Research map / Large language models / Language, safety and interpretability
The Silent Thought: Modeling Internal Cognition in Full-Duplex Spoken Dialogue Models via Latent Reasoning
Donghang Wu, Tianyu Zhang, Yuxin Li, Hexin Liu, Chen Chen, Eng Siong Chng, Yoshua Bengio
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add latent reasoning during the listening phase (ablation) | FLAIR · MMSU | accuracy: 50.2 -> 56.2 | Improved |
| add latent reasoning during the listening phase (ablation) | FLAIR · Llama Questions | accuracy: 73.0 -> 78.0 | Improved |
| add latent reasoning during the listening phase (ablation) | FLAIR · OpenbookQA | accuracy: 72.9 -> 74.2 | Improved |
| add latent reasoning during the listening phase (ablation) | FLAIR · SDQA | accuracy: 54.4 -> 56.2 | Improved |
| add latent reasoning during the listening phase (ablation) | FLAIR · WebQuestions | accuracy: 41.7 -> 43.0 | Improved |
| add latent reasoning during the listening phase (ablation) | FLAIR · TriviaQA | accuracy: 53.8 -> 51.2 | Hurt |
| add latent reasoning during the listening phase (ablation) | FLAIR · CommonEval | GPT score: 3.54 -> 3.65 | Improved |
| add latent reasoning during the listening phase (ablation) | FLAIR · AlpacaEval | GPT score: 3.80 -> 3.85 | Improved |
| add latent reasoning during the listening phase (ablation) | FLAIR · Impatient turn-taking latency | latency (s): 0.33 -> 0.39 | Hurt |
| add latent reasoning during the listening phase (ablation) | FLAIR · Impatient barge-in latency | latency (s): 0.49 -> 0.46 | No clear effect |
| add latent reasoning during the listening phase (ablation) | FLAIR · Full-Duplex-Bench barge-in | GPT-4o score: 4.08 -> 4.22 | Improved |
| add latent reasoning during the listening phase (ablation) | FLAIR · Full-Duplex-Bench barge-in | TOR: 89.0 -> 92.0 | Improved |
Closest papers by meaning
- STITCH: Simultaneous Thinking and Talking with Chunked Reasoning for Spoken Language Models (ICLR 2026 · similarity 0.90)
- Can Speech LLMs Think while Listening? (ICLR 2026 · similarity 0.89)
- Decoupling Conversational Dynamics in Full-Duplex Spoken Models through Reinforcement Learning (NeurIPS 2026 · similarity 0.89)
- Memory-T1: Reinforcement Learning for Temporal Reasoning in Multi-session Agents (ICLR 2026 · similarity 0.87)
- StateTree: Enhancing Long-term Dialogue Reasoning via Reinforcement Learning (NeurIPS 2026 · similarity 0.86)
- Think in Latent, Explain in Language: Self-Explainable Latent Reasoning (ICML 2026 · 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.
