Polysemy and the lexicon
Compromise #2: Having it both ways
There is a second kind of compromise between rampant polysemy and austere monosemy and it is one which has been particularly favored in cognitive linguistic accounts. We allow both fine-grained semantic distinctions and more general, schematic meanings which capture the commonality amongst groups of specific meanings. The different meanings, at various levels of schematicity, are typically displayed in a network. A particularly fine illustration of the methodology is provided by Tuggy’s (2003) analysis of the Nawatl verb kı̄sa (though many other examples, to be sure, could also be cited; see Palmer, Figure 2 in Word classes and grammatical functions in Cognitive Grammar). The network model is clearly an advance on the “half-way” solution, in that the analyst is not forced to choose between an array of distinct senses and an overarching schematic sense, but simply incorporates both. Note, for example, that the network model can easily handle cases of less than fully productive polysemization. Specific senses, even when predictable from general principles, would be included in the network representation; the fact that an otherwise expected sense is not attested would be captured, quite simply, by the absence of the relevant sense.
I have characterized and presented arguments for the network approach to polysemy elsewhere (Taylor 1995). The network model cannot, however, be regarded as the last word on polysemy. In recent years, the model has in fact come in for some substantial criticism. These are both methodological and conceptual in nature:
(i) Methodological problems. Suppose two linguists are working independently on the same word and each makes a proposal for a polysemy network. Suppose, furthermore, that the proposed networks differ in important respects. One linguist identifies a large number of highly specific senses which the other ignores, or there is disagreement about how particularized senses are to be allocated to more schematic senses and how these schematic senses are to be characterized. What criteria do we apply for evaluating the two accounts? The example is by no means hypothetical. Consider, for example, the still ongoing story of over.4 What has driven the discussion of this word, for more than two decades since Brugman’s (1981) pioneering dissertation, is disagreement over these methodological questions. How many different senses does the word really have and how are we to decide the matter? What is the content of these different senses? Should we accord representational status to highly particularized senses? To what extent is it feasible to collapse different uses under more schematic representations and what is the content of the proposed schemas? Further questions have been, how these different sense clusters relate to each other, and which, if any, of the proposed senses can be regarded as the most central.
(ii) The second issue concerns the ontological status of networks. Is a network supposed to have psychological reality? Are networks meant as hypotheses about speakers’ knowledge of words? If so, how can the psychological reality of a proposed network be tested? If not, what is a network supposed to be representing? What is the relevance of the networks in language acquisition? At what point in the network is a word accessed in production and reception? These are still very much open questions (Sandra & Rice 1995; Hallan 2001; Rice 2003; Langacker 2006).
4. The literature on over is vast. For some recent contributions, see Tyler and Evans (2001), Queller (2001), and Deane (2005), and the references cited therein.
