TEMPER: Testing Emotional Perturbation in Quantitative Reasoning

TL;DR

This paper investigates how emotional framing in questions affects large language models' quantitative reasoning, revealing that emotional language degrades accuracy and neutralization can mitigate this effect.

Contribution

The authors develop a controlled emotion translation framework and create Temper-5400, a benchmark to evaluate emotional impact on reasoning models.

Findings

Emotional framing reduces reasoning accuracy by 2-10 percentage points.

Neutralizing emotional variants recovers most of the lost performance.

Non-emotional paraphrases do not cause accuracy degradation.

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.