Reproduce “EmoLASP: Emotion Recognition with Language Models and Answer Set Programming” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Thao Le, Michael Thielscher

EmoLASP addresses conversational emotion recognition by asking a language model to propose candidate valence, arousal, and dominance scores, then using a rule-based reasoner to select among them based on recent predicted emotions. On text-only IEMOCAP, llama3.3:70b achieved MAE of 0.81 without dialogue history in its prompt, versus 0.91 with full history and no reasoner; this is one benchmark comparison.

Key insights

  • Declarative reasoning, specifically Answer Set Programming, can substitute explicit dialogue history for prompt-only LLMs, significantly lowering input token costs while maintaining or improving prediction accuracy.
  • Combining language models with symbolic reasoning offers a method to enforce logical consistency and contextual continuity in dimensional emotion predictions, mitigating the inherent instability of purely neural approaches.
  • Language models, even when augmented with reasoning, continue to struggle with accurately predicting extreme emotional states for subtle, short, or metaphorical utterances, often defaulting to mid-range VAD scores.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.