mlsanityData to 14 Oct 2026

CtD: Composition through Decomposition in Emergent Communication

arXiv 2601.10169 ↗First posted 15 Jan 2026cs.AIcited by 1

Not on the map. Why?

Abstract (arXiv)

Compositionality is a cognitive mechanism that allows humans to systematically combine known concepts in novel ways. This study demonstrates how artificial neural agents acquire and utilize compositional generalization to describe previously unseen images. Our method, termed "Composition through Decomposition", involves two sequential training steps. In the 'Decompose' step, the agents learn to decompose an image into basic concepts using a codebook acquired during interaction in a multi-target coordination game. Subsequently, in the 'Compose' step, the agents employ this codebook to describe novel images by composing basic concepts into complex phrases. Remarkably, we observe cases where generalization in the `Compose' step is achieved zero-shot, without the need for additional training.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 51 of its 86 references that are in the map. First posted on arXiv 15 Jan 2026.

Builds on

51 of 86 references.

References · 51

  1. 2017Attention Is All You Needcited by 60,361
  2. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  3. 2014Microsoft COCO: Common Objects in Contextcited by 17,608
  4. 2018Representation Learning with Contrastive Predictive Codingcited by 7,777
  5. 2017Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networkscited by 4,897
  6. 2017Neural Discrete Representation Learningcited by 4,712
  7. 2016Categorical Reparameterization with Gumbel-Softmaxcited by 2,739
  8. 2020Concept Bottleneck Modelscited by 1,032
  9. 2020Mastering Atari with Discrete World Modelscited by 1,007
  10. 2022Toy Models of Superpositioncited by 971
  11. 2016Learning to Communicate with Deep Multi-Agent Reinforcement Learningcited by 575
  12. 2019Compositionality decomposed: how do neural networks generalise?cited by 265
38 more
  1. 2021Curriculum Learning: A Surveycited by 259
  2. 2017Multi-agent Reinforcement Learning in Sequential Social Dilemmascited by 209
  3. 2018TarMAC: Targeted Multi-Agent Communicationcited by 208
  4. 2016Multi-Agent Cooperation and the Emergence of (Natural) Languagecited by 185
  5. 2020Emergent Multi-Agent Communication in the Deep Learning Eracited by 170
  6. 2017Emergence of Language with Multi-agent Games: Learning to Communicate with Sequences of Symbolscited by 125
  7. 2020Compositionality and Generalization in Emergent Languagescited by 103
  8. 2018Emergence of Linguistic Communication from Referential Games with Symbolic and Pixel Inputcited by 102
  9. 2017Natural Language Does Not Emerge 'Naturally' in Multi-Agent Dialogcited by 86
  10. 2019Measuring Compositionality in Representation Learningcited by 84
  11. 2020Compositional Languages Emerge in a Neural Iterated Learning Modelcited by 75
  12. 2016Reasoning About Pragmatics with Neural Listeners and Speakerscited by 68
  13. 2020Multi-agent Communication meets Natural Language: Synergies between Functional and Structural Language Learningcited by 66
  14. 2019Ease-of-Teaching and Language Structure from Emergent Communicationcited by 65
  15. 2018Emergent Communication through Negotiationcited by 62
  16. 2021The paradox of the compositionality of natural language: a neural machine translation case studycited by 59
  17. 2023Online Clustered Codebookcited by 55
  18. 2023Codebook Features: Sparse and Discrete Interpretability for Neural Networkscited by 48
  19. 2018How agents see things: On visual representations in an emergent language gamecited by 45
  20. 2023Break It Down: Evidence for Structural Compositionality in Neural Networkscited by 44
  21. 2019EGG: a toolkit for research on Emergence of lanGuage in Gamescited by 42
  22. 2021Emergent Communication of Generalizationscited by 42
  23. 2022Compositional Generalization in Unsupervised Compositional Representation Learning: A Study on Disentanglement and Emergent Languagecited by 42
  24. 2017ShapeWorld - A new test methodology for multimodal language understandingcited by 37
  25. 2018Compositional Obverter Communication Learning From Raw Visual Inputcited by 30
  26. 2022Emergent Communication: Generalization and Overfitting in Lewis Gamescited by 28
  27. 2021Interpretable agent communication from scratch (with a generic visual processor emerging on the side)cited by 25
  28. 2021Iterated learning for emergent systematicity in VQAcited by 21
  29. 2020A Rate-Distortion view of human pragmatic reasoningcited by 19
  30. 2018Emergence of Communication in an Interactive World with Consistent Speakerscited by 18
  31. 2021Emergent Graphical Conventions in a Visual Communication Gamecited by 16
  32. 2018Learning to Communicate Implicitly By Actionscited by 11
  33. 2020ReferentialGym: A Nomenclature and Framework for Language Emergence & Grounding in (Visual) Referential Gamescited by 11
  34. 2021Expressivity of Emergent Language is a Trade-off between Contextual Complexity and Unpredictabilitycited by 11
  35. 2022Emergent Quantized Communicationcited by 7
  36. 2024Concept-Best-Matching: Evaluating Compositionality in Emergent Communicationcited by 7
  37. 2024From Frege to chatGPT: Compositionality in language, cognition, and deep neural networkscited by 7
  38. 2022An Analysis of Discretization Methods for Communication Learning with Multi-Agent Reinforcement Learningcited by 6

50 most cited of 51 All 51 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Investigating the Development of Task-Oriented Communication in Vision-Language Modelscited by 2