Research map / Large language models / Language, safety and interpretability
Hierarchical Text Generation and Planning for Strategic Dialogue
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 |
|---|---|---|---|
| Pretrain discrete sentence representations to predict dialogue continuation and actions | Negotiation dialogue generation · Negotiation test | perplexity: 5.46 -> 5.24 | Helped |
| Pretrain discrete sentence representations to predict dialogue continuation and actions | Negotiation self-play · Self-play negotiations | score vs. RNN: 4.68 -> 6.75 | Helped |
| Add hierarchical sentence-level recurrence over messages | Negotiation dialogue generation · Negotiation test | perplexity: 5.47 -> 5.21 | Helped |
| Add rollout planning (ablation) | Full model · Self-play negotiations | score vs. no rollouts: 5.08 -> 7.81 | Helped |
| Use distinct latent intents for rollout candidates (ablation) | Full model with rollouts · Self-play negotiations | score vs. baseline rollouts: 6.57 -> 7.36 | Helped |
| Fine-tune only the intent predictor with reinforcement learning while keeping language parameters fixed (ablation) | Full model · Self-play negotiations | reward at human-like language quality | Helped |
| Apply reinforcement learning to all model parameters (ablation) | Full model · Self-play negotiations | language quality | Hurt |
| Add rollout planning (ablation) | Full model · Negotiations against humans | score vs. human: 5.42 -> 7.45 | Helped |
| Add rollout planning (ablation) | RNN · Negotiations against humans | score vs. human: 5.30 -> 6.99 | Helped |
| Fine-tune with reinforcement learning (ablation) | Full model · Negotiations against humans | score vs. human: 5.42 -> 6.26 | Helped |
| Fine-tune with reinforcement learning (ablation) | RNN · Negotiations against humans | score vs. human: 5.30 -> 6.01 | Helped |
| Use discrete latent representations rather than the direct hierarchical continuous connection (ablation) | Negotiation dialogue generation · Negotiation test | perplexity: 5.21 -> 5.24 | No clear effect |
Closest papers by meaning
- Simulation-Free Hierarchical Latent Policy Planning for Proactive Dialogues (AAAI 2025 · similarity 0.89)
- Plug-and-Play Policy Planner for Large Language Model Powered Dialogue Agents (ICLR 2024 · similarity 0.88)
- Decoupling Conversational Dynamics in Full-Duplex Spoken Models through Reinforcement Learning (NeurIPS 2026 · similarity 0.87)
- Learning to Clarify: Multi-turn Conversations with Action-Based Contrastive Self-Training (ICLR 2025 · similarity 0.87)
- Memory-T1: Reinforcement Learning for Temporal Reasoning in Multi-session Agents (ICLR 2026 · similarity 0.87)
- Controlling Large Language Model with Latent Action (ICML 2025 · similarity 0.87)
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.