Proceedings of the Annual Meeting of the Cognitive Science Society

Semantic shift, or the diachronic variation in word meaning, is a topic commonly discussed in historical and cognitive linguistics. Previous work has typically focused on characterizing semantic change in a population, but how word meanings shift over time in individuals is underexplored. We propose a computational framework for analyzing diachronic semantic shift at the individual level by adapting techniques for population-level semantic change. Using speaker-labelled text corpora, our framework utilizes contextualized word embeddings to identify predictable patterns in semantic shift among individual speakers. We discover that frequency predicts the rate of semantic shift over time across speakers, while polysemy does not. Additionally, nouns exhibit higher semantic stability over adjectives and verbs. These patterns partially mirror existing findings on regularity in historical semantic change, suggesting shared tendencies across individual and population levels. Our work bridges computational approaches to historical semantics with the diachronic modeling of personal lexicons.