Fillmore_Frame Semantics 原版

Chapter 10 Frame semantics Frame semantics Charles J. Fillmore 1. Introduction With the term ‘frame semantics’ I have in mind a research program in empirical semantics and a descriptive framework for presenting the results of such research. Frame semantics offers a particular way of looking at word meanings, as well as a way of characterizing principles for creating new words and phrases, for add- ing new meanings to words, and for assembling the meanings of elements in a text into the total meaning of the text. By the term ‘frame’ I have in mind any system of concepts related in such a way that to understand any one of them you have to understand the whole structure in which it fits; when one of the things in such a structure is introduced into a text, or into a conversation, all of the others are automatically made available. I intend the word ‘frame’ as used here to be a general cover term for the set of concepts variously known, in the literature on natural language understanding, as ‘schema’, ‘script’, ‘scenario’, ‘ideational scaf- folding’, ‘cognitive model’, or ‘folk theory’.1 Frame semantics comes out of traditions of empirical semantics rather than formal semantics. It is most akin to ethnographic semantics, the work of the anthropologist who moves into an alien culture and asks such questions as, ‘What categories of experience are encoded by the members of this speech community through the linguistic choices that they make when they talk?’ A frame semantics outlook is not (or is not necessarily) incompatible with work and results in formal semantics; but it differs importantly from formal semantics in emphasizing the continuities, rather than the discontinuities, between language and experience. The ideas I will be presenting in this paper represent not so much a genuine theory of empirical semantics as a set of warnings about the kinds of problems such a theory will have to deal with. If we wish, we can think of the remarks I make as ‘pre-formal’ rather than ‘non-formalist’; I claim to be listing, and as well as I can to be describing, phenomena which must be well understood and carefully described before serious formal theorizing about them can become possible. In the view I am presenting, words represent categorizations of experience, and Originally published in 1982 in Linguistics in the Morning Calm, Linguistic Society of Korea (ed.), 111–137. Seoul: Hanshin Publishing Company. Reprinted with permission. 374 Carles. J. Fillmore each of these categories is underlain by a motivating situation occurring against a background of knowledge and experience. With respect to word meanings, frame semantic research can be thought of as the effort to understand what reason a speech community might have found for creating the category represented by the word, and to explain the word’s meaning by presenting and clarifying that reason. An analogy that I find helpful in distinguishing the operation and the goals of frame semantics from those of standard views of compositional semantics is between a grammar and a set of tools – tools like hammers and knives, but also like clocks and shoes and pencils. To know about tools is to know what they look like and what they are made of – the phonology and morphology, so to speak – but it is also to know what people use them for, why people are interested in doing the things that they use them for, and maybe even what kinds of people use them. In this analogy, it is possible to think of a linguistic text, not as a record of ‘small meanings’ which give the interpreter the job of assembling these into a ‘big meaning’ (the meaning of the containing text), but rather as a record of the tools that somebody used in carrying out a particular activity. The job of interpreting a text, then, is analogous to the job of figuring out what activity the people had to be engaged in who used these tools in this order. 2. A private history of the concept ‘frame’ I trace my own interest in semantic frames through my career-long interest in lexical structure and lexical semantics. As a graduate student (at the University of Michigan in the late fifties) I spent a lot of time exploring the co-occurrence privileges of words, and I tried to develop distribution classes of English words using strings of words or strings of word classes as the ‘frames’ within which I could discover appropriate classes of mutually substitutable elements. This way of working, standard for a long time in phonological and morphological investigations, had been developed with particular rigor for purposes of syntactic description by Charles Fries (Fries 1952) and played an important role in the development of ‘tagmemic formulas’ in the work of Kenneth Pike (Pike 1967), the scholars who most directly influenced my thinking during this period. Substitutability within the same ‘slot’ in such a ‘frame’ was subject to certain (poorly articulated) con- ditions of meaning-preservation or structure-preservation, or sometimes merely meaningfulness-preservation. In this conception, the ‘frame’ (with its single open ‘slot’) was considered capable of leading to the discovery of important function- ing word classes or grammatical categories. As an example of the workings of such a procedure, we can take the frame consisting of two complete clauses and a gap between them, as in John is Mary’s husband – he doesn’t live with her. The substitution in this frame of but and yet suggests that these two words have (by Chapter 10: Frame semantics 375 this diagnostic at least) very similar functions; insertion of moreover or however¬ suggest the existence of conjunctions functioning semantically similarly to but and yet but requiring sentence boundaries. The conjunctions AND and OR can meaningfully be inserted into the frame, but in each case (and in each case with different effect) the logical or rhetorical ‘point’ of the whole utterance differs importantly from that brought about by but or yet. In each of these cases, what one came to know about these words was the kind of structures with which they could occur and what function they had within those structures. In the early sixties, together with William S.-Y. Wang and eventually D. Ter- ence Langendoen and a number of other colleagues, I was associated with the Project on Linguistic Analysis at the Ohio State University. My work on that project was largely devoted to the classification of English verbs, but now not only according to the surface-syntactic frames which were hospitable to them, but also according to their grammatical ‘behavior’, thought of in terms of the sensitivity of structures containing them to particular grammatical ‘transformations.’ This project was whole-heartedly transformationalist, basing its operations at first on the earliest work on English transformational grammar by Chomsky (1957) and Lees (1961), and in its later stages on advances within the theory suggested by the work of Peter Rosenbaum (Rosenbaum 1967) and the book which established the standard working paradigm for transformationalist studies of English, Chomsky (1965). What animated this work was the belief that discoveries in the ‘behavior’ of particular classes of words led to discoveries in the structure of the grammar of English. This was so because it was believed that the distributional properties of individual words discovered by this research could only be accommodated if the grammar of the language operated under particular working principles. My own work from this period included a small monograph on indirect object verbs (Fillmore 1961) and a paper which pointed to the eventual recognition of the transformational cycle as an operating principle in a formal grammar of English (Fillmore 1963). The project’s work on verbs was at first completely syntactic, in the sense that what was sought was, for each verb, a full account (expressed in terms of subcat- egorization features) of the deep structure syntactic frames which were hospitable to it, and a full account (expressed in terms of rule features) of the various paths or ‘transformational histories’ by which sentences containing them could be transformed into surface sentences. The kind of work I have in mind was carried on with much greater thoroughness by Fred Householder and his colleagues at Indiana University (Householder et al 1964), and with extreme care and sophis- tication by Maurice Gross and his team in Paris on the verbs and adjectives of French (Gross 1975). In the late sixties I began to believe that certain kinds of groupings of verbs and classifications of clause types could be stated more meaningfully if the struc- 376 Carles. J. Fillmore tures with which verbs were initially associated were described in terms of the semantic roles of their associated arguments. I had become aware of certain American and European work on dependency grammar and valence theory, and it seemed clear to me that what was really important about a verb was its ‘semantic valence’ (as one might call it), a description of the semantic role of its arguments. Valence theory and dependency grammar did not assign the same classificatory role to the ‘predicate’ (or ‘VP’) that one found in transformationalist work (see, e.g., Tesnière 1959); the kind of semantic classifications that I needed could be made more complete and sensible, I believed, if, instead of relying on theoreti- cally separate kinds of distributional statements such as ‘strict subcategorization features’ and ‘selectional features,’ one could take into account the semantic roles of all arguments of a predication, that of the ‘subject’ being simply one of them. Questioning, ultimately, the relevance of the assumed basic immediate-constitu- ency cut between subject and predicate, I proposed that verbs could be seen as basically having two kinds of features relevant to their distribution in sentences: the first a deep-structure valence description expressed in terms of what I called ‘case frames’, the second a description in terms of rule features. What I called ‘case frames’ amounted to descriptions of predicating words that communicated such information as the following: ‘Such-and-such a verb occurs in expressions containing three nominals, one designating an actor who performs the act desig- nated by the verb, one designating an object on which the actor’s act has a state- changing influence, and one designating an object through the manipulation of which the actor brings about the mentioned state change.’ In symbols this state- ment could be represented as [— A P I], the letters standing for ‘Agent’, ‘Patient’ and ‘Instrument’. Actually, the kind of description I sought distinguished ‘case frames’ as the structures in actual individual sentences in which the verbs could appear from ‘case frame features’ as representations of the class of ‘case frames’ into which particular verbs could be inserted. In the description of ‘case frame features’ it was possible to notice which of the ‘cases’ were obligatory, which were optional, what selectional dependencies obtained among them, and so on (see Fillmore 1968). We were developing a kind of mixed syntactic-semantic valence description of verbs, and we noticed that the separate valence patterns seemed to character- ize semantic types of verbs, such as verbs of perception, causation, movement, etc. Within these syntactic valence types, however, it seemed that some semantic generalizations were lost. There seemed to be important differences between give it to john and send it to chicago that could not be illuminated merely by showing what syntactic rules separate give from send, just as there seemed to be semantic commonalities between rob and steal�¬buy and sell�¬enjoy and amuse, etc., which were lost in the syntactic class separation of these verbs. My ultimate goal in this work in ‘case grammar’ (as the framework came Chapter 10: Frame semantics 377 to be called) was the development of a ‘valence dictionary’ which was to differ importantly from the kinds of valence dictionaries appearing in Europe (e.g., Helbig and Schenkel 1973) by having its semantic valence taken as basic and by having as much as possible of its syntactic valence accounted for by general rules. (Thus, it was not thought to be necessary to explain, in individual lexical entries, which of the arguments in a [V A P I] predication of the type described above was to be the subject and which was to be the object, since such matters were automatically predicted by the grammar with reference to a set of general principles concerning the mapping from configurations of semantic cases into configurations of grammatical relations.) Although the concept of ‘frame’ in various fields within cognitive psychology appears to have origins quite independent of linguistics, its use in case grammar was continuous, in my own thinking, with the use to which I have put it in ‘frame semantics’. In particular, I thought of each case frame as characterizing a small abstract ‘scene’ or ‘situation’, so that to understand the semantic structure of the verb it was necessary to understand the properties of such schematized scenes. The scene schemata definable by the system of semantic cases (a system of semantic role notions which I held to be maximally general and defining a mini- mal and possibly universal repertory) was sufficient, I believed, for understanding those aspects of the semantic structure of a verb which were linked to the verb’s basic syntactic properties and to an understanding of the ways in which differ- ent languages differently shaped their minimal clauses, but they were clearly not adequate for describing with any completeness the semantic structure of the clauses containing individual verbs. This theory of semantic roles fell short of providing the detail needed for semantic description; it came more and more to seem that another independent level of role structure was needed for the semantic description of verbs in par- ticular limited domains. One possible way of devising a fuller account of lexical semantics is to associate some mechanism for deriving sets of truth conditions for a clause from semantic information individually attached to given predicates; but it seemed to me more profitable to believe that there are larger cognitive structures capable of providing a new layer of semantic role notions in terms of which whole domains of vocabulary could be semantically characterized. My first attempt to describe one such cognitive structure was in a paper on ‘Verbs of judging’ (Fillmore 1971) – verbs like blame�¬accuse�¬criticize – for which I needed to be able to imagine a kind of ‘scene schematization’ that was essentially different from the sort associated with ‘case frames’. In devising a framework for describing the elements in this class of verbs, I found it useful to distinguish a person who formed or expressed some sort of judgment on the worth or behavior of some situation or individual (and I called such a person the Judge); a person concerning whose behavior or character it was relevant for the 378 Carles. J. Fillmore Judge to make a judgment (I called this person the Defendant); and some situa- tion concerning which it seemed relevant for the Judge to be making a Judgment (and this I called simply the Situation). In terms of this framework, then, I chose to describe accuse as a verb usable for asserting that the Judge, presupposing the badness of the Situation, claimed that the Defendant was responsible for the Situ- ation; I described criticize as usable for asserting that the Judge, presupposing the Defendant’s responsibility for the Situation, presented arguments for believ- ing that the Situation was in some way blameworthy. The details of my descrip- tion have been ‘criticized’ (see esp. McCawley 1975), but the point remains that we have here not just a group of individual words, but a ‘domain’ of vocabulary whose elements somehow presuppose a schematization of human judgment and behavior involving notions of worth, responsibility, judgment, etc., such that one would want to say that nobody can really understand the meanings of the words in that domain who does not understand the social institutions or the structures of experience which they presuppose. A second domain in which I attempted to characterize a cognitive ‘scene’ with the same function was that of the ‘commercial event’ (see Fillmore 1977b). In particular, I tried to show that a large and important set of English verbs could be seen as semantically related to each other by virtue of the different ways in which they ‘indexed’ or ‘evoked’ the same general ‘scene’. The elements of this schematic scene included a person interested in exchanging money for goods (the Buyer), a person interested in exchanging goods for money (the Seller), the goods which the Buyer did or could acquire (the Goods), and the money acquired (or sought) by the seller (the Money). Using the terms of this framework, it was then possible to say that the verb buy focuses on the actions of the Buyer with respect to the Goods, backgrounding the Seller and the Money; that the verb sell focuses on the actions of the Seller with respect to the Goods, backgrounding the Buyer and the Money; that the verb pay focuses on the actions of the Buyer with respect to both the Money and the Seller, backgrounding the Goods, and so on, with such verbs as spend, cost, charge, and a number of others somewhat more peripheral to these. Again, the point of the description was to argue that nobody could be said to know the meanings of these verbs who did not know the details of the kind of scene which provided the background and motivation for the categories which these words represent. Using the word ‘frame’ for the structured way in which the scene is presented or remembered, we can say that the frame structures the word-meanings, and that the word ‘evokes’ the frame. The structures I have mentioned so far can be thought of as motivating the categories speakers wish to bring into play when describing situations that might be independent of the actual speech situation, the conversational con- text. A second and equally important kind of framing is the framing of the actual communication situation. When we understand a piece of language, we Chapter 10: Frame semantics 379 bring to the task both our ability to assign schematizations of the phases or components of the ‘world’ that the text somehow characterizes, and our ability to schematize the situation in which this piece of language is being produced. We have both ‘cognitive frames’ and ‘interactional frames’, the latter having to do with how we conceptualize what is going on between the speaker and the hearer, or between the author and the reader. By the early seventies I had become influenced by work on speech acts, performativity, and pragmatics in general, and had begun contributing to this field in the form of a number of writings on presuppositions and deixis (see, e.g., Fillmore 1975). Knowledge of deictic categories requires an understanding of the ways in which tenses, person marking morphemes, demonstrative categories, etc., schematize the communi- cating situation; knowledge of illocutionary points, principles of conversational cooperation, and routinized speech events, contribute to the full understand- ing of most conversational exchanges. Further, knowing that a text is, say, an obituary, a proposal of marriage, a business contract, or a folktale, provides knowledge about how to interpret particular passages in it, how to expect the text to develop, and how to know when it is finished. It is frequently the case that such expectations combine with the actual material of the tex