Modelling metaphorical meaning: A systematic test of the predication algorithm

Metaphors, such as lawyers are sharks, are seemingly incomprehensible when reversed (i.e. sharks are lawyers). For this reason, Kintsch (Psychonomic Bulletin & Review, 7(2), 257-266, 2000) argued that computational models of metaphor processing need to account for the non-reversibility of metaphors, and demonstrated success with his computational model, the "predication algorithm," in simulating metaphor comprehension in a way that is consistent with human cognition. Predication is an ostensibly directional algorithm because its equation is asymmetric such that semantic properties of the vehicle (e.g., sharks) are added to the topic (e.g., lawyers) rather than vice versa. Although predication has been accepted as a viable algorithm for simulating metaphor processing, one of its core assumptions - that the semantic processing of metaphor is directional - has not been systematically tested, nor has it been systematically tested against multiple rival algorithms in simulating metaphor comprehension. To that end, we tested the predication algorithm's performance and that of a set of rival algorithms in simulating metaphor comprehension and distinguishing between canonical (e.g., lawyers are sharks) and reversed (e.g., sharks are lawyers) metaphors. Our findings indicate (1) the predication algorithm is comparable to simpler, rival algorithms in simulating metaphor comprehension, and (2) despite the beliefs about the directionality of the predication algorithm, it produces surprisingly similar simulations for canonical metaphors and their topic-vehicle reversals. These findings argue against predication, at least as implemented in Kintsch's (2000) algorithm, as a viable model of metaphor processing. Implications for computational and psycholinguistic approaches to metaphor are discussed.