Richard Hudson. 1990. English Word Grammar. (Blackwell)

Richard Hudson. 1990. English Word Grammar. (Blackwell)

1

GENERAL INTRODUCTION

1.1 SOME RECENT TRENDS IN LINGUISTICS

1 should like to set the intellectual scene for this book by highlighting a number of trends that have been noticeable in linguistics during the 1980s. It would be hard to claim that the trends in this list are the most important ones for this decade, in some objective sense for instance, in terms of the number of adherents, or even in terms of the quality of the debate about them though this is clearly true of some of them. They all belong in the list for one reason: they coincide with the premises of Word Grammar. Section 1.2 willexplain In outline how this is so, and the rest of the book will expand on the details. For the present, the list is a jumble of apparently unrelated, or even conflicting,trends. Some of these trends already have names, but for others 1 have had to invent a name. The list is as follows:

a lexicalism - the tendency to shift explanation from facts about constructions to facts about words;

bwholism the tendency to minimize the distinction between the lexicon and the rest of the grammar;

ctransconstructionism the tendency to reduce the number of rules that are each specific to just one construction;

dpolyconstructionism the tendency to increase the number of particular constructions that are recognized in a grammar;

erelationism the tendency to refer explicitly to grammatical relations, and even to treat these, rather than constituent structure, as primary;

fmonostratalism the tendency to reject the transformational idea that a sentence's syntactic structure cannot be shown in a single structural representation;

gcognitivism the tendency to emphasize the similarities and continuities between linguistic and nonlinguistic knowledge;

himplementationism the tendency to implement grammars in terms of computer programs.

(a) Lexicalism. Chomsky's paper 'Remarks on nominalization' (1970) is generally acknowledged as the start of lexicalism, since it established that transformational rules are unsuitable for explaining the relations between some kinds of partially similar structures (e.g. that between destroy the city and destruction of the city). Instead, he suggested, such relations should be shown in the lexicon, as facts about the headwords concerned (i.e. about destroy and destruction). This idea was widely welcomed, and a good deal of interesting work has since been generated under the banner of 'lexicalism' (e.g. the papers in Hoekstra et al. 1980, and in Moortgat et al. 1981).

Like many other isms, however, lexicalism is a slogan rather than a coherent theory, and its main achievement is probably to have given further support to Chomsky's original claim that some syntactic relations between sentences are hard to explain in terms of transformations. There is still no clear account of partial similarities between lexical items (e.g. those between destroy and destruction); various kinds of rules have been suggested notably, redundancy rules and wordformation rules but far less attention has been paid to their scope and formal properties than to those of phrase structure rules or transformations. It is not even clear that all such similarities should be described in terms of rules (Hudson 1976b).

Nor is it clear whether the focus of 'lexicalism' is on words qua lexical items or qua wordlength units i.e. on their specificity or on their brevity. In the first case the emphasis is on the fact that, say, one verb differs idiosyncratically from another, while in the second it is on the fact that the structure of a whole sentence can be deduced from facts about the individual words in it. The individuality of words has been emphasized especially in some work e.g. on morphology but the possibility of deriving syntactic information from single words has been stressed by some theories of syntax which describe themselves as 'lexicalist,' such as Categorial Grammar (e.g. Uszkorelt 1986, Karttunen 1986) and Headdriven Phrase Structure Grammar (Pollard and Sag 1987).

(b) Wholism. Ever since Chomsky's Aspects of the Theory of Syntax was published (1965), most linguists have found it selfevident that the lexicon is a distinct component of the grammar. It is worth remembering that this assumption was by no means obvious before 1965 e.g. no such distinction is recognized in Chomsky 1957 and even in the 1980s there were linguists who questioned it. Some have done so within a broadly transformational frame of reference (e.g. Pesetsky 1985), but for others it is an important part of their general theoretical stance (e.g. Langacker 1987). The number of people who support this view seems if anything to be increasing, so we may call it a trend. 'Wholism' seems a reasonable name for the view that lexical facts are of the same type as more general facts.

The existence of the wholist trend has implications for lexicalism, if the latter depends on moving facts from the rest of the grammar into the lexicon. If there is no boundary between the two, the move is obviously meaningless. Indeed, one of the weaknesses of the lexicalist movement has been its general failure to define precisely what the lexicon is, and how it differs from the rest of the grammar (Hudson 1983). However, as we have seen it is possible to interpret lexicalism as an approach to grammar in which words are basic and the boundary around the lexicon plays no part. This view is perfectly compatible with wholism.

(c) Transconstructionism. The impetus for this movement, like that for lexicalism, also came from Chomsky, especially in his Lectures on Government and Binding (1981). Once again, the main point was to show the unsuitability of transformations for dealing with certain types of phenomena, namely for showing partial similarities among constructions. If we apply the term 'construction' to general patterns like 'sentence whose main verb is passive', then Chomsky argues that it is wrong to account for each construction separately, by means of a purposebuilt transformation (as in typical transformational analyses of the 1960s and 1970s). The weakness of the earlier approach is that it misses generalizations which apply across constructions, and therefore falls to explain the shared properties of constructions. For example, a passive transformation which moves the verb's object into the subject position misses the similarities between passive verbs and adjectives similarities of distribution, and the inability to have a nounphrase as complement. If, on the other hand, a lexical rule changes passive verbs into adjectives, then the other facts about passives can be made to follow from general principles, some of which also apply to adjectives.

This trend has also generated a great deal of very interesting work, though a lot of the arguments are based on evidence which is extremely theory internal and hard to apply to other theories. The search for crossconstructional similarities has been quite fruitful, though somewhat selective (e.g. the rather obvious similarities between 'raising' and 'subjectcontrol' verbs, e.g. seem and try respectively, remain unexplained), and no doubt the present diversity of analyses will diminish as the debate continues. The main achievement, taking a crosstheory perspective, has probably been to set a new standard for success in analysis: all similarities among constructions must be brought out and, if possible, each should be given a unitary account which applies to all the constructions concerned.

(d) Polyconstructionism. This trend is most clearly represented by the Berkeley 'Constructional Grammar' approach (Fillmore et al. 1988, Lakoff 1987). One of the characteristics of this approach is the essentially negative observation that most other linguists pay attention to a rather small subset of the total set of constructions allowed by a language, an observation which is supported by a large collection of interesting and challenging examples of structures that are not covered by most linguists' grammars (e.g. constructions containing the more ... the more ... ). More positively, they suggest that many of these constructions are such that entries for lexical items need to be allowed to refer to other items whichare not their sisters, but their 'nieces' (or even more distantly related), and that grammatical theory should be revised to make this possible. Typical examples of this need are verbs which require some particular preposition as head of a sister prepositional phrase (e.g. depend, which requires on).

This trend is not simply the converse of transconstructionism, though it emphasizes the differences among constructions rather than their similarities. It is conceivable that a grammar in the tradition of Government and Binding (GB), with transconstructional principles, could eventually be extended to include all the constructions of the language, and could do so without invoking constructionspecific rules or principles. This remains to be demonstrated, and the evidence presented so far by advocates of constructiona1 grammar certainly seems to suggest that some constructionspecific rules are necessary. Nevertheless it is perhaps best at present to see the two trends as essentially complementary: transconstructionism emphasizes the need to bring out similarities among constructions, while polyconstructionism emphasizes their diversity and the need to cover the

full range.

(e) Relationism. In traditional grammar, the structure of a sentence is defined essentially in terms of binary 'grammatical relations' (alias 'grammatical functions') such as 'subject' and 'object' between individual words, to which information about the classmembership of the words concerned may be added. Many current theories, which all derive from Bloomfield's 'ImmediateConstituent' analysis (1933), reject this view in favour of an analysis in terms of constituent structure. These theories treat grammatical relations, which are relatively abstract, as derivative rather than basic; what they are derivable from is the very simple partwhole relations of constituent structure, which is basic. However, by no means all theorists in the recent history of grammar have accepted this idea, and there seems to be a tendency for grammatical relations to be given a more prominent (and basic) place in recent theories.

One manifestation of this tendency is the almost universal tendency for constituencybased theories to distinguish the head daughter of a constituent from its other daughters. This is of course one of the main principles of Xbar syntax, on which many current theories are based, and the notion 'head' often figures prominently in discussions. It is true that the sentence structures of 'pure' Xbar theory do not show explicitly which daughters are heads, since the head can be identified only from the configuration of nodes plus their features (Zwicky 1986). However, the phrasestructure rules of Generalized Phrase Structure Grammar (GPSG) indicate explicitly which daughter is the head (Gazdar et al. 1985), and in Headdriven Phrase Structure Grammar (HPSG Pollard and Sag 1987) heads are marked as such even in sentence structures. Furthermore, Categorial Grammar (CG), which is becoming increasingly popular, distinguishes functors explicitly from arguments both in sentence structures and in the grammar/lexicon, a distinction which is closely related, at least in simple cases, to the distinction between heads and nonheads (Zwicky 1985, Hudson 1987a).

An even clearer manifestation of 'relationism' is the rise of new theories in which particular kinds of nonhead are distinguished explicitly in sentence structures and in the grammar. The theories of the 1960s of which this was true e.g. Tagmemics (Pike 1982) and Systemic Grammar (Butler 1985) have been joined more recently by, among others, Relational Grammar (Blake 1990), Functional Grammar (Dik 1978), Corepresentational Grammar (Kac 1978), ArcPair Grammar (Johnson and Postal 1980) and LexicalFunctional Grammar (Bresnan 1982a).

Yet another sign of the trend towards taking grammatical relations as basic is the increased interest in dependency theory among Englishspeaking linguists (Hudson forthcoming a). This theory has of course been familiar for a long time to linguists in Germany and Eastern Europe, and ultimately it seems to be a direct continuation of the European grammatical tradition of the past two thousand years, but until recently it has been ignored by most linguists in North America and Britain. The basic concepts such as 'modifier' and 'adjunct' to say nothing of 'head" discussed above are now introduced in elementary textbooks such as Matthews (1981), Atkinson et al. (1982) and Huddleston (1988), and a number of theories are based on dependency notions e.g. Anderson's localist theory of case (1977), Starosta's Lexicase (1988) and my own earlier theory, DaughterDependency Theory (Hudson 1976a).

(f) Monostratalism. Transformational grammar introduced the idea that a sentence's syntactic structure could best be shown by means of a series of individually quite simple structures, which are mapped onto one another by transformational rules. Although this still constitutes the dominant school of thought, there has always been a good deal of opposition to it from monostratal theories. Some of these date from the 1960s (Tagmemics, Stratificational Grammar, Systemic Grammar) and others from the 1970s (e.g. Functional Grammar, DaughterDependency Grammar) but more recent alternatives have entered the 'mainstream' literature: LexicalFunctional Grammar (LFG), GPSG, HPSG and CG.

(g) Cognitivism. The structuralist tradition of linguistics in this century has generally emphasized the differences between linguistic and nonlinguistic structures, in the belief that language is fundamentally different from everything else, whether we view it as an abstract structure or as a kind of knowledge. This tradition is particularly clear in the 'MIT School' led by Chomsky and Fodor (e.g. Chomsky 1986a, Fodor 1983).

However, according to Tanenhaus (1988), most psychologists who work on language believe that 'language processing and language acquisition can best be understood within the same framework as other cognitive processes'. Moreover an alternative view has been developing among linguists too over the past decade or so, according to which linguistic knowledge is a particular case of general knowledge. This movement has been called Cognitive Linguistics (Dirven 1988, Taylor 1989), in recognition of the fact that linguistic knowledge is claimed to share the cognitive structures found in other kinds of knowledge. The case has been argued strongly in relation to meaning, where the distinction between 'dictionary' and 'encyclopedic' meaning is at issue (Haiman 1980, Jackendoff 1983, Langacker 1985), but it can also be applied to the categories of grammar. For some time linguists have been aware of the awkward fuzziness of many of these categories (e.g. Ross 1973), but it has been argued more recently that this fuzziness shows that grammatical concepts have similar properties to nonlinguistic concepts (Lakoff 1987, Taylor 1989, which is a particularly accessible survey of the whole movement).

As suggested by the example just given, cognitivism has generally involved an emphasis on the fuzziness and openendedness of language, in reaction against what is seen as the unrealistically neat picture offered by most linguistic theories. One result of this stress on fuzziness has been a lack of a clear research paradigm for those who are sympathetic with the general principles but who still want to write grammars. However, one fruitful direction in which cognitivism has pushed linguists is towards lexicalism and polyconstructionism, since these trends stress the sheer quantity of information which a speaker must know. (This link is explicit, for example, in Lakoff 1987.) Another benefit is the link established between linguists and those cognitive psychologists who argue that cognitive categories have the characteristics of prototypes; this link has generated some very useful theoretical and empirical work (surveyed in the works referred to above), which has pointed the way to a new 'clean' paradigm in which fuzziness can be accommodated. We might grace this particular aspect of cognitivism with the name 'prototypism', though it has to be admitted that the two trends are so closely linked as to be virtually inseparable.

Cognitivism has important implications for the theory of language acquisition, because very little specifically linguistic mental apparatus has to be assumed. This is a view which Tanenhaus (1988) calls 'linguistic minimalism': Linguistic minimalists assume that there is a relatively transparent mapping between semantic or conceptual notions and grammatical structure and thus little, if any, innate linguistic structure is required. This position contrasts with 'linguistic autonomy', according to which 'the mapping is too complex for the child to accomplish without some innate linguistic knowledge or specialized skills for extracting syntactic regularities'.

These contrasting ideas about the relation between linguistic and nonlinguistic knowledge also have different implications for how linguists work. A cognitivist is constrained by the hypothesis that any linguistic construct should have nonlinguistic analogs, at some level of generality; if one is forced to recognize some kind of rule or pattern in language for which no analog outside language is apparent, then the hypothesis is to that extent disconfirmed, so a cognitivist tries hard to avoid such assumptions. In contrast a linguistic autonomist is free to postulate linguistic concepts of any kind whatsoever; indeed, the more peculiar and specialized these concepts are, the better it is for the hypothesis of linguistic autonomy. It is easy to think of examples from the literature, from 'subiacency' and the 'empty category principle' of GB, to the 'footfeature principle' of GPSG. Of course the existence of these suggestions in the literature proves nothing other than that it is possible to describe language in its own terms; they tell us nothing about whether this is necessary. The only way of proving anything about the necessity of a uniquely linguistic analysis is by trying very hard to do the opposite. Success refutes the hypothesis, but failure can be interpreted either as support for the hypothesis, or as evidence that one has not tried hard enough.

(h) Implementationism. Many linguists see the implementation of grammars on computers as a natural development of the goal of generativity a goal which, once again, was defined by Chomsky (1957). If a grammar is generative to the extent that it is explicit, as I think most linguists accept, then a fully generative grammar should be amenable to computer implementation. Of course, explicitness in itself is no guarantee of truth, and it is quite possible to waste a good deal of time over the formal details of a grammar whose fundamental principles can be shown to be wrong. Nor does the existence of a computer implementation prove anything about a non-computerized version of 'the same' grammar unless the relation between the two systems is very clear. A host of important (and interesting) questions arise in this connection in general, how can we decide which parts of a computer program correspond to theoretical and descriptive constructs about language, and which are simply implementational devices?