|Polysemy, homonymy and reference
Groupe de recherche sur les fonctionnements discursifs EA 1339 LDL
Université Marc Bloch
22 rue Descartes
Tél : 33 (0)3 88 41 59 62
Email : firstname.lastname@example.org
Abstract: Homonymy and polysemy are two well-known semantic problems. Bank in river bank and Bank of England are homonymous: they share no meaning whatsoever; they function as two totally unrelated words. River bed and hospital bed seem to be somehow semantically linked: it is a case of polysemy. This paper first examines how the problem is usually treated. Dictionaries list polysemes under one entry and homonyms under several, although there are marked differences between dictionaries. Semantic theories tend to explain homonymy an polysemy in terms of metaphor and metonymy, or in terms of a subsuming cognitive element with specific meanings triggered by the context or by rules. We offer a typology of polysemy and an explanation in terms of reference: the engine of meaning is our desire to grasp and convey our shared experience with the help of polysemous lexical items.
Résumé: L’homonymie et la polysmémie sont des problèmes sémantiques bien identifiés. Bank dans Bank of England et river bank sont homonymes : ils ne partagent pas le moindre élément sémantique ; river bed et hospital bed semblent sémantiquement liés: il s’agit d’un cas de polysémie. Dans cet article, nous commençons par examiner la manière dont le problème est traité dans la littérature. Les dictionnaires regroupent les polysèmes sous une seule entrée et les homonymes sous plusieurs, bien qu’il y ait des différences notables entre dictionnaires. Les théories sémantiques expliquent la polysémie en termes de métaphore et de métonymie, ou bien en termes d’entités cognitives subsumantes avec des sens particuliers déclenchés par des règles ou le contexte. Nous proposons une typologie de la polysémie et une explication en termes de référence : le moteur du sens se trouve dans notre désir de saisir et communiquer notre expérience partagée à l’aide d’unités lexicales polysémiques.
Polysemy, homonymy and reference
Homonymy and polysemy are two well-known semantic problems. Bank in river bank and Bank of England are homonymous: they share no meaning whatsoever; they function as two totally unrelated words. River bed and hospital bed seem to be somehow semantically linked: it is a case of polysemy. The problems posed by homonymy and polysemy are probably at the very heart of semantics: what exactly is it that words may or may not share ? how come we can mean different things with the same word. This paper begins with a brief examination how the problem is usually treated, followed by a tentative explanation in terms of reference.
1. Lexicographical treatment of homonymy and polysemy
Lexicographers are keenly aware of polysemy and homonymy because semantic closeness and referential dispersion are especially obvious when words are considered out of context. In everyday speech our usage of words is almost never ambiguous. On saying, or hearing, "The Bank of England has lowered its rates", the other meanings of bank are probably not even considered. Dictionaries, on the other hand, consist of out of context words and phrases. Homonyms are usually listed under different entries and polysemes under one entry only, but not always. Some dictionaries lump homonyms together and others separate polysemes, which may be an indication that the boundary between polysemy and homonymy is not clear cut.
2. Semantic treatment of polysemy
Dictionaries are not meant to explain language from a theoretical point of view: all they do is offer a snapshot of usage at one particular moment. Theories of polysemy, on the other hand, usually rest on one of two hypotheses:
i) there is a literal meaning from which the other meanings are derived (a linear explanation)
ii) there is a core meaning with specific senses triggered either by the context or by rules (a subsuming explanation).
2.1 The linear explanation: literal and derived meanings
According to this point of view, words do possess a literal meaning, all other meanings are merely derived and figurative. For example, the literal meaning of mouse is the rodent; a derived meaning is the computer mouse. A bed is "a piece of furniture that you lie on"1 (literal); it is something flat at the bottom of something else (a river bed) or a place where something can be found in abundance (a shellfish bed, a bed of roses) in a figurative way.
But literal meanings are not always so easy to spot. For example, a position can be a physical position (a crouched position), a psychological position, a stand, a point of view (the Soviet position on German unity), or a social position, a job (his position as Speaker). Which one is the literal meaning ? We may be inclined to think it is the physical sense, but we are clearly not as sure as with mouse or bed.
Another problem is the link between literal and derived meanings ? What does it consist of? Let us consider the word knocker, which can mean door knocker, someone who knocks, or (not very nicely) women's breasts. I asked my native English-speaking informants2 if they felt these meanings were somehow linked and if they could formulate these links. All informants felt that they were indeed linked. The person meaning was definitely considered as the literal meaning. The door-knocker meaning was explained in terms of metonymy (the object used to knock is named after the person who is doing the knocking). As for the breast meaning, a wealth of links were offered:
metaphorical links to the door-knocker meaning
- breasts resemble some door knockers
- breasts protrude like door knockers
a metonymical link to the door-knocker meaning
- breasts are something one grabs (or feels like grabbing) like a door-knocker
metaphorical links to senses of to knock
- the sexual impact of breasts may knock you over
- when women run, breasts may move up and down, which resembles the act of knocking on doors
- breasts knock together
Thus the linear theory fails on two counts: i) there is no definite way of deciding which is the literal meaning; and ii), the link between literal and derived meanings cannot always be specified with certainty, even when it is established.
2.2 The subsuming theory : core and specific meanings
The subsuming theory assumes that words are endowed with a core meaning and that specific meanings are triggered either by the context or by generative rules. This means that understanding is an interpretation and that meaning is the result of some sort of unconscious calculus. Let us examine these assumptions.
According to this point of view, a particular interpretation of a word is selected by the context. For example, in Bank of England, the financial institution meaning is triggered by the words of England. All other meanings are blocked. But how does the process work ? Let us consider the following Italian sentence, taken from (Velardi & Pazienza 1988):
L'associazone degli industriali ha approvato un nuovo piano di investimenti nel Mezzogiorno
The association of manufacturers has approved a new investment plan in the South of Italy.
The authors point out that if all word meanings in a sentence depend on the other words, the interpretation process can hardly get started. For example, industriali could be an adjective or a verb, piano could be an adjective or an adverb (part-of speech ambiguity). Mezzogiorno could mean noon, south, or South of Italy; piano could mean plan, project, the floor of a building, a musical instrument; investimenti could mean financial investment or accident (lexical ambiguity). In the phrase un nuovo piano di investimenti for example, the selection of the plan meaning of piano can be achieved if the investment meaning of investimenti has already been established; but the selection of the investment meaning rests on the plan meaning of piano. The selection process, if there is such a thing, is clearly more complex than mutual influence. Such a very general and unspecified use of the context is certainly too powerful an explanation. It provides no insight into how the selection of meanings is actually done. What is needed here is a theory of how we use the context to select meaning.
One idea is that interpretation rules have to be spelled out. For example, according to Pustejovsky (1991, 1993, 1995), words are naturally ambiguous: they possess something Pustejovsky calls logical polysemy, where logical is endowed with the meaning analytical philosophers have given to logics, i.e. the entity which structures the universe and the mind. If polysemy is "natural" then there must also be a "natural" way of selecting senses. Pustejovsky hypothesises a number of very general predicative mechanisms, such as type coercion, which govern the phrase and sentence levels.
For example, let us consider one of Pustejovsky's favourite examples :
1) Mary began a book
2) Mary began reading a book
3) Mary began writing a book
4) Mary bought a book
Sentence 1 is interpreted as 2 or 3 according to the context. If we know that Mary is a writer, 3 may be correct, else it is 2. But how is this possible ?
Pustejovsky describes word meanings in terms of qualia roles For example, this is the qualia structure of book (Pustejovsky 1995) :
Constituve : pages (z), ...,
Formal : physobj (x), ...,
Telic : read (P, y, x), ...,
Agentive : write (T, w, x), ...,
This representation captures the fact that a book is a physical object (formal role), that it is made of pages (constitutive role), that it is something one reads (telic role) and that it has been written by somebody (agentive role). Book is thus viewed as an object with procedural aspects (reading and writing). The interpretation of sentence 4 is fairly straightforward: to buy needs an object as an argument and book is formally an object. But begin normally expects a process. For sentence 1 to be well-formed, the type coercion mechanism must be able to select a procedural role in the qualia structure (the telic and agentive roles happen to offer such meanings). It then coerces the type of book from object to process.
This theory of Pustejovsky's has been extensively criticised, for example in (Kleiber 1999), but this is not the place to go deeply into the matter. Relevant to our problem is the fact that the type coercion explanation does not say anything about how we actually use the context to select meanings; it only explains how a sentence can be well-formed after the context has been used. In other words, it does not say anything about why some semantic entity selects book as an argument of begin, which is considered illegal3, but only how some mechanism steps in to fix the problem. Also, it is difficult to imagine how a device of that type could explain why we should interpret 1 as 2 (reading) or 3 (writing). Finally, the type coercion explanation addresses a very narrow kind of polysemy, called referential polysemy, but more of that in a later section.
The subsuming explanation assumes that a word somehow contains all its possible meanings, and that some interpretation rule is able to select the "correct" meaning according to the context. But is there always such an all-encompassing subsuming entity? For example, is there a subsuming entity containing the computer and the rodent meanings of mouse.
In the end, the context and subsuming theories both fail for the same reason: they are unable i) to define the original (literal or subsuming) semantic entity, and ii) to explain the links between the original and (derived or subsumed) meaning of a word. What is lacking, in a nutshell, is the engine which powers the semantic process. No explanation in terms of rules can be totally satisfying because explaining some linguistic phenomenon A by a set of rules B begs the question of how B is set in motion, and so on in an endless regress. For example, how does the type coercion device select the formal, telic or agentive role of book? Do we need yet another set of rules?
Language is probably not mainly a matter of interpretation, of calculus. It is probably not a rule-based device, and if it is, then only marginally. In The Blue and Brown Book, Wittgenstein wonders why is it that we tend to explain meaning in terms of rules. Is it not because we are trying to solve problems which are only artefacts4 created by a false conception of language?
The engine of language is our desire to talk, or think, about objects. The following sections offer an alternative view based on reference. I shall first look at the evidence, and then try and formulate a general theory of homonymy and polysemy.
3. The evidence
Let us examine a few examples. Some have been taken from the British National Corpus Sampler CD-ROM5, some from other sources, some have been made up for the purpose.
1. She could hear the piano (sound)
2. She polished the piano (piece of furniture)
3. Elizabeth could hear voices through the open door (opening)
4. They painted the door (panel)
5. ... serving as an open door to the East (channel)
6. The bank was flooded yesterday (building)
7. The bank was very nice and understanding (personnel)
8. The bank was founded in 1990 (institution)
9. I am the bank (when playing Monopoly)
10. A blood bank, a memory bank (a place where something is stored)
11. A river bank (the rising ground bordering a river)
12. We were protected by a bank of about two feet high (a small flat mound)
13. Also a ridge, an undersea elevation, etc.
14. I saw armed men in a crouched position by the swimming pool (physical)
15. He could become Speaker, a position of some honour but no great responsibility (job)
16. The Soviet position on German unity (point of view, stand)
17. The compartment resisted the fire for an hour (German : aushalten)
18. The rebels resisted the Russians (German : widerstehen, Widerstand leisten)
19. She worked hard (French : travailler)
20. The lift doesn’t work (French : fonctionner)
21. I worked on him to come to the wedding (influence s.o.)
22. I just bought a book on boring postcards (object)
23. Malcolm wrote a book on boring postcards (text)
24. Jenny is a bookkeeper at Barclay’s bank (job)
25. We booked in at the hotel (US check in)
26. We booked our tickets this morning (buy, reserve a seat)
27. Les oiseaux volent (birds fly)
28. On nous a volé tout notre argent (all our money was stolen)
29. L’aigle vola un lapin (the eagle caught a rabbit)
30. Une bande de papier (a strip of paper)
31. Une bande de criminels (a gang of criminals)
32. Le navire donne de la bande (the ship is listing)
These examples show that polysemy is not a single homogeneous phenomenon.
4.1 Referential polysemy
Let us now consider a first subset of examples:
- piano 1 (sound), 2 (piece of furniture)
- door 3 (opening), 4 (panel)
- bank 6 (building), 7 (personnel), 8 (institution)
- book 22 (object), 23 (text)
In piano 1 and 2, both usages refer to the same object, but viewed from two different points of view. The same is true for the other examples as well. Bank for example can be considered as a cue which conjures up a complex object, of which one aspect in particular is retained (either the building, personnel or institution meaning) without suppressing the others. Since all meanings are linked by the object they refer to, this sort of polysemy may be called referential polysemy.
Non-linguists hardly notice any polysemy at all in those examples, probably because it seems quite obvious that a piano should be a music producing piece of furniture and a door a panel on hinges which may serve as a temporary opening through a wall. Professional linguists have nevertheless given it various explanations, for example Pustejovsky and his qualia roles (see section 2). Other theories of referential polysemy include Langacker's Cognitive Grammar where the difference between 1 and 2 is explained in terms of active zones (1984): the sound is active in 1, the piece of furniture in 2.
D.A. Cruse's (1996) explains the difference between 6, 7 and 8 in terms of facets. Bank refers to an object with at least three facets : the premises, the personnel, the institution. The advantage is that an intermediary semantic level has been introduced between the object and the contextual meanings, which allows for a distinction between real polysemy (the facets) and contextual variations (the usages of each facet). Problems with this theory include the number of facets (can it be precisely stated?) and their discreteness (to what extent do they overlap?). For example, in I hate this bank, which facet is concerned? Is it the personnel or the institution, or even the building?
G. Kleiber (1999) has put forward a referential theory called integrated metonymy whereby under certain conditions a part of an object can stand for the whole. For example, let us consider The Americans landed on the moon, and My trousers are dirty. Both sentences are true although not all Americans landed on the moon, and my trousers may only be stained. Some Americans stand for the whole nation; one stain on only one small part of the trousers is enough to think of the whole as being dirty.
These theories are all theories of metonymy. They aim to explain how we are able with a single word to focus on the many parts of an object, or to the object as a whole, or to some entity linked to the object. The answers basically rest on a very ordinary feature of thought: selecting one aspect of an object in particular does not exclude the other aspects and the whole. They remain in the background. If one polishes the piano, it does not mean that it ceases to be a music instrument, and vice-versa. When book is "coerced" into an event, it remains an object.
4.2 Lexical polysemy
This is what the layman has in mind when he thinks of polysemy. How come operation can refer to two very different things such as a surgical and a military operation? How come we have the feeling that their meanings are somehow linked ?
The evidence in section 3 can be separated into two subsets, according to whether or not we are able to trace a metaphorical or metonymical link between usages.
Metonymical and metaphorical links
In the subset below, some of the senses are clearly derived from others, and we are able to identify the original meaning. For example senses 3 and 4 of door can be lumped together because they both refer to the same object (the object door). We then consider them in relationship to usage 5, where door is clearly not a physical object but some abstract concept meaning opening or channel. The door usage in 5 is clearly a metaphor. We do not doubt for a minute that the physical meaning comes first and the abstract meaning is second. Here is a list of such lexical polysemy in the corpus :
door 3, 4 : physical object
5 : channel, opening : metaphor
bank 6, 7, 8: physical object
9 : Monopoly : metaphor
10 : store something : metaphor
work 19 : physical object
20 : humans work => machines work : metaphor
21 : change things =>change s.o. : metaphor
book 22, 23: physical object
24 : accounts are written down in books6: metonymy
25 : one’s name is written in a book : metonymy
No obvious metonymical or metaphorical links
position : 14 (physical), 15 (social), 16 (psychological)
resist : 17 (German aushalten), 18 (German widerstehen)
The position example has already been mentioned. The case is even stronger with resist, where it is certainly impossible to say which meaning is literal and which is derived.
Group 1 : One is able to trace a historical link between usages. Some word (for example bank) first refers to some object (for example the financial institution), and then some other object is referred to by the same name because of some resemblance with the first object (for example blood bank). It is impossible, or difficult, or artificial to conjure up a subsuming entity here (for example something subsuming bank and blood bank).
Group 2 : We are unable to trace the history of the word, or we are not certain. We could make up a theory, for example that the physical meaning of position came first, then the psychological meaning (a position in the mind), and then the sociological meaning (a position in society), but the link is clearly not obvious. We do not doubt that blood bank has a metaphorical link with bank; we are not so sure with position, and at a loss with resist. We feel there is a link, but not a horizontal one, rather a vertical one, a subsuming link, of which the usages of position and resist are only exemplars.
There is a strong case for a subsuming theory here. The aim of this paper is to offer a referential explanation of the phenomenon, whereby the subsuming entity is not cause, but consequence of usage.
4.3 A referential explanation of the subsuming hypothesis
Suppose we make use of a word, position for example, in reference to different types of objects, for example the physical, psychological and sociological positions. These objects are somehow lumped together by the fact that they are referred to by the same word (position). But beyond a shared signifier, we infer that these objects also share some intrinsic property, which the signifier is able to select.
Suppose p stands for position, c for crouched, s for Soviet, j for job, and x for any property a position cannot normally have, for example blue. Then,
if p(c) and p(s) and p(j) but not p(x) (with x = blue for example)
then c, s, j (but not x) share some element (e)
and p has the capacity to select element (e)
Element (e) is thus endowed with some separate existence. But it has to be noted that it is very difficult to put words on element (e): what is it exactly that the positions share?
This argument that the subsuming element (of position for example) is consequence of usage, and not cause, is only valid synchronically of course: the subsuming element (e) is construed by speakers of existing words. Diachronically, the subsuming element can be the cause of a new usage, if its first user believes that some new entity is endowed with that element. If other speakers agree, it becomes a neologism. Yet the fact remains that the subsuming element can only be produced by usage. The alternative view is untenable. It would mean that referring is a matter of mapping objects, as they appear to our consciousness, on a pre-determined semantic content of the lexicon. In other words, it would mean that our referring activity is caused by intrinsic features of some sort of genetic lexicon waiting in our brains to name objects. When someone first used free in connection with electron, they probably did so because it seemed an unbound electron shared something with other unbound objects, not because the meaning of free was there expecting scientist to discover electrons. When someone named the pointing device of a computer after the rodent, they did so because the thing looked like a mouse, not because the meaning of mouse made it possible.
Our usage-based theory rests on two premises, which will be presently put to the test.
1. Words have us surmise that the objects they refer to are somehow related
2. The subsuming entity is deduced from usage, it is not cause of usage
If "words have us surmise that the objects they refer to are somehow related", then objects referred to by different words should not be obviously related, even if they are objectively close. Resist is used with two different meanings in examples 17 and 18:
- resist1: some object resists the effect of some potentially dangerous object
- resist2: someone resists someone
In German there are two words for these meanings of resist: aushalten (17) and widerstehen (18). The dictionary offers two separate entries and mentions no link between them. My native speaking informants of German confirm: aushalten and widerstehen share no semantic element whatsoever. Yet for English speakers resist1 and resist2 do share something. Now fight off, endure and ward off can be considered as semantically close to resist, at least closer than eat and sleep. The difference between resist1 and resist2 is probably not greater than between them and respectively endure and fight off (or ward off), and yet resist1 and resist2 seem to be more obviously related than resist2 and ward off for example. Thus, if objects are not referred to by the same word, we do not surmise an obvious relation between them.
Another example is the German word Schuld, which means debt, fault, responsibility and guilt. Do German speakers believe there is an intrinsic link between debt and guilt, at least more so than speakers of French and English? My informants feel there is, but it would have to be looked into more closely from a cultural point of view.
If "a subsuming entity is deduced from usage", how come we do not deduce one from homonyms? This point is examined in the following section.
Bank 6, 7 8 have been lumped together because they are referentially polysemic (see section 4.1); we now lump them with their lexical polysemes 9 (Monopoly) and 10 (blood bank), and consider them in relationship with homonyms 11 (a river bank), 12 (a small flat mound) and 13 (an undersea elevation). Here's is a list of the homonyms in the corpus :
- bank: [(6, 7, 8), 9, 10], 11, 12 , 13
- book: [(22, 23), 24, 25], 26
- voler: 27, 28
- bande: 30, 31, 32
If premise 2 is correct, why do we not perceive a link between homonyms, while we do so with polysemes? Here too an explanation in terms of reference can be put forward, this time in terms of absence of a common reference.
a) There was a referential link once but it is forgotten
Bank-mound, bank-financial institution and river bank actually have a common origin, the Gothic word benc, which originally referred to a small mound. Then it was also used metaphorically for river bank, and also as a metaphor for a low table, especially those where bankers used to exchange money, which gave Italian banca, French banque, and English bank. Bankruptcy originates in banqueroute, i.e. "bank on the road".
Voler originally only meant flying. In the Middle Ages, eagles and falcons were used for hunting. In L’aigle vole le lapin, voler means to catch. Voler then meant stealing because the act of stealing resembles the way a falcon catches its prey. It is clearly a metaphor. Nowadays, falconry is not practised anymore, except by enthusiasts, and the original link is forgotten. Voler-flying and voler-stealing are considered by most speakers as homonyms.
b) The meanings were never related
The three usages of bande in 30, 31, 32 have three different origins. Bande (strip) comes from Gothic binda which means link (see English bind, German binden). Bande (gang) comes from Gothic banda which means flag, standard (see Italian bandera, English banner, French bannière), and then in Italian, a group of soldiers (banda). Bande (list) comes from Provençal banda, which means side.
So, either the relationship is forgotten because the object which gave birth to the metaphor or the metonymy has disappeared, or there never was one. The difference between polysemy and homonymy can be explained as follows :
1. When a word denotes two or more objects, then we tend to surmise some link between them.
2. When we check the link, we examine the denoted objects. Sometimes we find there is indeed some degree of resemblance (and the words are polysemous), sometimes we do not (and they are homonyms).
According to this view, words are not containers of linguistic or conceptual sub-entities. Single words or phrases happen to be able to denote a variety of objects, sometimes because of some factual or formal resemblance, sometimes by pure chance. If the resemblance between the objects seems motivated, we tend to consider the referring words as semantically related by a set of sub-entities. If not, we think of them as homonyms.
So the real question is, why do we think there is a resemblance between the Soviet position and a crouched position or between resist1 and resist2, and not between river bank and Bank of England ? This would need quite an exhaustive semiotic and cultural analysis. We live in an ever-changing world of physico-cultural objects loosely connected to an unlimited set of signs : some may refer to the many aspects of an object (referential polysemy), some to a number of resembling objects (lexical polysemy), others to unconnected objects (homonymy). Mankind has kept trying, at least since Aristotle, to organise and comprehend the real world, essentially by matching it with language. Language is a repository of knowledge gained by naming and relating objects. This is why we tend to believe that language and the real world are isomorphic, that language maps the world. It is a fallacy of course: isomorphism is a goal, not a given; it is the stuff of science and philosophy. When words refer to separate objects, we quite naturally believe that the objects and the words must share some features. Yet it fails with homonyms, because of some real world discrepancies or because we have forgotten why our forebears named these objects so. And we wonder: why are these two very different things referred to by the same word? We name the mystery homonymy, and add it to the repository. Homonymy is disappointed polysemy; we resort to it reluctantly. Most French speakers agree that we could, albeit with some difficulty, construe a semantic resemblance between bande de voleurs and bande de papier. Although they are historically not related at all, they seem to share, very vaguely, a notion of something long, something like a file. If one tries hard enough, one can even come up with a subsuming cognitive element for voler, as was done by S. Ikeda (1995), quoted in (Kleiber 1999). Our preference for polysemy seems very powerful. When no obvious relation between objects is available, we are able to construe one very easily (and often creatively).
Polysemy and homonymy can be entirely explained in terms of reference.
Objects (for example a piano) can be viewed from a number of points of view (for example as a music instrument or a piece of furniture). The link between these usages is clearly the object as a whole.
Polysemy is lexical, according to our definition, when words refer to objects which we think of as being somehow related. There are two types of such polysemy :
- linear polysemy, when we are able to trace a linear link, either metonymical or metaphorical, between the original object and a new object named after it (for example mouse-rodent and computer-mouse)
- subsuming polysemy, when usage has created an accepted common subsuming element (as in position and resist).
Words are thought of as homonyms, when the object which once linked two usages has culturally ceased to exist (for example when falconry disappeared, the metaphorical link between the two meanings of voler disappeared too), or when the link itself was forgotten (as between a bank and a low table). In other cases, objects were never actually related in any way, they just happen to share a signifier (as bande).
In short, the engine of meaning is reference, i.e. our desire to grasp and communicate our experience through language. The real world is not a given. We become aware of it through living in it, being part of it, observing it, naming it, thinking it. The denotation of new objects is not usually a planned business, it just happens. Polysemy and homonymy are the names of ingrained discrepancies in our referring activity: polysemy is referential when one object is linked to several usages of a word; polysemy is lexical when several resembling objects are linked to several usages of a word; the phenomenon is called homonymy when several non resembling objects are linked to several usages of a word.
D.A CRUSE (1995) : "Polysemy and related phenomena from a cognitive linguistic viewpoint". In Computational Lexical Semantics, Patrick Saint-Dizier & Evelyne Viegas eds. Studies in NLP. Cambridge University Press.
D.A. CRUSE (1996) : "La signification des noms propres de pays en anglais" in Rémi-Giraud S. et Rétat P. (eds), Les mots de la nation, Lyon, Presses Universitaires de LYON, 93-102.
FRATH Pierre (2001) : "Quelques considérations sur le verbe commencer..." Journal of French Language Studies, 2001 (forthcoming)
IKEDA S. (1995) : Essai d'unification des valeurs du verbe VOLER précédé d'une étude et d'une analyse des concepts, thèse de doctorat, Université de Paris IV.
KLEIBER Georges (1999) : Problèmes de sémantique. La polysémie en questions. Presses universitaires du Septentrion.
LANGACKER R.W. (1984) : "Active zones", Proceedings of the Annual Meeting of the Berkeley Linguistic Society, 10, 172-188.
PUSTEJOVSKY James (1991) : The Generative Lexicon . Computational Linguistics 17(4).
PUSTEJOVSKY James (1993) : "Type Coercion and Lexical Selection". In Semantics and the Lexicon. J. Pustejovsky ed., Kluwer Academic Publishers. Dordrecht.
PUSTEJOVSKY James (1995) : "Linguistic Constraints on Type Coercion". In Computational Lexical Semantics, Patrick Saint-Dizier & Evelyne Viegas eds. Studies in NLP. Cambridge University Press.
VELARDI Paola, PATIENZA Maria-Teresa & DE GIOVANETTI Mario (1988) : « Conceptual Graphs for the Analysis and Generation of Sentences ». In IBM Journal of Research and Development 32(2) 251-267.
WITTGENSTEIN Ludwig (1951, 1965) : Le carnet bleu et le carnet brun. Traduit par Guy Durand. Gallimard