Elamur Research⌕
☰
home / papers / 2026 / Modeling Multi-Dimensional Cognitive States in Large Language Models under Cognitive Crowding

Modeling Multi-Dimensional Cognitive States in Large Language Models under Cognitive Crowding

Paper · 2026 · Lin Zhong, Siyu Zhu, Zizhen Yuan et al.

Summary

Most LLM work on human cognition treats emotion, stance or intent as separate tasks. Psychology, however, views these dimensions as one interacting system. The authors build CognitiveBench, an English Twitter (X) dataset with expert labels for four dimensions at once: emotion, thinking style, stance and intent. On this benchmark, LLMs do well on any single dimension but fail badly when they must predict all four together.

A Gromov δ-hyperbolicity analysis shows the label space is strongly tree-like. The authors propose that hierarchical states need exponentially growing room, while Euclidean embeddings grow only polynomially. Distinct states therefore end up overlapping, an effect they name "Cognitive Crowding."

Their fix, HyCoLLM, has two parts:

  • a Hyperbolic Cognitive Network on the Poincaré ball, trained with a geometry-aware contrastive loss;
  • Hyperbolic Guided Alignment Tuning (HGAT), which pulls the LLM's internal representations toward that hyperbolic structure.

With this approach, an 8B model surpasses strong baselines, including GPT-4o. The work matters because it reframes joint cognitive modeling as a problem of representation geometry.

Key points

  • CognitiveBench data: 6,514 posts kept from 48,496 raw posts (about 13.4% retained).
    • Four topics: China-United States Trade, US Election, DEI and Federal Reserve Interest Rates.
    • Each post was labeled by three of 29 domain experts, and only posts with at least 2/3 agreement were kept.
  • Label taxonomy: each dimension follows an established theory.
    • Emotion: 9 emotions based on Plutchik's model.
    • Thinking: 8 styles based on dual process theory.
    • Stance: Support, Oppose or Unclear.
    • Intent: 7 labels based on speech act theory.
  • Joint-prediction failure: GPT-4o reaches only about 5.7% accuracy when predicting all four dimensions together. The dataset's relative δ is about 1%, which indicates a strong hierarchy.
  • Cross-topic transfer: using CUT for training and the held-out FRIR topic for evaluation, HyCoLLM scores best, with a Hamming Loss of 12.5% and PMA@4 of 11.2%. Stance transfers poorly across topics.
  • Cost and scaling: HyCoLLM adds no inference latency and only modest training cost. It scales from 1.7B to 32B models. Code, data and weights are available on GitHub.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksNeuroscience · 186 worksCognitive Neuroscience · 185 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,894 works

Cite

Cite as Lin Zhong, Siyu Zhu, Zizhen Yuan, Jinhao Cui, Xinyang Zhao, Lingzhi Wang, Hao Chen and Qing Liao. 2026. Modeling Multi-Dimensional Cognitive States in Large Language Models under Cognitive Crowding. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1017

@inproceedings{zhong2026modeling,
  title = {{Modeling Multi-Dimensional Cognitive States in Large Language Models under Cognitive Crowding}},
  author = {Lin Zhong, Siyu Zhu, Zizhen Yuan, Jinhao Cui, Xinyang Zhao, Lingzhi Wang, Hao Chen and Qing Liao},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1017},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1017}
}

References

40 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, and 1 others. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774.
  2. Abu Ubaida Akash, Ahmed Fahmy, and Amine Trabelsi. 2025. Can large language models address open-target stance detection? In Findings of the Association for Computational Linguistics: ACL 2025, pages 971– 985.
  3. Maryam Amirizaniani, Elias Martin, Maryna Sivachenko, Afra Mashhadi, and Chirag Shah. 2024. Can llms reason like humans? assessing theory of mind reasoning in llms for open-ended questions. In Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, pages 34–44. find
  4. Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, and Ion Stoica. 2023. Efficient memory management for large language model serving with pagedattention. In Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles.
  5. John Langshaw Austin. 1975. How to do things with words. Harvard university press. find
  6. Yingjie Li, Tiberiu Sosea, Aditya Sawant, Ajith Jayaraman Nair, Diana Inkpen, and Cornelia Caragea. 2021. P-stance: A large dataset for stance detection in political domain. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 2355–2365.
  7. Steven J Breckler. 1984. Empirical validation of affect, behavior, and cognition as distinct components of attitude. Journal of personality and social psychology, 47(6):1191. find
  8. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, and 1 others. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arX…
  9. Esra Dönmez and Agnieszka Falenska. 2025. “i understand your perspective”: Llm persuasion through the lens of communicative action theory. In Findings of the Association for Computational Linguistics: ACL 2025, pages 15312–15327. find
  10. Alice H Eagly. 1993. The psychology of attitudes. Thomson Wadsworth. find
  11. Gino Franco Fazzi, Julie Skoven Hinge, Stefan Heinrich, and Paolo Burelli. 2025. Don’t get too excited-eliciting emotions in llms. In Proceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems, pages 1–9. find
  12. Margherita Gambini, Caterina Senette, Tiziano Fagni, and Maurizio Tesconi. 2024. Evaluating large language models for user stance detection on x (twitter). Machine Learning, 113(10):7243–7266. find
  13. F Grasso, S Locci, G Siragusa, and L Di Caro. 2024. Ecoverse: An annotated twitter dataset for eco-relevance classification, environmental impact analysis, and stance detection. Preprint, arXiv:2400.00000. find
  14. Nikesh Gyawali, Iustin Sirbu, Tiberiu Sosea, Sarthak Khanal, and 1 others. 2024. Gunstance: Stance detection for gun control and gun regulation. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 12027–12044. find
  15. Daniel Kahneman. 2011. macmillan. find
  16. Nils Klüwer, Irina Nalis, and Julia Neidhardt. 2025. Context over categories: Implementing the theory of constructed emotion with llm-guided user analysis. In Proceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems, pages 1–7. find
  17. Maryam Sadat Mirzaei, Kourosh Meshgi, and Satoshi Sekine. 2023. What is the real intention behind this question? dataset collection and intention classification. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13606–13622. find
  18. Saif Mohammad, Svetlana Kiritchenko, Parinaz Sobhani, Xiaodan Zhu, and Colin Cherry. 2016. Semeval-2016 task 6: Detecting stance in tweets. In Proceedings of the 10th international workshop on semantic evaluation (SemEval-2016), pages 31–41.
  19. Hongbin Na. 2024. Cbt-llm: A chinese large language model for cognitive behavioral therapy-based mental health question answering. arXiv preprint arXiv:2403.16008. find
  20. Robert Plutchik and Henry Kellerman. 1980. Emotion, theory, research, and experience. Academic press. find
  21. Qiumei Pu, Fangli Huang, Fude Li, Jieyao Wei, and Shan Jiang. 2025. Integrating emotional features for stance detection aimed at social network security: A multi-task learning approach. Electronics, 14(1):186. find
  22. John R Searle. 1976. A classification of illocutionary acts1. Language in society, 5(1):1–23. find
  23. Natalie Shapira, Mosh Levy, Seyed Hossein Alavi, Xuhui Zhou, Yejin Choi, Yoav Goldberg, Maarten Sap, and Vered Shwartz. 2024. Clever hans or neural theory of mind? stress testing social reasoning in large language models. In Proceedings of the 18th Conference of the European Chapter of the Associati… find
  24. Phillip Shaver, Judith Schwartz, Donald Kirson, and Cary O’connor. 1987. Emotion knowledge: further exploration of a prototype approach. Journal of personality and social psychology, 52(6):1061. find
  25. Muzafer Sherif and Carl I Hovland. 1961. Social judgment: Assimilation and contrast effects in communication and attitude change. find
  26. Parinaz Sobhani, Diana Inkpen, and Xiaodan Zhu. 2017. A dataset for multi-target stance detection. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 2, Short Papers, pages 551–557.
  27. P Taylor, J Nicholas Hobbs, Javier Burroni, and Hava T Siegelmann. 2015. The global landscape of cognition: hierarchical aggregation as an organizational principle of human cortical networks and functions. Scientific reports, 5(1):18112. find
  28. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, and 1 others. 2023. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971.
  29. John C Turner and Penelope J Oakes. 1986. The significance of the social identity concept for social psychology with reference to individualism, interactionism and social influence. British journal of social psychology, 25(3):237–252. find
  30. Apoorva Upadhyaya, Marco Fisichella, and Wolfgang Nejdl. 2023. A multi-task model for emotion and offensive aided stance detection of climate change tweets. In Proceedings of the ACM Web Conference 2023, pages 3948–3958. find
  31. Leandro von Werra, Younes Belkada, Lewis Tunstall, Edward Beeching, Tristan Thrush, Nathan Lambert, Shengyi Huang, Kashif Rasul, and Quentin Gallouédec. 2020. Trl: Transformer reinforcement learning. https://github.com/huggingface/trl. ↗
  32. Yun Wan and Qigang Gao. 2015. An ensemble sentiment classification system of twitter data for airline services analysis. In 2015 IEEE international conference on data mining workshop (ICDMW), pages 1318–1325. find
  33. Boran Wang, Yiming Liu, Haoyu Tian, Rui Hua, Kai Chang, Jianan Xia, Xinyu Dai, Zhuliang Gao, Sitong Liu, Rui Wang, and 1 others. 2025. Llm4deu: fine tuning large language model for medical diagnosis in outpatient and emergency department visits of neurosurgery. Tsinghua Science and Technology, 30(6)… find
  34. Jiayin Wang, Fengran Mo, Weizhi Ma, Peijie Sun, Min Zhang, and Jian-Yun Nie. 2024. A user-centric multi-intent benchmark for evaluating large language models. arXiv preprint arXiv:2404.13940. find
  35. Fang Xu, Tianyu Zhou, Tri Nguyen, Haohui Bao, Christine Lin, and Jing Du. 2025. Integrating augmented reality and llm for enhanced cognitive support in critical audio communications. International Journal of Human-Computer Studies, 194:103402. find
  36. Vikas Yadav, Zheng Tang, and Vijay Srinivasan. 2024. Pag-llm: Paraphrase and aggregate with large language models for minimizing intent classification errors. In Proceedings of the 47th international ACM SIGIR conference on research and development in information retrieval, pages 2569–2573. find
  37. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025. Qwen3 technical report. arXiv:2505.09388. arXiv preprint Tan Yu, Jingjing Wang, Jiawen Wang, Jiamin Luo, and Guodong Zhou. 2024. Towards emotion-enriched text-to…
  38. Zhao Zhang, Yiming Li, Jin Zhang, and Hui Xu. 2024. Llm-driven knowledge injection advances zero-shot and cross-target stance detection. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 2: Short… find
  39. Lin Zhong, Lingzhi Wang, Xu Yang, and Qing Liao. 2025. Comapoi: A collaborative multi-agent framework for next poi prediction bridging the gap between trajectory and language. In Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 17… find
  40. Minghui Zou, Ronghui Guo, Sai Zhang, Xiaowang Zhang, and Zhiyong Feng. 2025. Tat: Improving stance detection on social media through thought alignment with llms. In Companion Proceedings of the ACM on Web Conference 2025, pages 1558–1562. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhong-2026-modeling-multi-dimensional-cognitive-states/