Beyond Neutrality: Detecting Polarity Divergence in AI-generated Dictionary Definitions Via Sentiment Screening
International Journal Of Lexicography · 16.01.2026 · 39
DOI: 10.1093/ijl/ecag041 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal Of Lexicography, online erschienen am 02.11.2026. Verfasst von Xi Chen und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Generative AI (GenAI) offers efficiency gains for lexicography but raises epistemological concerns, as its probabilistic consensus may flatten semantic complexity. This study investigates the boundaries of GenAI’s capacity in generating lexical definitions. Comparing 260 U.S. political lemmata from the Oxford English Dictionary (OED) with definitions produced by DeepSeek-3.2, Gemini-3, and ChatGPT-5, we analysed sentiment polarity with two instruments of different architecture: a discrete three-way classifier (Stanza) at sentence level, and a continuous valence measure (VADER) applied to whole definitions. Macro-level aggregate screening showed no source differing significantly from the OED. Micro-level comparisons, however, revealed systematic divergences: the two instruments disagree systematically, and that disagreement makes the distinction between descriptive polarity and polarity divergence observable. AI definitions were generally more concise, but some sacrificed semantic completeness. Crucially, divergences varied across models and lemmata, suggesting data- and model-specific influences rather than generalized political bias. This study extends sentiment analysis from descriptive tool to epistemological probe, reconfiguring the human–AI division of labour: the lexicographer’s core task shifts from comprehensive revision to targeted anomaly detection, treating anomalies as diagnostic signals requiring subsequent in-depth qualitative review.
Zitieren
APA
Chen, X., & Zheng, X. (2026). Beyond Neutrality: Detecting Polarity Divergence in AI-generated Dictionary Definitions Via Sentiment Screening. International Journal Of Lexicography, 39. https://doi.org/10.1093/ijl/ecag041
Vancouver
Chen X, Zheng X. Beyond Neutrality: Detecting Polarity Divergence in AI-generated Dictionary Definitions Via Sentiment Screening. International Journal Of Lexicography. 2026;39. doi:10.1093/ijl/ecag041
BibTeX
@article{Chen2026,
author = {Chen, Xi and Zheng, Xuelong},
title = {Beyond Neutrality: Detecting Polarity Divergence in AI-generated Dictionary Definitions Via Sentiment Screening},
journal = {International Journal Of Lexicography},
year = {2026},
volume = {39},
doi = {10.1093/ijl/ecag041}
}
EUVIMED