Metaphor, a figure of speech and a cognitive device, offers a powerful way to explain one conceptual domain in terms of another. Particularly successful metaphorical mappings are conventionalized through frequent use and lose their creative quality. They become sense extensions of polysemous words. Our dataset project captures such metaphors with ten regular polysemy patterns that manifest repetitively in the meaning structures of English words. Regular metaphor, unlike its counterpart regular metonymy, has not previously received a dedicated dataset, and we intend to close this gap. The dataset under construction features naturalistic sentences extracted from a general language corpus and is manually annotated with sense labels for metaphorically extended polysemes. Its intended use is to support linguistic, cognitive, and computational investigations into patterns of meaning in polysemy, while accounting for its complexity, regularity, continuity, and heterogeneity. We see neural language models as an excellent experimental ground for such research because they are able to show both distributional (continuous) and symbolic (discrete) behavior in language processing and representation. In this paper, we reflect on how these systems tally.

Linguistic accounts show that a word’s polysemy structure is largely governed by systematic sense alternations that form overarching patterns across the vocabulary. While psycholinguistic studies confirm the psychological validity of regularity in human language processing, in the research on large language models (LLMs) this phenomenon remains largely unaddressed. Revealing models’ sensitivity to systematic sense alternations of polysemous words can give us a better understanding of how LLMs process ambiguity and to what extent they emulate representations in the human mind. For this, we employ the measures of surprisal and semantic similarity as proxies of human judgment on the acceptability of novel senses. We focus on two aspects that have not received much attention previously —metaphorically motivated patterns and the continuous nature of regularity. We find evidence that surprisal from language models represents regularity of polysemic extensions in a human-like way, discriminating between different types of senses and varying regularity degrees, and overall strongly correlating with human acceptability scores.