Revisiting the Sound–Meaning Systematicity: A Fine-Grained Computational Analysis

Arbitrariness is a foundational principle in linguistics established by de Saussure (1916), yet a number of recent studies report nonarbitrary form–meaning associations, a phenomenon called systematicity. However, they usually overlook some possible confounders, primarily morphology and etymology, and rely on relatively coarse phonemic and semantic representations. In this work, we re-evaluate whether form–meaning correlation persists beyond known confounders with a fine-grained approach, and examine such correlation in English, French, Hindi, and Chinese. Starting with a rigorously filtered lexicon that excludes morphological and etymological variants, we use phonemic representations based on distinctive features and semantic representations from contextualized language model embeddings to calculate phonemic and semantic distances, respectively. By comparing our results against a randomized control baseline, we confirm the presence of weak but consistent systematicity across the examined languages, aligning with previous literature. Furthermore, we contribute a new method to explicitly quantify the distinct contributions of morphology and etymology to observed systematicity. While morphology and etymology are known as factors of systematicity, we find that their effects differ by language and do not necessarily inflate the observed correlation. Moving beyond the global correlations reported in prior literature, we identify and document a “locality effect”—form–meaning correlation is stronger within local semantic neighborhoods compared to the whole lexicon (Gutiérrez, Levy, and Bergen 2016)—which is inconsistent across languages. While English and Hindi show higher local systematicity, this effect is weak or even negative in Chinese and French, suggesting a cross-linguistic divergence in sound–meaning relationship.