Nature Publishing Group (2003)


Kent Bach

San Francisco State University, San Francisco, CA, USA


Theories of Meaning

Philosophical Issues about Meaning

Meaning and Communication

Relevance to Cognitive Science


Sentences have meanings, and speakers mean things in using them. The meaning is of a sentence is determined by the meanings of its constituents, together with its syntactic structure, but what a speaker means in using it is often not determined by what it means, since he may mean something more or something else.



Language is used to express thoughts and to represent aspects of the world. What thought a sentence expresses depends on what the sentence means, and how it represents the world also depends on what it means. Moreover, it is ultimately arbitrary, a matter of convention, that the words of a language mean what they do. So it might seem that what they mean is a matter of how they are used. However, they need not be used in accordance with their literal meanings. One can speak nonliterally, and convey something other than what the sentence means ('The look on his face spoke volumes'), or speak indirectly, and convey something more than what the sentence means ('I wonder if you know the time'). Linguistic communication requires knowledge of linguistic meaning, on the part of both the speaker and his audience, but it requires extralinguistic knowledge as well.


Theories of Meaning

Words have meanings, and sentences have meanings. Assuming a principle of semantic compositionality, according to which the meaning of a sentence is determined by the meanings of its constituents and its syntactic structure, the aim of a semantic theory is to give an account of how meanings of complex expressions are determined by the meanings of their simplest meaningful constituents. A theory of meaning should explain what meanings are.


Linguistic Meaning

It is plausible to suppose, at least as a first approximation, that what a declarative sentence means is what is asserted by a speaker using it literally, and that what the speaker thus asserts is the belief he is thereby expressing. So, for example, if you utter the sentence, 'Giraffes do not use periscopes', you are asserting, and expressing the belief, that giraffes do not use periscopes. If you are sincere, you actually believe what you are asserting, and that is the very thing that comprises the meaning of the sentence you uttered. Call this thing a proposition, in this case the proposition that giraffes do not use periscopes.

That leaves open what sort of thing propositions are. Presumably they are abstract entities, both mind-and language-independent, and whereas some philosophers think of them as structured, composed of the semantic values of sentence constituents, others think of them as sets of circumstances or possible worlds with respect to which a given sentence is true (see FORMAL SEMANTICS AND CATEGORIAL GRAMMAR; POSSIBLE WORLDS SEMANTICS). Some are reluctant to speak of propositions at all and prefer to speak of the truth conditions rather than the propositional contents of sentences. To remain neutral on this issue, let us use 'proposition' to mean the truth-conditional content of a (declarative) sentence, whether this is something which has a truth condition or simply is the truth condition. A proposition is something that different people can believe or assert and that different sentences, in the same or different languages, can mean. The truth or falsity of sentences, beliefs, and assertions depends on the truth or falsity of their propositional contents. So, in particular, the truth or falsity of the sentence, 'Giraffes do not use periscopes', or of the belief expressed and the assertion made in uttering it, depends on the truth or falsity of the proposition that giraffes do not use periscopes.

It is widely thought that the meaning of a sentence is determined by the meanings of its constituents and how they are arranged syntactically (in linguistic terms, the meaning, or semantic interpretation, of a sentence is a projection of its syntax at the level of LOGICAL FORM). From this perspective, it is reasonable to identify the meaning of a constituent of a sentence with the contribution it makes to the propositional contents of sentences in which it occurs. Different sorts of expressions make different sorts of contribution. For example, whereas 'giraffes' and 'periscopes' mean things of certain types, 'use' means a relation between agents and instruments, and 'not' is a negation operator.

Some words clearly have meanings and yet their meanings are not clear. With vague words like 'red', 'old' and 'rich', there is no clear boundary between what they do and do not apply to (see VAGUENESS). They appear to have borderline cases, although it has been argued (most notably by Williamson, 1994) that the boundaries are definite but unknowable. Also, different words can have the same meaning. For example, in one of their respective senses, 'correct' and 'right' are synonymous. So are 'teacher' and 'instructor', and 'consume' and 'devour'. And, of course, a word can have more than one meaning. Typical examples of ambiguous words are the nouns 'bat', 'club' and 'stroke', the verbs 'cut', 'draw' and 'lie', and the adjectives 'hard', 'high' and 'true'.

The occurrence of an ambiguous word in a sentence, as in 'George never goes near a golf club', renders the sentence ambiguous. But sentences can also be structurally ambiguous, as with 'German history teachers are pedantic'. Its structural ambiguity is clear from the difference between '[German history] teachers are pedantic' and 'German [history teachers] are pedantic'. Less apparent is the source of the structural ambiguity in 'The chicken is ready to eat', which can mean that the chicken is ready to do some eating or that the chicken is ready to be eaten. However, if we posit the covert presence of empty categories before and after the infinitive 'to eat', a plausible explanation for the ambiguity is that different empty categories are tied to 'the chicken', as indicated by the different indexing in 'The chicken1 is ready e1 to eat e2' and in 'The chicken1 is ready e2 to eat e1' (Chomsky, 1986).

Finally, the references of many words that are univocal in meaning can vary from use to use. These include personal pronouns like 'I', 'you', and 'she', temporal terms like 'today', 'tonight', and 'last week' (also tense indicators), demonstratives like 'this', 'those', 'here', and 'there, and even relational nouns like 'enemy', neighbor', and 'disciple'. See INDEXICALS AND DEMONSTRATIVES.



There are two fundamentally different, though not unrelated, conceptions of semantics in the literature. One conception takes semantics to be concerned with the linguistic meanings of expressions (words, phrases, sentences). On this conception, sentence semantics is a component of grammar. It assigns meanings to sentences as a function of the meanings of their semantically simple constituents, as supplied by their LEXICAL SEMANTICS, and their constituent structure, as provided by their SYNTAX. The other conception takes semantics to be concerned with the truth-conditional contents of sentences (or, alternatively, of utterances of sentences) and with the contributions expressions make to the truth-conditional contents of sentences in which they occur. The intuitive idea underlying this conception is that the meaning of a sentence, the information it carries, imposes a condition on what the world must be like in order for the sentence to be true.

Now the linguistic and the truth-conditional conceptions of semantics would come to the same thing if, in general, the linguistic meanings of sentences determined their truth conditions, and they all had truth conditions. Many sentences, however, are imperative or interrogative rather than declarative. They do not have truth conditions but compliance or answerhood conditions instead. Even if only declarative sentences are considered, in a great many cases the linguistic meaning of a sentence does not uniquely determine a truth condition. One reason for this is ambiguity, lexical or structural. The sentence may contain one or more ambiguous words, or it may be structurally ambiguous. Or the sentence may contain indexical elements. Ambiguity makes it necessary to relativize the truth condition of a declarative sentence to one of its senses, and indexicality requires relativization to a context. Moreover, some sentences, such as 'Jack was ready' and 'Jill had enough', though syntactically well-formed, are semantically incomplete. That is, the meaning of such a sentence does not fully determine a truth condition, even after ambiguities are resolved and references are fixed (Bach, 1994; Sperber & Wilson, 1995). Syntactic completeness does not guarantee semantic completeness.

The apparent conflict between these two conceptions of semantics can be resolved if it is supposed that they have different subject matters: linguistic semantics targets sentences, whereas truth-conditional semantics targets utterances of sentences. But it should not be supposed that an utterance encodes anything that is not encoded by the sentence itself. Information available in a context of utterance may be tied to constituents of the sentence as uttered, but it is not encoded. Such information is combined with the encoded information to produce an interpretation of the utterance. Also, the speaker may convey additional information when speaking indirectly or nonliterally. So we need to distinguish information encoded by a sentence, information tied to its utterance, and information conveyed in uttering it. The first is the province of linguistic semantics, the second the province of truth-conditional semantics, and the last the province of PRAGMATICS.


What are Meanings?

A traditional answer to this question, going back to Plato, is the so-called 'idea' theory. It regards meanings as concepts and complexes of concepts. The meaning of a word (or a morpheme) is the concept (or concepts, if it is ambiguous) conventionally associated with it (a full account of what this involves would be partly philosophical, partly psychological, and partly sociological). The meaning of a phrase is a complex concept made up of the simpler concepts associated with the words it contains, and the meaning of a sentence is the thought made up of the concepts expressed by its constituents, in accordance with how they are arranged syntactically.

A different answer is given by the so-called 'thing' theory: meanings are the references of expressions, as opposed to the concepts they express. 'Things' here include not only particular objects and events but also properties, relations and other abstract objects. The proposition expressed by a sentence consists of a structured entity made up of the referents of its constituent expressions. In the simplest sort of case, the sentence 'Koko is hungry' expresses the proposition that Koko is hungry, whose constituents are Koko and the property of being hungry.


Philosophical Issues about Meaning

The apparent conflict between the idea theory and the thing theory can be resolved by viewing them as answering two different questions: what confers meanings on words, and what comprises the meanings of words? This resolution distinguishes the cognitive contents of words, the ideas or concepts they express, and their semantic contents, what things (or properties or relations) they stand for. It respects the fact that words are used to talk about things, not ideas, and yet are used to express ideas.


Sense and Reference

The most influential implementation of a two-tiered approach has been Frege's (1892). He distinguished the sense of an expression from its reference. The sense is the expression's 'cognitive value', but it also determines the expression's reference, if any. The sense of an expression imposes a condition that the reference must satisfy and does not depend on whether or not there is a reference (see SENSE AND REFERENCE). As Frege explains, 'the thought remains the same whether "Odysseus" has reference or not'. The sense of an expression is its contribution to the thought expressed by a sentence in which it occurs. But words are used to refer to objects of thoughts, not ideas of those objects. Even so, since the same object can be thought of in different ways, under different 'modes of presentation', how the object is thought of, hence the sense of the expression used to refer to it, enters into the thought expressed. This explains, suggests Frege, how it is possible to think or say that Elton John is a singer and yet sincerely deny that Reginald Dwight is a singer, even though Elton John is Reginald Dwight.

By supposing that expressions have both sense and reference, Frege could explain how expressions that are distinct in meaning can have the same reference and how referring expressions without reference, like 'Odysseus', can still be meaningful. Nevertheless, one might ask whether what determines the reference of a referring expression, such as a proper name, an indexical or a demonstrative, is part of the semantic content of sentences in which the expression occurs. According to the doctrine of direct reference (Kaplan, 1989), such expressions refer directly to particular individuals. This is not to deny that there are conditions that something must meet to be the referent but only that such conditions are not part of semantic content. So, for example, if someone says 'I love you', he is using 'I' to refer to himself and 'you' to refer to his listener. However, the fact that he is the speaker and the fact that the other party is the listener do not enter into what the speaker says (see REFERENCE, THEORY OF; DIRECT REFERENCE).



It seems that some sentences are true (or false) solely in virtue of their meanings. For example, to understand the sentence 'All sofas are couches' is to know that it is true. In contrast, one can understand 'All snails are slow' without knowing it to be true (its truth depends on an empirical fact about snails). The first sentence is evidently analytic, the second synthetic. However, despite the glaring intuitive difference between the two, the analytic-synthetic distinction was vigorously attacked by Quine (1951) and others. Quine argued that this distinction cannot be explained in an illuminating way (i.e., without appeal to equally 'obscure' notions like synonymy, translation and, indeed, meaning itself), and he suggested that it is, at best, a matter of degree. Grice and Strawson (1956) appealed to the obvious intuitive difference between, e.g., the two sentences mentioned above to maintain that the analytic-synthetic distinction, whatever it is, is genuine.

Even if Grice and Strawson are right about this, it doesn't follow that analytic truths are true by definition. For it is arguable that words, most words anyway, do not have definitions, at least of the sort that provide singly necessary and jointly sufficient conditions for their application. Wittgenstein challenged the Platonic assumption that all the items to which an unambiguous word applies must have something in common. In the aptly titled, 'Against Definition', Fodor et al. (1980) marshal a variety of examples and considerations to show that Wittgenstein's famous example of 'game' (1953, 66) is not a special case. This raises the question of what word meaning (and knowledge of it) does consist in. On that see LEXICAL SEMANTICS.


Meaning Skepticism

The notion of analyticity, or of synonymy, is no more puzzling than the notion of meaning itself. If a sentence has a certain semantic content, there is no reason why another sentence cannot have the same semantic content and thereby be synonymous with it. Quine's attack on analyticity is ultimately a form of meaning skepticism. Wittgenstein (1953) and Kripke (1982), developing Wittgenstein's ideas, have offered other reasons for such skepticism (the complex debates they have generated are reviewed in Hale and Wright, 1997, chs. 8 and 14-17), arguing that meaning cannot be fixed by facts about individual speakers. They appeal to social facts to explain how meaning can be determinate, but it seems that the same skeptical reasons, if valid at all, apply to this social explanation.


Are Meanings in the Head?

Putnam (1975) used an ingenious thought experiment to argue that at least some linguistic meanings are not 'in the head' (and do not supervene on what is therein). Imagine a distant planet, Twin Earth, where everything is just like it is here, except that there the clear, tasteless liquid that falls from the skies, fills the seas and quenches thirst is composed not of H2O but of some other stuff XYZ. It is 1750 and people on both planets could not, were they given the opportunity, tell the difference between H2O and XYZ. There is no psychological difference between the two populations with respect to the word 'water'. Nevertheless, claims Putnam, 'water' means something different on each planet. Accordingly, an Earthling suddenly transported to Twin Earth would incorrectly apply the word 'water' to the common liquid he sees there (relative to what it means on Earth), which is not water but what we (being in the know) might call 'twin water'. Similarly, a Twin Earthling moved to Earth would wrongly apply the word 'water', as used on Twin Earth, to the common liquid he sees here, which is not twin water but water (H2O). So, Putnam concludes, the reference of 'water' depends on the underlying nature of the clear, plentiful liquid around us and not just on being how we think of it.

In another thought experiment (Burge, 1979), an arthritic patient called Art complains to his doctor, 'My arthritis has spread to my thigh'. Nothing in his acquisition of the term 'arthritis' has kept him from supposing that this inflammatory disease can occur in the limbs as well as the joints. Meanwhile, his Twin Earth counterpart Bart registers a similar complaint. There, however, the term 'arthritis' is used to refer to an inflammatory disease of either the joints or the limbs. Bart's exposure to the term 'arthritis' is the same as Art's, but, given how it is used on Twin Earth, he understands it correctly. Now, observes Burge, when Art says that his arthritis spread to his thigh he is speaking falsely, but when Bart says that, using 'arthritis' as it is used on Twin Earth, he is speaking truly. Therefore, what they are saying about themselves is different, even though there is no subjective difference between them. What they mean by 'arthritis' is partly an social matter.

These thought experiments have met with considerable enthusiasm but also with neglected criticism, e.g. by Unger (1983) and Bach (1987, ch. 13). For one thing, Putnam does not explain why the term 'water' should be (and was in 1750) relevantly different from such terms as 'earth', 'air' and 'fire'. These terms do not denote natural kinds: there is no chemical restriction on the sort of soil to which 'earth' can apply, 'air' is not restricted to particular mixtures of nitrogen, oxygen, and carbon dioxide found in Earth's atmosphere, and 'fire' does not apply only to flame-producing chemical reactions that involve oxidation. As for 'water', although it applies (on Earth) to H2O but not to XYZ, is this because of what 'water' means or because, as a consequence of what we have learned from modern chemistry, we count as water only stuff that is chemically like these samples? It was a discovery that this stuff, unlike earth, air, and fire, is a chemical natural kind. It is not evident why this should show that the term 'water' has a different sort of meaning, with its reference determined in a different way, than 'earth', 'air' and 'fire'.

The arthritis argument depends essentially on the supposition that one can have beliefs with meanings one 'incompletely understands'. It assumes, for example, that Art not only misunderstands the word 'arthritis' but operates with the concept arthritis rather than with some broader concept ('tharthritis') that he mistakenly associates with the word. So, it might be objected, Art understands the term 'arthritis' in precisely the same way as Bart does, and has the very same belief, namely that his tharthritis has spread to his thigh. Whatever evidence there is that he also believes that his arthritis has spread to his thigh is overridden by his idiosyncratic grasp of the term 'arthritis'. We are not tempted to say that he believes that he has inflammation of the joints in his thigh, and this should disincline us to suppose that he means that his arthritis has spread to his thigh. Because of how he misuses the term 'arthritis', he means what Bart means by it.


Meaning and Communication

The distinction between speaker's and linguistic meaning is needed to accommodate the fact that what a speaker means in uttering a sentence can, and often does, diverge from what is meant by the sentence he utters (even if it is neither vague nor ambiguous). A speaker can mean something other than what the sentence means, as in using 'Nature abhors a vacuum' nonliterally, or something more, as in using 'It's getting cold in here' indirectly to ask someone to close the window. In general, utterances can be literal, nonliteral, or indirect (Bach & Harnish 1979, ch. 4), depending on the relationship between the what the sentence means and what the speaker means in uttering it. Insofar as the two diverge, which is most of the time, successful communication requires the listener to fill in the gap inferentially.


Speaker's Meaning vs. Linguistic Meaning

Grice had two further reasons for invoking the distinction between speaker's and linguistic meaning. First, he thought that linguistic meaning could be reduced to speaker's meaning. This reductive view has not gained wide acceptance, partly because of the extreme complexity of its detailed formulation (for example, see Grice, 1989, ch. 6, and Schiffer, 1972) and also because it requires the controversial assumption that language is essentially a vehicle for communicating thoughts and not a medium of thought itself. Still, many philosophers would concede that mental content is more fundamental than linguistic meaning, and perhaps even that semantics reduces to the theory of mental content.

Grice invoked the distinction between speaker's and linguistic meaning also to combat extravagant claims, made by so-called 'ordinary language' philosophers, about various important philosophical terms, such as 'believes' and 'looks'. For example, it was suggested that believing implies not knowing, because to say, e.g., 'I believe that dolphins communicate' is to imply that one does not know this, or to say 'Jenny's eyes look blue' is to imply that Jenny's eyes might not actually be blue. However, as Grice (1989, ch. 2) pointed out, what carries such implications is not what one is saying but that one is saying it, rather than the stronger 'I know that dolphins communicate' or 'Jenny's eyes are blue' (see IMPLICATURE). He also objected to certain ambiguity claims, e.g., that 'or' has an exclusive as well as inclusive sense, as in 'I would like an apple or an orange', by pointing out that it is the use of 'or', not the word itself, that carries the implication of exclusivity. Grice's Modified Occam's Razor ('Senses are not to be multiplied beyond necessity') cut back on a growing conflation of (linguistic) meaning with use, and has since helped linguists appreciate the importance of separating, so far as possible, the domains of semantics and pragmatics.


Communicative Intentions

Grice's concept of speaker's meaning (1989, chs. 5, 14, & 15) was an ingenious refinement of the crude idea that communication is a matter of intentionally affecting another person's psychological states. He discovered that there is a distinctive, rational means by which the effect is achieved: by way of getting one's audience to recognize one's intention to achieve it. The intention includes, as part of its content, that the audience recognize this very intention by taking into account the fact that they are intended to recognize it. A communicative intention is thus a self-referential, or reflexive, intention. Grice observes that 'this seems to involve a reflexive paradox, but it does not really do so' (1989, p. 219), but he does not explain why not.

So consider his example of deliberately frowning to communicate that one is displeased. Since frowning is a natural sign of displeasure, frowning deliberately might lead another to believe that one is displeased. The simplest scheme would be to misleadingly intend the other person to take one's deliberate frown for a spontaneous one and thereby take it as evidence that one is displeased. But what if one's frowning is obviously deliberate? Then it would lack the evidential value of natural frowning. However, one could exploit the common knowledge that frowning is a sign of displeasure and intend one's frowning to be taken as indicating displeasure, provided the other person supposes one intends it to be so taken. Part of what the other person is to take into account in order to infer that one is displeased is that one intends one's frowning as indicating that. This may seem to be circular, to involve a 'reflexive paradox', but Grice does not mean that one's audience has to already know what one's intention is in order to figure out what it is. He means, rather, that the audience has to take into account that one's intention, whatever it is, is intended to be recognized.

In communicating, linguistically or otherwise, one is expressing some PROPOSITIONAL ATTITUDE or other psychological state. In performing the speech act of asserting, one is expressing a belief, in requesting one is expressing a desire, in making an offer one is expressing a conditional intention, in apologizing one is expressing regret for something one did, and in thanking one is expressing gratitude for something the listener did. In general, different types of SPEECH ACTS may be classified by the type of psychological state they express (Bach and Harnish, 1979, ch. 3), although expressing such a state does not imply that one is in it--one could be insincere.


Relevance to Cognitive Science

Cognitive science touches on meaning at many points, some of which are covered in separate articles: CONCEPTS AND CONCEPT ACQUISITION; LANGUAGE ACQUISITION; LANGUAGE PROCESSING; MENTAL REPRESENTATION; PARSING. A few other topics will be briefly addressed here.


Language Understanding

Normal speakers understand tens of thousands of words of their language. Understanding a word is knowing its meaning, but what does that involve? A natural suggestion is that this involves two things: possessing a certain concept and associating that concept with the word. Merely possessing the concept is not enough--the concept must be tied to the word somehow. However, it would be implausible to require that the speaker believe that the concept is the meaning of the word, for that would require him to have concepts of both the concept and the word. Something less cognitive and more associative seems more plausible.

Whatever understanding a word amounts to, part of what it involves is knowing how the word can fit into a sentence, e.g., whether it is a noun, a verb, or some other part of speech. Many words are compounds, like 'newspaper', 'hotbed' (notice the difference between the location of the stress in 'hotbed' and 'hot bed') and 'toadstool' ('mushroom' has the look of a compound but isn't), or is otherwise composed of meaningful parts, stems with prefixes and/or endings, such as 'composed', meaningful', 'parts', 'prefix' and 'ending' (see LEXICAL PROCESSING; LEXICON). Also, lexical knowledge, which includes information about the form and meaning of a word and syntactic constraints on its occurrence, must be combined with knowledge of how words are put together to form sentences. Finally, all of this purely linguistic knowledge must be supplemented with knowledge about how what people mean can go beyond what their words mean. Such knowledge, the subject of PRAGMATICS, is not, strictly speaking, linguistic knowledge.


Pragmatic Processing

Although there is ample research on phonological, lexical, and syntactic, and semantic processing, there has been little work on pragmatic processing, that is, on how people manage, when uttering a sentence, to make themselves understood and how, when hearing a sentence, they manage to understand what the speaker means. Grice's (1989, ch. 2) theory of conversational implicature and the associated maxims of conversation provides a broad framework for identifying the factors that contribute to successful communication, from both the speaker's and the hearer's sides, but it does not explain in detail how this process works.

One issue pertaining to pragmatic processing has begun to be addressed. Recanati (1995) has distinguished different models of the hearer's processing of an utterance in cases where what the speaker means is an enriched version of what the sentence means, e.g., 'I've had breakfast [today]' and 'Mary has [exactly] three children', or where the speaker intends to be implicating something. On one model, the hearer computes the proposition strictly and literally (the 'minimal' proposition) expressed by the utterance before arriving at a candidate for what the speaker means; on another, this proposition is computed, but not necessarily first; on the third, 'local processing' model, this proposition is not computed at all. Bezuidenhout and Cutting (2000) have tested these models experimentally. Their findings tend to favor the second model over the third and to disconfirm the first, the 'literal-first' model.



It is simplistic to suppose that knowing the meaning of a word consists in associating the 'right' concept with it. One obvious complication is that a great many words are ambiguous, thereby expressing more than one concept, but this only shows that the relation of concepts to words is many to one (actually, it is many to many, since many words have synonyms). A further complication is that with a great many words it is unclear how many meanings they have. Take common words like 'go', 'get', 'keep', 'put', 'on', 'in', 'from' and 'to'. Their variable uses might make it seem that their meanings are not fixed from use to use. Ruhl (1989) maintains that each such word does have a core meaning, but that this meaning is too impoverished to comprise what the speaker can mean in using the word. That is, in any particular use the meaning is always enriched in one way or another (see Ravin and Leacock (2000) for variations on this sort of view, applied to a variety of lexical items).

Consider the words 'go', 'from' and 'to' as they occur in the following sentences:

            Max went from Wilmington to Washington.

            The road went from Wilmington to Washington.

            The concert went from 8 to 10.

            Max went from irritated to outraged.

            The house went from Max to his wife.

On a simplistic conception of linguistic meaning, it might seem that only in the first sentence, which involves movement from one place to another, are the words 'go', 'from' and 'to' used literally, and that their uses in the other sentences are in various ways 'extended', hence nonliteral uses. However, it seems plausible to suppose that the words are used literally in all these sentences, despite the fact that they are used differently from one sentence to the next. What is the existence of those various uses shows is that the meanings of these polysemous terms are more abstract than the simplistic movement model would suggest.

A great many seemingly univocal words turn out to have distinct but related forces when applied to things of different sorts. Consider the adjective 'sad' as it modifies these different nouns:

            sad person / sad face / sad song / sad episode

A sad face is not sad in the way that a sad person is. Rather, it expresses the person's sadness. A sad song isn't sad in the way that a sad person is, and need not express any particular person's sadness, such as the writer's or the performer's (nor need it make the listener feel sad). And a sad episode isn't itself sad--it causes sadness. Similarly, notice the variable import of the adjectives 'fast', 'generous' and 'conscious', as they occur in the following phrases:

            fast car / fast track / fast race

            generous donor / generous gift

            conscious being / conscious state

Obviously the most plausible import of each of these adjectives varies with the noun it modifies. Indeed, only in the first phrase in each case does the adjective express a property that directly belongs to what the noun denotes. Considering that such adjectives have counterparts in other languages with the same different but related uses, their variable import is due to polysemy, not ambiguity (linguistic coincidence). Pustejovsky (1995) proposes that such polysemy involve 'co-compositionality': what varies from case to case is not a term's semantic properties but how those properties interact with those of the term it combines with. Although his ambitious theory, intended primarily as a computational model, is a definite improvement over what he calls 'sense enumeration lexicons', it seems to conflate pragmatic plausibility with semantic possibility. Although a speaker is not likely to use, e.g., 'sad song' to mean song experiencing sadness, 'fast track' to mean track that is moving fast, or 'generous gift' to mean gift that gives generously, arguably these improbable interpretations of those phrases are nevertheless generated by the grammar. It does not seem to be a semantic fact (a fact about meanings of words) that songs do not experience sadness, tracks do not move (much less move fast), and gifts are not generous in their giving.

The above examples of adjectival modification raise questions about the nature of  compositionality (see Partee 1995). So do the following noun-noun compounds:

            child abuse / drug abuse

            election nullification / jury nullification

            slalom skiing / snow skiing / heliocopter skiing

            jellyfish / goldfish / catfish

            clipboard / diving board / bread board / game board / headboard

Such examples make it clear that compositionality is not as straightforward as it might seem, for there are different ways in which the meanings of words can combine. See LEXICAL SEMANTICS.


Concepts vs. Conceptions

Cognitive scientists, such as prototype theorists and developmental psychologists, often refer to people's conceptions of various types of things as 'concepts'. This usage is misleading, since conceptions are much richer than concepts and play different roles. Conceptions play a role in how people group and categorize things, judge similarities and differences between things, and form theories of things of different types, but concepts are what people associate with words in virtue of which words mean what they do. The difference is clear if we consider that people can have different conceptions of a given type of thing and yet use the same word to refer to it. For example, they can associate the same concept with the word 'tree' but have different conceptions of trees.



Bach K (1987) Thought and Reference. Oxford: Oxford University Press.

Bach K (1994) Conversational impliciture. Mind & Language 9: 124-162.

Bach K and Harnish RM (1979) Linguistic Communication and Speech Acts. Cambridge: MIT Press.

Bezuidenhout A L and Cutting J C (2000) Literal meaning, minimal propositions and pragmatic processing. Journal of Pragmatics 000: 000-000.

Burge, T (1979) Individualism and the mental. In: French P, Uehling T and Wettstein H (eds) Midwest Studies in Philosophy, vol. 4, pp. 73-121. Minneapolis: University of Minnesota Press.

Chomsky N (1986) Knowledge of Language. New York: Praeger.

Fodor J A, Garrett MF, Walker E T and Parkes C (1980) Against definition. Cognition: 8: 1-105.

Frege G (1892) On sense and reference. Reprinted in: Harnish RM (ed) (1994) Basic Topics in the Philosophy of Language, pp. 142-160. Englewood Cliffs, NJ: Prentice-Hall.

Grice HP (1989) Studies in the Way of Words. Cambridge, MA: Harvard University Press.

Grice HP and Strawson PF (1956) In defense of a dogma. Philosophical Review 65: 141-158.

Hale B and Wright C (eds) (1997) A Companion to the Philosophy of Language. Oxford: Blackwell.

Kaplan D (1989) Demonstratives. In: Almog J, Perry J and Wettstein H (eds) Themes from Kaplan, pp. 481-563. Oxford: Oxford University Press.

Kripke S (1982) Wittgenstein on Rules and Private Language. Cambridge, MA: Harvard University Press.

Partee Barbara (1995) Lexical semantics and compositionality. In Gleitman L and Liberman M (eds.), An Invitation to Cognitive Science, Volume 1: Language, 2nd edition, pp. 311-60. Cambridge, MA: MIT Press.

Pustejovsky J (1995) The Generative Lexicon. Cambridge, MA: MIT Press.

Putnam H (1975) The meaning of 'meaning'. In: Gunderson K (ed) Language, Mind and Knowledge, pp. 131-193. Minneapolis: University of Minnesota Press.

Quine WV (1951) Two dogmas of empiricism. Philosophical Review 60: 20-43.

Ravin Y and Leacock C (2000) Polysemy: Theoretical and Computational Approaches. Oxford: Oxford University Press.

Recanati F (1995) The alleged priority of literal interpretation. Cognitive Science 19: 207-232.

Ruhl C (1989) On Monosemy: A Study in Linguistic Semantics. Albany, NY: State University of New York Press.

Schiffer S (1972) Meaning. Oxford: Oxford University Press.

Sperber D and Wilson D (1995), Relevance, 2nd ed. Oxford: Blackwell.

Unger P (1983) Philosophical Relativity. Minneapolis: University of Minnesota Press.

Williamson T (1994) Vagueness. London: Routledge.

Wittgenstein L (1953) Philosophical Investigations. New York: Macmillan.


Further Reading

Bach, K (1999) The semantics-pragmatics distinction: what it is and why it matters. In: Turner K (ed), The Semantics-Pragmatics Interface from Different Points of View, pp. 65-84. Oxford: Elsevier.

Cruse DA (1986) Lexical Semantics. Cambridge, UK: Cambridge University Press.

Davidson D (1984) Essays on Truth and Interpretation. Oxford: Oxford University Press.

Kasher A (ed) (1998) Pragmatics: Critical Concepts. London: Routledge.

Keefe R and Smith P (1996) Vagueness: A Reader. Cambridge, MA: MIT Press.

Kripke S (1980) Naming and Necessity. Cambridge, MA: Harvard University Press.

Larson R & Segal G (1995) Knowledge of Meaning. Cambridge, MA: MIT Press.

Levinson S (2000) Presumptive Meanings: The Theory of Generalized Conversational Implicature. Cambridge, MA: MIT Press.

Lyons J (1995) Linguistic Semantics: An Introduction. Cambridge, UK: Cambridge University Press.

Quine WV (1960) Word and Object. Cambridge, MA: MIT Press.

Searle J (1979) Expression and Meaning. Cambridge, UK: Cambridge University Press.


ambiguous#Having several unrelated meanings

analytic#True solely in virtue of meaning

communicate#Express a psychological state

compositionality#Determination of the meaning of a complex expression by the meaning of its constituents and their syntactic relationship

indirect speech act#Performance of one speech act by way of performing another

nonliteral utterance#Use of a sentence that means one thing to mean something else

polysemous#Having several related meanings

proposition#The truth-conditional content of a sentence, statement, belief, etc.

synonymous#Having the same meaning as another expression

vague#Not having a clear range of application