Research map / Large language models / Language, safety and interpretability
MultiTalk: A Highly-Branching Dialog Testbed for Diverse Conversations
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Fine-tune GPT-2 M on MultiTalk conversation paths with loss on the target utterance | GPT-2 M · MultiTalk diverse generation test | PPL: 41.95 -> 13.21 | Helped |
| Fine-tune GPT-2 M on MultiTalk conversation paths | GPT-2 M · MultiTalk diverse generation test | Matching BLEU-4: 0.14 -> 0.28 | Helped |
| Fine-tune GPT-2 M on MultiTalk conversation paths | GPT-2 M · MultiTalk diverse generation test | Diversity: 3.78 -> 3.93 | No clear effect |
| Condition generation on the requested emotion | Emotion-Aware Model · Emotion-conditioned generation test | Emotion accuracy (average): 0.14 -> 0.51 | Helped |
| Train the Emotion-Aware Model on an emotion-oversampled dataset (ablation) | Emotion-Aware Model · Emotion-conditioned generation test | Emotion accuracy (average): 0.51 -> 0.54 | Helped |
| Train the Emotion-Aware Model on an emotion-oversampled dataset (ablation) | Emotion-Aware Model · Emotion-conditioned generation test | Eval PPL: 7.53 -> 8.33 | Hurt |
| Train the Lookahead Model to target the partner's next-response emotion | GPT-2 M · Theory of mind generation test | Eval PPL: 7.11 -> 6.62 | Helped |
| Train the Lookahead Model to target the partner's next-response emotion | GPT-2 M · Theory of mind generation test | Emotion accuracy (average): 0.14 -> 0.15 | No clear effect |
| Condition the Emotion-Aware Model directly on the target emotion e | Emotion-Aware Model · Theory of mind generation test | Emotion accuracy (average): 0.14 -> 0.16 | Helped |
| Train the Lookahead Model on the emotion-oversampled dataset (ablation) | Lookahead Model · Theory of mind generation test | Eval PPL: 6.62 -> 7.30 | Hurt |
Closest papers by meaning
- DialectLLM: A Dialect-Aware Dialog[ue] Generation Framework Beyond Standard American English (NeurIPS 2026 · similarity 0.88)
- TA&AT: Enhancing Task-Oriented Dialog with Turn-Level Auxiliary Tasks and Action-Tree Based Scheduled Sampling (AAAI 2024 · similarity 0.88)
- StateTree: Enhancing Long-term Dialogue Reasoning via Reinforcement Learning (NeurIPS 2026 · similarity 0.87)
- Improving Open-Domain Dialogue Response Generation with Multi-Source Multilingual Commonsense Knowledge (AAAI 2024 · similarity 0.86)
- Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond) (NeurIPS 2025 · similarity 0.86)
- Learning to Clarify: Multi-turn Conversations with Action-Based Contrastive Self-Training (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.