REFLECT: Role-Aware Detection of Sentiment, Emotion, Politeness, and Formality
REFLECT puts semantic roles inside transformer attention to test whether dialogue roles help detect communicative meaning.
The paper maps sentiment, emotion, politeness, and formality to Semantic Role Labeling patterns, then uses role-aware attention rather than treating roles as an added feature. It reports statistically reliable gains over BERT-based baselines for emotion, politeness, and formality across four benchmark datasets. Sentiment gains were marginal and not significant, which the authors read as evidence that sentiment depends more on lexical choice than participant-role structure. The authors also report improved performance and traceability on real-world mental-health data. ArXiv · AI/CL/LG's note
The paper maps sentiment, emotion, politeness, and formality to Semantic Role Labeling patterns, then uses role-aware attention rather than treating roles as an added feature. It reports statistically reliable gains over BERT-based baselines for emotion, politeness, and formality across four benchmark datasets. Sentiment gains were marginal and not significant, which the authors read as evidence that sentiment depends more on lexical choice than participant-role structure. The authors also report improved performance and traceability on real-world mental-health data. ArXiv · AI/CL/LG's note
score 4