RICHARD HUDSON SOME BASIC ASSUMPTIONS ABOUT LINGUISTIC AND NON-LINGUISTIC KNOWLEDGE One of the most interesting questions which relate to the use of 'frames' in understanding human cognition is whether 'linguistic' knowledge is itself structured in this way. If we may assume that at least some parts of our encyclopedic knowledge are best represented by means of frame-type structures (including scripts, scenes, and so on), then what does this suggest about knowledge of language? On the one hand, we might say that this proves that language is unique, because linguistic knowledge does not involve frames; so the relevance of frames to the study of language lies just in the fact that they provide a way of formalizing the non-linguistic knowledge which needs to be referred to by the pragmatics. On the other hand, we might equally well draw a very different set of conclusions. For those of us who are struck by the similarities between linguistic and non-linguistic structures, it might seem reasonable to ask, as a research project, whether, or to what extent, it is possible to describe linguistic structures in terms of frames. Until we have tried to describe language in this way, we cannot be at all sure that language is as unique as has often been assumed; and if we find characteristics of language which remain stubbornly different from everything outside language, then we may be fairly confident that in these respects language is indeed unique. In this brief note I shall assume that the onus is on those who believe that linguistic and non-linguistic knowledge are different to produce evidence for their view, rather than on those of us who believe that they are at least in large measure similar. There is in any case surely no need to provide research evidence for the claim that there are some similarities after all, one of the main relations to which linguists refer is linear order, and another is that of subclassilication, neither of which is at all restricted to language. Against this background assumption, I shall try to show that two other assumptions follow - QUADERNI Dl SEMANTICA I Vol. VI, no. 2, December 1985 284 naturally, namely (1) that the meaning of a linguistic expression is a cognitive structure with no special properties to distinguish it from other cognitive structures - i.e. there is no special object that we could call a 'linguistic semantic structure'; and (2) that linguistic structures themselves have the properties of frames. These assumptions underlie the theory called 'Word grammar' described in Hudson [1984]. (1) Imagine some Being designing the human mind, and wondering how to relate the bit of the mind responsible for linguistic expressions to the bit responsible for understanding non-linguistic experiences. Arrangements have already been made for dealing with strings of phonemes and words (language), and also for dealing with restaurants, birthday parties, moon rocks, and so on (non-language). Surely it would be obvious to this Being that the easiest way to make humans use language would be to allow them to exploit the non-linguistic structures directly, as meanings of linguistic expressions. Any extra layer of structure, in which the entities analyzed were 'meanings', but in which their analysis was not that of ordinary non-linguistic cognition, would simply get in the way. This being so, we need very strong and compelling evidence indeed for the existence of such a layer. To my mind, the existing discussions in the literature fall short of this requirement, as they appear to work on the assumption that the onus of proof lies on the other side. Apart from such general, aprioristic, considerations as these, there is considerable evidence in favour of the view that linguistic structures are linked directly to general cognitive structures. Perhaps the clearest and easiest evidence comes from the area of 'semantic valency', where we are concerned with the 'argument structure' of a word: donate and give both involve at least three arguments (a giver, a gift, and a recipient), rain involves (arguably) no arguments at all, fond involves two (a feeler and a phenomenon towards which the feeling is directed), and father involves two (the person referred to plus at least one child). One of the main problems which arises in working on this area is that linguists find it difficult to arrive at a 'specifically linguistic' classification, which they could call the set of 'semantic cases' of the language. How many such categories should we distinguish, and which specific categories should we assign to specific constructions? How should the categories be defined - syntactically, or in relation to the world? Are the categories universal or specific to particular languages? Theoretical and descriptive linguists have devoted a good deal of attention to these questions, and a number of interesting analytical frameworks have been proposed. However, the whole discussion seems to be made more difficult than it need be by the assumption which is often at least implicit, that the argument structure of a word is a specifically linguistic object. It is surprisingly rare to find an analysis in which it is recognized that the 'semantic' argument structure of a word is bound to be in a one-to-one relation to the argument structure of the concept to which it is related. For example, the arguments of write must be the same as whatever elements are mentioned in our cognitive frame for the concept of writing (a writer, an instrument, a surface, a message, a code and perhaps a few more). It is inconceivable (to me, at any rate) that 285 there could be a word which referred to a concept X, and whose nonsyntactic argument structure was distinct from the structure of X. (Of course, we must distinguish between these semantic/cognitive arguments, and the syntactic valency of the word concerned – for example, some of the arguments of write need not be expressed overtly by means of a syntactic expression, but this has nothing to do with the semantic structure. Thus, the fact that no instrument is referred to explicitly in John wrote a letter does not mean that he may have written the letter without using an instrument). Once this basic connection between the argumentstructure of a word and the structure of its corresponding conceptual frame becomes clear, the linguist's analytical problems appear in a different light - his problems are the same as those of the cognitive psychologist working on conceptual structures, and the latter may in fact be better equipped to solve the problems than he is. (A very similar view has been advocated by Charniak [1981]). Other areas of linguistics in which evidence for the direct relevance of cognitive frames can be found include compounds and vague words like do. It could be argued that when nouns are combined into a compound (as in English tooth-brush and wire brush), the semantic structure is controlled just by one general principle: that the first noun should define a filler for some slot in the frame for the concept denoted by the second noun. Thus we may assume that the frame for "brush” contains a slot for the thing to be brushed (e.g. teeth), and another slot for the substance out of which the bristles are made (e.g. wire). No specifically 'linguistic' structure need be referred to at all by the rule for interpreting noun-noun compounds. As for vague words like do, we need to distinguish cases like do the beds, do the potatoes and do the exercise. If we recognize a different sense in the entry for do corresponding to each of these uses, we face problems, because there is at least a very large number of distinct uses, each involving a different kind of action, and it may even be that the list is open-ended (e.g. do a photo can be contextualized). On the other hand, if we recognize only one sense, we are left with the problem of distinguishing between the truth conditions for e.g. do the beds and do the potatoes, since the actions needed to satisfy the conditions are obviously different in the two cases. The frames for concepts are relevant here if we assume that the frames for 'bed' and 'potato' include some kind of specification of the action performed by humans when acting on the object concerned. The 'semantic' structure of do can then be given in the lexicon as (roughly) 'whatever action is defined in slot X of the frame for the concept denoted by the object'. Such an entry can only be formulated if the linguist can refer directly to cognitive structure, and an intervening 'linguistic' structure would be of no help, and might even be a positive hindrance. (2) If we assume that our Being had designed the general conceptual workings of the human mind before starting on language (which is surely a reasonable assumption), it is hard to see any reason why the Being should develop a completely new set of structures for dealing with linguistic expressions. On the contrary, it would be most natural if the Being exploited the existing structures as far as was possible, and only introduced the minimum number of new 286 types of structure. Now, if we believe that non-linguistic knowledge is structured in terms of frames, we should look for such structures in language; and we find them. The most obvious example of a frame-like structure is the word, which combines a number of properties (a meaning, a set of syntactic properties such as a class-membership and some syntactic valency, and an internal structure in terms of morphemes or phonemes, and perhaps some other properties too). We might even apply the distinction between 'frames' and 'scripts' to words, and say that words are scripts because they have an internal temporal structure (e.g. when saying in you first say the /i/ and then the /n/). Like other scripts, they are located in an 'isa' hierarchy, since each word is related to some general class of words (defined by names like 'preposition', 'auxiliary verb', and so on); and like other scripts, they only define the normal patterns, from which deviations of various kinds are possible. Thus, in normally ends in [n], but it may end in [m] if followed by a [b] (e.g. in bed); it is normally a preposition, but it may be used as a noun (e.g. He scored an 'in'). To the extent that there are differences between wordscripts and other kinds of scripts, they seem to involve the details of the slot structures rather than the very general principles controlling their structures and their relations to other structures. (Similar points could probably be made about various other kinds of linguistic elements, such as phonemes, but I think they are most clear in the case of the word, which I consider to be the most important linguistic element of all). What I have tried to do is to show that it would be surprising if linguistic structures were not closely related to cognitive structures of the kind often described in terms of frames, and that this applies not only in the sense that linguistic structures exploit frames (as their meanings), but that they themselves are frames. In the days when we had very little idea what the structure of 'thought' was, it was reasonable to maximize the uniqueness of language, because we did know quite a bit about the structure of language, and there was no need to lower our standards to the vague speculations to which those working on 'thought' in general were prone. Now that there are more interesting models for thought (notably those associated with the word 'frame'), the situation is quite different, and it should be generally recognized by linguists as well as others that we should assume that linguistic structures are the same as non-linguistic structures until we find convincing evidence to the contrary. REFERENCE S Charniak, E. [1981], The case-slot identity theory, in «Cognitive Science» 5, pp. 285-92. Hudson, R.A. [1984], Word Grammar, Oxford, Blackwell. 287