TEMPER: Testing Emotional Perturbation in Quantitative Reasoning

2026-04-09

AI Summary

This study investigates whether affective expression impairs the quantitative reasoning capabilities of large language models while preserving all numerical and semantic content. To this end, the authors develop a controllable emotional rewriting framework that transforms neutral questions into emotionally charged yet semantically equivalent variants and systematically evaluate its impact across 18 mainstream models. They report the first evidence that emotional style alone can reduce reasoning accuracy by 2–10 percentage points, and demonstrate that neutralizing inputs at inference time effectively restores performance—a benefit not observed with non-emotional rephrasings. Beyond revealing the detrimental effect of emotion on model robustness, this work introduces a lightweight mitigation strategy and releases a general-purpose, controllable style-transfer evaluation framework.

Abstract

Large language models are trained and evaluated on quantitative reasoning tasks written in clean, emotionally neutral language. However, real-world queries are often wrapped in frustration, urgency or enthusiasm. Does emotional framing alone degrade reasoning when all numerical content is preserved? To investigate this, a controlled emotion translation framework is developed that rewrites problems into emotional variants while preserving all quantities and relationships. Using this framework, Temper-5400 (5,400 semantically verified emotion--neutral pairs) is constructed across GSM8K, MultiArith, and ARC-Challenge, and evaluated on eighteen models (1B to frontier scale). Two core results emerge: First, emotional framing reduces accuracy by 2-10 percentage points even though all numerical content is preserved. Second, neutralizing emotional variants recovers most of the lost performance, showing both that the degradation is tied to emotional style rather than content corruption and that neutralization can serve as a lightweight inference-time mitigation. Non-emotional paraphrases cause no such degradation, implicating emotional content rather than surface-level changes. Beyond emotion specifically, the benchmark construction procedure provides a general framework for controlled stylistic translation and robustness evaluation.

Problem

• emotional perturbation
• quantitative reasoning
• large language models
• reasoning robustness
• controlled stylistic translation

Innovation

• emotional perturbation
• controlled stylistic translation
• quantitative reasoning robustness
• emotion-neutral benchmark
• inference-time neutralization