Pith. sign in

REVIEW 2 major objections 5 minor 32 references

EmCom-Diffusion measures how much visual content emergent languages encode by regenerating each source image from its message and scoring the match.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-12 00:11 UTC pith:AEBAJIMX

load-bearing objection Solid new EmCom evaluation method with careful triplet tests; the generative-bottleneck claim is overstated but the baselines and experiments still make the metric useful. the 2 major comments →

arxiv 2607.03752 v1 pith:AEBAJIMX submitted 2026-07-04 cs.CV cs.AIcs.CLcs.MA

EmCom-Diffusion: Probing Visual Reflection in Emergent Languages via Image Generation

Haruumi Omoto , Tadahiro Taniguchi This is my paper
classification cs.CV cs.AIcs.CLcs.MA
keywords emergent communicationvisual reflectiontext-to-image diffusionevaluation metricsreferential gameimage reconstructionMS-COCO
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

When agents invent a private language to talk about images, how much of the actual visual scene ends up in the messages? Existing checks only use proxies—fixed concept lists, natural-language captions, distance correlations, or success at a referential game—and can miss content that is present or credit content that is not. This paper introduces EmCom-Diffusion: fine-tune a pretrained text-to-image diffusion model on pairs of images and emergent messages, then generate images from the messages alone and compare them to the originals with perceptual similarity. On MS-COCO with a referential game, the method ranks real emergent languages above random and fixed tokens across three visual encoders, and it still separates visually similar images in settings where translation, concept matching, topographic similarity, and game accuracy fail. A reader cares because it offers a direct, annotation-free way to ask how much of the physical world is reflected in a language that formed from scratch.

Core claim

Visual reflection—the recoverability of an image’s visual content from its emergent message without using the original speaker–listener pair—can be measured directly by fine-tuning a pretrained text-to-image diffusion model on (image, message) pairs and scoring the perceptual similarity between the generated reconstruction and the original image. On MS-COCO with a Referential Game, EmCom-Diffusion correctly ranks Random/Fixed < Emergent Language < natural-language captions across CLIP, DINOv2, and SigLIP, and in targeted tests it captures visual distinctions that CBM, supervised translation, TopSim, and R@1 miss or spuriously credit.

What carries the argument

EmCom-Diffusion: fine-tune a text-conditional diffusion model so it reconstructs each image from its emergent message alone, then score average perceptual similarity between reconstructions and originals on held-out pairs. The generative bottleneck is the instrument: content absent from the message should be absent from the generated image.

Load-bearing premise

The method assumes a fine-tuned image generator plus a perceptual similarity score can recover what the message encodes without inventing missing details from its own prior or discarding content the message actually carried.

What would settle it

If reconstructions from random or fixed tokens scored as high as those from trained emergent messages on held-out images, or if EmCom-Diffusion failed to prefer the visually similar candidate in the caption-controlled and edit-distance-matched triplets while the other metrics succeeded, the claim that it measures recoverable visual content would be falsified.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Visual reflection can be scored without human concept inventories or paired captions.
  • Caption translation, concept matching, distance-rank correlation, and distractor-dependent game accuracy systematically mis-estimate what emergent messages encode.
  • Generative reconstruction yields a stable information ranking (random/fixed < emergent language < natural-language captions) across multiple vision encoders.
  • The same pipeline can be run across training to track how world-reflecting structure forms in a language grown from scratch.
  • Regularities that never map to English nouns remain measurable because the comparison target is the image itself.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the method holds beyond referential games and MS-COCO, it could become a standard check whenever someone claims an emergent protocol is grounded rather than merely task-successful.
  • Tracking the score over training epochs could show whether visual reflection appears early then drifts, or accumulates—something formal compositionality metrics do not capture.
  • The larger EL-versus-random gap on a vision-only encoder hints that emergent codes may lock onto coarse object structure more readily than fine layout; token-level attribution would test that.
  • The same generative probe could be applied to other discrete visual codes to measure preserved content without human labels.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper defines visual reflection as the extent to which emergent-language messages preserve recoverable information about their source images without further appeal to the original speaker–listener pair. It proposes EmCom-Diffusion: fine-tune a pretrained text-to-image diffusion model on (image, emergent-message) pairs (Eq. 1) and score reflection as the average perceptual similarity s between the generated reconstruction and the original image (Eq. 2). Instantiated with a Referential Game on MS-COCO, Experiment 1 shows the expected ranking Random/Fixed < EL < SD-NL across CLIP-img, DINOv2 and SigLIP, plus multi-seed controls (Tables 2–3, Fig. 2). Experiment 2 constructs targeted triplets that expose known limitations of CBM, supervised translation, TopSim and R@1 and shows EmCom-Diffusion retains more signal (Tables 4–6). The authors conclude that the generative comparison captures visual content the four baselines miss or spuriously credit.

Significance. If the generative score is a faithful external decoder of visual content, EmCom-Diffusion would supply a missing, annotation-free instrument for a central open problem in emergent communication—what the language actually encodes about its visual inputs—beyond formal properties such as compositionality or task success. The work is concrete: it ships a clear formal definition (Eqs. 1–2), a reproducible pipeline (code link, Appendix A–C), multi-encoder and multi-seed controls, and carefully constructed contrastive tests against four established metrics. Even if the absolute score is partly prior-dependent, the relative ranking and the triplet results already give the community a usable generative probe that existing proxies lack.

major comments (2)
  1. §2.2 generative-bottleneck argument and Eq. (2): the claim that content absent from m is necessarily absent from ˆx (and therefore lowers s) is under-supported. Table 2 already shows that random/fixed tokens produce non-degenerate images (CLIP-img ≈0.49, non-zero Vendi/Recall), so the Stable Diffusion prior supplies substantial visual content independent of m. Fine-tuning on (x,m) pairs can further teach the model to map sparse tokens onto full MS-COCO-like scenes by exploiting dataset regularities rather than recovering only what m encodes. The multi-seed analysis (Table 3) and random/fixed baselines control only for input-independent plausibility, not for message-conditioned hallucination of unencoded attributes. Without an ablation that isolates prior contribution (e.g., frozen vs. fine-tuned generator, or controlled attribute-masking of m), the assertion that EmCom-Diffusion is a fai
  2. §3.1 / Appendix A and §3.3: the speaker uses a frozen DINOv2 ViT-B/14 backbone, and DINOv2 is also one of the three readout encoders for the EmCom-Diffusion score and the primary readout in the triplet tests of Experiment 2. Although CLIP and SigLIP are also reported in Table 2, the strongest discrimination claims (Tables 4–5) rest on the DINOv2 cosine. This shared backbone introduces a mild circularity risk: the metric may preferentially recover features already privileged by the speaker’s visual encoder rather than arbitrary visual content. A control that freezes a different backbone for the speaker (or reports the full triplet tables under CLIP/SigLIP readouts) is needed to confirm that the advantage over baselines is not backbone-specific.
minor comments (5)
  1. Limitations section: the dependence on the generator prior and the single-game/single-domain scope are acknowledged, but the text still asserts that the generative bottleneck “fundamentally restricts information hiding.” Soften that language to match the empirical caveats already present in Table 2.
  2. Table 2 / Fig. 2: qualitative examples show individual EL samples occasionally outscoring SD-NL; a short note that these are sample-level fluctuations (as the caption already hints) would prevent over-reading.
  3. §3.3: the visual-similarity threshold τ=0.7 is justified as the 99th percentile, but sensitivity of the triplet accuracies to τ (or to an alternative object-category match) is not reported; a one-paragraph robustness check would strengthen the claim.
  4. Notation: L_diff in Eq. (1) is never expanded; a brief reminder that it is the standard latent-diffusion denoising loss would help readers outside the diffusion literature.
  5. Appendix C: CBM inventory is restricted to the 80 MS-COCO categories via dominant-instance count; this design choice is reasonable but should be flagged when claiming that CBM “cannot” capture content outside the inventory, since a richer inventory might change the gap.

Circularity Check

1 steps flagged

No significant circularity: EmCom-Diffusion is an independent generative probe validated against random/fixed baselines and triplet tests that do not reduce to the definitions of CBM, Translation, TopSim, or R@1.

specific steps
  1. other [Appendix A (speaker) + §3.1 / Eq. (2) (DINOv2 readout)]
    "The speaker is a frozen DINOv2 ViT-B/14 [23] encoder followed by a cross-attention module... We compute it with three encoders... CLIP-img..., DINOv2..., and SigLIP..."

    The same frozen DINOv2 family appears both as the speaker’s visual front-end that produces the messages and as one of the three perceptual similarity functions used to score reconstructions. This is a mild architectural self-reference, not a definitional loop: the score is still s(G(m), x) rather than a quantity defined from the speaker’s own embeddings, multi-encoder results (CLIP-img, SigLIP) preserve the same ranking, and random/fixed baselines remain near floor on DINOv2. It does not force the central claim by construction.

full rationale

The paper defines visual reflection as recoverability of image content from messages by an external decoder independent of the speaker–listener pair (§2.1), then instantiates that decoder as a fine-tuned text-to-image diffusion model whose score is perceptual similarity s(G_θ*(m), x) (Eqs. 1–2). This is a methodological proposal, not a first-principles derivation that claims to force a unique result from axioms. Experiment 1 ranks EL above random/fixed tokens and below SD-NL on held-out images; Experiment 2 constructs triplets that deliberately neutralize the proxies of the four baselines (caption-similarity control for Translation, matched edit distance for TopSim/CBM, distractor-scheme variation for R@1) and shows EmCom-Diffusion retains signal where those proxies collapse. None of these steps is self-definitional: the metric is not defined in terms of CBM/TopSim/R@1 scores, no parameter is fitted to a subset and then re-reported as a prediction of a closely related quantity, and no uniqueness theorem or ansatz is imported from the authors’ prior work to forbid alternatives. The only mild self-referential element is architectural reuse of a frozen DINOv2 backbone in both the speaker and one of three readout encoders, plus the fact that the generator is fine-tuned on the same (x, m) pairs it later scores; both are controlled by multi-encoder reporting, multi-seed analysis, and random/fixed baselines, and neither makes the ranking or the triplet accuracies true by construction. Limitations candidly note the pretrained prior’s contribution and the single-game/single-domain scope. Score 1 reflects only that minor architectural overlap, not load-bearing circularity.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 1 invented entities

The central claim rests on standard ML tooling (latent diffusion, perceptual encoders, referential games) plus the domain assumption that generative reconstruction quality is a faithful external measure of visual content. Free parameters are ordinary training hyperparameters; no new physical constants or fitted universal scales appear. The only invented entity is the named metric itself.

free parameters (3)
  • message length K and vocabulary size V
    Default K=8, V=256 chosen for the main tables; alternative (K=16,V=64) shown only qualitatively. Capacity setting affects how much visual content can be packed into messages.
  • diffusion fine-tuning steps / LoRA lr / guidance scale
    10k steps, lr 1e-4, guidance 7.5, DDIM 50 steps are hand-chosen; no sensitivity sweep is reported for the EmCom-Diffusion score itself.
  • visual-similarity threshold τ=0.7 for triplet construction
    Defined as the 99th percentile of pairwise CLIP similarities; used to label 'visually similar' pairs in Experiment 2. Different τ would change the triplet sets.
axioms (4)
  • domain assumption A pretrained text-to-image diffusion model fine-tuned on (image, message) pairs can serve as an external decoder whose reconstruction fidelity measures visual content recoverable without the original speaker-listener pair.
    Stated as the central idea of §2.2 and Eq. (2); the generative-bottleneck argument is the load-bearing justification.
  • domain assumption Perceptual cosine similarity under CLIP / DINOv2 / SigLIP is a valid proxy for visual reflection between original and reconstructed images.
    Instantiates the function s in Eq. (2); acknowledged that CLIP prefers salient foregrounds.
  • domain assumption Referential-game training on MS-COCO produces emergent messages that are a fair test case for visual-reflection metrics.
    Common EmCom setup (Appendix A); Limitations note that other games/domains remain untested.
  • standard math Standard latent-diffusion denoising loss and LoRA fine-tuning preserve enough of the pretrained prior to allow meaningful conditioning on novel discrete tokens.
    Taken from the Stable Diffusion / LoRA literature; used without re-derivation.
invented entities (1)
  • EmCom-Diffusion score (Eq. 2) no independent evidence
    purpose: Named quantitative measure of visual reflection obtained by averaging perceptual similarity of diffusion reconstructions to source images.
    The paper's primary contribution; independent evidence is the ranking and triplet experiments, not an external physical measurement.

pith-pipeline@v1.1.0-grok45 · 16635 in / 3121 out tokens · 22703 ms · 2026-07-12T00:11:22.798896+00:00 · methodology

0 comments
read the original abstract

Measuring the extent to which emergent languages encode the visual content of their inputs is an open problem. We refer to this property as visual reflection: the extent to which emergent messages preserve information about their source images that can be recovered without appeal to the speaker-listener pair that produced them. Existing metrics measure it only indirectly, through proxies such as human-defined concept inventories, natural-language captions, structural distance correlations, or Referential Game accuracy, each of which can either miss visual content the message encodes or credit content it does not. We propose EmCom-Diffusion, an evaluation framework that measures visual reflection directly: it reconstructs each input image from its emergent message and compares the reconstruction with the original image itself, rather than with human-defined targets. Concretely, it finetunes a pretrained text-to-image diffusion model on (image, emergent-message) pairs and scores visual reflection as the perceptual similarity between the reconstructed and original images, operating generatively rather than discriminatively. Instantiating it on MS-COCO with a Referential Game, we validate the metric against random and fixed-token baselines under three pretrained visual encoders, and compare it against four existing metrics (CBM, supervised translation, TopSim, and R@1). EmCom-Diffusion captures visual content the other metrics miss.

Figures

Figures reproduced from arXiv: 2607.03752 by Haruumi Omoto, Tadahiro Taniguchi.

Figure 1
Figure 1. Figure 1: Overview of our EmCom-Diffusion framework. (a) Agents play a communication game (here, a Referential Game) on images, producing emergent languages as sequences of discrete tokens. (b) We fine-tune a pretrained image generation model on pairs of images and emergent languages. (c) Generated images are compared with the original images using a suite of evaluation metrics to assess what visual information emer… view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative comparison. CLIP-img cosine similarity to the original is given be￾low each generated image (omitted for the Original column, trivially 1.0). The two EL columns (K=8, V =256 [default] and K=16, V =64) illustrate that the qualitative behavior is not specific to a single capacity setting. SD-NL outperforms EL on average ( [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

32 extracted references · 5 canonical work pages

  1. [1]

    arXiv preprint arXiv:1308.3432 (2013) 14 H

    Bengio, Y., Léonard, N., Courville, A.: Estimating or propagating gradi- ents through stochastic neurons for conditional computation. arXiv preprint arXiv:1308.3432 (2013) 14 H. Omoto and T. Taniguchi

    Pith/arXiv arXiv 2013

  2. [2]

    In: Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing

    Bouchacourt, D., Baroni, M.: How agents see things: On visual representa- tions in an emergent language game. In: Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. pp. 981–985 (2018). https://doi.org/10.18653/v1/D18-1119

    work page doi:10.18653/v1/d18-1119 2018

  3. [3]

    Artificial Life12(2), 229–242 (2006)

    Brighton, H., Kirby, S.: Understanding linguistic evolution by visualizing the emergence of topographic mappings. Artificial Life12(2), 229–242 (2006). https://doi.org/10.1162/artl.2006.12.2.229

    doi:10.1162/artl.2006.12.2.229 2006

  4. [4]

    In: Findings of the Asso- ciation for Computational Linguistics: ACL 2024

    Carmeli, B., Belinkov, Y., Meir, R.: Concept-best-matching: Evaluating compositionality in emergent communication. In: Findings of the Asso- ciation for Computational Linguistics: ACL 2024. pp. 3186–3194 (2024). https://doi.org/10.18653/v1/2024.findings-acl.189

    work page doi:10.18653/v1/2024.findings-acl.189 2024

  5. [5]

    In: Proceedings of the 58th Annual Meeting of the Association for Computational Lin- guistics

    Chaabouni, R., Kharitonov, E., Bouchacourt, D., Dupoux, E., Baroni, M.: Compositionality and generalization in emergent languages. In: Proceedings of the 58th Annual Meeting of the Association for Computational Lin- guistics. pp. 4427–4442. Association for Computational Linguistics (2020). https://doi.org/10.18653/v1/2020.acl-main.407

    doi:10.18653/v1/2020.acl-main.407 2020

  6. [6]

    In: International Confer- ence on Learning Representations (2022)

    Chaabouni, R., et al.: Emergent communication at scale. In: International Confer- ence on Learning Representations (2022)

    2022

  7. [7]

    Transactions on Machine Learning Research (2023)

    Friedman, D., Dieng, A.B.: The Vendi score: A diversity evaluation metric for machine learning. Transactions on Machine Learning Research (2023)

    2023

  8. [8]

    In: Inter- national Conference on Learning Representations (2024)

    Gurnee, W., Tegmark, M.: Language models represent space and time. In: Inter- national Conference on Learning Representations (2024)

    2024

  9. [9]

    WORD10(2-3), 146–162 (1954)

    Harris, Z.S.: Distributional structure. WORD10(2-3), 146–162 (1954). https://doi.org/10.1080/00437956.1954.11659520

    doi:10.1080/00437956.1954.11659520 1954

  10. [10]

    In: Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing

    Hessel,J.,Holtzman,A.,Forbes,M.,LeBras,R.,Choi,Y.:CLIPScore:Areference- free evaluation metric for image captioning. In: Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. pp. 7514–7528. Association for Computational Linguistics (2021). https://doi.org/10.18653/v1/2021.emnlp- main.595

    doi:10.18653/v1/2021.emnlp- 2021

  11. [11]

    Advances in Neural Information Processing Systems30(2017)

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., Hochreiter, S.: GANs trained by a two time-scale update rule converge to a local Nash equilibrium. Advances in Neural Information Processing Systems30(2017)

    2017

  12. [12]

    In: International Conference on Learning Representations (2017)

    Jang, E., Gu, S., Poole, B.: Categorical reparameterization with gumbel-softmax. In: International Conference on Learning Representations (2017)

    2017

  13. [13]

    In: Proceedings of the Workshop on Cognitive Modeling and Compu- tational Linguistics

    Kouwenhoven, T., Peeperkorn, M., Van Dijk, B., Verhoef, T.: The curious case of representational alignment: Unravelling visio-linguistic tasks in emergent com- munication. In: Proceedings of the Workshop on Cognitive Modeling and Compu- tational Linguistics. pp. 57–71. Association for Computational Linguistics (2024). https://doi.org/10.18653/v1/2024.cmcl-1.5

    work page doi:10.18653/v1/2024.cmcl-1.5 2024

  14. [14]

    Advances in Neural Information Processing Systems32(2019)

    Kynkäänniemi, T., Karras, T., Laine, S., Lehtinen, J., Aila, T.: Improved precision and recall metric for assessing generative models. Advances in Neural Information Processing Systems32(2019)

    2019

  15. [15]

    arXiv preprint arXiv:2006.02419 (2020)

    Lazaridou, A., Baroni, M.: Emergent multi-agent communication in the deep learn- ing era. arXiv preprint arXiv:2006.02419 (2020)

    Pith/arXiv arXiv 2006

  16. [16]

    In: International Conference on Learning Repre- sentations (2017)

    Lazaridou, A., Peysakhovich, A., Baroni, M.: Multi-agent cooperation and the emergence of (natural) language. In: International Conference on Learning Repre- sentations (2017)

    2017

  17. [17]

    Proceedings of the AAAI Conference on Artificial Intelligence39(22), 23231–23239 (2025)

    Levy, I., Paradise, O., Carmeli, B., Meir, R., Goldwasser, S., Belinkov, Y.: Unsupervised translation of emergent communication. Proceedings of the AAAI Conference on Artificial Intelligence39(22), 23231–23239 (2025). https://doi.org/10.1609/aaai.v39i22.34489 EmCom-Diffusion: Probing Visual Reflection 15

    work page doi:10.1609/aaai.v39i22.34489 2025

  18. [18]

    In: Proceedings of the 42nd Annual Meeting of the Association for Computational Linguistics (ACL-04)

    Lin, C.Y., Och, F.J.: Automatic evaluation of machine translation quality using longest common subsequence and skip-bigram statistics. In: Proceedings of the 42nd Annual Meeting of the Association for Computational Linguistics (ACL-04). pp. 605–612. Barcelona, Spain (2004). https://doi.org/10.3115/1218955.1219032

    doi:10.3115/1218955.1219032 2004

  19. [19]

    In: Eu- ropean Conference on Computer Vision

    Lin, T.Y., et al.: Microsoft COCO: Common objects in context. In: Eu- ropean Conference on Computer Vision. pp. 740–755. Springer (2014). https://doi.org/10.1007/978-3-319-10602-1_48

    doi:10.1007/978-3-319-10602-1_48 2014

  20. [20]

    Advances in Neural Information Processing Systems36, 34892–34916 (2023)

    Liu, H., Li, C., Wu, Q., Lee, Y.J.: Visual instruction tuning. Advances in Neural Information Processing Systems36, 34892–34916 (2023)

    2023

  21. [21]

    In: International Conference on Learning Representations (2019)

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. In: International Conference on Learning Representations (2019)

    2019

  22. [22]

    Advances in Neural Information Processing Systems34, 17994–18007 (2021)

    Mu, J., Goodman, N.: Emergent communication of generalizations. Advances in Neural Information Processing Systems34, 17994–18007 (2021)

    2021

  23. [23]

    Transactions on Machine Learning Research (2024)

    Oquab, M., et al.: DINOv2: Learning robust visual features without supervision. Transactions on Machine Learning Research (2024)

    2024

  24. [24]

    Autonomous Agents and Multi-Agent Systems39(1) (2025)

    Peters, J., et al.: Emergent language: A survey and taxonomy. Autonomous Agents and Multi-Agent Systems39(1) (2025). https://doi.org/10.1007/s10458- 025-09691-y

    doi:10.1007/s10458- 2025

  25. [25]

    In: International Conference on Machine Learning

    Radford, A., et al.: Learning transferable visual models from natural language su- pervision. In: International Conference on Machine Learning. pp. 8748–8763 (2021)

    2021

  26. [26]

    In: Proceedings of the 19th International Conference on Autonomous Agents and MultiAgent Systems

    Resnick, C., Gupta, A., Foerster, J., Dai, A.M., Cho, K.: Capacity, bandwidth, and compositionality in emergent language learning. In: Proceedings of the 19th International Conference on Autonomous Agents and MultiAgent Systems. pp. 1125–1133. International Foundation for Autonomous Agents and Multiagent Sys- tems (2020)

    2020

  27. [27]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 10684–10695 (2022). https://doi.org/10.1109/CVPR52688.2022.01042

    doi:10.1109/cvpr52688.2022.01042 2022

  28. [28]

    Song,J.,Meng,C.,Ermon,S.:Denoisingdiffusionimplicitmodels.In:International Conference on Learning Representations (2021)

    2021

  29. [29]

    Advanced Robotics0(0), 1–26 (2026)

    Taniguchi, T., Ueda, R., Nakamura, T., Suzuki, M., Taniguchi, A.: Generative emergent communication: Large language model is a collective world model. Advanced Robotics0(0), 1–26 (2026). https://doi.org/10.1080/01691864.2026.2661958

    work page doi:10.1080/01691864.2026.2661958 2026

  30. [30]

    arXiv preprint arXiv:2505.17091 (2025)

    Verma, P., Pilanci, M.: Large language models implicitly learn to see and hear just by reading. arXiv preprint arXiv:2505.17091 (2025)

    arXiv 2025

  31. [31]

    In: International Conference on Learning Representations (2022)

    Yao, S., Yu, M., Zhang, Y., Narasimhan, K., Tenenbaum, J., Gan, C.: Linking emergent and natural languages via corpus transfer. In: International Conference on Learning Representations (2022)

    2022

  32. [32]

    In: Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision

    Zhai, X., Mustafa, B., Kolesnikov, A., Beyer, L.: Sigmoid loss for language image pre-training. In: Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision. pp. 11975–11986 (2023). https://doi.org/10.1109/ICCV51070.2023.01100

    doi:10.1109/iccv51070.2023.01100 2023