Predicative Metaphors Are Understood as Two-Stage Categorization: Computational Evidence by Latent Semantic Analysis

In this paper, we address the problem of how people understand predicative metaphors such as “The rumor flew through the office”, and argue that two-stage categorization is the process via which predicative metaphors are understood. In the two-stage categorization process, the verb of a predicative metaphor (e.g., fly) evokes an intermediate category, which in turn evokes a metaphorical category of action or state to be attributed to the target noun (e.g., rumor), rather than directly creating a metaphorical category as argued by Glucksberg’s (2001) categorization theory. We test our argument by means of computer simulation experiment in which the meanings of predicative metaphors are computed from the representations of the verb and the noun in a multidimensional semantic space constructed by latent semantic analysis. In the simulation, three algorithms for predicative metaphor comprehension, i.e., two-stage categorization, categorization and comparison, are compared in terms of how well they mimic human interpretation of 30 predicative metaphors. The simulation result was that the two-stage categorization algorithm best mimicked human interpretation of predicative metaphors, but the comparison model yielded the best performance in the case of less apt metaphors. These findings suggest that predicative metaphors, in particular apt metaphors, are understood via a two-stage categorization process, but less apt metaphors may possibly be understood via a comparison process.