The lexicon adapts to competing communicative pressures: Explaining patterns of word similarity

Cross-linguistically, lexicons tend to be more phonetically clustered than required by the phonotactics of the language; that is, words within a language are more similar to each other than they need to be. In this study, we investigate how this property evolves under the influence of competing communicative pressures: a production-side pressure to re-use more easily articulated sounds, and a comprehension-side pressure for distinctiveness of wordforms. In an exemplar-based computational model and a communication experiment using a miniature artificial language, we show that natural-language-like levels of clustering emerge from a trade-off between these pressures. With only one pressure at work, the resulting lexicons tend to inhabit an extreme region of the possible design space: production pressures alone give rise to maximally clustered lexicons, while comprehension pressures alone give rise to maximally disperse lexicons. We also test whether clustering emerges more strongly for high-frequency items, but our results lend support only to a weak relationship between frequency and clustering. Overall, this study adds to a growing body of evidence showing that mechanisms operating at the level of individual language users and individual episodes of communication can give rise to emergent structural properties of language.