Epistemological Pitfalls in Metaphor Comprehension: A Comparison of Three Models and a New Theory of Metaphor

Abstract

[Introduction] If metaphor is to be viewed as a fundamental cognitive agency, as recent work suggests, what ramifications does this view have for a model of semantic memory? This paper presents a computational treatment of metaphor comprehension, named Sapper (see Veale & Keane 1993, 1994), which is built upon a parallel, adaptive, and learning network model of semantic memory. Sapper is a hybrid symbolic/connectionist model which views the interpretation of novel metaphors as a process of connectionist bridge-building, a process which subsequently alters the activation dynamics between different conceptual schemata in semantic memory, thereby causing these schemata to interact (following Black, 1962) in a representationally dynamic fashion. Sapper employs a bottom-up approach to metaphor comprehension, one which encourages the existing structure of semantic memory to shape and accommodate the most natural interpretation for each concept juxtaposition. In this way, the entirety of contingent background knowledge is brought to bear on the interpretation process. However, the Sapper mechanism combines the base-filtering stage of interpretation with the formation of initial match hypotheses in a single connectionist phase, thereby significantly curtailing the sweep of the matching process and side-stepping the factorial death that models such as the Structure Mapping Engine (SME - see Gentner 1983) can all too easily fall victim to. In fact, this paper will provide empirical evidence that SME is fundamentally unsuited to the interpretation of a broad class of metaphors that rely on an object-centred, as opposed to predicate-centred, representation. These metaphors, which often find linguistic expression as noun: noun comparisons, depend mainly upon the adequate representation of object partonomies and taxonomies, rather than the representation of actions and events toward which models such as SME are inherently biased. This evidence casts serious epistemological doubts on the validity of these models as cognitive theories of human metaphor comprehension.