Toward an Exemplar-based Computational Model for Cognitive Grammar 1 Usage-based Models of Language Structure

An exemplar-based computational framework is presented which is compatible with Cognitive Grammar. In an exemplar-based approach, language acquisition is modeled as the incremental, data-oriented storage of experiential patterns, and language performance as the extrapolation of information from those stored patterns on the basis of a language-independent information-theoretic similarity metric. We show that this simple architecture works for many aspects of phonological, morphological, and morphosyntactic acquisition and processing. Furthermore, we sketch how the approach may also work for syntactic processing. A central insight of the approach, based on the results of computational modeling experiments, is that abstraction of representations is not only unnecessary to achieve generalization (i.e. to make the system productive, and to make it gòbeyond' the learned patterns), but even harmful, and that useful language-independent metrics can be found for deening similarity in the context of language processing. In the generative tradition, generality is achieved by means of abstraction, and the representations of choice to describe these abstractions are rules. This implies that redundancy and the storage of individual instances are to be avoided, except for exceptions to the generalizations expressed in rules. In Langacker, 1991 (Chapter 10), this methodology is critically examined, and cognitive grammar is described as an alternative usage-based model of language structure. In the latter, bottom-up, approach, patterns (rules, generalizations) and (redundant) instantiations of those rules are assumed to co-exist in the grammar, describing phenomena at all levels of generality, from exceptionless regularities to idiosyncratic exceptions. Rules are presumed to be necessary for the computation of novel instantiations. In the remainder of this paper we will introduce an exemplar-based approach to language acquisition and processing. The approach is in large part compatible with Lan-gacker's usage-based model, but is more radical in its "maximalism": language knowledge is supposed to consist only of "instantiations" (exemplars); there is no role for explicit abstractions corresponding to (sub)regularities. We will argue on the basis of computational modeling experiments that the adoption of abstractions (rules, patterns), taken as necessary for explaining generalization and productivity in both the generative and the cognitive grammar approach, is misguided. Furthermore, the exemplar-based approach contributes to making cognitive grammar ideas more concrete by providing computational operationalisations of both acquisition and processing in such a framework.