The DOI matches its Crossref record; 64 of the 68 references it cites were found in a registry or an index; 42 of the 65 sources its claims cite were checked for support.
Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity
Open accessAbstract
Two prominent, yet contrasting, theoretical views are available to characterize the underlying drivers of language evolution: on the one hand, task-specific utility maximization; on the other hand, task-agnostic communicative efficiency. The latter has recently been grounded in an information-theoretic tradeoff between communicative complexity and informativeness, known as the Information Bottleneck (IB) principle. Here, we integrate these two views and propose an information-constrained emergent communication framework that trades off utility, informativeness, and complexity. To train agents within our framework, we develop a method, called Vector-Quantized Variational Information Bottleneck (VQ-VIB), that allows agents to interact using information-constrained discrete communication embedded in a continuous vector space. We test this approach in three domains and show that pressure for informativeness facilitates faster learning and better generalization to novel domains. At the same time, limiting complexity yields better alignment with actual human languages. Lastly, we find that VQ-VIB outperforms previously proposed emergent communication methods; we posit that this is due to the semantically-meaningful communication embedding space that VQ-VIB affords. Overall, our work demonstrates the role of cognitively-motivated optimality principles in inducing aspects of human-like communication among artificial agents.
Overview
CoChat Summary
TL;DR
The paper integrates utility maximization with the Information Bottleneck tradeoff between informativeness and complexity into an information-constrained emergent communication framework, trained via a new method called VQ-VIB. Across color naming, open-domain image naming, and 2D navigation, informativeness pressure improved learning and generalization, while complexity limits produced more human-like communication.
Key findings
- In the color reference game, VQ-VIB_N was near the IB-optimal bound and more human-like than utility-only baselines: human-agent gNID 0.151 (0.00) vs. Onehot 0.42 (0.16), VQ-VAE 0.38 (0.15), and IB-optimal 0.18 (0.10); efficiency loss was 0.024 (0.00) vs. 0.06 (0.01) and 0.08 (0.01).
- In ManyNames open-domain communication, increasing informativeness improved out-of-distribution generalization; for C = 32, VQ-VIB_N with n = 4 reached up to roughly 60% OOD accuracy, while Onehot and Proto peaked at approximately 17% and 13%.
- Limiting complexity improved alignment with human language: relative representation alignment with GloVe peaked when EC informativeness matched English response informativeness, and increasing n lowered distortion but worsened alignment.
- In 2D navigation, increasing λ_I led to faster convergence and higher reward, increasing λ_C led to less complex communication and coarser spatial discretization, and VQ-VIB achieved greater utility for the same complexity than other architectures.
Methods
- The authors propose the ICEC objective trading off utility, informativeness, and complexity, and implement VQ-VIB (VQ-VIB_N and VQ-VIB_C) using stochastic embedding/quantization, learnable codebooks, variational complexity bounds, and an entropy tie-breaking loss.
- Color experiment: speaker observed noisy target colors and listener selected target vs. distractor; λ_U = 1.0, λ_I = 1.0, then λ_C was increased after 40 epochs for 50 epochs; results compared to WCS languages and IB-optimal systems over five runs.
- ManyNames experiment: 25,000 images with roughly 36 English responses each, 442 topnames, 20% of topnames used for training and semantically distinct test images held out; ResNet 512-d features plus VAE noise; C = 2 during training, evaluated at C = 16 and 32; λ_U = 1, λ_C = 0.01, varied λ_I. 2D navigation used MADDPG with speaker observing target location, listener observing only its own location, reward as negative Euclidean distance, and varied λ_I or annealed λ_C.
Limitations
- Stated in the paper: it is unclear how human speakers and listeners could estimate informativeness without direct access to others’ mental states, motivating future Theory of Mind work; agents were trained with a single utility function, leaving multi-task utility as future work; and rigorously defining how “meaningful” or “semantic” an EC space is remains challenging.
- Evident from the text: the experiments are simulated reference games and a simplified 2D navigation world, not direct tests of human language evolution; alignment with human language is measured indirectly through WCS color data, GloVe embeddings, and English naming responses.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures






















Verification
Claims the paper makes
27 Valid · 3 Not valid · 16 Not checked · 155 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“While some studies in the machine-learning emergent communication literature have focused on other reinforcement learning settings (Foerster et al., 2016;Mordatch & Abbeel, 2018, e.g.,), they have not made direct contact with general, cognitively-motivate principles that are believed to underlie human language, such as IB.”
§ Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity > OPEN MIND: Discoveries in Cognitive Science
What the sources say
- Supportedcites Foerster et al., 2016 source
The Abstract (p. 1) identifies this as a study of agents learning communication protocols through deep multi-agent reinforcement learning, supporting its use as an example of another RL setting. Section 6.4 (p. 8) relates channel noise to the discreteness of human language, but the paper does not make a direct connection to IB or a comparable cognitively motivated principle; discussing a question about language is not the same as applying such a principle. (LLM, conf=0.85)
- Supportedcites Mordatch & Abbeel, 2018 source
The citation identifies Mordatch and Abbeel (2018) as an example of an emergent-communication study in a reinforcement-learning setting. The Related Work section (p. 2) places their work alongside reinforcement-learning approaches to learning communication protocols, and the description of their multi-agent environment (p. 1) specifies a shared reward. The subsequent assessment about contact with principles such as IB is the citing authors’ own characterization, not the point this citation is offered to identify. (LLM, conf=0.91)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“However, efficient communication approaches have so far focused on closed domains (Kemp et al., 2018), and it has not been clear how to extend this approach to open domains.”
§ Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity > OPEN MIND: Discoveries in Cognitive Science
What the sources say
- Supportedcites Kemp et al., 2018 source
In Extensions and Future Directions (p. 15), the paper says the work reviewed has mostly considered one semantic domain at a time and presents extending efficiency analysis to larger portions of the lexicon as future work. This supports the citing sentence’s characterization of the approaches as focused on closed domains. (LLM, conf=0.90)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Lastly, Lin et al. (2021) indirectly explores the role of informativeness in guiding emergent communication, by using a reconstruction loss to generate communication encodings.”
§ Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity > Emergent Communication
What the sources say
- Supportedcites Lin et al. (2021) source
Our copy supports both halves of the paraphrase. The 'Communication Autoencoder' section states the decoder reconstructs the input state from the communication message and that the auxiliary objective is a reconstruction loss, jointly optimized with the listener's policy-gradient loss — i.e. communication encodings are generated by a reconstruction loss. The 'Positive Signaling and Positive Listening' and 'Communication Embedding' sections then examine what that forces on the messages (agents 'are intrinsically optimized for positive signaling', 'can effectively transmit information to other agents'), and Section 5.4 is literally titled 'The Role of Autoencoding', so the paper does probe the role of information-bearing messages in emergent communication without framing it as 'informativeness' as such — consistent with the citing sentence's 'indirectly explores'. The paper's own construct name (grounding/positive signaling rather than informativeness) is a wording difference, not a content mismatch. (LLM, conf=0.82)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“This theoretical IB tradeoff, where β represents the importance afforded to informativeness relative to complexity, yields two important insights: 1) at least in simple systems, one can derive optimal communication schemes for varying β, and 2) one can compare human naming systems to IB-optimal systems by measuring their informativeness and complexity values (where IB-optimal systems achieve”
§ Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity > The Information Bottleneck for Semantic Systems
What the sources say
- Supportedcites Zaslavsky et al. (2018) sourceExperimental — accuracy not yet measured
In the information-plane discussion (indexed under Abstract; page unavailable), the paper says each language has measurable complexity and accuracy values that can be plotted against the curve defined by IB-optimal languages. It identifies accuracy with informativeness, supporting the proposed comparison for color-naming systems. (LLM, conf=0.94)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“First, tradeoffs between informativeness and complexity dictate important aspects of task-agnostic compression; such compression has been widely used in studying human naming systems (Mollica et al., 2021;Zaslavsky et al., 2018Zaslavsky et al., , 2019Zaslavsky et al., , 2021)).”
§ Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity > INFORMATION-CONSTRAINED EMERGENT COMMUNICATION
What the sources say
- Partly supportedcites Mollica et al., 2021 sourceExperimental — accuracy not yet measured
The theoretical-framework passage describes a trade-off between informativeness and complexity in efficient data compression, and the Discussion relates the results to prior work on color and kin naming. The paper studies grammatical markers in a communicative setting, however; it does not establish that these trade-offs dictate aspects of task-agnostic compression. (LLM, conf=0.78)
- Partly supportedcites Zaslavsky et al., 2018 sourceExperimental — accuracy not yet measured
The Abstract says that color-naming systems compress meanings by trading off lexicon complexity against accuracy, and the Discussion says this trade-off explains much cross-language variation. But the paper analyzes compression for a specified communication scenario involving a speaker and listener; it does not establish the claim about task-agnostic compression. (LLM, conf=0.82)
- Partly supportedcites Zaslavsky et al., 2019 sourceExperimental — accuracy not yet measured
The Theoretical framework and predictions (p. 2) describes efficient naming as a tradeoff between encoding complexity and communicative accuracy, and the Results (p. 4) apply that tradeoff to container naming. The paper does not establish the broader claim about task-agnostic compression: its account is framed around communicating and reconstructing meanings. (LLM, conf=0.83)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
107 more claims · 5 with their cited sources checked
Create a free account to see all 201 claims and every source verdict.
- Probably not a substantial claim. 47
- It may state two separate claims as one. 51
- The quote only partly states this claim. 2
- Borderline: one of the tests for a substantial claim is unclear. 1
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity > Experiment Setup
“Using a pre-trained EC speaker, we fit a linear mapping from EC vectors for images to GloVe embeddings of an English label for the image (drawn from the responses associated with each image in the dataset).”
§ Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity > Functional Alignment Analysis
“In some of our ManyNames experiments, we trained a linear model to map from GloVe embeddings to EC vectors.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity > Experiment Setup
“Both agents achieved reward equal to the negative Euclidean distance from the listener to the target, so utility was maximized if the listener moved straight to the target.”
§ Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity > Results
“Each curve in Figure 12 represents the team utility (calculated as the average distance from the listener to the goal) over the course of training VQ-VIB N and VQ-VIB C teams with different λ I values, averaged over five trials.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity > Vector-Quantized Variational Information Bottleneck
“VQ-VIB C uses a stochastic Q ψ , however, by sampling a discrete representation according to probabilities based on the negative Euclidean distance from z to each ζ i .”
§ Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity > OPEN MIND: Discoveries in Cognitive Science
“For VQ-VIB C , we could directly measure and penalize the entropy of the categorical distribution, as it was calculated while sampling tokens: ℙ(ζ i |m) ∝ e ‖ζ i -z(m)‖ 2 .”
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474
Glove: Global Vectors for Word Representation
Pennington, J., Socher, R., & Manning, C. (2014) · Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.3115/v1/d14-1162 · Cited by 34,071 · cited 2×
“…deep-learning method we develop, which enables artificial agents to communicate using discrete symbols that are embedded in a continuous vector space, much like word embeddings (Pennington et al., 2014), while being guided by pressure to optimally trade off utility, informativeness, and complexity.”
Reinforcement Learning: An Introduction
Sutton, R. S., & Barto, A. G. (1998) · IEEE Transactions on Neural Networks · doi:10.1109/tnn.1998.712192 · Cited by 27,225
Advances in Neural Information Processing Systems 19
(2007) · The MIT Press eBooks · doi:10.52202/085713 · Cited by 15,539
Advances in Neural Information Processing Systems 29
Teymur, O., Zygalakis, K., & Calderhead, B. (2016) · Cited by 13,887
Proceedings of the 24th international conference on Machine learning
(2007) · doi:10.1145/1273496 · Cited by 11,630
2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
(2017) · doi:10.1109/cvpr35066.2017 · Cited by 7,141
Proceedings of the 21st International Conference on Neural Information Processing Systems
Koller, D., Schuurmans, D., Bengio, Y., & Bottou, L. (2008) · Cited by 4,461
Colman (2022) · doi:10.1073/pnas · Cited by 4,370
Simple Statistical Gradient-Following Algorithms for Connectionist Reinforcement Learning
Williams, R. J. (1992) · Machine Learning · doi:10.1007/bf00992696 · Cited by 3,416 · cited 4×
“Results from our agents, trained with REINFORCE, are depicted in Figure D.1.”
Influential
Research works printed but not linked: 3
- Chaabouni, 2021, Emergent communication at scale
- Kay, 2009, The World Color Survey
- Malloy, 2022, Resource-rational cognitive modelling: An information-theoretic approach
Cited by
Iterated language learning is shaped by a drive for optimizing lossy compressionPreprint
Imel, N., Culbertson, J., Kirby, S., & Zaslavsky, N. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.05.23.655237 · Cited by 5
“This bias might be weaker without an explicit need to communicate, and more generally, can be beneficial for social agents (Tucker et al., 2022, 2025).”
Self-citationAssessing Pressures Shaping Natural Language Lexica
Bruneau‐‐Bongard, J., Chemla, E., & Brochhagen, T. (2025) · Cognitive Science · doi:10.1111/cogs.70145 · Cited by 3
Metadata
- Authors
- Tucker, M., Shah, J., Levy, R., & Zaslavsky, N.
- Affiliations
- Anthropic (United States) (US)
- Massachusetts Institute of Technology (US)
- Department of Brain & Cognitive Sciences, Massachusetts Institute of Technology
- New York University (US)
- Corresponding author
- Tucker, Mycal
- Venue
- Open Mind
- Publisher
- MIT Press
- Published
- 2025
- Volume
- 9
- Pages
- 418-451
- ISSN
- 2470-2986 (ISSN-L)
- PMC
- PMC11984795
- OpenAlex
- W4409248873
- Cited by
- 2
- References
- 59
- Citations by year
- 2025: 2
- FWCI
- 12.87
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Open Mind (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion, cc-by)
- Funders
- Simons Center for the Social Brain, Massachusetts Institute of Technology
- Quest for Intelligence, Massachusetts Institute of Technology
- Topics
- Social Sciences › Cultural Studies › Language and cultural evolution
- Physical Sciences › Computer Science › Artificial Intelligence › Evolutionary Algorithms and Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Reinforcement Learning in Robotics
- Keywords
- emergent communication, language evolution, information bottleneck theory, informativeness, communicative complexity, utility maximization, variational information bottleneck, vector quantization, artificial agents, principle of optimality
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-01, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-01 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity. https://app.cochat.ai/papers/10.1162/opmi_a_00188
- The paper
- Tucker, M., Shah, J., Levy, R., & Zaslavsky, N. (2025). Towards Human-Like Emergent Communication via Utility, Informativeness, and Complexity. Open Mind. https://doi.org/10.1162/opmi_a_00188
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.