Prototype semantics

from Wikipedia, the free encyclopedia

The prototype semantics is a special semantics that was derived from the prototype theory that was developed in the 1970s by Eleanor Rosch and colleagues. It combines the knowledge of psychology and linguistics and describes the quantitative gradation of the membership of entities in categories . It postulates prototypes as central normative category elements, with a more or less large “distance” between the other category members. This differs from the conventional way of thinking that one can always qualitatively unambiguously decide on the membership of a category, namely by specifying the necessary conditions. Typical for this classification by definition is e.g. B. “If it has feathers and a beak and can fly, then it is a bird.” However, experiments showed that humans classify some bird species (especially songbirds) significantly more as birds than marginal species such as chicken, ostrich or penguin. So there are category members who can be considered typical representatives, while the classification of atypical representatives is much more difficult. The effect of the quantitative gradation occurs even with apparently clear categories such as “bachelor” or “even number”.

Prototype semantics as a cognitive science

Prototype semantics not only play an important role in the area of ​​lexical semantics. It was also the "origin of the cognitive paradigm shift in linguistics " (Blank, 2001: 44). She made it clear what role human cognition plays in language production. “A person's ability to speak is a specific part of cognition : it is a human- specific mental ability that is constitutive for many of our general cognitive abilities. In this sense, cognition is the more general term and includes language ”(Schwarz, 1992: 36). An essential area of ​​prototype semantics is the research of categorization processes, i.e. the mental processing of information .

The beginnings: Basic Color Terms

In the languages ​​of the world there are noticeable differences in which colors have their own names and which do not, and some linguists began to ask why: “ For example, Italian knows two different types of blue ( azzurro, blu ), and French has two types of Braun ( marron, brun ), the Cymric spoken in Wales does not distinguish between 'blue' and 'green' ”(Blank, 2001: 46). From a structuralist point of view, this observation can be interpreted as proof that these designations are arbitrary in the sense of Ferdinand de Saussure . According to Taylor (compare Taylor, 1995: 7), the said color names are equivalent within a system - even if they are used differently. A speaker of any language, for example, perceives several shades of red, but does not perceive one of them to be “redder” than the other.

However, studies of perceptual psychology have shown that all people around the world distinguish almost exactly eleven basic colors, even if there are no names of their own for them in their language:

  • "It now appears that the different languages ​​have a different number of color categories in their vocabulary, but that there is a universal set of exactly eleven basic color categories from which the eleven or fewer names of each language are selected." (It appears now that, although different languages ​​encode in their vocabularies different numbers of basic color categories, a total universal inventory of exactly eleven basic color categories exists from which the eleven or fewer basic color terms of any language are always drawn. Berlin / Kay, 1969: 2)

The just mentioned term basic color terms, to which the research work by Berlin and Kay was limited, also plays into the area of ​​so-called basic level terms (basic terms, see below). Their results can be summarized as follows (compare Blank, 2001: 45):

  1. There are central and marginal representatives of one color .
  2. Since speakers of almost all languages ​​recognize the same color tones as the central ones, even if the limits of the linguistic classification are set differently, these central representatives are to be regarded as universal.
  3. Basic color terms are neither hyponyms of a color word ( e.g. turquoise ), nor morphologically complicated (e.g. light green ), nor technical language ( e.g. cyan ) or restricted to certain collocations ( e.g. blonde , which can only be associated with hair or beer).
  4. These basic color words show a hierarchy among themselves. Languages ​​with only two basic color words differentiate between black and white (or light and dark). For languages ​​with three colors, red always comes next, followed by yellow or green etc.

These observations were attacked by structuralists in the early 1970s, but Eleanor Rosch managed to confirm and at the same time expand the investigations by Berlin and Kay. She carried out various experiments with a group of English-speaking people and a group of the Dani, a people in Papua New Guinea whose language only knows two color names, “namely mola, white u. all warm colors (red, orange, yellow, pink, purple) 'and mili' black u. all cold colors (blue, green) '. ”(Blank, 2001: 46) Above all, it was tested to what extent typical representatives of a color, so-called focal colors , were categorized as such by the various groups. It turned out that the congruence of the two groups, although their languages ​​could hardly be more different, was astonishingly high. In addition, the colors that were not categorized more precisely for the Dani were given “new” names that were to be learned from them. The result showed that central representatives of a color were learned faster than marginal ones. (Compare Taylor, 1995: 11f) It was thus proven that “color terminology turns out to be much less arbitrary than the structuralists maintain. Color […] is instead 'a prime example of the influence of underlying perceptual-cognitive […] categories.' (Heider 1971: 447) "(Taylor, 1995: 15)

Prototype semantics

The standard version

In the so-called "standard version" of the prototype theory the basic concepts of prototypicality are family resemblance , conciseness , cue validity (assignment validity Prototypikalitätsgrad), hedges and basic level terms of importance:


Labov's now famous cup experiment was an example of the proof of a prototype among different representatives of a category (compare Taylor, 1995: 40). He presented his test subjects with various drawings of vessels and asked them to name them. It turned out that the categories of cup and bowl partially merged and that, above all, the relationship between width and depth of the vessel depended on whether there was a handle or what the contents of the vessel were. Thus there were "central representatives who also showed the usually associated (encyclopedic) features such as [with handle], [for coffee] etc., but also those whose only correspondence with these 'prototypes' was the property of being vascular." ( Blank, 2001: 46) This gave rise to "the concept of the prototype as the best specimen or example, the best representative or central element of a category" (Kleiber, 1993: 31). So it is a kind of prototypicality in which a particular speaker forms the center of a category. In addition, there is also a second type of prototypicality. Take the bird category as an example . The representatives belonging to this category are all birds in the biological sense. None of them is “more or less” a bird. “In contrast to Labov's cups, the individual bird species themselves form sub-categories: we have an independent conception of sparrows, owls or ostriches” (Blank, 2001: 46f). For example, in various attempts to match, the central representatives of the bird category (such as sparrows or blackbirds ) were identified as such significantly faster than marginal representatives (such as penguin or ostrich ). From this the following conclusion could be drawn by Eleanor Rosch:

  • The inner structure of many natural categories consists of the prototype of the category (the clearest representatives, the best examples) and the non-prototypical examples, which are arranged in a ranking that extends from the best to the less good examples. (Rosch, 1975: 544)

This results in a consequence that should not be underestimated, which is referred to as the fuzziness or fuzziness of categories. “The boundaries of a category are often not clearly defined […]; in such cases, the question about the assignment to a category [...] can only be answered with a conditional yes or no. ”(Kortmann, 1999: 175) This assertion is briefly illustrated here using an example by Lakoff (compare Kleiber, 1993: 35) be clarified. Thus, the following statements are graded as true or false, rather than absolutely.

  • (a) A sparrow is a bird. (true)
  • (b) A chick is a bird. (less true than a)
  • (c) A penguin is a bird. (less true than b)
  • (d) A bat is a bird. (false or far from being true)
  • (e) A cow is a bird. (absolutely wrong)

This blurring arises from the fact that a relatively obvious feature such as [can fly] also applies to the bat, which brings it back to Vogel. On the other hand, the penguin, which, in contrast to the bat, is a bird from a biological point of view, does without this important characteristic and is thus far from being a “typical bird”. This has also been proven experimentally. In experiments that were similar to the above-mentioned Labov's, the test subjects were asked to divide terms such as sparrow, chick , etc. into categories. It turned out that peripheral representatives were classified significantly more slowly than central ones, which underlines the vagueness of the categories just mentioned.

Family resemblance, conciseness and cue validity

The problem of fuzziness just mentioned can be solved with the help of the so-called concept of family resemblance , which was coined by Ludwig Wittgenstein . There are representatives within a category who have no or only a few features in common, from which the aforementioned fuzziness results. Wittgenstein now set up his concept based on the game category as follows:

  • For example, consider what we call 'games'. I mean board games, card games, ball games, fighting games, etc. What do they all have in common? [...] if you look at them, you will not see something that would be common to all, but you will see similarities, relationships, and indeed quite a number. [...] And the result of this observation is now: We see a complicated network of similarities that overlap and cross one another. [...] I cannot characterize these similarities better than with the word 'family similarities'; because in this way the various similarities that exist between the members of a family overlap and intersect: stature, facial features, eye color, gait, temperament etc. etc. (Wittgenstein, 1997: 66f.)

This theory can be applied to the Vogel example as follows. There are important characteristics to categorizing a speaker as a bird, such as [can fly] or [has feathers]. However, these do not have to be common to all, as can be seen from the comparison between sparrow and penguin, since "the overall structure shows the overlaps and overlaps characteristic of the model of family resemblance." Consequences that lead to the extension of the standard version of the prototype semantics.

The theory just described shows that there is a high degree of family resemblance between central representatives of a category, as they have many central or distinctive features in common. As Blank (Blank, 2001: 47f.) Also states, these provide less information about whether a speaker belongs to this or that category, but rather how close the prototype is. In addition, intensity , frequency , familiarity, good shape and information content contribute to the conciseness of a characteristic. The prototype is not primarily dependent on a specific individual language, but on the influence of the outside world, i.e. the special encyclopedic knowledge . Therefore prototypes can differ from region to region. While in Central Europe the sparrow is the prototypical bird, in South America this is possibly the toucan .

In this context, it is important to briefly address the term cue validity . It shows how often a certain characteristic is assigned to a category, i.e. the degree of family similarity. Characteristics with high cue validity are therefore crucial for the categorization of a speaker. "For the bird category , this would be the feature [capable of flying], for example, and therefore it is more difficult for us to recognize flightless birds such as the penguin as a bird." (Blank, 2001: 47) Such probabilities are determined with the help of experiments in which the test subjects have to assign characteristics to certain categories. A high number of mentions results in a high cue validity .

Hedge expressions

Another phenomenon in the field of prototype semantics is that of hedge expressions or so-called hedges . The boundaries of categories are not always clearly defined and clearly recognizable. Taylor (compare Taylor, 1995: 68-74) introduces the terms expert categories and folk categories on this point . The latter are the categories that are used in daily life and usually have blurred boundaries. Expert categories, on the other hand, are sharply demarcated from one another. Taylor gives the odd number category as an example. For mathematicians (i.e. experts), all odd numbers are the same and the category is therefore not structured as a prototype. To a non-mathematician (i.e. non-experts), however, the number 3 appears to be a more prototypical representative of the odd numbers category than the number 447. The category here is clearly structured as a prototype, since we often have to deal with smaller numbers in everyday life and therefore these central representatives of the odd numbers category arise. In order to put existing or missing expert knowledge into perspective, we use formulations such as, strictly speaking , in the broadest sense , actually , somehow . “A (proto) typical conflict speaker is, for example, the whale: it can swim and lives exclusively in the sea; in a popular categorization it would pass as a whale 'fish'; for experts, on the other hand, it is a marine mammal. ”(Blank, 2001: 48) Due to a different level of expert knowledge, according to Blank, the following statements can now be made:

  • (a) Strictly speaking, the whale is a mammal, even if it lives in the sea.
  • (b) In the broadest sense, the whale is a fish because it lives in the sea.

In this example, hedges are used, on the one hand, to delimit a category more sharply (a) and, on the other hand, to add a certain degree of truth to a false statement (b). However, hedge expressions can also be used to express whether the speaker is a central or peripheral representative of a category (Taylor, 1995: 77). The continuation of the above example:

  • (a) Strictly speaking, the whale is a mammal.
  • (b) The monkey is a typical mammal.

The hedge used in (a) indicates a marginal representative of a category, while the hedge used in (b) indicates a central representative. A final function of hedges to be discussed at this point is the ability to use them to make the boundaries of categories more flexible. Here, too, the “whale” should serve as an example.

  • (a) The whale is a fish.
  • (b) In the broadest sense, the whale is a fish.

While (a) is false, the hedge used here means that (b) is at least no longer obviously false, but rather true in a certain way. The fish category was expanded to include the term hedge in the broadest sense . Taylor summarizes the functions of Hecken presented here as follows: “Hedges require us to distinguish between central and peripheral members of a category […], as well as between different degrees of non-membership in a category […]. They show that category boundaries are flexible […]. "(Taylor, 1995: 80)

Basic level terms (basic concepts, basic categories)

In addition to the horizontal, co-hyponymic level of the prototype semantics , the vertical, hyponymic level , i.e. the division into superordinate and subordinate categories, is another aspect of Rosch's theory. She suggests three levels for the subdivision of the vertical axis (Rosch, 1975):

  • superordinate level, e.g. tier
  • Basic level (e.g. Vogel)
  • subordinate level, e.g. eagle

The words used at the basic level now indicate, similar to those in Chap. 2, different linguistic features (compare Taylor, 1995: 49f): They are usually short and mostly monomorphemic ( e.g. table), whereas on a subordinate level they are often compound words ( e.g. kitchen table ). The two phenomena just mentioned relate to the basic and subordinate level. At the higher level, several phenomena can be identified, which, however, depend on the language. An example of a feature of higher-level German categories : While the higher-level categories of fruit and vegetables are grammatically neutral, a gender-related specification is made on the basic and subordinate levels (e.g. apples, carrots). In addition, there is another characteristic characteristic of basic level terms: “These categories are psychologically fundamental in the sense that they have the highest information density in cognitive processing , as can be seen, for example, in the speed of recognition and categorization ('Look, a (e) ...! '), which expresses the ability to be visualized or the earliness in language acquisition . "(Kortmann, 1999: 176)

From this statement, the conclusion can be drawn that the basic categories also play a fairly important role in communication, which is also true in practice. “Look, an animal!” Is too imprecise in many situations and is also rarely used. “Look, a bird!” Is informative enough. So you can "refer to the world relatively precisely with basic concepts, but still so general that as many speakers as possible are recorded." (Blank, 2001: 50) Kleiber (compare Kleiber, 1993: 60) uses the example of the chair category another feature of basic categories. It is often the case that with basic categories, as opposed to higher-level categories , we have a clear motor program for how we deal with it. We know exactly how to act when we want to use a chair (we have to sit on it). If you now look at the higher-level furniture category , you will see that there is no clear motorized concept for using a piece of furniture, as this term is too general. One can only guess whether one is perhaps sitting on it, stowing something in it, etc. However, these ideas are controlled by the basic categories of chair, cupboard, etc., because one always imagines such a piece of furniture. The following conclusion clearly emerges from the above considerations: Basic level terms "are the first and most natural forms of categorization" (Lakoff, 1987: 49)

The extended version

The extended version of the prototype semantics emerged, among other things, from Wittgenstein's considerations on the subject of family similarities. Using the example of the game category , he shows that representatives of a category often do not have any common properties, but are only connected to one another via a network of overlapping properties (i.e. family similarities) (compare Blank, 2001: 50), so that, for example, the following Chain gives:

  • AB BC CD DE. (compare Kleiber, 1993: 120)

For example, the first and last link no longer have anything in common, but they belong to the same category. "On this basis, the conception of the prototype as the best representative of an extensional category is abandoned [...]." (Blank, 2001: 50) As a consequence of these considerations, the following two theses remained (compare Kleiber, 1993: 113):

  • There are only prototypical effects: the prototype as a representative of the categorical terms and as the structural basis of the category no longer exists.
  • The relation that connects the different representatives of the same category is that of family resemblance for categories of any kind.

So another concept had to be created that took these theses into account. One could not avoid including the figurative meanings of a word in the conception of categories. “The starting point of this conception of a category is suddenly no longer a more or less clearly delimited group of speaker classes or world knowledge aspects, but the corresponding word in a single language!” (Blank, 2001: 50-51) So you have to do that with Vogel, for example figurative meanings of the word, like looking at airplane or strange person . The above statements correspond to the so-called idealized cognitive models (ICM) by George Lakoff . Various prototypical effects are combined in order to cover the entire Vogel concept . This example also clearly shows that you have to limit yourself to individual languages. In English, for example, bird is a slang term for a young woman. In addition to its actual meaning, the Italian uccello is a vulgar term for the penis. The corresponding ICMs would look very different again. With the help of this new model, the problem of polysemy could finally be solved. "The application of the concept of family resemblance to prototype theory paves the way for a theory of multi-referential categorization or polysemy." (Kleiber, 1993: 120) Because the prototype has been omitted and its reduction to prototypical effects, aircraft does not appear to be an extremely peripheral representative von Vogel, but takes on its own position.

  • In this case, the extended version no longer represents a theory of the category structure, but a theory of the semantic structure of polysemic lexemes . It does not show how a category (or a term) can be structured, but how a word can refer to different categories without having to postulate the existence of a common category that encompasses these different categories. (Kleiber, 1993: 130)

This extended version concept also takes into account the previously mentioned hedge expressions . Formulations as in the broadest sense can therefore also indicate the existence of transferred meanings. Thus, aircraft or bat although no birds in the biological, but just in a figurative sense, and therefore must be included in the concept.

Criticism of the prototype semantics

The standard version of the prototype semantics does not explicitly refer to single-language phenomena, but examines mental concepts across languages. Different categorizations in certain languages ​​are ignored, although the cultural backgrounds of the languages ​​hardly differ. As Blank (compare Blank, 2001: 52f.) Describes, Spanish and Portuguese distinguish between a large and a small bird by the two lexemes ave (sp./pt.) And pájaro (sp.) And pássaro (pt.) Respectively although other Romance languages do not make this distinction. It also seems rather unlikely that nature on the Iberian Peninsula differs so drastically from that in Central Europe that one could attribute this phenomenon to it. This clearly shows that ignoring the individual language level, as is the case in the standard version, is not really justifiable. Another difficulty emerges when examining the basic level terms .

  • The solution developed as part of the standard version can no longer (or only with difficulty) be applied to the prototypes of the superordinate categories. This is because these can no longer be viewed as specimens with the best properties (the typical properties) of the category, as is the case with the prototypes of the basic categories, because the salient properties of the category are not concentrated in them. Their status as best specimen is based on the fact that they appear frequently in the speaker's (direct or indirect) experience. (Kleiber, 1993: 98)

If one looks at the fruit category , for example , it is noticeable that blueberries or mirabelle plums are by no means classified as prototypical representatives, although they combine many characteristics with a high cue validity , which should actually mean that they are not marginal representatives of this category are designated. Rather, this classification is based on the degree of familiarity with these fruits, which is much higher for apples or pears. (Compare Kleiber, 1993: 98f) This makes it clear that one cannot always rely on cue validity and conciseness with regard to prototypicality, but one or the other can be disregarded, which does not necessarily lead to clarity and simplicity on this one Area leads. Although some weaknesses of the standard version (for example the failure to observe the individual language level) have been eliminated in the extended version of the prototype semantics, it is not spared from criticism. According to Blank, it is a misinterpretation of Wittgenstein :

  • Indeed, it is naive to believe that all games must have something in common because they are all called a game. Lakoff concludes from this that all games must belong to the same cognitive category because they are all called game! Our example VOGEL shows that this is too short-sighted: the penguin is certainly an atypical bird, but it is one; BAT, AIRPLANE and COMIC MAN are independent concepts [...]. They are simply not birds […]. (Blank, 2001: 53)

Subsequently, Blank continues to criticize, pointing out that Wittgenstein's concept can only describe the relationship between the members of a category, but by no means cross-category relationships. He also accuses the extended version of blurring the clearly existing advantages of the standard version. (compare Blank, 2001: 54)

From a conceptual theory or communication semantic point of view, the following is critical of prototype semantics: “The introduction of a prototypical structure in a classification ... falsifies thinking in terms, because it wants to mark a certain application as (proto-) typical for the term. This implies a gradual weighting that is not compatible with the discontinuous terminology of a language. "The prototype has no conceptual justification, but only a" existence as a contingency-related cliché "that" reflects rather than relevant the subjectivity of the speaker Properties of the object ”. Similarly, it is said that the prototype model, apart from its "empirical touch", has as a "fundamental weakness" that "the question [remains] unanswered, why these and not other properties and relationships are perceived and why these particular characteristics are perceived as typical or be recorded together. " The "generalization and structuring potential" of every term in the term development would be ignored. The prototypical features are only "surprising" because the competing "theoretical reconstruction of the concepts inevitably neglects the individual peculiarities, the concrete fixations and situational limitations of the idiosyncratic concepts". The persuasiveness of the prototype theory is based solely on the fact that, viewed from an individual psychological perspective - contrary to the assumption of the component model - for specific persons, terms "do not usually consist of systematic and highly structured feature constructions like explicit definitions."

In a different perspective it is pointed out that the applicability to abstract predicates was not explicitly asserted either by Putnam or Rosch. Knowledge of the meaning is always assumed when describing a stereotype. Ultimately, the assumption of stereotypes or prototypes compared to the feature semantics "only amounts to a slightly different form of reference to semantic features" and does not help in problematic cases.


One of the main features of the standard version of the prototype semantics is, above all, the presence of a central representative of a category (i.e. a prototype), which is considered the best of the same. Based on this, the category is structured and based on the degree of similarity to it, it is decided whether a speaker belongs to the category. Another remarkable finding is the vertical division using basic level terms. It turns out that on a certain basic level between the superordinate and subordinate level there is the greatest possible information density with sufficient generality, which leads to a further prototypical effect, namely to an increased use of these lexemes in everyday life. The two essential points of the extended version are, on the one hand, the departure from the classic concept of the prototype and its reduction to so-called prototypical effects; on the other hand, only Wittgenstein's concept of family resemblance applies here as a structuring principle. From this, Lakoff develops his so-called ICMs, which are supposed to encompass all (also transferred) meanings of a word and are consequently more complex. This also automatically results in an expansion to single-language phenomena, since metaphorical meanings are only seldom and then to a lesser extent cross- lingual . "Therefore one can say that the extended version breaks with the basic principles of the standard version." (Kleiber, 1993: 140)

See also

Individual evidence

  1. M. Eysenck, M. Keane: Cognitive Psychology . Psychology Press, Hove (UK), 2000
  2. a b Mudersbach, Klaus: Terms in the view of the language user. In: Wille, Rudolf (ed.): Conceptual knowledge processing: basic questions and tasks. BI-Wiss.-Verl .: Mannheim [u. a.], 1994, p. 117 (143)
  3. Seiler, Thomas Bernhard: Concepts of Concept: Analyzes and conceptions of concepts in psychological research. In: Ganter, Bernhard; Rudolf Wille; Karl Erich Wolff (Ed.): Contributions to concept analysis. BI-Wissenschaftsverlag: Mannheim, Vienna, Zurich, 1987, ISBN 3-411-03157-3 , p. 95 (110 f.).
  4. All quotations from Seiler, Thomas Bernhard: Concepts of Concept: Analyzes and conceptions of concepts in psychological research. In: Ganter, Bernhard; Rudolf Wille; Karl Erich Wolff (Ed.): Contributions to concept analysis. BI-Wissenschaftsverlag: Mannheim, Vienna, Zurich, 1987, ISBN 3-411-03157-3 , p. 95 (111).
  5. Busse, Dietrich: Semantics. W. Fink, Paderborn 2009 (UTB 3280), p. 52
  6. Busse, Dietrich: Semantics. W. Fink, Paderborn 2009 (UTB 3280), p. 56
  7. a b Busse, Dietrich: Semantics. W. Fink, Paderborn 2009 (UTB 3280), p. 57



  • Andreas Blank (2001): Introduction to lexical semantics, Tübingen.
  • Dietrich Busse: Semantics. W. Fink, Paderborn 2009 (UTB 3280), pp. 49-59.
  • Volker Harm: Introduction to Lexicology. WBG, Darmstadt 2015 (Introduction to German Studies), ISBN 978-3-534-26384-4 , pp. 41–46.
  • Georges Kleiber (1993): Prototype semantics: An introduction, Tübingen.
  • Bernd Kortmann (1999): Linguistics. Essentials, Berlin.
  • M. Mangasser-Wahl (2000): Prototype theory in linguistics: application examples, method reflection, perspectives. Tübingen: Stauffenburg. ISBN 3-86057-706-9
  • M. Mangasser-Wahl (2000): From prototype theory to empirical semantics. Shown using the example of women categorizations. Frankfurt: Peter Lang.
  • Monika Schwarz (1992): Introduction to cognitive linguistics, Tübingen.
  • Monika Schwarz, Jeanette Chur (2007): Semantics (5th ed.). Tübingen: G. Narr, pp. 46–53.
  • Ludwig Wittgenstein (1997): " Philosophical Investigations ", in: Writings. Vol. 1, Frankfurt, 225-580.


  • Brent Berlin / Paul Kay (1969): Basic Color Terms: their Universality and Evolution, Berkeley.
  • Joachim Grzega (2003): On Using (and Misusing) Prototypes for Explanations of Lexical Change, Word 54: 335-357.
  • Eleanor Heider (= Rosch) (1971): 'Focal' color areas and the development of color names, Developmental Psychology 4: 447-455.
  • George Lakoff (1987): Women, fire and dangerous things: What categories reveal about the mind, London.
  • Eleanor Rosch (1975): Cognitive reference points, Cognitive Psychology 7, 532-547.
  • John R. Taylor (1995): Linguistic categorization: Prototypes in linguistic theory, Oxford.