Stance-aware Definition Generation for Argumentative Texts

ACL ARR 2025 February Submission4441 Authors

Abstract: Definition generation models trained on dictionary data are generally expected to produce neutral and unbiased output. However, previous studies have shown that generated definitions can inherit biases from both the underlying models and the input context. This paper examines how stance-related bias in argumentative data influences generated definitions, demonstrating that even dictionary-trained models can produce outputs that reflect subjective or emotive framing. Additionally, we explore the intentional generation of persuasive definitions, which express an opinion about the target word based on argumentative usage examples. Through this study, we provide new insights into bias propagation in definition generation and its implications for argument mining and other Natural Language Processing applications.

Paper Type: Long

Research Area: Sentiment Analysis, Stylistic Analysis, and Argument Mining

Research Area Keywords: definition generation, argument mining, sentiment analysis, stance detection

Contribution Types: Publicly available software and/or pre-trained models, Data analysis, Position papers

Languages Studied: English

Submission Number: 4441