The Alternative to the Classical Model

1. This rereading has already in part begun, and informs, for example, the work of Kleiber 1990, Tsohatzidis 1990, and Geeraerts 19X9.

2. Cf. in particular Lakoff 1987.

3. Cf. for example Kleiber 1990.

4. Cf. Geeraerts 1992.

5. Cf. Rosch 197), I975< '977. '978; Mervis and Rosch 1981; Rosch ei al. 1976,

6. This does not, however, necessarily mean rejecting representation by means ol .1 breakdown into features, In this reipecti prototype semantics is .1 representational theory

Notes to Pages 99-102.

which uses componential analysis, even though it permits models which are not exclusively propositional.

7. Remember that Hjelmslev uses the Danish term mening.

8. As this interaction may vary from culture to culture, the general principles will be applied differently in various cultures, giving rise to different lexical systems.

9. Besides being valid for natural world entities, the same considerations hold true for any other semantic area, ranging from our actions to our feelings or psychic states. The lexis of emotions, for instance, renders pertinent only some forms in the passionate con­tinuum of affective states, forms which are probably perceived to be more significant to subjects or more frequently encountered.

10. For comment on and criticism of the semantic structure of Porphyry's tree, cf. Eco 1984.

11. Berlin 1978; Berlin et al. 1973, 1974. Regarding this issue, see also Cardona 1990.

12. See Hunn (1976), Randall (1976), and Wierzbicka (1985). The claim of universal­ity seems particularly questionable to me. It seems more probable that in our culture the most salient level is that of terms like tree, flower, bird, fish (which, in Berlin's terms, are life forms), and not that of the taxa of genera, such as oak, daisy, swallow, catfish. This naturally depends on the particular nature of interaction with the environment developed by a par­ticular culture; it is not surprising that less industrialized cultures than ours are based on a finer categorization than cultures which no longer have such a need, and have therefore perhaps lost the capacity to distinguish between different types of tree, fish, bird, etc.

13. The three starred categories—fish, bird, and tree—were initially classified as superordinate categories, following the criteria of Berlin who did not classify them as (ba­sic) genera but as life forms. Experimental data showed, however, that at least for English speakers, these categories possessed all the characteristics of the basic level: fish, bird, and tree were more salient than tuna, redbreast, and pine (Rosch et al. 1976). In order to make these data compatible with their scheme, Rosch and her collaborators simply shifted the classification of levels, calling the level that corresponded to tree the basic level instead of the one relating to oak. In this way, they were able to continue referring to the basic level, simply adding that "for biological taxonomies, the basic level appeared to be the next higher level in the taxonomy than had been initially proposed by anthropological and linguistic-taxonomic evidence" (Rosch et al. 1976: 392). This solution has a number of drawbacks. Above all, as Wierzbicka (1985:160) observed, if one shifts the level of intermediate abstrac­tion at will, the entire concept of an intermediate level of abstraction ends up losing all empirical content. Second, and perhaps more importantly, it is not true that for all biologi­cal taxonomies the level of the "basic objects" corresponds to the next higher level in the taxonomy: in the case of animal and dog, it is genus—dog and not animal—which com­prises the basic level. We therefore find that both dog and cat and tree and bird are located on the same basic level, but it is sufficient to analyze more closely the semantic organization of the two terms to identify significant differences. What is different is the relation be­tween the category and its components: while dog is a category of individuals, tree is a category which includes subcategories; the same terms are not at the same level of abstrac­tion. Although from the point of view of the vertical structure of the classification tree or bird are superordinate categories, like animal or furniture, the parallelism does not hold true from the point of view of psychological saliency. This means that in cognitive terms the basic level is not homogenous; this certainly does not invalidate the existence and psy­chological importance of a basic level in the categorial hierarchy, but does make it neces­sary to devote greater attention to internal differences within that level. Although cat is perceived as a unitary category, tree is not, even though both are positioned at the basic level.

14. Cf, for example, Lakoff 1987 and Kieiber 1990.

15. "Basic objects arc the most inclusive categories whose members: (a) possess sig-

Notes to Pages 103-109

nificant numbers of attributes in common, (b) have motor programs which are similar to one another, (c) have similar shapes, and (d) can be identified from averaged shapes of members of the class" (Rosch et al. 1976: 382).

16. In experiments, subjects were shown images (for example, of a greyhound or a golden delicious apple) together with different denominations corresponding to the three levels of abstraction (animal—dog—greyhound; fruit—apple—golden delicious) and asked to say as quickly as possible whether the term and the image corresponded. The reaction times were shortest when the term corresponded to the basic level. The picture of a grey­hound is recognized more quickly as the image of a dog than as the image of an animal or a greyhound. Therefore, it is at the basic level that subjects identify the members of the category most rapidly (Rosch et al. 1976).

17. In reality, the issue is more complex, and the possibility of a unitary image varies according to the specific base categories which are examined. Take the example of dog. however stylized the image of a dog, it always remains that of a particular breed: a near-wolfhound, a near-boxer, etc. It seems difficult to completely erase these traces, even at a high level of stylization. For example, the images in "Beware of the dog" signs, which should be a generic dog, are actually more similar to certain breeds than to others.

18. Tversky 1986; Tversky and Hemenway 1984.

19. On the basis of this, a new characterization of the three levels is proposed: basic categories belonging to the same superordinate category (for example, table, chair, lamp in the furniture category) differ from one another in terms of part o/features but share other attributes; while subordinate categories of the same basic category (for example, various kinds of chair or table) have the part in common but differ in terms of other attributes. For example, all fish (basic category) are distinct from other animals (superordinate) be­cause they possess scales, fins, and gills. Goldfish and salmon share these properties bill differ in size, color, and the fact that the former are kept in domestic aquariums and arc not considered edible, while the latter live in rivers and are eaten. This chatacterization of the levels seems to me the most problematic part of Tversky's hypothesis. Although the example she gives is convincing, it is difficult to generalize. If we consider, for instance, dogs and cats, which are both basic categories like fish, their composition in parts docs not seem to be sufficiently differentiated. Both have a head, four paws, tail, ears, fur, but it is not at this level that we identify the difference between dogs and cats. The basic category in this case does not distinguish in tetms of part «/"properties. There are also difficulties in the case of artifacts: chair, table, and lamp, basic categories of the superordinate furniture, are not initially distinguished in terms of the diffetent composition of their parts, but above all in terms of their different function. This is an abstract property, starting from which the form is then differentiated.

20. Previous experiments concerned only superordinate categories, which are mas­tered from the age of four upward, while there had been no verification of the ability to categorize at the basic level, which begins to appear around the two-year mark (Mervis 1984).

21. Cf. for example, Lakoff 1987, Johnson 1987.

22. The corporeal schema can also be further analyzed in terms of structured dinicn-sions such as high/low, internal/external, etc. For a discussion of the way in which space is structured, starting from our body and its dimensions, see Violi 1991.

23. Cf. Violi 1996b.

24. I am naturally not referring here to pathological situations, which could give rise to involuntary movements.

25. In this respect the concept of event is very similar (albeit more circumscribed) 10 the concept of intcriextiial frames proposed by Kco 1979 and the concept ol script proposed by Schank and Abelson 1977.

26. Examples ol this type arc the novels ol Kobbc (irillcl and I'halr tin regard.

Notes to Pages no—113

5. The Horizontal Dimension of Categories

1. Prototypes are "the clearest cases of category membership defined operationally by peoples judgements of goodness of membership in the category" (Rosch 1978: 36).

2. Basic tetms were defined according to vatious criteria, such as their composition (simple lexemes like red and not compounds like dark red) or the fact that their name was the name of a color and not of an object (ra/and not terracotta).

3. For example, geometric forms (Rosch 1973, 1974) or facial expressions indicating emotions (Ekman 1971).

4. "Perception of typicality differences is, in the first place, an empirical fact of people's judgments about category membership. It is by now a well-documented finding that subjects overwhelmingly agree in theit judgments of how good an example or clear a case members are of a categoty, even for categories about whose boundaries they disagree" (Rosch 1983: 36).

5. Greater attention would allow the cognitive, cultural, and anthropological aspects of prototypes to be linked, enabling the comparison of psycholinguistic and sociolinguistic data, which is essential for a complete theory of meaning. To give an example: even within the same culture it is possible to think of subcultures with differentiated prototypical sys­tems, above all those relating to particular semantic areas; these differences will be con­nected to different linguistic uses and semantic shifts of the same terms, as occurs in slang. An adequate explanation of these linguistic phenomena must take account of the cognitive aspects of the different categorizations and the socio-cultural variables. From this point of view, I believe it is also possible to reformulate the terms of discussion regarding linguistic relativism, overcoming the false dichotomy between cognitive univetsalism and cultural relativism.

6. For example, Rosch and Mervis 1975 mentions prototypic effects.

7. Cf. De Mauro 1994b.

8. This perspective appears to be compatible not only with instructional semantic models, like the one held by Eco (1984), in which lexical meaning is a set of instructions for the relative contextual insertions, but also with pragmatically oriented models like that of Sperber and Wilson 1986. In this case too, comprehension is seen not purely as decoding but as an inferential process in which the code in the strict sense of the term has a much weaker and more limited status than in classical models.

9. It is interesting to note that these two positions co-exist in the early works of Eleonor Rosch, as the following two passages reveal:

"Many natural categories are internally structured into a prototype (clearest cases, best examples) of the category with nonprototype members tending towards an order from better to poorer examples" (Rosch 1975: 544).

However, in the same year we find explicit references to family resemblance as the internal principle of categorial organization:

"The purpose of the present research was to explore one of the majot structural prin­ciples which, we believe, may govern the formation of the prototype structure of seman­tic categories. This principle was first suggested in philosophy; Wittgenstein (1953) argued that the referents of a word need not have common elements in order for the word to be understood and used in the normal functioning of language. He suggested that, rather, a family resemblance might be what linked the various referents of a word. A family resem­blance relationship consists of a set of items of the form AB, ВС, CD, DE. That is, each item has at least one, and probably several, elements in common with one or more other items, but no, or few, elements are common to all items" (Rosch and Mervis 1975: 574-575)-

(ieeraens (1992) also stresses llie mixed features of the theory, while Klcibcr (1990) stresses the representational reading of the early phase.

Notes to Pages 114-136

10. This idea is also based on the use of a patticular logic of fuzzy sets, created by Lofti Zadeh (1965).

11. In reality these two conditions are not sufficient to delimit the categoty; the platypus, for instance, which has a beak and lays eggs, is a mammal. However, this does not change the substance of the argument, because to make the scheme acceptable it suffices to add the property of not breastfeeding theit young.

12. Someone might ask why I treat the two properties that Geeraerts implicitly views as being necessary (having a beak or bill and being oviparous) in a different way, being more inclined to "relax" the first but not the second. In actual fact, I believe that there is a crucial difference in status between the two features, which I will explain fully later.

13. As Geeraerts (1988b: 222) observes: "The apple is not a prototypical fruit because we talk more about apples than about mangoes, but because we experience apples more often than we encounter mangoes (and this fact, in turn, may be the reason why we talk more about apples)."

14. According to Rosch and Metvis (1975: 575), the cue validity of a certain property in a category is obtained by dividing the frequency of that property in the category by the total frequency of the same property in all other pertinent categories.

15. Cf. Hjelmslev [1954] 1959, [1957] 1959.

16. Indeed, one might reasonably ask to what extent the fact that 7 is perceived to be a better example of an odd number than 8,461 concerns the meaning of odd numbers.

17. Lakoff (1987) seems to take this line, and Kleiber (1990) also accentuates the rift between the first version of rhe theory, which he defines as standard, and the second which he defines as extended. For a less extreme position, see Geeraerts 1992.

18. I have substituted the Dutch term vers (fresh) with newspaper, and I have added fruit to the four terms analyzed by Geeraerts (bird, fresh, red, odd numbers), in order to provide an example of prototypic prototypicality.

19. In fact, the prototype of bird refers to an individual definable by properties that are different from other individual members of the category (referentially, therefore, we have discrete units). In the case of red, the prototype is a focal point on a scale, and we do not have discrete entities nor the possibility of indicating specific properties that distin­guish discretely one instance of red from another (properties like "fly/cannot fly" in the case of bird). Finally, the prototypic exemplats of odd number are specific cases that can be discretely differentiated ftom non-prototypic exemplars, but which can in no way be distinguished from them in terms of defining propetties (they are all odd numbers in ex­actly the same way).

20. Cf. Craig 1986.

21. Lakoff himself, for example, in analyzing the categories of bayi and balan, con­siders their central and prototypic instances to be respectively man and woman. For a critique of this analysis, see Mylne 1995, who argues that the system of nominal classifica­tion in Dyirbal is not constructed around real entities in the world, but in relation to bands of properties which define the categories.

22. Another example is numeral classifiers. In almost all East Asian languages (Japa­nese, Vietnamese, Burmese, Chinese, etc.), numeral expressions require a modifier which specifies the class of what is being counted. In Japanese, for example, the classifiers vary according to whether objects are round, oblong, made of paper (books and newspapers), machines, buildings, and so on, right down to a classifier for tights and socks.

23. According to Fillmore (1982b: 36) a theory of prototype semantics allows, in deed, demands, that we separate the primary meanings ol linguistic terms from derived meanings.

24. This concept is very similar to il not the same as that which some writers includ ing Putnam call stereotype. I lowever, I prefer the expression typical meaning because the term stereotype is often used with a different and more sociological meaning.

л. < Ir.nlv. dlffereni pans ol speech will raise different problemi,

Notes to Pages 141-147

6. The Structure of Semantic Properties

1. Cf. Coseriu 1974 and Pottier 1974. Classemes are very general sense components that are common to terms belonging to different lexical fields; they represent the systematic part of meaning and tend to be not only lexicalized but also grammaticalized. Examples of classemes are opposites like animate/inanimate, male/female, and such like. Semes, on the other hand, are the minimal distinctive features of meaning, specific and idiosyncratic for each term, like those that distinguish chair from armchair in Pottier's analysis. Greimas (1966) uses the term classeme with a different meaning to indicate contextual semes, that is, semes that recur in discourse and assure its isotopy. Semes, on the other hand, are mini­mal units at the level of content, closer to the semantic markers of Katz and Fodor.

2. Cf. Bolinger 1965, Haiman 1980, Eco 1976.

3. Certainly, within the properties belonging to the prototypic instance of the cate­gory, it is possible to distinguish differences in terms of the parameter "gradual versus discrete." The standard height and width of a cup, like a color range, are gradual, while having wings or feathers or a striped coat, in the case of birds and tigers, is not. However, this difference can be traced back not to different properties but to the overall structural description of the category, and concerns whether it is fuzzy or defined. Cups and glasses, and probably all artifacts, have fuzzy boundaries in terms of their formal and perceptual properties, while natural kinds, at least as we conceptualize them, do not. But it is by no means certain that a non-fuzzy category only presents typical and non-gradual proper­ties. Despite Jackendoff s claim, the color of certain natural kinds, for example lemons, is gradual and can range from yellow to green.

4. Other distinctions have been drawn by Coleman and Kay (1981) between proto­typical properties and typical properties (limited, however, just to the analysis of the verb lie), by Lakoff (1987), and by Wierzbicka (1985). The last suggests distinguishing between essential and non-essential properties on the basis of a diagnostic test which is, to say the least, problematic: essential properties are those that correspond to the formula "imagining things (animals, creatures) of this kind, people would say these things about them," while for non-essential properties the formula is the things "that people could say." The difference between the two relies on the questionable distinction between would and could. In all of Wierzbicka's detailed analyses of natural kinds, there is not a single essential property, with the sole exception, for terms corresponding to the basic level like bird or tree, of the fact that there are different types in the category. All the other properties (including, for example, "oviparous" and "having a beak or bill" for birds) are considered non-essential. The confusion between the two types of property is even more glaring in the analysis of certain terms like mouse in which all the properties are listed under the heading "things that people would/could say," thus annulling any distinction between the essential and the non­essential.

5. Cf. Cruse (1986). Cruse distinguishes between five different types of feature: cri-terial, expected, possible, unexpected, excluded. Criterial features correspond to my essential conditions, while expected and possible features correspond, with some exceptions that I will discuss, to typicality. Neither unexpected nor excluded features are part of meaning and their differing degrees of unacceptability result from their respective opposition to typical properties and essential properties.

6. This is the main difference from Wierzbicka, for whom essential properties are positive membership criteria because they constitute "the smallest set of features which, taken together, ensure than any object which has them will be generally recognized as a member of the category in question" (1985: 60).

7. Naturally, the fact that whales are mammals is not a matter of cultural choice, but concerns the nature of these animals; what is cultural is the decision to select as con-.stilulivc ibis property rather than others. In other words, although the ontological level

Notes to Pages 147-158

in certain cases motivates the selection of properties, it does not determine the semantic structure.

8. Putnam (1975) hypothesized the existence of a twin earth in which certain enti­ties called cats, entirely similar in appearance to our cats, were in reality robots. The argu­ment was used by Putnam to sustain the thesis of direct reference also for natural kinds.

9. Clearly, the decision to redefine cat would probably also imply a transformation of the meaning of animal: if cats are not animals, I no longer really know what an animal is, while classifying the whale as a mammal rather than a fish does not involve modifying the meaning of these categories.

10. See also Kleiber (1990: 123) on this point.

11. For a similar treatment, cf. Fillmore 1982b and Lakoff 1987.

12. Cf. Eco (1984: 84-85): "The system of hyperonyms provided by a dictionary rep­resents a way to save 'definitional energies'. . . . The function of hyperonyms in a lexical system depends exactly on the epistemological decisions that govern the life of a culture. We can make up dictionary-like representations in order to save definitional energies in any context in which certain 'central' assumptions of a cultutal system are taken for granted."

13. "Artifacts" should be understood here in a broad sense, similar to the class that Lyons (1977) defines as "cultural kinds terms" to distinguish them from natural kinds. Basically, the category includes artificial concrete objects like chair, piano, telephone, church, car, house, etc.

14. Cf. Lakoff 1973.

15. Naturally this is true also for some of the properties defined as essential for natu­ral kinds (for example, "animal" for cat). However, in natural kinds it is also possible to find essential properties (which are thus not erasable) that positively define a single class. For example, in Geeraerts' proposal, the combination of "oviparous" and "having a beak or bill" delimits univocally and specifically the class of birds.

16. Lavoro comes from the Latin labor: fatigue, toil, suffering (Semerano 1994).

17. Cf. Semerano 1994.

18. Even though naturally there may be responsibilities regarding the rules of a game.

19. One can very easily imagine a culture in which punching one another is consid­ered pleasurable and thus classified as play.

20. Lakoff (1987) distinguishes in these cases between Idealized Cognitive Model on which the prototype depends—and the classical structure of the category, based on necessary and sufficient membership conditions.

21. Bachelor: "unmarried man"; single: "unmarried. Pertaining to the unmarried state" {Random House Unabridged Dictionary of the English Language).

22. Greimas and Courtes 1979: 62.

23. Cf. Cruse 1986: 16.

24. Cruse (1986) also indicates a difference between canonical and non-canonical features. A canonical feature is, for example, a dog having four legs; this is an erasable feature but would indicate an anomalous individual. A non-canonical feature such as "be­ing able to fly" for birds, on the other hand, if canceled, would indicate the presence of an atypical but not a malformed exemplar. The difference, according to Cruse, is clear in sentences like (1) and (2):

1. The typical bird can fly.

2. ? The typical dog has four legs, where, in Cruses view, (2) is anomalous.

7. The Delimitation of Lexical Meaning

1. Cf. Iл.. i<)X.|.

i. CI'. Violi 1998,

Notes to Pages 159-173

3. In this second sense, the encyclopedia becomes very similar to Foucault's concept of the episteme of a given age.

4. The differences at a textual and interpretative level between these two types of property will be analyzed in chapter 9.

5. This thesis can already be seen in Saussure {CGL: no).

6. The overall encyclopedia relating to the word whale is naturally subject to indi­vidual variation and not necessarily all members of a given culture possess it to the same degree. It is, however, possible to imagine an average encyclopedic competence that an individual must possess in order to belong to a given culture, even though this concept should then be further articulated into more diversified sub-competences. (The encyclope­dic competence expected of a university professor is not generally the same as that expected of a traffic policeman, although their actual competences might hold some surprises.)

7. The idea of a competence that is assumed to be shared is close to the idea of "naive semantics" as the basis for the understanding of natural language, which has been advanced in the field of artificial intelligence (cf. Dahlgren 1988), and in anthropology (cf the notion of commonplace knowledge in Tyler 1978). Also, in linguistics Apresjan has already sustained that the semantics of a linguistic sign reflect the naive idea of a thing, property, action, etc., a naive idea that forms and is consolidated in time through repeated experience, or we could say, through the experience of our continual interaction with the world and with the surtounding environment. "The naive picture of the world, which slowly forms itself throughout centuries, and which includes a naive geometry, a naive physics, a naive psychology and so on, reflects the material and spiritual experience of a people. . . . The task of a lexicographer . . . consists in revealing the naive picture of the world hidden in the lexical meaning of words and reflecting it in a system of explications" (Apresjan 1974: 57; quoted in Wierzbicka 1985: 197).

8. The correspondence is only partial because the average semantic competence of a speaker will certainly not correspond to the entire dictionary of a language.

9. Note that the test functions as a lexical ctiterion only in a positive, and not a negative, sense. In fact, a nightingale also emits a sound that we attribute a particular value to and which we may wish to indicate in the representation (i.e., consider part of semantic competence) even though there is no specific term to describe it in the lexicon.

10. Cf. Hjelmslev [1957] 1959.

11. Wierzbicka talks about different possible concepts for the same referents, but in order to avoid the ambiguity of this formulation I prefer to talk about different compe­tences.

12. Cf. Violi 1996c, I996d. To arrive at the formation of the habit, understood both as a disposition toward action and as a moment of stabilization of the semiosis in a general form, Peirce identifies a hierarchy of interpretants involving two primary levels (emotional interpretant and energetic interpretant) that are still single acts, and so connected to the individual field. There is also a third level, the logical or final interpretant, manifested in the habit and of a general order, that corresponds to meanings or concepts. The crucial point in this hietarchy of interpretants is that the level of generality is not irremediably separate from the levels of individual knowledge that precede and produce it, but actually develops from them. Access to subjective experience remains open and fluid, leaving open the possibility of return, modification, and circularity. The individual level of signification thus remains a potential that is open to possible new reformulations of sense. I have sug­gested thinking of the relation between the habit and individual interpretants as the rela­tion between general form and the individual path of access to that form; although the concepts represent the general form, the paths by which they are formed in each individual varies, because individual experience is always different. In other words, access to the gen­erality of concepts follows individual paths and modalities, and a trace remains of these modalities at the level of those singular specifications of concepts that are determined by our specific experience ami that wc could consider to be free variants on the plane of con-

Notes to Pages 174—189

tent. Viewing the connection between the levels of individual experience and generality in this way also permits a better understanding of phenomena of change and transformation of meanings. These are the result of the stabilization of a habit; however, the habit is never static and defined once and for all, but can be ttansformed following different experiences that subjects have, precisely because the connection between individual experiences and their generalization is always open to possible reformulations. On the basis of new experi­ences, new individual interpretants are produced; that is, new representations that can in turn produce new habits, which flourish as new general meanings.

13. See, for example, Johnson-Laird 1983, Moravcsik 1981, and the many works de­voted to the subject by Marconi (1987, 1991, 1995), now collected in Marconi 1997.

14. "One's inability to pick out, under normal empirical conditions, members of the extension of a given term does not show that one lacks an adequate mental representation of the associated intension" (Moravcsik 1981:13).

15. The interest of a distinction like this one also lies in the possibility of introduc­ing a criterion of cultural differentiation, although this perspective is not mentioned by Putnam.

16. This naturally does not mean, as I have observed and as Marconi has also under­lined, that the content of individual competence has to be homogenous in terms of format with the content of dictionary entries.

8. Regularity and Context

1. Cf. Eco 1979.

2. Cf. Duranti and Goodwin 1992 for a reading of the notion of context from an anthropological perspective. Particularly well documented is the variety of positions relat­ing to die extension of context even within the same ethno-anthropological field: while authors such as Schegloff are in favor of as restricted a use as possible, others such as Cicourel extend context to take in even the institutional structures within which interac­tions take place. For a discussion from a semiotic perspective of issues concerning context, cf. Casetti 1994.

3. Cf. Gorfein 1989.

4. This idea is present in the concept of codified contextual selections in Eco 1976.

5. If I have understood correctly, this is the line taken by Almeida. In fact, after having suggested that words function as an "addresses" or "research signals"—a view very similar to the one I have put forward here—he proposes to consider all terms, even syn-categorematic ones, as rigid designators of themselves. "All this should reinforce the idea that words do not contain their meaning—apart from being self-designatory—any more than they find it in the language system" (Almeida 1992: 13; my translation).

6. One distinction does, however, need to be clear. Since the seventies, various disciplines, in particular sociology and anthropology on the one hand and cognitive psy­chology and artificial intelligence on the other, have used the concept of the frame or schema. They do not, however, use these concepts in the same way. In social anthropology, the notion of the frame was introduced by Bateson (1972) to explain how it is that indi­viduals not only exchange messages possessing explicitly recognizable content, but also metacommunicative signals that indicate how and at what level that message must be in terpreted. Similarly, in the animal kingdom forms of metacommunication often signal, lor example, that a certain kind of behavior should be interpreted as a game and not as a form ol aggression. It is this idea of an "aim" that suggests the term frame, which highlights the general framing ol the type of activity that is being carried out. This is the sense in which it is used by I lymes 1974, ⁽lollman 197.1, and generally in the ethnography of language.

The same term is used very differently in cognitive psychology and artificial inlelli gence, Hew it emphtllwe OUt Organization ol knowledge rathei than the spei ifit.ilion of

Notes to Pages 189—190

the kind of interaction. In psychology the first formulation of the notion of schema prob­ably dates back to the famous work of Battlett (1932) on memory, in which he argued that memory is not merely cumulative but is also constructive in nature. In the seventies and eighties, studies of the schematic structures of knowledge proliferated, both in cognitive psychology and in artificial intelligence, starting with the first definition of frame proposed by Minsky (1975) who suggested that it was a data structure to represent a stereotyped situation. See also Bobrow and Collins 1975; Charniak 1975; Schank and Abelson 1977; Rumelhart and Ortony 1977; Thomdyke 1977. For a partial review of these concepts, see Violi 1982.

7. Rumelhart (1980: 34), for example, underlines this point: "A schema theory is basically a theory about knowledge. It is a theory about how knowledge is represented and about how that representation facilitates the use of the knowledge in particulat ways. Ac­cording to schema theories, all knowledge is packaged into units. These units are the sche­mata. Embedded in these packets of knowledge is, in addition to the knowledge itself, information about how thus knowledge is to be used."

8. Besides various works by Fillmore (1976a, 1976b, 1977, 1982b, 1985), see Metzing 1980, Eikmeyer and Rieser 1981, De Beaugrande and Dressier 1981, Wilks 1980, Lehrer and Kittay 1992.

9. So, for example, the verb run can be inserted into a schema that envisages just one case, the agentive—understood as the case of the author of the action expressed by the verb and typically felt to be animate. The verbs move and open can be inserted into a schema that envisages the agentive case and the objective case (concerned with whatever is affected by the action or state expressed by the verb); a verb like give involves a schema with three positions, that includes the agentive, the objective, and the dative, the case of the animate being affected by the state or action of the verb. This, at least, was the formulation of Fillmore 1968. The list of cases and their denominations wete to undetgo numerous modifications, as can be seen in Fillmote 1977.

10. The representation of only case roles creates in this respect a restrictive semantics (cf. Ravin 1990).

11. In reality, Fillmore 1976b distinguishes four separate levels: scenes refer to the ex­periences, actions, objects, and perceptions of the real world, and individuals' memory of them; schemata indicates the reference frameworks through which actions, events, and ob­jects are categorized; frames designate the specific lexical-grammatical possibilities that in a given language permit the description of the categories and the relations identified at the level of the schemata; finally, there are the models, the overall representations that an intet-preter must construct in order to give sense to a particular text. This terminology was to be modified over the course of time.

12. In Fillmore's model, the description of the valencies of a verb correspond in part to what in other semantic models are contextual restrictions. Other semantic information traditionally connected to contextual restrictions is also inttoduced here as descriptions of the scene rather than as special types of restriction existing between terms, or classes of terms, in given grammatical constructions.

13. The semantic frame also contains specific information relating to the temporal and aspectual determinations of a given verb. For example, in English there are particular pieces of information associated with the use of the progressive, the past, and the present, for each predicate. In the case of write, the progressive form denotes an activity in progress, the past a completed act, the present simply a profession. The semantic representation must also give an account of the effects produced by particular tenses and verbal aspects, so as to explain why Harry writes is understood as an assertion regarding Harry's profession, while Harry drinks, although the same tense is used, cannot be interpreted in the same way. In Fillmore's model, temporal and aspectual specifications like these arc inserted into the representation of the scene associated with the term and not in a separate representation of

Notes to Pages 190-207

the temporal system. Unlike other models, frame semantics does not have two distinct representations for the tense system and for the meaning of the term, and consequent rules of projection for a possible amalgam of the two systems, but a single representation regard­ing the prototypical scene associated with a term, in which all the necessary information on tenses and verbal aspects is grouped together.

14. Fillmore (1985: 2.33) is very clear on this point: "In this respect, frame semantics can be said to take a much more encyclopedic view of meaning than is common. In par-ticular, it does not seek to dtaw an a priori distinction between semantics proper and (an idealized notion of) text understanding; rather it sees the units and categories of language as having come into being in the first place to serve the purposes of communication and understanding."

15. As Fillmore (1985: 223) says, "What holds such word groups togethet is the fact of their being motivated by, founded on, and co-structured with, specific unified frame­works of knowledge, or coherent schematizations of experience, for which the general word frame can be used."

16. Fillmore subscribes to this position in relation to frames, as does Fauconnier (1985: 1) who, referring to his mental spaces, says that they "are not part of the language itself" even though "language does not come without them."

17. In Fillmore's frame semantics, for example, the concept of contrast cotresponds to that of lexical framing.

18. In Greimasian semiotics, a narrative program is defined as a change of state ef­fected by a subject and which regards anothet subject (the two may coincide). Cf. Greimas and Courtes 1979: 297.

19. See, for example, the detailed analysis of risk in Fillmore and Atkins 1992.

20. In truth, some of the criticisms leveled by semioticians at Fillmore's concept of case concerned this aspect of causal structure; in theit view, Fillmote does not reach down to the level of deep actantial sttuctute, which is considered more pertinent, but remains tied to the surface level of grammatical representation. This confusion of levels creates difficulty in defining once and for all in a categotical way the numbet and denomination of cases. On this point, see Petitot 1985.

xi. Cf. Fillmore 1977.

22. This is the same problem of the level of abstraction that we saw for cases.