Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Incorporating Pragmatic Reasoning Communication into Emergent Language
Yipeng Kang, Tonghan Wang, Gerard de Melo
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 two-sided IBR pragmatic reasoning | Emergent language agents · Referential game, Challenge | accuracy: 53.0$\pm$2.6 -> 80.6$\pm$2.8 | Improved |
| Use GameTable-sequential equilibrium selection | GameTable pragmatic model · Referential game, Challenge | accuracy: 74.6$\pm$2.0 -> 94.0$\pm$0.6 | Improved |
| Use probabilistic RSA instead of deterministic IBR | Two-sided pragmatic reasoning · Referential game, Challenge | accuracy: 80.6$\pm$2.8 -> 55.9$\pm$1.9 | Hurt |
| Use two-round IBR instead of converged IBR (ablation) | IBR pragmatic model · Referential game, Overall | accuracy: 95.5$\pm$0.6 -> 95.6$\pm$0.5 | No clear effect |
| Use two-round RSA instead of converged RSA (ablation) | RSA pragmatic model · Referential game, Overall | accuracy: 91.9$\pm$0.9 -> 93.1$\pm$0.8 | Hurt |
| Give the speaker no weight for consistency with its prior (SampleL λ=0) (ablation) | SampleL · Referential game, Overall | accuracy: 90.1$\pm$0.9 -> 90.1$\pm$1.1 | No clear effect |
| Use one-sided SampleL rather than baseline decoding | Emergent language agents · Referential game, Challenge | accuracy: 53.0$\pm$2.6 -> 54.8$\pm$2.9 | No clear effect |
| Use GameTable instead of IBR | Two-sided pragmatic reasoning · Referential game, Challenge | accuracy: 80.6$\pm$2.8 -> 74.6$\pm$2.0 | Hurt |
| Use a reconstructed virtual opponent model rather than an exact copy | IBR · Referential game with virtual opponent | accuracy: 80.6$\pm$2.8 -> 68.6$\pm$1.6 | Hurt |
| Use a reconstructed virtual opponent model rather than an exact copy | GameTable · Referential game with virtual opponent | accuracy: 74.6$\pm$2.0 -> 58.1$\pm$2.2 | Hurt |
| Use a reconstructed virtual opponent model rather than an exact copy | GameTable-sequential · Referential game with virtual opponent | accuracy: 94.0$\pm$0.6 -> 69.9$\pm$1.8 | Hurt |
| Add pragmatic reasoning to communication | StarCraft II ally agents · 1o2r_vs_4r | battle win rate under message dropping | Improved |
Closest papers by meaning
- Bridging semantics and pragmatics in information-theoretic emergent communication (NeurIPS 2024 · similarity 0.90)
- Improving Cooperation in Language Games with Bayesian Inference and the Cognitive Hierarchy (AAAI 2025 · similarity 0.89)
- Multi-agent KTO: Enhancing Strategic Interactions of Large Language Model in Language Game (NeurIPS 2025 · similarity 0.87)
- Lewis's Signaling Game as beta-VAE For Natural Word Lengths and Segments (ICLR 2024 · similarity 0.87)
- Semantics and Spatiality of Emergent Communication (NeurIPS 2024 · similarity 0.87)
- Narrowing the Collaboration Gap, Probably (NeurIPS 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.
