"Kissine offers a new theory of speech acts which is philosophically sophisticated and builds on work in cognitive science, formal semantics, and linguistic typology. This highly readable, brilliant essay is a major contribution to the field."
EDITORS: Sugayama, Kensei; Hudson, Richard A. TITLE: Word Grammar SUBTITLE: New Perspectives on a Theory of Language Structure PUBLISHER: Continuum International Publishing Group Ltd YEAR: 2005
Valeria Quochi, Istituto di Linguistica Computazionale CNR Pisa, Italy, and Dipartimento di Linguistica, University of Pisa, Italy
This book is a collection of 9 chapters by different authors, and presents research done within the Word Grammar framework from different linguistic perspectives: lexical semantics, morphology, syntax, and others. The aim is to introduce the reader recent developments in the theory, and to demonstrate how it can be applied to different languages and different language facts. What ties these works together is the idea that the dependencies between single words are fundamental for an understanding of language. This recent version of the theory presents considerable differences from the previous versions. The book is divided in two parts: the first collects 7 contributions that apply the theory to a range of phenomena and show its explanatory power; the second part addresses theoretically key issues related to the notions of head and dependency, and proposes significant innovations.
SUMMARY
The Preface, by Kensei Sugayama, is an overview of the roots, history, and theoretical basic notions of Word Grammar. Word Grammar (WG) is a dependency theory of grammar that also includes ideas from cognitive approaches to language. The Key notion in WG is the word-word dependency: both syntax and semantics are built on this notion. WG is further defined as a monostratal model of language that rejects the notion of Phrase Structure. Language is seen as a network of knowledge which is part of general cognition, and in which grammatical relations and functions are represented explicitly. The preface also provides a detailed summary of the contents of the book.
The first Chapter is a general introduction by Richard Hudson, the founder of the theory. The chapter introduces the historical background and subsequent developments of the theory, as well as its most important tenets. WG is described as a general theory of language structure, which is primarily concerned with syntax, but which also covers other areas of linguistics, except phonology and language acquisition. The central unit of analysis is the word, and sentence structure is represented in the form of dependencies between single words. Words are also the basic lexical/semantic units and the basic unit for contextual analysis. The language network is seen as composed by emergent sub-networks, each characterized by a distinct (set of) link(s). Links in the network are classified and labelled, and are treated as second order concepts, which means that they can be grouped into larger classes and can be learned exactly in the same way as ordinary concepts. Section 4, discusses the fundamental property of Default Inheritance. Inheritance hierarchies are the only means of classifying concepts, since in WG feature structures are not accepted. Multiple inheritance, additionally, allows for words to be classified along different dimensions. Utterances in WG are also represented as networks having the same formal characteristics as the permanent cognitive network, but are temporary. Section 7, 8 and 9 are devoted to the explanation of the sub-networks of morphology, syntax and semantics respectively. Finally, issues related to processing matters are discussed together with a brief introduction to the incremental parsing system used to test the theory.
Chapter 2, by Creider and Hudson, deals with case agreement in Ancient Greek and its implications for a theory of covert elements in WG. Previous versions of WG rejected the notion of covert elements completely. This analysis, instead, favours the introduction of empty elements in WG, in order to deal with special situations of case agreement in Ancient Greek, where a predicative adjective or a predicative noun agrees with the subject of its clause, even if the subject is covert. This happens regularly in infinitival constructions. A new notion of Null Element in WG is contrasted both to the PRO-element of generative approaches and to the ''potential subcat list'' of HPSG. Two sections are dedicated to the traditional description of accusative and non-accusative subjects in Ancient Greek. The traditional rule that infinitives in Ancient Greek have accusative subjects is fully accepted. Case marking of infinitives in Ancient Greek is used as a language fact that provides strong evidence in favour of the need of null elements in WG: because morphological case is considered a purely morpho-syntactic property, it is not possible to explain the data postulating a subject argument in the semantic structure alone. There needs to be a syntactic supporting subject. Finally, the WG account is shown as preferable to other accounts, in that it makes use of the general machinery available in the theory to account for the specific facts of understood entities. In particular, the author shows how a particular attribute, ''quantity'', which is generally employed to account for how types are realised into tokens in utterances, can account also for Null Elements in syntax.
Chapter 3, by Sugayama, proposes a WG analysis of Japanese null complements of transitive verbs, through a contrastive analysis of two cases: the English 'eat' and the Japanese 'taberu'. The claim is that the current WG analysis of English understood objects of transitive verbs like 'eat' is not suitable for Japanese verbs like 'taberu', because there is a semantic difference in the suppressed object between the two languages. Sugayama claims that this semantic difference is not adequately represented in the theory and proposes the introduction of an additional rule to the grammar of Japanese that would account such linguistic facts. The author gives a WG standard description and representation of the English transitive verbs like 'eat', and then an analysis of the Japanese translation-equivalent verbs with the proposed necessary innovation. From a semantic perspective, English and Japanese transitive verbs differ in the behaviour of their null object complements. While in English an understood object of 'eat'-like verbs must be unknown to the speaker, therefore indefinite, the understood object of Japanese 'taberu' must be known to the speaker. Object complements in fact are freely and regularly omitted in Japanese, as long as they are definite. The proposed innovation to WG is, thus, the introduction of two nodes for the concepts of definiteness and indefiniteness, which are then be appropriately linked to the referents of null objects.
Chapter 4, again by Sugayama, the English 'be to' construct from a WG perspective. The main claim is that the verb 'be' in the construction is a ''non-prototypical'' instance of a modal verb, that the sense of the 'be+to' construction is determined by the sense of the infinitival clause that follows, and that the 'be+to' construct is not a lexical unit because there is a syntactic and semantic gap between the two elements. The fact that the construct may have both epistemic and non-epistemic readings is accounted for in WG by representing the verb 'be' as both an instance of the lexical verb 'be' and an instance of 'modal verb'; this allows it to inherit properties from both. The sense of the construct is said to derive from the sense of the following infinitive clause, although a core sense for the construct can be identified independently of it. The very nature of the event, however, is determined by the clause that follows the verb. Evidence that supports the claim that 'be+to' is not a lexical unit: the separability of the two elements, the possible omission of 'to' in question tags, and the fact that no phonetic fusion appears to take place (cfr. 'ought to', 'want to'). The advantage of the WG analysis proposed here is that it assigns the same syntactic and semantic structure to both epistemic and non-epistemic readings: 'be' is the syntactic and semantic head of the construct and functions like a raising verb, and 'to' and the infinitive clause are represented as its sharer dependents, i.e. dependents that share their subjects with it.
In Chapter 5, Jasper Holmes attempts to formulate an adequate account of linking which exploits the machinery of WG, without the need to stipulate subcategorization representations. The main assumption is that the meaning of verbs alone is not sufficient to determine semantic structure; context also plays an important role. In this approach, linking rules are seen as generalizations over correspondences between syntactic and semantic relationships. Therefore, linking rules will link classes of syntactic dependency relations with classes of semantic associations. To explain and show to potentiality of this account the author describes and represents the behaviour of various examples of the English double object syntax and of the equivalent constructs in German. Three linking rules that account for the mappings to the three main syntactic arguments (subjects, direct objects and indirect objects) are presented and discussed. One of the major consequences of this analysis is that syntactic dependencies have meanings, and that these meanings serve to constrain the possible combinations of syntactic elements and to determine the interpretations of compositional structures. The second part of the paper is dedicated to the presentation of an Event-Type Hierarchy and its fundamental role in linking. Correlated with this topic, is also the importance and nature of thematic/semantic and argument roles. Two kinds of semantic associations are, thus, distinguished in this approach: Participant Roles and Argument Roles. Participant Roles carry thematic content and are an open-ended class; Argument roles are more schematic and limited, and are determined by the force-dynamic properties of the event to which they are associated. These are the associations that characterize the semantic sub-network.
Chapter 6 by Eva Eppler is an attempt to account for English-German code-mixing data within the framework of Word Grammar. The chapter starts with a brief introduction to code-switching/code-mixing studies, and reviews the major approaches based on phrase structure grammars to the phenomenon. The main claim here is that such approaches are not adequate enough to account for intra-sentential syntactic code-mixing phenomena because they are too restrictive, and don't account for her data. The study presented is done on a bilingual corpus of English/German spontaneous speech transcriptions and combines both quantitative (distributional) and qualitative analysis of intra-sentential code-mixing of subordinate clauses of cause ('because-' and 'weil' clauses). WG is claimed to be a more suitable framework because it rejects the notion of constituent structure. The word-word dependencies, instead, are preferred because they allow putting less restrictive constraints on the type of mixing allowed/disallowed in a language. Another advantage of WG is that it also has the possibility of integrating sophisticated sociolinguistic information about speakers within the network. Since code-mixing is strongly influenced by social factors, this characteristic is particularly appealing. The choice to analyse subordination of cause is motivated by the interesting contrast in word order presented by the two languages under investigation. The results obtained support the null hypothesis: that is, mixed codes seem to conform to the grammar of the language to which they belong.
In chapter 7 Maekawa confronts three different grammatical frameworks with the analysis of linear order asymmetries between main and subordinate clauses in wh-interrogatives, namely Word Grammar, Constructional Head Phrase Structure Grammar (HPSG), and Linearization-based HPSG. His study favours this last approach. All the three approaches are discussed in some detail, and in each case a very brief introduction to the theory-particular machinery for handling word order is given. Then a set of problematic cases of wh-interrogatives is analysed under the three approaches. The reasoning is that both WG (Hudson 1990, 2005) and Constructional HPSG (Ginzburg and Sag 2000) have some weaknesses in accounting for extraction/topicalization data, whereas Linearization-based HPSG provides a neat explanation for the English data. The claim is that the weaknesses shown by the first two frameworks strongly depend on their assumptions about grammatical structure and their machineries for dealing with word order. Linear order in such approaches is tied to the combinatorial structure, and this is claimed to prevent an adequate account of certain discontinuities like those found in topicalized wh-interrogatives. Linearized HPSG, instead, provides a straigntforward account of those facts because, through the attribute DOMAIN, linear order is kept separated from combinatorial mechanisms and local trees.
In Chapter 8 Rosta claims that two different types of heads need to be recognized in syntactic structure, namely Structural heads and Distributional Heads. The distinction is needed in order to properly account for a number of data in which the distribution of a phrase is determined not by its head, but by one of its dependents. This behaviour is called ''hypocentricity'' by the author. After a definition of the two notions of structural and distributional heads, Rosta analyses various English hypocentric constructions: that-clauses, extent-operators and other adverbials like 'just' and 'only', pied-piping, degree words, attributive adjectives, determiner phrases, the 'type-of' construction, ''inside-out'' interrogatives. Such analyses are used to introduce and discuss the proposed innovations to the theory, in particular, the modification of the notion of dependency relation and the introduction of empty categories. The advantages of these innovations are also discussed in relation to the treatment of adjuncts, subject relations in conjoined predicates, and complements. Given the wide range of language fact that show the same behaviour, hypocentricity is held to be pervasive in language, and therefore, the new version of the theory seems necessary. The new theory is claimed to be simpler, in that it reduces the range of mechanisms fot representing syntactic structure, but is still consistent with the basic tenets of the original version. In this revision syntactic structure is seen as a network of words linked by unlabelled branches, forming a tree; and additional supportive links are needed to deal with discontinuities in syntax.
Chapter 9 by Gisborne addresses the theoretical notion of subject dependency, and argues in favour of a three-way distinction in subject-dependency types. The paper focuses on the analysis of the so-called Locative Inversion data, and claims that in such cases the subject properties are split between the locative- and the noun phrase. First, the main properties of subjects as discussed in the literature are reviewed, and some of those properties are retained as useful diagnostics of subjecthood. Semantic Role, argumenthood of intransitives and co-reference are the three diagnostics considered the most useful general diagnostics, while subject-auxiliary inversion, agreement and extraction are held to be the most critical diagnostics for English. Then, the author thoroughly reviews Bresnan's (1994) LFG account of locative inversion data. Such data is taken as evidence of the fact that in locative inversion constructions several subject properties can be found both on the locative complements and on the theme noun phrase. The distinction in three types of subject proposed in the LFG account is also judged useful for an adequate description of the fact, and therefore the author proposes a revision of the theory of Word Grammar in this sense. The problem of standard WG is that it has only one domain of structure, i.e. the dependency relations, and only one type of subject dependency. Instead, the analyses previously presented demonstrate the need to have at least three types of subjects-- lexical, syntactic and morphologic subjects--by virtue of the different properties that characterize them. The WG account is, finally, claimed to present more advantages with respect to LFG in that it posit a simpler structure: while LFG treats the distinction within 3 domains of structure, WG deals with them in one single domain.
CRITICAL EVALUATION
The book as a whole gives an interesting insight into the state-of-the-art Word Grammar framework and the possible fields of application other than syntax. The contribution by its founder is valuable, in that it puts the theory into perspective and projects it into the future challenges of the language science field. The chapter dedicated to the introduction of empty elements in syntax is very interesting and courageous, considering that it goes counter to previous claims of strict adherence to the facts, and WG's nature as fully surfacist theory. Although the analysis is fascinating, the issue remains controversial, and though the argumentation is well structured and supported, it is still not fully convincing.
Chapter 3 on the treatment on Null Complement in Japanese provides an interesting solution for the Japanese facts. However, it is not so clear how and why the definiteness parameter would be set differently for different languages, or why it should be systematically overridden in English. Moreover, the argumentation on the advantage of WG for verb final languages with free word order, because of the final position of the head of all dependencies is not very clear and seems to suggest that for languages with different characteristics a WG approach is not advantageous.
Chapter 4, on the 'be+to' construction is an interesting attempt to account for both syntactic and semantic features of a construct; however, it shows a few weaknesses, and all in all does not seem very convincing. The examples brought to support the claim look like newspaper headlines, and therefore most probably elliptical or hyperbolic sentences, where the 'be' element could be said to be understood. The diagrams are often difficult to read, and the explanations provided are sometimes not very helpful.
The same problems with diagrams and explanation are also present here and there in chapter 5, on linking in WG. The chapter is however very interesting, for it treats such an important area of linguistics. However, it seems a WG 'translation' of a Construction Grammar approach to linking (Goldberg 1995, 2002). Of particular interest is the criteria on the basis of which the two different types of roles are distinguished, which also exploit the force-dynamic properties of events.
Chapter 6 on code-mixing is a difficult paper to follow. The terminology used is not clearly defined and the basic notions on code-mixing are not well explained. Section 3 in particular seems not very well structured: the author continuously mixes theoretical motivations with the methodology and with part of the analysis.
Chapter 7 is well structured and interesting in the comparison between three approaches on exactly the same data. The introduction to the basics of the theories, esp. of HPSG, however, is only sketchy. Therefore, the chapter is difficult to understand in details for anybody with no previous knowledge of HPSG.
Chapter 8 presents considerable modifications to the theory, and, although it claims to stay close to the original tenets, the outcome looks far from current WG. The idea of distinguishing between structural and distributional heads is stimulating, but the elimination of dependency types and the introduction of ''supporting'' relations that would account for hypocentricity appear controversial.
Chapter 9 also deals with theoretical issues, related to the very notion of subject relation. The proposed innovations to the theory are motivated by evidence coming from an LFG analysis of split-subject in locative inversion constructs and other recent studies. The distinction into three types of subject in the WG analysis, however, seems to force the linguist to make decisions as to whether a particular phenomenon is syntactic, lexical or morphological, which is often controversial.
REFERENCES
Bresnan, J.W. (1994) 'Locative Inversion and the Architecture of Grammar'. Language, 70, 72-131.
Ginzburg, J. and I.A. Sag (2000) Interrogative Investigations. Stanford: CSLI.
Hudson, R. A. (1990) English Word Grammar. Oxford: Blackwell.
--- (2005) Word Grammar. http://www.phon.ucl.ac.uk/home/dick/wg.html
ABOUT THE REVIEWER:
ABOUT THE REVIEWER
Valeria Quochi is a Ph.D. Student in Linguistics of the University of Pisa,
Italy. She graduated in English and German Language and Literature at the
University of Pisa. Among her interests are Computational lexicography,
cognitive, data-driven approaches to language, the syntax-semantic
interface. She is currently working on the acquisition of semi-productive
constructions in Italian, from a constructional perspective.
