Polysemy and the lexicon
Alternative perspectives on word meaning
In view of the kinds of problems that arise when one attempts to give a detailed semantic description of a word, it comes as no surprise that different scholars (and lexicographers) often fail to converge on a common answer to the question of just how many different meanings a word has and what these meanings are supposed to be (Fillmore & Atkins 2000). Given this state of affairs, there are, broadly speaking, two approaches that an analyst might take. One would be to continue refining the criteria by which two uses are to be regarded as elaborations of the same meaning or representative of distinct polysemes. But given the less than unanimous outcomes of research to date, it is not surprising that a number of scholars have recently come to question the traditional notion of polysemy (namely, as the association of a single word-form with more than one distinct meaning), and the reification of meaning which the notion entails (Allwood 2003; Geeraerts 1993; Zlatev 2003; Langacker 2006). It may be well to stand back a little from the discussion and consider the theoretical issues that drive it.
Elsewhere (Taylor 2003a, c) I have suggested that a desire to associate a word with a fixed number of distinct meanings derives from a specific conceptualization of the object of investigation. I have referred to this (Taylor, in press, a) as the “generative” model of language, or, more colloquially, as the “dictionary + grammar book” model. According to this model, knowledge of a language resides in two components. One component is the dictionary, or lexicon, which lists the basic meaning-bearing units of the language (prototypically, the words). The other component is the syntax, which lists the rules whereby smaller elements can be combined. The rationale for this view of linguistic knowledge is explained in the following remarks by Jackendoff:
Since the number of possible utterances in a human language is unlimited, language users cannot store them all in their heads. Rather, … knowledge of language requires two components. One is a finite list of structural elements that are available to be combined. This list is traditionally called the “lexicon,” and its elements are called “lexical items”. … The other component is a finite set of combinatorial principles, or a grammar. To the extent that speakers of a language (or a dialect) are consistent with one another … we can speak of the “grammar of a language” as a useful approximation to what all its speakers have in their heads. (Jackendoff 2002: 39)
For a first approximation, the lexicon is the store of words in long-term memory from which the grammar constructs phrases and sentences. (Jackendoff 2002: 130)
Baker is even more explicit in his presentation of the model:
Suppose someone were to ask you what English is. … You might say that English is the set of sentences that are constructed by combining the following ingredients (you hand her a massive dictionary that lists all the English words) according to the following rules of grammar (you hand her an equally massive English grammar). … [B]etween them they would tell the reader how to make any conceivable English sentence. (Baker 2001: 53–4)
There are several corollaries to the dictionary + grammar book model. One is the expectation that complex expressions exhibit compositionality. According to the compositionality principle, complex expressions are formed by the assembly of smaller units, such that the properties of the whole can be predicted from the properties of the parts and the manner of their combination. Now, for compositionality to go through, it is necessary that each of the constituent parts has a set of fixed properties which it is able to contribute to the whole. This, in turn, requires that words have fixed meanings. Moreover, since it is manifestly the case that some words can make different semantic contributions depending on the context of their use, the postulation of polysemy becomes imperative. Concretely: the normal interpretation of I went round the guests, introducing myself, if it is to be derived compositionally, must make reference to a meaning of round which is different from the meaning that is accessed in the compositionally derived reading of I walked round the lake.
The compositionality principle is deeply ingrained in our thinking about language. Indeed, compositionality is accorded the status of an axiom, needed in order to account for our ability to produce (and readily understand) novel combinations of words. This, indeed, is the point which Jackendoff makes, at the beginning of the above citation. Three considerations, however, weaken the compositionality imperative (and the generative model of language with which it is associated):
(i) Full compositionality is rarely the case. There is a sense in which every expression, even the most banal, when used in a specific situation to symbolize the speaker’s momentaneous conceptualization, will inevitably take on properties that go beyond, or which may be at variance with, its strictly compositional value. This would be the case even if highly specific senses were attributed to a word. The point was recognized by the two-level theorists, and, indeed, was one of their motivations for recognizing a “conceptual” (or situation-specific) level of representation, as distinct from a “semantic” level. The point has also been made by Langacker (though within a different theoretical framework). See, for example, his remarks on the expression the football under the table (Langacker 1987: 279–82), as the expression would be understood in different utterance contexts, and the more extended discussion of the example in Taylor (2002: 106–109).
(ii) In spite of the pre-eminence of the compositionality principle, it is generally acknowledged that compositionality fails to apply in the case of idiomatic expressions. In fact, idioms are commonly defined against the backdrop of compositionality– an idiom, namely, is an expression whose (conventionalized) meaning is not compositional (see, e.g., Chomsky 1980: 149, cited in Nunberg et al. 1994: 498). Let the cat out of the bag (‘reveal the secret’) is an idiom, since the meaning of the expression cannot be worked out on the basis of the meanings which cat and bag have elsewhere in the language (more precisely: from the meanings which cat and bag have, when they are not being used idiomatically). The approach, needless to say, presupposes that one can determine, for any use of a word, whether its use is idiomatic or non-idiomatic. This is a requirement which falls out, once again, from the dictionary + grammar book model, and its various entailments.
(iii) The distinction between the idiomatic and the non-idiomatic may not be so clear-cut, however. Interestingly, much of the relevant research has been conducted, not so much by linguistic theoreticians, but by scholars with a more practical orientation– second language acquisition researchers and lexicographers (see, e.g., Moon 1998; Wray 2002). The emerging view is that a very great deal of a person’s knowledge of a language may consist in knowledge of multi-word expressions– not only idioms as such, but phraseologies, formulaic phrases, collocations, clichés, quotations, catch phrases, and such like. At the same time, the scope of the idiomatic has been extended to include not only the prototypical “non-compositional” idioms, of the kind let the cat out of the bag, but all manner of expressions which, by dint of their frequent use, have become entrenched in the minds of language users and which are not plausibly assembled afresh on each occasion of their use. Even expressions whose semantics are not at all remarkable (e.g. I wouldn’t do that if I were you) may still be regarded as “idiomatic”, to the extent that they are learned, and used, as pre-formed chunks. A particularly fruitful line of study has centered on constructions, specifically, so-called constructional idioms, and the predisposition of certain words to occur in certain constructions, together with the tendency for constructions to select the words which instantiate them (Stefanowitsch & Gries 2003; Goldberg 2006). On this approach, one facet of knowing a word would be knowing the word’s distribution within constructions. Knowing the word dint, for example, involves knowing that the word occurs almost exclusively in the complex preposition by dint of (Taylor, in press, b).
Searching for “the meaning of a word” may be a futile exercise, once we take account of the contexts in which the word is used. This is because contextualized uses may take on semantic values that cannot be compositionally derived. The process, moreover, can be recursive, in that a contextualized use may itself have contextual variants. As we consider the properties of ever larger configurations, the semantic contribution of component words will become less and less of a concern.
