Natural Language Does Not Emerge ‘Naturally’ in Multi-Agent Dialog
Open accessThe DOI matches its Crossref record; 23 of the 23 references it cites were found in a registry or an index.
Abstract
A number of recent works have proposed techniques for end-to-end learning of communication protocols among cooperative multi-agent populations, and have simultaneously found the emergence of grounded human-interpretable language in the protocols developed by the agents, learned without any human supervision!In this paper, using a Task & Talk reference game between two agents as a testbed, we present a sequence of 'negative' results culminating in a 'positive' one -showing that while most agent-invented languages are effective (i.e.achieve near-perfect task rewards), they are decidedly not interpretable or compositional.In essence, we find that natural language does not emerge 'naturally', despite the semblance of ease of natural-language-emergence that one may gather from recent literature.We discuss how it is possible to coax the invented languages to become more and more human-like and compositional by increasing restrictions on how two agents may communicate.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Despite the flexibility of such methods, traditional EC agents often exhibit undesirable properties, such as being overly-complex (Chaabouni et al., 2019), slow to converge (Eccles et al., 2019;Lin et al., 2021), or unable to generalize to novel inputs (Kottur et al., 2017).”
From Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity (2025) · cites this paper as Kottur et al., 2017
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Based in part by a desired to mirror the complexity-limited discrete nature of words, some recent EC works seek to induce complexity-limited discrete communication among agents (Kottur et al., 2017).”
From Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity (2025) · cites this paper as Kottur et al., 2017
The cited source says what the claim attributes to it.
In the citing paper
“In traditional EC work, agents are trained in cooperative multi-agent environments to maximize a utility function (sometimes called the reward) (Havrylov & Titov, 2017;Kottur et al., 2017;Lazaridou et al., 2018;Lowe et al., 2017;Wang et al., 2020).”
From Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity (2025) · cites this paper as Kottur et al., 2017
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Adam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946
Strategic information transmission
Crawford, V. P., & Sobel, J. (1982) · Econometrica · doi:10.2307/1913390 · Cited by 4,293
Convention: A Philosophical Study.
Llewelyn, J. E., & Lewis, D. K. (1970) · The Philosophical Quarterly · doi:10.2307/2218418 · Cited by 3,121
Building End-To-End Dialogue Systems Using Generative Hierarchical Neural Network Models
Serban, I., Sordoni, A., Bengio, Y., Courville, A., & Pineau, J. (2016) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v30i1.9883 · Cited by 1,615
Learning to Communicate with Deep Multi-Agent Reinforcement Learning
Foerster, J. N., Assael, Y. M., de Freitas, N., & Whiteson, S. (2016) · arXiv · doi:10.48550/arxiv.1605.06676 · Cited by 1,217 · cited 1×
“, 2014) and is witnessing a recent resurgence (Wang et al., 2016; Foerster et al., 2016; Sukhbaatar et al., 2016; Jorge et al., 2016; Lazaridou et al., 2017; Havrylov and Titov, 2017; Mordatch and Abbeel, 2017; Das et al., 2017b) – is situated language learning.”
BackgroundLearning End-to-End Goal-Oriented DialogPreprint
Bordes, A., Boureau, Y.-L., & Weston, J. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1605.07683 · Cited by 774
The Ubuntu Dialogue Corpus: A Large Dataset for Research in Unstructured Multi-Turn Dialogue Systems
Lowe, R., Pow, N., Serban, I., & Pineau, J. (2015) · Proceedings of the 16th Annual Meeting of the Special Interest Group on Discourse and Dialogue · doi:10.18653/v1/w15-4640 · Cited by 748
Procedures as a representation for data in a computer program for understanding natural language
Winograd, T. A. (1971) · DSpace@MIT (Massachusetts Institute of Technology) · Cited by 520
Das, A., Kottur, S., Gupta, K., Singh, A., Jain, D., Moura, J. M. F., Parikh, D., & Batra, D. (2017) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · Cited by 453
Self-citationEmergence of grounded compositional language in multi-agent populations
Mordatch, I., & Abbeel, P. (2018) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v32i1.11492 · Cited by 440
GuessWhat?! Visual Object Discovery through Multi-modal Dialogue
De Vries, H., Strub, F., Chandar, S., Pietquin, O., Larochelle, H., & Courville, A. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.475 · Cited by 369
The evolution of syntactic communication
Nowak, M. A., Plotkin, J. B., & Jansen, V. A. A. (2000) · Nature · doi:10.1038/35006635 · Cited by 358
Learning Cooperative Visual Dialog Agents with Deep Reinforcement Learning
Das, A., Kottur, S., Moura, J. M. F., Lee, S., & Batra, D. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.321 · Cited by 356
Self-citationIterated learning and the evolution of language
Kirby, S., Griffiths, T., & Smith, K. (2014) · Current Opinion in Neurobiology · doi:10.1016/j.conb.2014.07.014 · Cited by 313
Learning Multiagent Communication with Backpropagation
Sukhbaatar, S., Szlam, A. D., & Fergus, R. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1605.07736 · Cited by 261
Multi-Agent Cooperation and the Emergence of (Natural) Language
Lazaridou, A., Peysakhovich, A., & Baroni, M. (2016) · arXiv · doi:10.48550/arxiv.1612.07182 · Cited by 231
Image-Grounded Conversations: Multimodal Context for Natural Question and Response GenerationPreprint
Mostafazadeh, N., Brockett, C., Dolan, B., Galley, M., Gao, J., Spithourakis, G., & Vanderwende, L. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1701.08251 · Cited by 209
Emergence of Language with Multi-agent Games: Learning to Communicate with Sequences of Symbols
Havrylov, S., & Titov, I. (2017) · arXiv · doi:10.48550/arxiv.1705.11192 · Cited by 195
Emergence of Grounded Compositional Language in Multi-Agent Populations
Mordatch, I., & Abbeel, P. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1703.04908 · Cited by 190
Learning Language Games through Interaction
Wang, S. I., Liang, P., & Manning, C. D. (2016) · Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p16-1224 · Cited by 132
An ISU dialogue system exhibiting reinforcement learning of dialogue policies
Lemon, O., Georgila, K., Henderson, J., & Stuttle, M. N. (2006) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.3115/1608974.1608986 · Cited by 95
Dialog-based Language LearningPreprint
Weston, J. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1604.06045 · Cited by 94
Learning Cooperative Visual Dialog Agents with Deep Reinforcement LearningPreprint
Das, A., Kottur, S., Moura, J. M. F., et al. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1703.06585 · Cited by 88
Self-citation
Research works printed but not linked: 4
- Vincent Crawford and Joel Sobel. 1982. Strategic in- formation transmission. Econometrica 50(6):1431-51.
- Emilio Jorge, Mikael Kågebäck, and Emil Gustavsson. 2016. Learning to play guess who? and inventing a grounded language as a consequence. In NIPS work- shop on deep reinforcement learning.
- Iulian Vlad Serban, Alessandro Sordoni, Ryan Lowe, Laurent Charlin, Joelle Pineau, Aaron Courville, and Yoshua Bengio. 2016b. A Hierarchical Latent Variable Encoder-Decoder Model for Generating Dialogues. arXiv preprint arXiv:1605.06069 . · arXiv
- Ronald J Williams. 1992. Simple statistical gradient- following algorithms for connectionist reinforcement learning. Machine learning 8(3-4):229-256.
Cited by
Deep Reinforcement Learning: An Overview
Li, Y. (2017) · arXiv · doi:10.48550/arxiv.1701.07274 · Cited by 762
“-to-end optimization with deep RL for goal-driven and visually grounded dialogue systems for GuessWhat?! game.Das et al.(2017) proposed to learn cooperative Visual Dialog agents with deep RL. See alsoKottur et al. (2017). SeePasunuru and Bansal(2017) for video captioning. 5.5 NEURAL ARCHITECTURE DESIGN Neural networks…”
BackgroundInfluentialLearning Cooperative Visual Dialog Agents with Deep Reinforcement Learning
Das, A., Kottur, S., Moura, J. M. F., Lee, S., & Batra, D. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.321 · Cited by 356
Self-citationLi, C. (2019) · Reinforcement Learning for Cyber-Physical Systems · doi:10.1201/9781351006620-6 · Cited by 188
Deep multiagent reinforcement learning: challenges and directions
Wong, A., Bäck, T., Kononova, A. V., & Plaat, A. (2023) · Artificial Intelligence Review · doi:10.1007/s10462-022-10299-x · Cited by 197
Learning when to Communicate at Scale in Multiagent Cooperative and Competitive TasksPreprint
Singh, A., Jain, T., & Sukhbaatar, S. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1812.09755 · Cited by 291
Metadata
- Authors
- Kottur, S., Moura, J., Lee, S., & Batra, D.
- Affiliations
- Carnegie Mellon University (US)
- Georgia Institute of Technology (US)
- Venue
- Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing
- Published
- 2017
- Pages
- 2962-2967
- OpenAlex
- W2730230212
- Cited by
- 164
- References
- 20
- Citations by year
- 2026: 4 · 2025: 6 · 2024: 14 · 2023: 13 · 2022: 12 · 2021: 32 · 2020: 43 · 2019: 25 · 2018: 8 · 2017: 7
- FWCI
- 356.92
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Conference on Empirical Methods in Natural Language Processing (EMNLP) (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Social Sciences › Cultural Studies › Language and cultural evolution
- Physical Sciences › Computer Science › Artificial Intelligence › Multi-Agent Systems and Negotiation
- Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems
- Keywords
- multi-agent communication, interpretability, compositional language
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.