A spectator's guide to syntactic theories

Links & Letters 1, 1994 25-50 A spectator's guide to syntactic theories Richard Hudson University Coiiege London Novernber 1992 ABSTRACT This article' is intended for non-speciaiists who would like to understand the state of play in syntactic theory. It introduces nine different syntactic theories which count as 'important' in some sense, and explains some of the assumptions that they make about sentence structure. It aiso discusses the various kinds of solutions that have been offered for one problem, that of discontinuities produced by topicalisation, and introduces a tenth theory which rests on fundamentally different assumptions. l. The aim of the garne Current work on syntax is like a garne of football, in which each good idea counts as a goal. There is just one set of goal-posts (though from time to time a group of players may try to shift them), and one goal-keeper: The Facts. The balls which the players try to get past the goal-keeper are theories, rules, principles - ideas about the workings of grammars of particular languages or of languages in general. In principle, the winner at any given time is the person or group with the highest number of goals scored; but in reality there's some confusion not only about the scoring system but even about which balls have scored at all. There's no referee, and new facts are coming to light al1 the time, so some of today's apparent goals may be disqualified by tomorrow's new facts. Most people who are actually playing the game seem to enjoy it - why else would they be playing? However, inexperienced spectators can find it al1 rather confusing, and it is for them that this article is intended. LetS start by looking at the balls. Unlike real football, there isn't just one, but an unlimited supply (so in this respect the game is more like playing with a pin-ball machine than football). Another difference is that the players 1. My thanks to And Rosta, Anat Ninio and an anonyrnous reviewer for comments on the first draft of this paper. 26 Links & Letters 1, 1994 Richard Hudson themselves have to make the balls before they try to kick them through the goal-posts, because each ball is an idea. For example, suppose you were working on the syntax of questions in English and you had noticed examples like the following. (1) a. It is hot in Barcelona. b. 1s it hot in Barcelona? You know that is is a verb, and that it is its subject, so your first idea might be that you form questions by putting the verb before its subject. Unfortunately this won't score a goal, because it comes up against the fact that the question c~rrespondin~ to (2a) is (2b) and not the form predicted by your idea, (2c) (which is ungrammatical, as shown by the *). (2) a. It gets hot in Barcelona. b. Does it get hot in Barcelona? c. * Gets it hot in Barcelona? Something is wrong with this idea, so you have to change it into a different one, and you go on changing it until it gets past the goal-keeping facts. It doesn't matter if someone else has already scored with that ball; we often reinvent one another's ideas, or reuse them, and in the long run al1 that counts is whether an idea is true, and not where it came from or who invented it. The aim of the game, then, is to work out ideas which fit the facts. Ideas come in many different shapes and sizes, differing both in generality and abstractness. Here is a sample of ideas of increasing degrees of generality: (3) a. Is must have a subject. b. In English, a tensed auxiliary verb must have a subject. c. In English, a tensed verb must have a subject. d. In English, a verb must have a subject. e. In every language, a verb must have a subject. Provided we can recognise a subject, it is relatively easy to decide whether (3a) is true, but it becomes increasingly difficult to check the claims as we go down the list. Similarly for abstractness; the following statements al1 relate to sentence (la). (4) a. Is consists of i followed by s. b. Is consists of a stem i followed by a suffix s. c. Is is the 3rd-singular present form of the verb BE. d. Is has a subject, it. e. Is is the 'main verb' of the sentence. A spectator's guide to syntactic theories Links & Letters 1, 1994 27 Obviously there is a limit to the number of goals that can be scored with - ideas that are very concrete and very specific; the prizes go to those who can climb these hierarchies. In general, the hierarchies are closely related: as claims become more general, they have to be stated in terms of more abstract concepts. And more importantly, the more general or abstract we get, the more interconnected each claim is with other claims - e.g., we cant decide whether every language requires its verbs to have subjects until we know whether every language has verbs and subjects, which we can't decide without clear notions of the concepts 'verb' and 'subject'. Every player faces this same need to match very specific facts with more general ones, but of course different players have different personalities and different interests. Some are happy to focus on a narrow range of facts with the aim of discovering more and better general facts underlying them; others prefer to keep a wide range of phenomena in their sights in the hope of discovering links arnong them; and others again have practical interests such as languageteaching which oblige them to consider the full range of facts, without necessarily pushing for much generalisation. Al1 these interests are legitimate and mutually supportive, but 1 think it is important to beware of claims that 'description and 'explanation' are alternatives. We can+ explain things we cant describe, nor can we describe things we can't understand. What sorts of facts are syntacticians responsible for, as syntacticians? We al1 agree that syntax is the study of how words combine with one another, so we are at least responsible for the study of word-order and for the choice of words where the choice of one word is dependent on some other word with which it cooccurs - for the fact that in (la) Barcelona must follow in, that is must be accompanied by a nominal such as it, and that this must be singular rather than plural. There is some disagreement about the extent of our responsibility for the internal structure of words; e.g. is it the syntactician's job to break gets into get plus S? There is also disagreement about where the game of syntax changes into semantics; an example of this disagreement would be whether we are playing syntax or semantics when we say that in Barcelona defines a location. 1 discuss this in section 4. These disagreements are important, as we shall see, but the shared ground makes it quite clear that basically we are al1 playing the same game, with the same aim. 2. Introducing the players The typical syntactician is an academic working in a university or research institute, but an increasing number of syntacticians are to be found in organisations with more immediate aims which include the production of computer software for handling ordinary language, translation of religious texts and diagnosis of medical conditions connected with language. An important area 28 Links & Letters 1, 1994 Richard Hudson of activity is the training of school teachers and preparation of materials for use in schools, where modern syntax has many possible applications. Even if we simplify the picture by ignoring syntacticians who have clear applications in view, there are considerable differences of which any spectator should be aware. Some syntacticians work on rather concrete and particular matters such as writing a dictionary of some language, or working out al1 the finer details of some small but complex area of English syntax (e.g., what are the rules for using the word AS?); others choose to concentrate on much more general and abstract things, tackling questions like 'What kinds of word-order rules are there?' Some concentrate on English, others work on some other language, while others again study a small range of phenomena across hundreds of languages at the same time. Some like to keep firmly within syntax, as narrowly defined, whereas others try to relate their work on syntax to more general issues and to the analysis of other parts of language. Al1 these differences are healthy and potentially productive (of new discoveries and insights, al1 potential goal-scorers) . Unfortunately there are other differences which are less welcome, and which are probably counterproductive because they quite unnecessarily prevent crossfertilisation of ideas. These differences include one which is symbolised by the slogans 'functional grammar' and 'formal grammar' (see Lightfoot 1992 and Thompson 1992 for clear and brief summaries of these ideologies). 'Functionalists' like to emphasise the social and communicative functions of language, whereas the 'formalists' tend to emphasise its formal patterns. For example, we might compare a functionalist aid a formalist answer to the question why moving is before it in (1) should produce a question. A functionalist (e.g., Halliday 1985) might answer that is carries the sentence's 'polarity', the contrast between positive and negative, so ifyou wanted to show that the sentence was 'about' the truth of the proposition that it is hot in Barcelona, it makes sense to start with the word that is most directly related to truth, namely is. In contrast, a formalist (Chomsky 1986) would look for a 'formal' explanation, in terms of formal patterns; e.g., one notices that the position of is looks like that of a subordinator such as that in (5). (5) a. 1 know that it is hot in Barcelona. b. I II 1s it - hot in Barcelona. The 'formal' explanation, then, is that is fills the slot which is also available for that. If is which is inverted round the subject competes for a single 'slot' with that, this would explain why inversion never occurs with that (or, for that . matter, with other subordinating words like whether, which presumably fit into the same slot): A spectator's guide to syntactic theories Links & Letters 1, 1994 29 (6) a. * 1 wonder whether is it hot in Barcelona. b. 1 wonder whether it is hot in Barcelona. Both the formal and the functional explanations have some plau~ibilit~, so which is right? The rhetoric around the distinction between functionalists and formalists sometimes suggests that the battle is between hearts and minds (heartless concentration on mere form, or mindless concentration on rather nebulous function). But this is nonsense, because when pushed every formalist would agree whole-heartedly that language has communicative and social functions, and every functionalist would also agree that language is a system of formal patterns. Both the formal patterning and the functions are among the data which the goal-keeper knows about, and the only ideas about language that will score are those which are compatible with al1 the relevant data. Returning to our example, it is quite posible, in principle, that both accounts of the position of is in (lb) are correct: what is and that have in common is that they indicate the relation of the clause's basic meaning to the context, and it is (functionally) helpful to the hearer to have this information before starting to process the rest of the sentence. If they share a (formal) slot in the sentence's structure, there may be a functional explanation for it. As a matter of fact, this particular formal analysis may not be right because that and inverted is can in fact cooccur, as in (7). (7) 1 know that only in summer ir it hot in Barcelona. But whatever the merits of this analysis, the point remains that the inverted verb and the subordinators like THAT are formally similar inter alia in that they both occur near the start of the clause, and we can't te11 in advance whether the explanation for these similarities will be in purely formal terms or purely functional ones or a mixture of the two. Formal and functionai approaches are complementary and mutually dependent, so spectators, as potential 'customers', should beware of high-pressure sales rhetoric. 3. Some important teams Syntax is complicated, and any syntactician, however modest their aims may be, needs an intellectual framework within which to locate each claim - a set of general categories (such as 'verb' or 'subject'), and a set of general ideas about how sentences are structured and about the workings of a grammar. There is a widely agreed set of general categories, many of which are traditional both in their names and in their meanings; but there is no generally agreed framework. Instead, there are a relatively large number of competing alternatives. 30 Links & Letters 1, 1994 Richard Hudson Faced with these choices, one option for a syntactician is to playa lone game using a more or less 'theory-neutral' set of assumptions but avoiding issues that require a more sophisticated set of assumptions. Another option is to follow the even more lonely road of building a coherent framework of oneS own. But the most popular option is to adopt one of the available frameworks. Although this requires time and effort, it is the easiest of the options because inventing your own theory is a life's work, and the 'theory-neutral' option is very limiting - rather like playing football with bare feet, when it is wise to think twice before kicking any given ball in case it damages your toes. A profes~ionall~ constructed theory is like a good pair of boots, which will allow the wearer to tackle al1 sorts of interesting problems that couldn't even be stated without a theory. Moreover, by opting for theory X you automatically join the ranks of syntacticians who use theory X, the theory X 'team'. This is important because you can communicate easily with other members of the same team, you can use their ideas, and (hopefully) they will use your ideas. The intellectual bonds may lead to friendships, jobs, fame, ... There are al1 sorts of reasons for joining a team, not al1 connected, or at least not directly, to the correctness of the tea& theory. What, then, are the most important teams in the syntax game? It al1 depends on what you mean by 'important', but by almost any standards of importance, the one that comes out on top is Government-Binding Theoy (GB, alias PrinciplesandParameters Theoy, alias, in the dim distant past, Transformational Grammar, Salkie 1990, Lyons 1970191). The tradition including GB and its precursors is the brain child of Noam Chomsky, who must be the most famous linguist (certainly the most famous syntactician) in the world, measured in terms of media coverage and general fame beyond the academic community. Partly because of its popularity, there is actually enormous diversity among practicioners of GB, and much more so than in any of the other theories which 1 shall mention below, so it would probably be more accurate to describe GB as a family of theories. After 35 years of theory-building Chomsky is still a fast worker, and his thinking is currently going through a particularly radical revision; so al1 of his followers are (by definition) to some extent out of date, and some are extremely so. Moreover, although he is clearly the leader of the team, many of the members are outstandingly good syntacticians and thinkers, capable of disagreeing with him on important issues. Consequently two articles both claiming to apply GB theory may well presuppose radically different, and incompatible, assumptions on key issues. The dominant position of GB becomes evident from the professional journals that regulady carry articles on syntax. GB is the only theoretical framework that needs neither justification nor explanation because it is generally assumed that the typical reader already understands, and probably accepts, some version of GB. The same is true ofconference papers presented to societies such as the Linguistic Society of America and the Linguistics Association of Great Britain. A spectator's guide to syntactic theories Links & Letters 1, 1994 31 Having said this, though, 1 must make it clear that other theories are - - important, in one sense or another, as well. Let's consider first the matter of social importance - the size, coherence and organisation of the teams. By this criterion we must recognise three other theories as important: Systemic (Functionai) Grammar (Halliday 1985, Butler 1985), Functional Grammar (Dik 1989, 1991, Siewierska 1991) and Cognitive Grammar (Langacker 1987, 1990, 199 1). Al1 these teams have a newsletter or a journal and annual meetings which are held in a variety of countries and involve hundreds of participants. The same is true of GB, which qualifies as important in any case. Interestingly the first three teams are strongly committed to the slogans of the 'functional' approach, even to the extent of denying the existence of formal patterns in syntax (in the case of Cognitive Grammar); while the GB team is equally dedicated to the 'formal' approach. One suspects that the slogans may have an important social function for the teams. Another test of importance is the extent of representation in the international refereed journals. Here too we find the GB team well ahead of its rivals, but in this case the rivals are a completely different set. A few years ago it could have been said that the main two were Lexical-Functional Grdmmar (Bresnan 1982, Wescoat and Zaenen 199 1) and Generalised Phrase-Structure Grammar (Gazdar et al 1985); these two, plus GB, were singled out for special treatment in two excellent introductory books, Sells (1985) and Horrocks (1987). More recently, though, the field has changed somewhat. LFG is still represented, but GPSG has generally been abandoned in favour of Head-driven Phrase-Structure Gmmmar (Pollard and Sag 1987). Borsley 199 1 is a good introduction to both GPSG and HPSG, as well as to GB. Meanwhile there has been an increase in output from two older theories, Categorial Grammar (Wood forthcoming) and Relational Grammar (Blake 1990, Aissen 199 1). In spite of the word 'Functional' in the name of LFG, al1 these theories tend to focus on formal patterning, though without precluding functional links to meaning, to processing and so on. The functional approach tends to be represented in the professional journals by rather theory-neutral analyses of large amounts of cross-linguistic data in the 'typological' tradition (Comrie 1981189). A third measure of importance is in terms of applications, i.e. 'spectators' of the syntax game who are basically not playing the game but using the ideas. This is the one criterion by which GB does not emerge as the leader; indeed, it would be fair to see that in most applications GB is nowhere in sight (though this may change if current activity in GB bears fruit). In the field of language teaching, the most widely applied theory of syntax is probably Systemic Functional Grammar (SFG). In computer applications too SFG is popular, as are also Lexical Functional Grammar, Generalised Phrase Structure Grammar, Head-driven Phrase Structure Grammar and Categorial Grammar (Winograd 1983). 32 Links & Letters 1, 1994 Richard Hudson None of these measures is inherently and directly connected to the question of truth (compatibility with al1 known facts). It is quite possible for a theory to be extremely popular, and widely used in the professional literature, and widely used in applications, without being the best available in objective terms - i.e., the nearest to the truth. It would probably be fair to say that most syntacticians apply the theory that they learned first, because that is the theory they understand best, and not because they have seriously compared it with its rivals. The theory learned first depends on the linguistics department where syntax was first studied, which in turn depends on al1 sorts of social, geographical, economic and academic considerations. Nor does publication in a refereed journal guarantee truth; al1 it means is that typical readers will relate easily to the contents. And would-be-appliers of syntactic theories tend to want a detailed and reasonably coherent and explicit description, rather than a theory as such. So it may be that the prize for truth could still go to the future descendant of a theory which is still out of sight in terms of our three criteria. (1 hope this is so, at any rate, for reasons which will become clear at the end of the article!) The general conclusion of this discussion is that al1 these theories are claimed, by their proponents, to be more or less true; how could it be otherwise? Moreover although their proponents typically have different interests, ultimately they are al1 going to have to mesh, sooner or later, with the same reality. Nor can we realistically divide the theories, as some people like to do, into those which aim at 'description' and those whose aim is 'explanation'. Once again, their proponents or users may differ but 1 don't know of any theory which provides a description of a range of facts without also trying to explain at least some of them. Advocates of GB are particularly prone to claim that in some sense GB is especially good at explaining things, or at least that explanation is particularly high on the agenda in GB work. In my experience this is simply false in both directions. The other theories al1 aim at explanation, and 1 know linguists who use GB simply as a descriptive framework, without paying much attention to its success in explaining. We have arrived at the following list of syntactic theories which count as important by one or more of the above criteria: Categorial Grammar Cognitive Grammar Functional Grammar Generalised Phrase Structure Grammar Government Binding Theory (GB) Head-driven Phrase Structure Grammar Lexical-Functional Grammar Relational Grammar Systemic Functional Grammar A spectator's guide to syntactic theories Links & Letters 1, 1994 33 4. Syntax and meaning Everyone agrees that words, when combined together, express meaning, and that the meaning that a string of words express varies according to how the words are combined. The meanings that are expressed form some kind of specifically 'semantic' structure, so we al1 agree that a string of words (let's cal1 it a sentence) has a semantic structure. We don't agree on what belongs in this semantic structure, but at least it must show the sense of each word, and whatever semantic relations there are among the words. We also agree that a sentence has a phonological structure, describable in terms of syllables, consonants, vowels and so on. The question for us to consider is what role syntax plays between the semantic and phonological structures. One of our theories, Cognitive Grammar, claims that there is no specifically syntactic structure, since each word is a Saussurean 'sign' with just two parts, a meaning (signifié) and a phonological form (signifiant). There are syntactic rules, but al1 they do is control the mapping relation between the phonological and semantic structures; for example, some rules control the order in which phonological structures occur. There are no specifically syntactic categories like 'verb' or 'subject', because al1 such categories are claimed to be definable in semantic terms. Speaking personally, 1 am convinced this claim is wrong, but 1 have to admit that it is surprisingly hard to argue the case against it. A particularly serious obstacle is the uncertainty about semantic structures; one of the characteristics of Cognitive Grammar semantic structures is how closely they correspond to syntactic structures in other theories - e.g., it is claimed that there is a semantic category that is shared by ail and only nouns, in spite of the apparent diversity among nouns like John, milk, beauty and explosion. At the other extreme, GB claims that the specifically syntactic structure of a sentence is very rich indeed, and incorporates information which other theories would put only in the semantic structure. For example, take sentence (8). (8) Everyone told one person their name. Does this mean the same as (9a) or (9b), and whose name was it that was told - the teller's, or the hearer's, or someone else's? (9) a. For everyone there was one person to whom they told their name. b. There was one person to whom everyone told their name. These are considered to be syntactic questions in GB, and the syntactic anaiysis of a sentence includes one sub-structure, the 'Logicai Form', whose purpose is to provide the answers. Rather strangely, there is very little discussion 40 Links & Lerters 1, 1994 Richard Hudson topicalised as when it is in its normal position. For example, omitting it leads in both cases to ungrammaticality, for the simple reason that likes needs an object. But aren't we in fact showing this relation by labelling syntax 'Obj' in diagram (17)? (1 should emphasise that most of the theories in our collection would not in fact use functional labels like 'Obj', so we are considering the most favourable possible analysis.) No, we aren't; and the reason is that although we may be marking syntax as the object of something, the label 'Obj' doesn't, in itself, te11 us what syntax is the object of. When it is in its normal position, we can te11 easily that it is the object of likes, because the convention is thar an element marked 'Obj' is the obje;t of the verb inside the same verb-phrase. But what if syntax is not inside any verb-phrase, as in the analysis in (17)? The convention breaks down. It would be reasonable to see this as making a mountain out of a very small mole-hill. Al1 we need is an extra convention, for topicalised objects: 'if something is labelled both Obj and Topic, treat it as the object of the next verb-phrase.' This would work for our current example, but unfortunately, it doesn't solve the problem because topicalisation can move an object out of its own clause into a 'higher' one. To see this, consider the next two examples. (1 9) a. Fred denies that he likes syntax. b. Syntax Fred denies that he likes. Here is a structure for the second of these examples: I i I v Syntax Fred denies rhat he likes. Once again 1 have made the most favourable assumptions that 1 could; but the problem remains. Syntdx is indeed labelled as the object, but the object ofwhat? A spectator's pide to syntactic theories Links & Letters 1, 1994 41 What we want to show is that it is the object of likes, but this is not what the diagram is actually showing. This is only one example of the kind of problem that faces any grarnmarian grappling with the complex realities of natural languages. Many of the other problems are similar to this one, arising from the possibility of using elements in positions other than the ones where they might normally be expected; but many of them involve other kinds of complexity. In the foiíowing section 1 shall hint at some of the solutions that have been suggested to this one problem, but some of these solutions are suficiently general to solve problems of many different kinds. 8. Some ways to enrich phrase structure The problem, then, is how to show that Syntax is both the topic of the top clause and the object of the lowest one in (19b): (1 9) b. Syntax Fred denies that he likes. Al1 the solutions that we shall consider in this section build on the view of sentence-structure as a hierarchy of increasingly large 'phrases', but enrich the basic view that 1 have sketched. However, it is important to repeat that they don't al1 start from the same basic view; in particular, some theories allow functional labels like 'subject', and others don't. This difference is important when it comes to enrichments. The best-known solution is probably the transformational one, adopted by GB (and, in a slightly different way, by Relational Grammar). Since a single structure wont aüow us to show both of the links that involve syntax, we use (at least) two structures, one showing syntax as the topic of the whole sentence, and the other showing it as object of likes. Each of these trees is a very simple structure, with the simplest kind of (non-functional) labels on its joints and twigs, but by combining two of these very simple trees we can deal with very complex sentence-patterns. These trees are bound together, as part of the total analysis of a single sentence, by means of rules called 'transformations' which transform one tree into another - thereby allowing an item like syntax to be moved out of its 'lower' object position into the 'higher' topic position. The latter position is where it actuaüy occurs - i.e., where we see it - in (19b), so it can be described as relatively concrete (or 'surface'), whereas the other position is much more abstract. The logic of this kind of analysis encourages linguists to accept increasingly abstract analyses in which neither the number nor the order of elements bears any simple ielation to the elements in the surface structure. To sirnplify somewhat, the hall-mark of GB analyses is their use of extremely abstract trees with very simple labels. (You will recall that 1 said earlier that GB builds a great deal of information into the geometry of trees.) 42 Links & Letters 1, 1994 Richard Hudson Some other theories solve the problem by enriching the labels rather than increasing the number and abstractness of the trees. In Generalised Phrase Structure Grammar, for example, we give (19b) a single surface tree similar to the following: (21) A NP A NP "r\ v "t ? A NP "r\ V NPINP Syntax Fred denies that he likes The main thing to notice about this tree is the series of extra labels which 1 have highlighted. In each case, the normal label (S, VP) is combined with 'INP' (to be read as 'slash NP'), which gives the extra information that a noun-phrase is missing from somewhere inside the sentence or verb-phrase concerned. These labels are like a trail of footprints to guide the rules which will eventually try to find an object for likes: if you can't find an object noun-phrase in the usual position, because it has gone elsewhere, try to find it by following its footprints up the tree. Technically, the slash label is a feature (a named variable which can have some value), and it is typical of the Generalised Phrase Structure Grammar approach to dealing with complexity by enriching the feature labels attached to a single concrete tree. Head-driven Phrase Structure Grammar and Categorial Grarnmar both take a rather similar approach to the problem. Lexical Functional Grammar adopts a third solution to the problem of enriching the basic structure. Like GB, it does not restrict the syntactic structure of a sentence to nothing but a single, surface structure. But unlike GB, it adds A spectator's guide to syntactic theories Links & Letters 1, 1994 43 just one extra structure, and this extra structure is labelled exclusively with functional categories like 'subject' (hence its name, 'functional structure'). The two structures offer an easy and natural analysis of the problem sentence: syntdx is at the start of the sentence in the surface structure, but in the functional structure it is the object of the lower clause. The two positions are connected by a special rule for finding a 'path', which has similarities both to the transformations of GB and to the slash 'footprints' of Phrase Structure Grammar. In short, those theories that have solved this problem have found three quite different kinds of solution, al1 ofwhich involve in some way enriching the kinds of structure that can be used for showing the syntactic analysis of a tree: a. more information can be shown in the trees by increasing the number and abstractness of the trees for each sentence, but without any increase either in the complexity of the individual subtrees or in the labelling; b. more information can be packed into the labels, without any increase in the number or abstractness of the trees; c. more information can be shown by adding one extra tree which uses functional labels. Needless to say, theoretical linguists would very much like to know which of these approaches is nearest to the truth; in principle this is a researchable question, because the different approaches can be tested against the facts (our goalkeeper, if you remember). But at present the outcome of the game is still completely open. 9. Advertiser's announcement: word grammar As 1 have already explained, linguists are al1 shooting at the same goal, which we can cal1 'truth', and we would al1 like to believe that either we personally, or the team to which we belong, is doing well in terms of goals scored. My own view is that al1 these teams, for al1 their importance, are missing a lot of goals, and 1 should like to end by introducing a group of pla~ers that 1 haven't mentioned so far. 1 think it would be fair to say that most members of the nine teams listed above would be united in considering this group unimportant (which is why 1 have ignored it so far), but it does pass one of our importance tests, and may even be capable of out-scoring the other tearns. Let me start by contrasting the beliefs of this group with the starting-point shared by al1 the others: the belief that the structure of a sentence is based on a hierarchy of phrases (clauses or sentences, noun-phrases, and so on). This view is rather like an analysis of a community of people which says that its structure can be defined only by identifying successively smaller sub-commu- nities. For example, the only relation that 1 can have to any of my colleagues, according to this theory, would be fellow-membership of groups such as 44 Links & Letters 1, 1994 Richard Hudson departments, faculties, committees and so on. There would be no way to say that 1 was a friend of X, or collaborating with Y, because these are direct relations between them and me. Direct relations between individuals have no place in this kind of analysis. Similarly for syntax: direct relations between words have no place in a theory which is based on phrases. This is not the only possible view, though most of its adherents pay little or no attention to any alternatives. Nor is its history very long, compared with the history of grammatical theorizing (which in Europe dates back to the Ancient Greeks); it was only introduced in the 1930's, by the American linguist Leonard Bloomfield (1933). The alternative is, of course, to recognise direct relations between words, and to take these word-word relations as the basis for syntactic analysis. This alternative has for a long time been accepted in the teaching of grarnmar in schools throughout Europe (though it has died, with al1 grammar-teaching in UK schools). This is especially true in Eastern and Central Europe. To see what difference these theoretical abstractions make, consider what the two approaches say about sentence structure. If we take the phrase-based approach, we say that a clause or sentence has a subject, which is a noun-phrase; but if we take the other approach, it is not a clause or sentence, but a verb (i.e., a word) that has the subject, and the subject is not a noun-phrase, but a noun. Both kinds of analysis allow the subject noun to be accompanied by other words (adjectives, determiners and so on), but the word-based analysis treats this entirely as a private matter (so to speak) between the noun and its companion words, without setting up a 'noun-phrase' to hold them al1 together. These word-word relations are called 'dependencies', because they are always asymmetrical, one word being the 'dependent' ofthe other, its 'head'. Traditional grammar uses words like 'govern', 'modify' and even 'belong' or 'take' to describe dependencies, as in the following statements: (22) a. In German, the preposition ZU governs a dative noun. b. Big modifies book in the combination big book. c. In I told him yesterday that syntax was fin, the word yesterday belongs to told). d. The verb OUGHT takes to (or a 'to-infinitive'). Various systems have been used for displaying dependency relations, including one which looks remarkably similar to the tree notation for phrases. Here, first, is a simplified 'phrase-marker' (i.e., a phrase-based tree) for It is hot in Barcelona: A spectator's guide to syntactic theories Links & Letters 1, 1994 45 It is hot in Barcelona The crucial thing to notice here is that there are labelled nodes in the tree not only for the five words (classified as 'noun', 'verb', 'adjective' or 'preposition'), but also for seven phrases. Now contrast this with what we might cal1 a 'word-marker' for the same sentence: It is hot in Barcelona In this diagram, there are no phrase nodes, and the vertical dimension shows dependency, not the part-whole relation of the phrase-marker. It is true that some phrases are implicit in the word-marker (e.g. you can see that in Barcelona is a phrase, whereas hot in is not), but they can stay implicit because, according to this approach, they are redundant. The labels on the diagonals name the relations (subject, complement, adjunct) between the words concerned. Diagrams like this word-marker are used by some linguists, but for various reasons 1 prefer a different notation, in which the direction of dependency is shown by arrows rather than by using the vertical dimension. In these diagrams, the arrow points from the head to the dependent: It is hot in Barcelona. The vertical arrow linking is to nothing at al1 shows that this word is the 'root' of the whole sentence, because it depends on no other word within the sentence (or anywhere else, for that matter). Once again, the thing to notice is the complete absence of any unit larger than a word - not even the sentence appears, as such, in the diagram. 1s this word-based approach really different from the phrase-based approaches? After all, phrases lurk beneath the surface of this diagram, as 1 have already pointed out, because the dependencies group the words into phrase-like clusters. Each of these 'phrases' has a single word as its root, just as the whole of this sentence is rooted in is; and similarly, in Barcelona counts as a phrase (rooted in in). Does it really matter then whether we show these patterns in terms of phrases, or in terms of word-word dependencies? Yes, it does. For one thing, it makes a radical difference to the way we analyse a sentence's structure. As we noted earlier, phrase-based approaches differ according to whether or not they recognise a phrase (the 'predicate' or 'verbphrase') consisting of the verb plus al1 its dependents except the subject. But if we start from dependencies, it is quite simply impossible to recognise such a phrase, even implicitly, because the subject is a dependent of the verb along with its other dependents. Does this limitation count for or against dependency analysis? If it really is essential to recognise verb-phrases, because there are facts which simply can't be stated in any other terms, dependency analysts should seriousIy consider abandoning the match (or switching sides). But if verbphrases can be dispensed with (given the other assumptions of dependency analysis), then we can celebrate a victory because our theory predicts that such phrases are dispensable - and is vindicated by the facts. The issues are complex, but on balance things look good for dependency grammar. Another difference between word-word dependencies and phrases as the basis for analysis lies in the treatment of discontinuous phrases , phrases which are split by other words which don't belong to them. These are very common, and are one of the main head-aches for any syntactician (because it is generally agreed that they are atypicalmost languages are organised in such a way that most phrases are continuous). We have already looked at one exarnple, produced by topicalisation as in (19b), Syntax Fred denies that he likes. Here the discontinuous phrase is syntax ... he likes, which is rooted in likes (notice that he is its subject, and syntax is its object). As we saw, a range of clever tricks have A spectator's guide to syntactic theories Links & Letters 1, 1994 47 been developed for dealing with examples like this, but one option is excluded in principle by most of the theories we have considered: that of allowing a phrase to be discontinuous. This exclusion is fundamental to most of the notations we considered; for instance, how could we use brackets to mark off a discontinuous phrase? Any brackets we put round syntdx ... be likes, for example, will also include words which are not part of this phrase (Fred denies that). Now let's suppose we start with dependencies as the basis for analysis. We can easily recognise that phrases are normally continuous; to achieve this, al1 we need to say is that dependency arrows normally mustn't cangle. But this is just the normal pattern, and we may well expect exceptions (just as we expect exceptions in other parts of the grammar, such as in the rules for forming past-tense verbs). Provided we can make it clear precisely which patterns are allowed to break the rule, we can allow phrases to be discontinuous. This leads to dependency diagrams like (26), in which we give syntax two different dependencies in the same structure, one to denies, as its 'topic', and one to likes, as its object. Rather conveniently, we can separate the arrows for these dependencies by drawing one above the line of word-class labels and the other below it. Syntax Fred denies that he likes. The social background to dependency analysis is quite important in understanding why it has had so little impact on theoretical syntax. Part of the explanation lies, 1 suspect, in its very success in school-teaching. In countries where it is the basis for grammar-teaching in schools, it is naturally accepted as unproblematic by those who apply it in academic research. In Europe a lot of highly regarded linguists use it; for example, a European mega-project on automatic translation, called Eurotra, uses dependency analysis as the basis for its syntactic processing. However, using something is not the same as theorising about it, and on the whole those who have used it have taken it for granted as something so familiar and well developed that there's no need to agonise about its theoretical bases. Since they publish few theoretical articles about it (least of al1 in English), it is not surprising that linguists who do theorise about grammar have little reason to think about it or to consider it as an alternative to phrase-structure. 48 Links & Letters 1, 1994 Richard Hudson In spite of this basicall~ non-theoretical stance of dependency analysts, dependency grammar has recently started to have more influence on western linpistic theories. Most obviously, many of the phrase-based theories that 1 described recognise the asymmetrical relations among the words in a phrase by picking out one of them as the phrase's 'head' (what 1 have been calling its root; some theories use other terms). This is true of Categorial Grammar, Generalised Phrase Structure Grammar, GB, Head-driven Phrase Structure Grammar (notice the name!) and Relational Grammar. GB has gone even fürther, as it includes an elaborate subtheory called 'government theory' (reflected in the name of the whole theory, 'Government and Binding theory'), which allows us to build what is (in al1 but name) a dependency structure on top of a phrase structure. These dependency features (and others) have been added during the last two decades, and have been presented as theoretical innovations, but they have rather obvious roots in traditional grammar. A few of us have gone even further, by developing theories in which dependency-like ideas aren't just glued onto phrase-based theories, but in which these ideas are fundamental. (1 should add that this process can be painfül, but stimulating; 1 myself started from a phrase-based theory, Systemic Functional Grammar, and was very surprised, and disturbed, to find how weak the evidence for the phrase-based approach was.) My own work has led to a theory called Word Grammar which 1 have described in a number of works including two books (Hudson 1984, 1990). Unfortunately, by none of our criteria does this theory count as important, but 1 think 1 have scored enough goals to make it worth pursuing the ideas in the theory, though some shots still bounce back to me (as they do, incidentally, for everyone else too!). Maybe it will turn out to be fundamentally wrong, and popular choice will have been vindicated; if so, only a handful of people (a few of my research students and myself) will have wasted their time. But we have to explore every idea, because at this stage of the game we really have no idea which ideas will eventuall~ turn into the Eventual Winner. 10. The great cup final Any reader with some experience of traditional grammar and none of modern theoretical grammar may well be wondering at this point why we are taking so long to get syntax sorted out, especiall~ since we are building on two thousand years of work dating back to Greek and Latin grammarians (who supplied notions like 'noun', 'tense' and 'subject'). The problem is that grammar turns out to be very, very complicated. The individual structures and rules are often reasonably simple, but what is complex is the way they interact not only with one another, but with virtually everything else in our minds as well - with what we know about the immediate past (for which notions like 'topic' are A spectator's guide to syntactic theories Links & Letters 1, 1994 49 relevant), with what we know about human behaviour, and with a good deal of what we know about the world at large. Added to which, grammar is ultimately located in the minds of individual speakers, each of whom is unique in the tiny details but who also form extremely complicated alliances called dialects and languages. Al1 these complexities are grist for the grammarian's mill. One thing is absolutely certain: there is no chance of a definitive, stable, universaíly-accepted general theory of syntax in the 20th century. We are certainly further from the beginning of the journey than we were, say, 30 years ago; but the end is still nowhere in sight, and it is only optimism that allows the belief that we are getting nearer to it al1 the time. References AISSEN, Judith (1981). "Relational Grammar". Droste, F. and Joseph, J. (eds). Linguistic Theory and Grammatical Description, 63-102. Amsterdam: John Benjamins. BW, Barry (1990). Relational Grammar. London: Routledge. BLOOMFIELD, Leonard (1933). Language. New York: Holt. BORSLEY, Robert (1991). Syntactic Theory: A Unijed Approach. London: Edward Arnold. BRESNAN, Joan (1 982). The Mental Representation of Grammatical Relations. Cambridge, Ma: MIT Press. BRIGHT, William (1992). Internatiunal Encyclopedia of Linguistics. Oxford: Oxford University Press. BUTLER, Christopher (1985). Systemic Linguistics: Theory and Applications. London: Batsford. CHOMSKY, Noam (1986). Barriers. Cambridge, Ma: MIT Press. COMRIE, Bernard (1 98 1/89). Language Universals and Linguistic qpology. Oxford: Basil Blackwell. DIK, Simon (1989). The Theoy ofFunctiona1 Grammar I: The Structure of the Clause. Dordrecht: Foris. DIK, Simon (1991). "Functional Grammar". Droste, F. and Joseph, J. (eds.). Linguistic Theory and Grammatical Description, 247-274. Amsterdam: John Benjamins. DROSTE, Flip and JOSEPH, John (eds.) (199 1). Linguistic Theory and Grammatical Description. Amsterdam: John Benjamins. GAZDAR, Gerald; KLEIN, Ewan; PULLUM, Geoffrey and SAG, Ivan (1985). Generalized Phrase Structure Grammar. Oxford: Basil Blackwell. HALLIDAY, Michael (1 985). An Introduction to Functional Grammar. London: Edward Arnold. HORROCKS, Geoffrey (1987). Generative Grammar. London: Longman. HUDSON, Richard (1984). Word Grammar. Oxford: Basil Blackwell.