Emotion is Not Just a Label: Latent Emotional Factors in LLM Processing

TL;DR

This paper investigates how latent emotional factors influence large language models' reasoning and attention, introducing a new dataset and regularization method to improve performance across emotionally diverse texts.

Contribution

It reveals the impact of emotion on attention geometry in transformers, introduces the AURA-QA dataset, and proposes an emotional regularization framework to enhance model robustness.

Findings

Emotion affects attention metrics like locality and entropy in transformers.

The proposed regularization improves QA performance across diverse emotional contexts.

Models trained with emotion-aware methods show better generalization under distribution shifts.

Abstract

Large language models are routinely deployed on text that varies widely in emotional tone, yet their reasoning behavior is typically evaluated without accounting for emotion as a source of representational variation. Prior work has largely treated emotion as a prediction target, for example in sentiment analysis or emotion classification. In contrast, we study emotion as a latent factor that shapes how models attend to and reason over text. We analyze how emotional tone systematically alters attention geometry in transformer models, showing that metrics such as locality, center-of-mass distance, and entropy vary across emotions and correlate with downstream question-answering performance. To facilitate controlled study of these effects, we introduce Affect-Uniform ReAding QA (AURA-QA), a question-answering dataset with emotionally balanced, human-authored context passages. Finally, an emotional regularization framework is proposed that constrains emotion-conditioned representational drift during training. Experiments across multiple QA benchmarks demonstrate that this approach improves reading comprehension in both emotionally-varying and non-emotionally varying datasets, yielding consistent gains under distribution shift and in-domain improvements on several benchmarks.