MASCOT (EMNLP 2026) trains multi-agent role-played dialogue, hitting 80.0 persona consistency
jiqizhixin · x · 2026-11-13
Researchers from Georgia Tech and MIT present MASCOT, a multi-agent socio-collaborative dialogue framework accepted to EMNLP 2026 Main. Key points:
- Phase I, persona-aware behavioral alignment: model feedback plus a learned reward model trains speaker agents to maintain distinct behavioral roles across emotional support quality, empathy consistency, relevance, and social contribution.
- Phase II, collaborative dialogue optimization: speakers are frozen and a Director policy learns who should speak next and which conversational strategy to follow.
On positive-emotion contexts from Empathetic Dialogues, reported persona consistency reaches 80.0 vs 65.9 for the strongest baseline — a 14.1-point gap on a model-judged scale, explicitly not a clinical outcome. Evaluation also covers meeting discussions and out-of-domain emotional-support settings.
More from Research
- Incremental research is now riskier than going all in on big ideas — fkasummer · 2026-11-13
- MIT & Meta paper: ReHarness revises agent memory notes with later outcomes, beats baselines — rohanpaul_ai · 2026-11-13
- Stanford's CS312 'Deep Learning Alchemy' opens quizzes and materials to the public — stanfordnlp · 2026-11-13
- Blogger Works Through Stanford CS31: Residuals Matter More Than Normalization — stanfordnlp · 2026-11-13
- Stanford Course Shares Data: AI Users Score Higher on Homework, Worse on Exams — stanfordnlp · 2026-11-13
- Cartwheel Claims Breakthrough in Hand-Object Kinematics Capture from Any Video — andrew_n_carr · 2026-11-13