationships, in our situation to do with lexical items.

2.

Human readable: Using intelligible tag names make it possible for the XML to be easily read by people as well as software.

3.

Machine-readable: XML was designed to be easy for computers to process. XML is completely compatible with Java, and is portable. Any application can process XML on any platform, as it is a platform-independent language.

4.

XML fully supports Arabic: We chose to create our data-source as XML files, for optimum support of different platforms. It was also easier as we used Arabic letters rather than Unicode inside the data-source.

5.

XML search engine: It is easy to extend the search sample to display more information about the search. Search via the Java API Document Object Model (DOM) was found to be the ideal tool for searching collections of XML documents, that is, our lexicon.



UniArab: RRG Arabic-to-English Machine Translation

326

6.2. Lexicon Interface In order to allow for robust user interaction with the lexicon, we use a graphical interface to capture the information for each part of speech. The user selects the part of speech of the word to be added, and is then presented with only the attributes relevant to the selected part of speech. The interface also limits the user’s selections to acceptable values and ensures that all attributes are filled. With this technique, we minimize the risk of human errors, and therefore the information is more accurate. The graphical interface is quicker and easier when a user adds a new word in the lexicon within the XML data source. Figure 8 shows the entry interface that is implemented as part of the UniArab system.

Figure 8: The Lexicon Interface of UniArab

7. Lexical Representation in UniArab Lexical frames represent the language-dependent lexicon. We use an XML data source to represent the UniArab lexicon. The lexicon creates pointing references to corresponding conceptual frames with associated attributes for each word. These frames also have relations which link them to verb class frames, which are organized hierarchically according to the particular language, here, Arabic and English. In Phase 3 of our Interlingua Bridge PARSEÆGENERATE framework, the UniArab system tokenizes a sentence into words, and then sends each word to the search engine within the Lexicon to query the category of each word plus determine all attributes associated with that word. The Lexicon returns the corresponding category and its attributes. The Morphology Parser (Phase 5) receives the word metadata and ensures that the properties of the words are consistent. The verb attributes, in particular, are of critical importance in correctly extracting sentence logical structure further down the processing chain, helping to answer the basic question “Who does what to whom?”. In free word order sentences of Arabic, multiple orders are possible



Brian Nolan and Yasser Salem

327

including VSO, VOS or SVO (Figure 9). The attributes of the verb define the gender of the subject. Given a masculine gender of the verb, for example, the Syntactic Parser will look for a masculine proper noun to make the actor for this sentence. If there is more than one masculine proper noun in such a case, then Modern Standard Arabic defines the first proper noun as the actor. The Morphology Parser will, in future research, be extended so that it can deal with words that are defined in multiple categories, deciding which should be processed. Meanwhile the Syntactic Parser, so far, has only been implemented for extracting word order, though it will be extended to deal with word ambiguities in future versions.

Figure 9: The linking of the Arabic clause under free word order

7.1. Lexical Properties The structure of the Lexicon including the properties stored for each word category is indicated in Figure 10. For all categories, an Arabic word is stored along with its English representation. There is an isomorphic mapping, importantly at the semantic level via the Interlingua Bridge (RRG) logical structures, from the source to the target language of noncomplex sentences that UniArab processes up to now. A level of word ambiguity is supported in the structure, with each possible case stored as a separate record. All search results will be passed to the Morphology Parser



UniArab: RRG Arabic-to-English Machine Translation

328

to decide which is taken. Since the verb is the key component when analysing using RRG, each verb has an associated logical structure (Figure 11), which is later used to determine the logical structure of the full sentence. The tense of the verb is also stored within its metadata along with the person.

Figure 10: Fragment - RRG lexical entry in XML for verb read

The verb type also stores the gender, which in Arabic must be either masculine or feminine; there is no neutral gender. The number property in Arabic can be singular, dual or plural. These properties help the Syntactic Parser analyse the sentence, since there must be agreement with the subject and verb, among other rules. We show a Java code fragment, in Figure 12, which determines the appropriate gender marking on an argument.



Brian Nolan and Yasser Salem

329

Figure 11: RRG lexicon in XML

8. UniArab Evaluations Evaluation of MT software is necessary in order to improve system performance and analyse potential problems and, of course, its accuracy and effectiveness. In the evaluation of UniArab we considered many different aspects of the MT system including quality of translation, time for translation, ability to add a new word in the lexicon of the system and resource utilization. The evaluation of MT systems is a difficult task. This is not only because many different metrics are involved, but also because translation is itself difficult. The first important aspect for a potential test is to determine the translational capability. Therefore, we needed to draw up a complete overview of the translational process, in all its different aspects. A good translation has to effectively capture the meaning. This involves establishing the size of the translation task, is it machine legible and if so, according to which standards? Current general function MT systems cannot translate all texts consistently. Output can have very poor quality. It is to be mentioned that the “subsequent editing required” increases, as translation quality gets poorer (Turian et al. 2003). Given the scale of the lexicon implemented in this work so far, we evaluate the effectiveness and accuracy of UniArab by comparison of output results against an ideal output produced by hand by a native Arabic L1 speaker. We created variants of Arabic sentences that represent all possible structures of the sentences that UniArab can translate. We then make a comparison between human-translated and machine-translated versions. At the moment, the lexicon is categorised into seven parts of speech. We have designed the GUI so that when adding a specific word to



UniArab: RRG Arabic-to-English Machine Translation

330

the lexicon, only the related options are presented to the user for that part of speech. This minimises errors when entering data. As our research extends, we expect to modify the categorisation of the lexicon to allow for more complicated word types.

Figure 12: Java code fragment that determines the appropriate gender marking

UniArab does not process ambiguous words or complex sentences, so far, in this research. This research focussed first on discovering whether the logical structure of a sentence, based on RRG can be used for translation. Hence, we decided to limit the scope of the project to exclude ambiguity resolution, since this is work in a new area that has not been



Brian Nolan and Yasser Salem

331

investigated before. We fully expect to expand the system to allow it to cope with ambiguity in the future. The system’s reliability and accuracy depends on the content of the lexicon in the XML data source and cannot handle words not in the lexicon. However, it manages this intelligently by determining the “x” and “y” argument slots in the logical structure and inserting the (unknown) Arabic nominal into the correct slot. This native Arabic word in then carried through to the English translation. to handle unknown words. UniArab does not process single words, even if those words are in its lexicon, because UniArab is built on the logical structure of verbs. The missing or unknown word can then be easily inserted into the lexicon. Therefore, for the processing of unrecognised Arabic words, where a word is not available in the lexicon, but the logic structure is recognised, then UniArab will output a correctly structured translation, but with the unknown Arabic word in its position within the English sentence (Figure 13). This makes the system resilient to slight misspellings (in nominals), which can be recognised and corrected by the human translator.

Figure 13: Processing unrecognised Arabic words

In our comparison with other translation systems we have used noncomplex sentences. While UniArab is limited to non-complex sentences and has appropriate coverage within these, we believe it is essential to reach high quality translation of these sentences in the first instance, in order to be able to expand to high quality translations of more complex sentences. We can see that the existing tools from Google and Microsoft



UniArab: RRG Arabic-to-English Machine Translation

332

cannot even achieve reasonable translations of simplex sentences, so how can we expect them to give high quality translations of larger text? We have found that small errors in the initial analysis of a sentence can cause huge errors in the final translation, so high quality analysis is very important. We have MT processing of non-complex sentences in Arabic and their equivalent translations in English. By non-complex we mean any clause that does not have a juncture relation, of any kind, in RRG terms. We have covered a representative broad selection of verbs across intransitive, transitive and ditransitive constructions in simplex sentences in active voice. Complex sentences are beyond the research scope to date, but we intend to address this in the next version. However, we do address copulalike nominative clauses in Arabic. We tested UniArab in many ways. We tested single sentences and multiple sentences. UniArab easily deals with more than one sentence as input and its output matches. That is, UniArab can accept and translate a text consisting of many sentences. Additionally, we entered random sentences together in one input or as individual sentences. In our testing and evaluation of UniArab, we subjected the UniArab System to a series of tests in a wide range of sentence categories. For each test we compared the results obtained through UniArab to those obtained when using translation engines from Google and Microsoft. We also presented a human-translated equivalent to each. In contrast, the Google and Microsoft translators gave mixed results. In many cases, sentence meaning was lacking, and even some basic constructs could not be translated. This, perhaps, is due to their focus on translating long sentences and paragraphs via statistical means rather than using semantically oriented linguistic structured to drive the translation. We highlighted this by comparing them to UniArab for longer compound sentences and found that they did indeed convey more of the meaning. These results suggest that RRG is a promising candidate for Arabic to English machine translation, and as the grammar is developed, the system should begin to cope with more complicated sentences. For non-complex sentences (intransitive, transitive and ditransitive) it clearly outperforms existing systems for the production of grammatically correct translations. In summary then, with respect to our evaluation, given the proof of concept work implemented so far, we were very careful to rigorously test and evaluate the performance of UniArab, and its accuracy in the fast production of grammatically correct sentences in the target language. We created a testbed of sets of sentences in Arabic to represent all of the



Brian Nolan and Yasser Salem

333

possible combinations of structures and possibilities for the sentences that we wanted UniArab to be able to translate. We then executed UniArab for these and compared our results with that of a human L1 Arabic translator. We also tested the Google and Microsoft automatic machine translation services with our data set of sentences to compare our UniArab results against all of these, with some very interesting and surprising results. Our testbed of grammatically correct sentences in Arabic and their equivalent translations in English have a good coverage and we tested UniArab with these. We additionally tested inputs of both single sentences and multiple sentences (as in a paragraph of Arabic text). UniArab is designed to easily deal with more than one sentence as input and its output correctly and grammatically matches.

9. The Accuracy of the Translations In this section we review the accuracy of the translations and compare the results of our system, UniArab based on the RRG linguistic model, with results from Google and Microsoft. While not rehearsing the complete set of evaluations here, in summary, our testing has include the following (8) testing criteria: (8)

UniArab Evaluation tests 1. Test-1: Copula with present progressive 2. Test-3: Verb noun - one argument in different tenses 3. Test-3: Generating the English copula verb ‘to be’ 4. Test-4: Free word order (V N N - first possibility) 5. Test-5: Free word order (V N N - second possibility) 6. Test-6: Free word order (N V N - third possibility) 7. Test-7: Pro-drop sentence 8. Test-8: Intransitive sentences 9. Test-9: DTV word order 10. Test-10: DTV word order (with prepositional phrase)

We provide indicative sample outputs, plus a screen capture that demonstrates the actual results, for a number of these in Appendix 1.



UniArab: RRG Arabic-to-English Machine Translation

334

10. Summary In this paper we have presented an Arabic-to-English machine translation system called UniArab, based on our implementation of an Interlingua Bridge framework that was programmed in Java with the lexicon built in XML, and which is based on the Rule and Reference Grammar model. We detailed the design of the system and how it was built to accommodate specifics of the Arabic language and the generation of English translations.

Figure 14: The RRG Interlingua for UniArab

We presented a high-level view of the system framework and defined our evaluation criteria for measuring system performance. We also talked about the challenges of machine translation, with a specific focus on those specific to the Arabic language. The main topic of investigation is the development of a framework for translating Arabic to English based on RRG. The framework is designed to demonstrate the capabilities of RRG as a base for machine translation. This work has shown that RRG facilitates the translation process from a specific source human language to other target languages.



Brian Nolan and Yasser Salem

335

Appendix-1: Evaluation test results Evaluation Test-1: Copula with present progressive In test-1, the output of the Google translator is faulty in respect of the tense and the marking of the “V-ing” form in English, and the non-use of the copula verb “to be”. Microsoft's MT failed to translate most of the sentence with respect to tense, copula vs. matrix verb and word order. UniArab successfully translates the sentence entirely. We show this sentence output correctly in the UniArab system.

Figure 15: Evaluation Test-1: Copula with present progressive

For each of the other diagnostic tests, we show the results following. It is easy to see the UniArab output in contrast to the other translation engines.



UniArab: RRG Arabic-to-English Machine Translation

336

Evaluation Test-2: Verb noun - one argument in different tenses

Figure 16: Evaluation Test-2: Verb noun - one argument in different tenses



Brian Nolan and Yasser Salem

Evaluation Test-3: Copula Verb ‘to be’ Figure 17: Evaluation Test-3: Copula Verb “to be”



337

UniArab: RRG Arabic-to-English Machine Translation

338

Evaluation Test-4: Free word order (V N N - first possibility)

Figure 18: Evaluation Test-4: Free word order (V N N - first possibility)



Brian Nolan and Yasser Salem

Evaluation Test-5: Free word order (V N N - second possibility)



Figure 19: Evaluation Test-5: Free word order (V N N - second possibility)



339

UniArab: RRG Arabic-to-English Machine Translation

340

Evaluation Test-6: Free word order (V N N - third possibility)

Figure 20: Evaluation Test-6: Free word order (V N N - third possibility)



Brian Nolan and Yasser Salem

Evaluation Test-7: Pro-Drop

Figure 21: Evaluation Test-7: Pro-Drop



341

UniArab: RRG Arabic-to-English Machine Translation

342

Evaluation Test-8: Intransitive sentences

Figure 22: Evaluation Test-8: Intransitive sentences



Brian Nolan and Yasser Salem

Evaluation Test-9: DTV word order

Figure 23: Evaluation Test-9: DTV word order



343

UniArab: RRG Arabic-to-English Machine Translation

344

Evaluation Test-10: DTV word order (with prepositional phrase)

Figure 24: Evaluation Test-10: DTV word order (with prepositional phrase)



Brian Nolan and Yasser Salem

345

References Al-Sughaiyer, I. A. and I. A. Al-Kharashi. 2004. Arabic morphological analysis techniques: A comprehensive survey. JASIST 55.3: 189–213. Alosh, Mahdi. 2005. Using Arabic: A Guide to Contemporary Usage. Cambridge: Cambridge University Press. Arciniegas, F., 2000. XML Developer’s Guide. McGraw-Hill Companies. Attia, M., 2004. Report on the introduction of Arabic to ParGram. Proceedings of ParGram Fall Meeting, Dublin, Ireland. Attia, M. A., 2008. Handling Arabic Morphological and Syntactic Ambiguity within the LFG Framework with a View to Machine Translation. Unpublished Ph.D. Dissertation, University of Manchester. Bateson, M. C. 2003. Arabic Language Handbook. Georgetown: Georgetown University Press. Bray, T., J. Paoli, C. M. Sperberg-McQueen, E. Maler, and F. Yergeau. 2008. Extensible Markup Language (XML) 1.0 (Fifth Edition). Brown, P., S. D. Pietra, V. D. Pietra and R. Mercer. 1993. The mathematics of statistical machine translation: parameter estimation. Computational Linguistics 19.2: 263–311. Holes, C., 2004. Modern Arabic: Structures, Functions, and Varieties. Georgetown: Georgetown University Press. Hutchins, W. J. 2003. Machine Translation: General Overview. Oxford: Oxford University Press. Hutchins, J. and H. L. Somers. 1992. An Introduction to Machine Translation. London: Academic Press. Izwaini, S. 2006. Problems of Arabic machine translation: evaluation of three systems. The British Computer Society (BSC), London, 118–148. Nolan, Brian (ed.) 2004. Linguistic theory and practice: description, implementation and processing. Proceedings of the RRG 2004 Dublin Conference. Available from the RRG website at: http://linguistics. buffalo.edu/people/faculty/vanvalin/rrg.html. Owens, J. 2006. A Linguistic History of Arabic. Oxford: Oxford University Press. Pustejovsky, James. 1995 The Generative Lexicon. Cambridge, MA: The MIT Press Ramsay, A. and H. Mansour. 2006. Local constraints on Arabic word order. 5th International Conference on NLP, FinTAL 2006, 447–457. Ryding, K. C. 2005. A Reference Grammar of Modern Standard Arabic. Cambridge: Cambridge University Press. Salem, Yasser, Arnold Hensman and Brian Nolan. 2008a. Implementing Arabic-to-English Machine Translation using the Role and Reference



UniArab: RRG Arabic-to-English Machine Translation

346

Grammar Linguistic Model. In Proceedings of the Eighth Annual International Conference on Information Technology and Telecommunication (IT&T 2008), Galway, Ireland. Available at: http://www.ittconference.ie/prevconf/2008/Conference%20Proceedings %20ITT08.pdf Salem, Yasser, Arnold Hensman and Brian Nolan. 2008b. Towards Arabic to English MT. ITB Journal 17: 20–31. Salem, Yasser and Brian Nolan. 2009a. Designing an XML lexicon architecture for Arabic machine translation based on Role and Reference Grammar. Proceedings of the 2nd International Conference on Arabic Language Resources and Tools (MEDAR 2009), Cairo, Egypt. —. 2009b. UNIARAB: a universal machine translator system for Arabic based on Role and Reference Grammar. Proceedings of the 31st Annual Meeting of the Linguistics Association of Germany (DGfS 2009). Schulz. Eckehard. 2005. A Student Grammar of Modern Standard Arabic. Cambridge: Cambridge University Press. Trujillo, A. 1999. Translation Engines: Techniques for Machine Translation. London: Springer. Turian, J. P., L. Shen, and I. D. Melamed. 2003. Evaluation of machine translation and its evaluation. Proceedings of the MT Summit IX, New Orleans, USA, 386–393. Van Valin, Robert D. Jr. (ed.) 1993. Advances in Role and Reference Grammar. Amsterdam and Philadelphia: John Benjamins. —. 2005. Exploring the Syntax-Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Versteegh, K. 2001. The Arabic Language. Edinburgh: Edinburgh University Press.





LANGUAGE INDEX A Anyin 146 Arabic (Modern) 177, 312-312, 317329, 331-334 Arabic, Classical 313 Atlantic Creoles 145 B Bamileke 159 Bengali 42, 44-45, 51-53 Bonggi 3, 7-8, 14, 17-18, 20, 22, 26, 28-30 C Catalan 256, 267 Chinese 256-258 Cora 219 Corachol 219, 232, 236 Czech 267 D Danish 267, 273 Djaru 42, 44-45, 51-52 Dutch 299-300 Dyirbal 248 E English 2-3, 5, 13, 30, 80-82, 105, 123, 141, 150, 159, 161, 177, 181-183, 190, 206, 218, 232, 240, 247, 251, 254, 256, 267-268, 312314, 317, 323-324, 326-327, 331335 Estonian 42, 45-46, 53-54 F Fa d’Ambu 152 Fijian 223 Finnish 6, 37, 42, 45-46, 53-54 French 105, 107, 108-111, 114, 118, 120, 129-131, 267

G Gă 161 German 105, 109, 112, 114, 117, 119, 123, 169, 171, 178-179, 181, 183-185, 187-188, 190-191, 193, 267 Greek 112 Guarijio 218-219, 223-226, 228, 235 H Haitian Creole 151, 156 Halkomelem 37, 42, 44 Hindi 42, 48-49, 54-56 Huichol 219, 222, 227-229, 234-235 Hungarian 267 I Icelandic 256 Igede 160 Inuktitut 42, 46-47, 53-54 Irish 64-66, 70, 72-73, 76, 79-86, 82-85, 88, 91-98 Italian 105, 107, 109-111, 114-116, 118-120, 124, 267 J Jabêm 149, 158, 162-163 Jamaican 161 Japanese 37, 258, 266-269, 271-272, 274, 276-277, 288-289, 291 K Kabardian 36-37, 42-44, 50-51, 56, 58 Kana 159 Keo 162 Khmer 158 Koiari 149, 153 Korean 267, 291 Kristang 160

348

Language Index

L Lakhota 248 Latin 105 Lealao Chinantec 223 Loniu 162 M Mayo 219 Melanesian Pidgins 145 N Nahuatl 219, 221-222, 229, 232, 236 Ngambay 5 Nyungar 42 O O’odham 219, 229-230 P Paamese 148, 160 Paicî 160 Palauan 42-44, 50-51, 56 Pima 219-220, 228-229 Pipil 219 Portuguese, Brazilian [BP] 104-107, 110-111, 113-114, 117-119, 121-123, 125, 129-131 Portuguese, European 107, 111 S Sakao 149 Saramaccan 146 Sesotho 267 Sign Language, American 251 Sign Language, British 247



Sign Language, Taiwan [TSL] 246, 248 Skou 145 Spanish 105, 107-110, 112-114, 116, 118-120, 123, 129, 131, 134-136, 141, 200-201, 206, 208-210, 212, 214-215, 229, 294-296, 298, 300, 303, 305, 307 Sranan 159 Swahili 6 T Tagalog 42, 47-48, 248 Taracahitan 219, Tarahumara 219, 222-224, 229 Tepehuan, Northern219 Tepehuan, Southern219-221, 229 Tepiman 219, 232, 236 Thai 148, 159-161 Tinrin 149, 151 Toba Batak 267 Tubar 219 Turkish 258 Twi 158, 162 U Uto-Aztecan 217-220, 223, 232, 242 V Vietnamese 162 Y Yagnob 42, 44, 50-51, 56, 58 Yaqui 217, 219-220, 222-228, 232, 234-239 Yoruba 159-161

SUBJECT INDEX A accusative 22, 36-39, 41-49, 51, 53-57, 88, 93, 188-189, 191-192, 203, 219, 225, 284, 288-289 acquisition, language 182, 247, 294294, 299, 302, 305, 307 activity calculus 104, 106 actor, see macrorole Actor-Undergoer Hierarchy 21, 66, 82, 85, 141, 154, 173, 175, 177, 183 adjective 3, 13, 30, 70, 76-79, 88, 90, 317 adjunct 14, 22, 29, 37, 44, 58, 106, 118, 129, 187-188, 191, 200-201, 203-205, 207-215, 232, 248, 269 affix 2-3, 5-6, 8, 18, 28, 30-31, 70-74, 86, 98, 179 derivational 30 inflectional 3, 30-31, 74 adverbs 76, 79, 225, 227, 249, 317 adverbial clauses, see subordinate clauses agentive role 81 (see also qualia theory) Aktionsart 2, 7, 11-14, 18, 25-26, 29-30, 81, 136, 169-170 accomplishment 7, 9, 12-13, 18, 25, 28-30, 81, 120, 136, 139, 170, 173 achievement 7, 10-12, 18, 2528, 81, 120, 135-136, 139, 170 active accomplishment 12, 2930, 136, 170, 173 activity 7-8, 10-12, 18, 21, 2526, 20-30, 81, 106, 123, 125126, 135-136, 138, 170-171, 208, 221, 223

semelfactive 170 state 7-8, 11-12, 18, 21-22, 25, 30, 81, 123, 135-136, 139-141, 170-172 ambiguity resolution 331 animacy 49 anticausative 104-107, 110, 112, 114, 118-131 anti-exhaustive listing, see exhaustive listing antipassive 124, 183 API 325 applicative 214-215 argument 11, et passim semantic 21, 139, 149, 203-204, 209-210, 314, 323-324 sharing 148-149 syntactic, see core arguments argument-adjunct 200-201, 203-205, 209, 211-212, 215, 218 argument realization 100, 169-170, 175-176, 183, 185, 189 argument structure 20, 108, 144, 150, 168-169, 171, 175-180, 183-185, 192-193, 201, 314 aspect 4, 11, 15-19, 25-26, 28-29, 31, 85, 87, 141, 144, 221, 234, 237, 246, 248, 250, 257, 323 attributes on the signature 83 automatic language translation 316 B broad consonant 91 bekommen-passive, see passive C case marking 22, 35-36, 41, 43, 45, 49-56, 58, 88, 185-186, 188-189, 191-192, 204

350

Subject Index

causation 117-118, 123 external 105, 108, 110, 122, 126, 131 internal 108, 112-113, 118, 122123, 125, 131 causative construction 104, 106-107, 109 circumfix 70, 86 clause 6, et passim layered structure of 14, 64, 66, 97, 202-203, 250, 268, 270, 295 clause chaining 143, 145, 149-150, 153-154 clause linkage 157, 217-219, 225, 229-230, 242 clause linkage marker 149, 154, 219, 224, 229, 242 clitic 48-49, 54-56, 66, 108, 115116, 119, 134-137, 141, 145, 206, 214-215, 296-297, 301-302, 304305, 307 complementation 230, 232 Completeness Constraint 20, 186 complex sentences, see juncture; nexus compounding 5, 65, 78-79 computational linguistics 313 conceptual frames 326 conjunction reduction 183-184, 209 consonant group 91 constituent projection, see projection constitutive role 80 (see also qualia, qualia theory) construction 22, et passim body 69-70 output 78, 85 repository 65, 67-68, 83 signature 67-68, 83, 187-188, 191 workspace 69, 84 Construction Grammar 106, 169, 171, 180 constructional idiom 180, 182 constructional schemas 66, 69, 79, 83-84, 97-98, 104, 106, 127-129, 131, 168-169, 171, 177, 179,



183-194 constructional template 65, 76, 98, 180 control relation 217, 219, 241 controller, see privileged syntactic argument [PSA] copula 188, 191, 250, 323-324, 332-333, 335, 337 core 14, et passim core argument 㻃 22, 24, 155-156, 187-188, 190-191, 200-201, 203207, 209-210, 212, 215, 221, 230, 232, 249, 290, 323-324 direct 22, 24, 203, 206, 210 oblique 200-201, 203-207, 209, 215 D dative 36-37, 39, 41-45, 47-52, 54-58, 88, 93-94, 117-119, 123, 129-131, 191-192, 206, 214-215, 228 dative construction 214 definiteness 45, 49, 85, 87, 93 deictic verb 148 dependant-marking 258 derivation 4, 25, 70, 86, 107, 317 zero 4, 73 determiner 90, 93-95 direct object, see object directional 87, 222, 250 ditransitive 57, 180, 332 construction 49, 176-177, 179180, 183, 185, 190-192, 332 document object model 325 double-marking 36, 44 E eclipsis 85, 94-96 endocentric 72 ergative 36-39, 41-51, 53-58, 174, 203 evaluation test 333, 335-344 evidential 87, 250 exhaustive listing 281, 287 anti- 281-282 exocentric 72

New Perspectives in Role and Reference Grammar

351

 exponent

5, 7, 20, 25, 28, 31

F feminine 55, 90-91, 93, 317, 328 focus 12, et passim contrastive 267, 269, 274-276, 279, 281 narrow 191, 266, 268-272, 274276, 278, 281-283, 287, 289, 291, 298, 300 predicate 269-270, 275, 282-284 restrictive 267-268, 274, 276278, 280, 282, 291 sentence 269-271, 277, 283, 287, 289, 296, 298, 300, 305, 307-308 focus domain 271, 298 actual 285 potential 268, 278 focus structure 12, 188, 191, 266269, 273-274, 278, 283, 285, 290-291, 298 defocusing effect 267, 282-284 focusing effect 286, 290 subordinate 266-268, 273 form-function correlation 179, 185 formal role 80 (see also qualia theory) frame-based lexeme constructions 84 free word order 313, 318, 327, 333, 338-340 Functional Grammar 202 fusion 76 morphological 73, 86, 93, 98 G gender 4, 7, 55, 72, 85, 87-88, 91, 93, 317, 321, 324, 327-328, 330 generation 59, 313, 316, 323-324, 334 language 324 sentence 323 syntax 323 generalized semantic roles, see macroroles

genitive 36-39, 41-56, 88, 90, 93-94 gradience 181 grammatical relation 21-22, 174, 204 (see also privileged syntactic argument) grammatical words, see words grammaticalization 184, 193, 248 H haben-passive, see passive head-marking 219, 221 I iconicity 201, 203 diagrammatic 45 idiom 181 illocutionary force 10-11, 15, 17-18, 25-26, 28, 87, 188, 191, 250, 257 incorporation 65 infix 8-11, 17, 19, 26, 28, 31, 70, 86, 98 inflection 4, 25-26, 55, 64-66, 70-72, 74, 85-88, 96-98, 246 information structure 294-295, 298, 300, 305 inheritance hierarchy of relations 83 inheritance network 98 Interclausal Relations Hierarchy 233 interclausal semantic relations 239 intransitive 50, 105-111, 113-115, 117-119, 121, 125, 127, 151, 173, 175, 179, 183, 185, 206, 209, 215, 296, 332-333, 342 Interlingua Bridge Framework 313, 334 J Java programming language 320 juncture 150, 157, 230-232, 332 (see also juncture- nexus relations) clausal 157, 230 core 150, 157, 230 nuclear 150, 230

352

Subject Index

ad-nuclear 234 juncture-nexus relation 219, 232 clausal coordination 231 clausal subordination 232 ad-clausal subordination 232, 237 core cosubordination 232 core subordination 232 ad-core subordination 217, 232, 237 nuclear cosubordination 217, 231-232, 236 ad-nuclear subordination 234 L left-detached position [LDP] 268, 270-271, 278, 280-281, 284, 288-289 lenition 85, 94-96 Lexeme Constructional Schema 79 lexeme formation, see morphology, lexeme lexeme store, see morphology, lexeme lexeme template inventory, see morphology, lexeme Lexical Constructional Model [LCM] 179-180 lexical decomposition 7, 11, 154, 170, 314 Lexical Functional Grammar [LFG] 202, 204 lexical rule 3, 13, 30, 134, 136-138, 141, 314 lexicon 2, et passim linking 3, et passim semantics to syntax 20-23, 25, 28, 31, 88, 105, 127-128, 145, 156-157, 168, 170, 174-175, 184-187, 189-190, 192-193, 289, 322-323 syntax to semantics 20, 23-25, 28, 290-291, 312, 320, 322 linking algorithm 16, 24, 66, 128, 171, 248, 269, 288-291, 312, 314, 320, 323



logical structure 11-13, 17-18, 20-22, 24-26, 30-31, 67, 69, 73, 82-84, 98, 136-139, 141, 154, 156, 170-176, 179, 183, 186, 207, 217, 242, 289-290, 312-315, 317, 321-324, 326-328, 330-331 M macrorole 21-22, 24, 38, 59, 128, 134, 140-141, 145, 154-155, 157, 170, 173-177, 179, 183, 193, 290 actor 9, et passim (Default) Assignment Principles 21, 139, 154, 185 undergoer 9, et passim non-macrorole 37, 44, 50-51, 5758, 141, 155 masculine 55, 91, 93, 317, 324, 327-328 meaning construction 64-66, 70-74, 76, 80, 84-85, 97-99, 179, metadata 312, 326, 328 meta-representation 312, 317, 323 metaphor 180 metonymy 180 middle 136-137 modality 4, 15, 18, 26, 87, 246246-248, 250-254, 256-258, 260 deontic 222, 227, 246, 251, 253254, 256-260 double 248, 256 epistemic 246, 250-260 root 87, 250 morphology 2-7, 16-17, 20, 26, 28-31, 35, 46, 58, 64-65, 67-68, 70, 80, 87, 97-99, 112, 129-130, 188, 191, 225, 288, 320, 322- 324, 326-328 base 2, et passim derivational morphology 181 inflectional morphology 3, 5, 7, 24-25, 29, 64, 71, 86 inflectional rules 2, 30 lexeme 3, et passim formation 71 store 64-65, 74, 76, 84, 97-

New Perspectives in Role and Reference Grammar

353

 98 template inventory 76 morpheme-based approach 3, 56, 30-31 morphemes 2, et passim morphological rules 17, 20, 30 morphosyntactic category 4, 18 morphosyntactic feature 2, 4-5, 8-11, 18, 23, 28, 30-31 morphosyntactic property 4-7, 18, 20, 25, 203, 269 PƗnini principle 28 paradigm-based approach 3, 17, 24, 28 process-based approach 2, 30 realizational approach 2-3, 5-6, 16, 20, 29-31, realization rule 18-19, 25 root 2, 5, 7-11, 13, 18-19, 28-30, 70, 73-74, 317-318 stem 4, 10-11, 18-19, 26, 28, 31, 70-71, 73, 87, 179, 222 verb morphology 2, 29 word-and-paradigm approach 5-6 word-form 4, 17 morphosyntax 69, 323-324 multi-clausal construction 143, 145, 149 multi-verb construction 154 mutation 5, 85, 94-96 N negation 6, 87, 213, 222, 225, 232, 235, 237, 248, 250, 299-300 neutralization 37 nexus 230-231 coordination 144, 149, 184, 217, 230-231, 235 cosubordination 217, 231-233, 236 subordination 144, 149, 217218, 230-232, 234, 236-237 nominative 4, 6, 22, 36-37, 39-41, 44-45, 48-49, 51, 54-55, 88, 93, 188-189, 191-192, 203, 219, 236-

237, 267, 269-270, 284, 288, 332 non-macrorole, see macrorole non-manual feature 246-247, 252253, 255-256 noun phrase 87, 291 layered structure of 64-65, 88, 97 nucleus 14-15, 202-203, 230-231, 234, 249, 257, 290, 323-324 complex 150-151, 153 number 4-7, 17, 55, 85, 87-88, 317, 321, 328 O object 48, 58, 88, 124-126, 131, 149, 155, 174, 284, 299-300, 304, 318 direct 88, 106, 174, 211, 272, 299, 305 indirect 155, 191, 215 primary 215 operator 15, 18, 25-26, 31, 86-88, 189, 192, 219-222, 224, 227, 230-233, 235, 237, 248-251, 254, 257-258, 260 clausal 87, 249-250, 254, 257 core 249-250, 254, 257 nuclear 15, 250 P passive 22, 77, 104, 131, 135, 137, 174, 177-179, 183, 187, 210 bekommen 181 haben 181 reflexive 104, 106, 122, 126 periphery 14, 185, 191, 201-205, 207-212, 232, 237, 249, 266-268, 289-291 phonological word, see word pragmatics 67-70, 129-130, 188, 191, 248, 294 discourse 294-295 post-core slot [PoCS] 187, 190, 268, 288-290 pre-core slot [PrCS] 187, 191, 268, 270-271, 278, 280-283, 288-290 prefix 6, 8-11, 17, 19, 26, 28, 70, 74,

354

Subject Index

76-77, 79, 86, 88, 187, 190, 318 preposition 22, 43-44, 70, 88-89, 93, 200-201, 203-209, 211-212, 215, 218, 229 complex 218, 229 compound 88 non-predicative 204-207, 211 predicative 201, 204, 208, 212 primary object, see object privileged syntactic argument [PSA] 22, 32, 66-67, 70, 96, 128-130, 136, 138-141, 156, 177, 185-186, 188-192, 204, 206, 209, 289 controllee 232, 236, 241 controller 204, 208, 210, 226, 241 pivot 204, 210-212, 214, 240, 307 productivity 168, 181, 183-184 projection 15, 248 constituent 14-15, 74, 86, 150, 233, 249, 251 focus structure 267, 269, 290-291 operator 15-16, 31, 86, 233, 248-251, 258, 260 proposition 191, 231, 246-247, 250251, 270, 282, 300, 318 proto-agent 123 proto-patient 123 Q qualia 65, 73, 81, 84 qualia theory 80, 84 R reanalysis 181 referential dependency 219, 241 resultative 120, 173, 182-183 right detached position [RDP] 268, 288-290 S semantics 7, et passim serial verb construction [SVC] 143146, 148-155, 157 asymmetrical 144



minor verb 144-148, 155, 157 main verb 144, 146-148, 153, 155 symmetrical 144 status 250, 254, 257 subject 5, et passim subordinate clause 206, 214, 218, 232, 237 adverbial clause 218, 225, 227, 229-230, 232 complement clause 218 subjectivity 246, 248, 251, 254-256, 260 suffix 2, 5-6, 17, 20, 36, 42, 50, 70, 74, 80-83, 86, 90, 96, 221-223, 225, 227-228, 235, 318 switch reference 153 syncretism 35-37, 39, 41-50, 53, 55, 57-58 case 35-37, 39-40, 42, 44, 48-49, 58 syntactic change 184 syntactic template 23, 128, 185, 190, 192, 213, 289 selection principle 129, 186, 289 syntax 4, et passim syntax-semantics interface 143, 168, 200 T telic role 80 (see also qualia theory) telicity 120-121, 123, 126, 169 thematic relation 170-176 addressee 144, 146, 157, 189 agent 30, 57-58, 80, 108, 125, 144-146, 151, 155, 158, 161-162, 170-173, 175, 188-189, 191-192, 210, 225 allative 39, 228 benefactive 48, 58, 191, 215, 228, 237 beneficiary 144, 146-147, 157, 228 comitative 37, 58 instrument(al) 27, 37, 39, 48, 58, 108, 146, 157, 226, 228, 237

New Perspectives in Role and Reference Grammar

355

 malefactive 191 patient 57, 123, 125-126, 131, 144, 170-173, 175, 188-189, 191-192 recipient 48-49, 57-58, 144-148, 155, 157-158, 161-162, 173, 177, 185, 191, 206, 214, 228 recipient-like R 144 R-type serialized P 145-147 source 144, 146, 157, 206-207, 212 theme 57, 140, 144-147, 154158, 161, 170, 172, 191, 210, 226 theme-like T 144 T-type serialized P 144-146 thematic relations continuum 171-172, 176 Theta-System 123 topic 188, 191, 266-278, 281, 287291, 294, 299-308 contrastive 266-267, 270, 274275, 279-280, 291 type coercion 84 U unaccusative 108-109, 129-130, 294298, 300, 305, 307 unergative 109, 130, 295, 300, 308 UniArab 312-314, 316-317, 320-322, 324-335

V valence 129, 171, 178-179, 183, 187, 232 verb class 2, 25-28, 30-31, 81, 144, 173, 175, 326 vocative 49, 88 W wh-question 183, 190 word formation 4, 30, 66, 70-71, 73 word formation rule 30 word tokenizer 320-321 word grammatical 66 layered structure of 64-65, 67, 70, 72, 74, 86, 97-98 phonological 66 workspace 64-65, 68-70, 84, 168, 184, 187-188, 190-191 X XML 312-313, 317, 320, 322, 324326, 331, 334 XML data source 322, 326, 331 Z zero anaphora 270, 277-278, 286, 289- 290 zero derivation 4, 73