TEMPER: Testing Emotional Perturbation in Quantitative Reasoning

ArXi:2604.07801v1 Announce Type: new 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.