Lexical semantic development of high-frequency polysemous words in L2 writing: An LLM-assisted sense-level analysis

Abstract

A key dimension of L2 writing development is learners’ ability to use words across a wider and more differentiated range of senses, yet this development remains difficult to capture at scale. This study proposes a scalable corpus-based approach to analyzing lexical semantic development at the sense level, using large language models for sense annotation and validating their performance against dictionary examples and human-annotated learner sentences. Focusing on 136 selected high-frequency polysemous lemmas, we compare sense distributions across 16 proficiency levels in EFCAMDAT with those in the written component of the BNC. Two main developmental patterns emerged. First, Jensen–Shannon distance between learner and native reference sense distributions decreased overall with proficiency, indicating increasing alignment with native reference usage. Second, entropy increased with proficiency, suggesting greater diversification of sense use. Exploratory item-level analysis showed that individual lemmas followed heterogeneous and sometimes non-linear trajectories, partly associated with differences in writing topic and genre. These findings suggest that L2 lexical development in writing involves not only changes in the frequency of word use but also the gradual expansion and differentiation of sense repertoires. The study also offers pedagogical implications for vocabulary instruction, writing development, and sense-aware assessment in second language education.