Elamur Research⌕
☰
home / papers / 2024 / ToMBench: Benchmarking Theory of Mind in Large Language Models

ToMBench: Benchmarking Theory of Mind in Large Language Models

Paper · 2024 · Zhuang Chen, Jincenzi Wu, Jinfeng Zhou et al.

Summary

Whether large language models possess a theory of mind (ToM) is disputed. Earlier tests gave conflicting results, covered only narrow aspects such as beliefs or emotions, relied on costly manual scoring of open answers, and often reused classic psychology materials that LLMs may have seen during training. ToMBench addresses these problems with a new benchmark written entirely from scratch. Authors trained by psychologists wrote everyday social stories in Chinese and paired each with a multiple-choice question that has one correct answer and several plausible distractors. This format allows automatic scoring. The items cover eight established ToM tasks and 31 abilities drawn from the ATOMS framework.

Ten popular LLMs were tested with plain and chain-of-thought prompting and compared with a human baseline. The GPT-4 models performed best but still scored more than 10 points below humans. The gap grew when questions required a fuller grasp of the social scenario. The benchmark therefore offers a cleaner, less contaminated way to measure social reasoning in LLMs.

Key points

  • Inventory: 2,860 original bilingual test samples. The Chinese items were written first, then translated into English with GPT-4 and manually reviewed. The dataset is released at github.com/zhchen18/ToMBench.
  • Tasks: the eight tasks include False Belief, Strange Story, Hinting, Faux-Pas Recognition, Unexpected Outcome, Scalar Implicature, Persuasion Story and Ambiguous Story.
  • Abilities: ATOMS defines 7 dimensions and 39 abilities. The authors dropped the visual Percept dimension and some mixed abilities, keeping 6 dimensions and 31 abilities. The eight tasks cover only 19 of these abilities, so extra items were written for the remaining 12.
  • Evaluation: 10 LLMs were tested with vanilla and CoT prompting. The human baseline came from 20 graduate students who are native Chinese speakers. The best model, GPT-4, trails humans by more than 10 percentage points. Attention analysis suggests the models depend on semantic associations rather than human-like reasoning.
  • Limitations:
    • Some abilities have as few as 20 samples.
    • The benchmark is text-only, with no visual ToM tests.
    • The English version may keep some cultural artifacts from the Chinese originals.

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

 Institutions

🎓Tsinghua UniversityBeijing · China333 works · 87,019 citations🎓Chinese University of Hong KongHong Kong · China126 works · 14,073 citations🎓Beijing Institute of TechnologyBeijing · China34 works · 3,652 citations🎓Tianjin UniversityTianjin · China41 works · 1,128 citations🎓Nankai UniversityTianjin · China24 works · 762 citations🎓Northwest Minzu UniversityLanzhou · China2 works · 202 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 worksPsychology · 1,552 worksDevelopmental and Educational Psychology · 499 works

Cite

Cite as Zhuang Chen, Jincenzi Wu, Jinfeng Zhou, Bosi Wen, Guanqun Bi, Gongyao Jiang, Yaru Cao, Mengting Hu, Yunghwei Lai, Zexuan Xiong and Minlie Huang. 2024. ToMBench: Benchmarking Theory of Mind in Large Language Models. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2024.acl-long.847

@inproceedings{chen2024tombench,
  title = {{ToMBench: Benchmarking Theory of Mind in Large Language Models}},
  author = {Zhuang Chen, Jincenzi Wu, Jinfeng Zhou, Bosi Wen, Guanqun Bi, Gongyao Jiang, Yaru Cao, Mengting Hu, Yunghwei Lai, Zexuan Xiong and Minlie Huang},
  booktitle = {Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2024},
  doi = {10.18653/v1/2024.acl-long.847},
  url = {https://doi.org/10.18653/v1/2024.acl-long.847}
}

References

74 references, 11 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, et al. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774.
  2. KOLMOGOROV AN. 1933. Sulla determinazione empirica di una legge didistribuzione. Giorn Dell’inst Ital Degli Att, 4:89–91. find
  3. James N Aronson and Claire Golomb. 1999. Preschoolers’ understanding of pretense and presumption of congruity between action and representation. Developmental Psychology, 35(6):1414. find
  4. Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, et al. 2023. Qwen technical report. arXiv preprint arXiv:2309.16609.
  5. Baichuan-Inc. 2023. Baichuan 2. Online. find
  6. Simon Baron-Cohen, Alan M Leslie, and Uta Frith. 1985. Does the autistic child have a “theory of mind”? Cognition, 21(1):37–46. find
  7. Simon Baron-Cohen, Michelle O’riordan, Valerie Stone, Rosie Jones, and Kate Plaisted. 1999. Recognition of faux pas by normally developing children and children with asperger syndrome or high-functioning autism. Journal of autism and developmental disorders, 29:407–418. find
  8. Cindy Beaudoin, Élizabel Leblanc, Charlotte Gagner, and Miriam H Beauchamp. 2020. Systematic review and inventory of theory of mind measures for young children. Frontiers in psychology, 10:2905. find
  9. Mark Bennett and Linda Galpert. 1993. Children’s understanding of multiple desires. International Journal of Behavioral Development, 16(1):15–33. find
  10. Helene Borke. 1971. Interpersonal perception of young children: Egocentrism or empathy? Developmental psychology, 5(2):263. find
  11. Sandra Bosacki and Janet Wilde Astington. 1999. Theory of mind in preadolescence: Relations between social understanding and social competence. Social development, 8(2):237–255. find
  12. Elisabeth EF Bradford, Ines Jentzsch, Juan-Carlos Gomez, Yulu Chen, Da Zhang, and Yanjie Su. 2018. Cross-cultural differences in adult theory of mind abilities: a comparison of native-english speakers and native-chinese speakers on the self/other differentiation task. Quarterly Journal of Experiment… find
  13. Michael Brambring and Doreen Asbrock. 2010. Validity of false belief tasks in blind children. Journal of Autism and Developmental Disorders, 40:1471– 1484. find
  14. Sébastien Bubeck, Varun Chandrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Kamar, Peter Lee, Yin Tat Lee, Yuanzhi Li, Scott Lundberg, et al. 2023. Sparks of artificial general intelligence: Early experiments with gpt-4. arXiv preprint arXiv:2303.12712.
  15. Marcella Caputi, Serena Lecce, Adriano Pagnin, and Robin Banerjee. 2012. Longitudinal effects of theory of mind on later peer relations: the role of prosocial behavior. Developmental psychology, 48(1):257. find
  16. Stephanie M Carlson and Louis J Moses. 2001. Individual differences in inhibitory control and children’s theory of mind. Child development, 72(4):1032– 1053. find
  17. Cristina Colonnesi, Carolien Rieffe, Willem Koops, and Paola Perucchini. 2008. Precursors of a theory of mind: A longitudinal study. British Journal of Developmental Psychology, 26(4):561–577. find
  18. Rhiannon Corcoran, Gavin Mercer, and Christopher D Frith. 1995. Schizophrenia, symptomatology and social inference: investigating “theory of mind” in people with schizophrenia. Schizophrenia research, 17(1):5–13. find
  19. Jean Decety and Philip L Jackson. 2004. The functional architecture of human empathy. Behavioral and cognitive neuroscience reviews, 3(2):71–100. find
  20. Susanne A Denham. 1986. Social cognition, prosocial behavior, and emotion in preschoolers: Contextual validation. Child development, pages 194–201. find
  21. Murray J Dyck, Kara Ferguson, and Ian M Shochet. 2001. Do autism spectrum disorders differ from each other and from non-spectrum disorders on emotion recognition tests? European child & adolescent psychiatry, 10:105–116. find
  22. John H Flavell, Frances L Green, Eleanor R Flavell, Malcolm W Watson, and Joseph C Campione. 1986. Development of knowledge about the appearance-reality distinction. Monographs of the society for research in child development, pages i–87. find
  23. Shahriar Golchin and Mihai Surdeanu. 2023. Time travel in llms: Tracing data contamination in large language models. arXiv preprint arXiv:2308.08493.
  24. Noah D Goodman and Andreas Stuhlmüller. 2013. Knowledge and implicature: Modeling language understanding as social cognition. Topics in cognitive science, 5(1):173–184. find
  25. Felice W Gordis et al. 1989. Young children’s understanding of simultaneous conflicting emotions. find
  26. Francesca GE Happé. 1994. An advanced test of theory of mind: Understanding of story characters’ thoughts and feelings by able autistic, mentally handicapped, and normal children and adults. Journal of autism and Developmental disorders, 24(2):129–154. find
  27. Paul L Harris, Kara Donnelly, Gabrielle R Guz, and Rosemary Pitt-Watson. 1986. Children’s understanding of the distinction between real and apparent emotion. Child development, pages 895–909. find
  28. G-Juergen Hogrefe, Heinz Wimmer, and Josef Perner. 1986. Ignorance versus false belief: A developmental lag in attribution of epistemic states. Child development, pages 567–582. find
  29. Albert Q Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, et al. 2023. Mistral 7b. arXiv preprint arXiv:2310.06825. find
  30. Michal Kosinski. 2023. Theory of mind may have spontaneously emerged in large language models. arXiv preprint arXiv:2302.02083.
  31. Sebastian Lapuschkin, Stephan Wäldchen, Alexander Binder, Grégoire Montavon, Wojciech Samek, and Klaus-Robert Müller. 2019. Unmasking clever hans predictors and assessing what machines really learn. Nature communications, 10(1):1096. find
  32. Matt Le, Y-Lan Boureau, and Maximilian Nickel. 2019. Revisiting the evaluation of theory of mind through question answering. In EMNLP. find
  33. Changmao Li and Jeffrey Flanigan. 2023. Task contamination: Language models may not be few-shot anymore. arXiv preprint arXiv:2312.16337. find
  34. Xiaomeng Ma, Lingyu Gao, and Qihui Xu. 2023a. Tomchallenges: A principle-guided dataset and diverse evaluation tasks for exploring theory of mind. arXiv preprint arXiv:2305.15068.
  35. Ziqiao Ma, Jacob Sansom, Run Peng, and Joyce Chai. 2023b. Towards a holistic landscape of situated theory of mind in large language models. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 1011–1031. find
  36. Cameron Robert Jones, Sean Trott, and Ben Bergen. 2023. Epitome: Experimental protocol inventory for theory of mind evaluation. In First Workshop on Theory of Mind in Communicating Agents. find
  37. Andrew N Meltzoff. 1995. Understanding the intentions of others: Re-enactment of intended acts by 18-month-old children. Developmental psychology, 31(5):838. find
  38. Melanie Killen, Kelly Lynn Mulvey, Cameron Richardson, Noah Jampol, and Amanda Woodward. 2011. The accidental transgressor: Morally-relevant theory of mind. Cognition, 119(2):197–215. find
  39. Mistral AI. 2023. Mixtral of experts: A high quality sparse mixture-of-experts. Online. find
  40. Hyunwoo Kim, Melanie Sclar, Xuhui Zhou, Ronan Bras, Gunhee Kim, Yejin Choi, and Maarten Sap. 2023. Fantom: A benchmark for stress-testing machine theory of mind in interactions. In EMNLP, pages 14397–14413. find
  41. Peter Kinderman, Robin Dunbar, and Richard P Bentall. 1998. Theory-of-mind deficits and causal attributions. British journal of Psychology, 89(2):191–204. find
  42. Jan-Christoph Klie, Bonnie Webber, and Iryna Gurevych. 2023. Annotation error detection: Analyzing the past and present for a more coherent future. Computational Linguistics, 49(1):157–198.
  43. Ariel Knafo, Carolyn Zahn-Waxler, Maayan Davidov, Carol Van Hulle, JoAnn L Robinson, and Soo Hyun Rhee. 2009. Empathy in early childhood: Genetic, environmental, and affective contributions. Annals of the New York Academy of Sciences, 1167(1):103– 114. find
  44. Anna M Kołodziejczyk and Sandra L Bosacki. 2016. Young-school-aged children’s use of direct and indirect persuasion: role of intentionality understanding. Psychology of Language and Communication, 20(3):292–315. find
  45. Henrike Moll, Cornelia Koring, Malinda Carpenter, and Michael Tomasello. 2006. Infants determine others’ focus of attention by pragmatics and exclusion. Journal of Cognition and Development, 7(3):411–430. find
  46. OpenAI. 2023a. Gpt-3.5-turbo-0613: Function calling, 16k context window, and lower prices. Online. find
  47. OpenAI. 2023b. New models and developer products announced at devday. Online. find
  48. Josef Perner and Heinz Wimmer. 1985. “john thinks that mary thinks that. . . ” attribution of second-order beliefs by 5-to 10-year-old children. Journal of experimental child psychology, 39(3):437–471. find
  49. Joan Peskin, Carly Prusky, and Julie Comay. 2014. Keeping the reader’s mind in mind: development of perspective-taking in children’s dictations. Journal of applied developmental psychology, 35(1):35–43. find
  50. Ann T Phillips, Henry M Wellman, and Elizabeth S Spelke. 2002. Infants’ ability to connect gaze and emotional expression to intentional action. Cognition, 85(1):53–78. find
  51. Bradford H Pillow. 1989. Early understanding of perception as a source of knowledge. Journal of experimental child psychology, 47(1):116–129. find
  52. Francisco Pons and Paul Harris. 2000. Test of emotion comprehension: TEC. University of Oxford. find
  53. David Premack and Guy Woodruff. 1978. Does the chimpanzee have a theory of mind? Behavioral and brain sciences, pages 515–526. find
  54. François Quesque and Yves Rossetti. 2020. What do theory-of-mind tasks actually measure? theory and practice. Perspectives on Psychological Science, 15(2):384–396. find
  55. Betty M Repacholi and Alison Gopnik. 1997. Early reasoning about desires: evidence from 14-and 18-month-olds. Developmental psychology, 33(1):12. find
  56. Maarten Sap, Ronan Le Bras, Daniel Fried, and Yejin Choi. 2022. Neural theory-of-mind? on the limits of social intelligence in large lms. ArXiv. find
  57. Maarten Sap, Hannah Rashkin, Derek Chen, Ronan LeBras, and Yejin Choi. 2019. Socialiqa: Commonsense reasoning about social interactions. arXiv preprint arXiv:1904.09728.
  58. Natalie Shapira, Mosh Levy, Seyed Hossein Alavi, Xuhui Zhou, Yejin Choi, Yoav Goldberg, Maarten Sap, and Vered Shwartz. 2023. Clever hans or neural theory of mind? stress testing social reasoning in large language models. arXiv preprint arXiv:2305.14763. find
  59. Herbert A Simon and Herbert A Simon. 1977. Spurious correlation: A causal interpretation. Springer. find
  60. Virginia Slaughter, Michelle J Dennis, and Michelle Pritchard. 2002. Theory of mind and peer acceptance in preschool children. British journal of developmental psychology, 20(4):545–564. find
  61. Patricia A Smiley. 2001. Intention understanding and partner-sensitive behaviors in young children’s peer interactions. Social Development, 10(3):330–354. find
  62. Kate Sullivan, Ellen Winner, and Natalie Hopfield. 1995. How children tell a lie from a joke: The role of second-order mental state attributions. British journal of developmental psychology, 13(2):191–204. find
  63. J Swettenham. 1996. Can children be taught to understand false belief using computers? child psychology & psychiatry & allied disciplines, 37 (2), 157–165. find
  64. THUDM. 2023. Chatglm3. Online. find
  65. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288.
  66. Tomer Ullman. 2023. Large language models fail on trivial alterations to theory-of-mind tasks. arXiv preprint arXiv:2302.08399. find
  67. Max J van Duijn, Bram van Dijk, Tom Kouwenhoven, Werner de Valk, Marco R Spruit, and Peter van der Putten. 2023. Theory of mind in large language models: Examining performance of 11 state-of-the-art models vs. children aged 7-10 on advanced tests. arXiv preprint arXiv:2310.20320. find
  68. Henry M Wellman and Karen Bartsch. 1988. Young children’s reasoning about beliefs. Cognition, 30(3):239–277. find
  69. Frank Wilcoxon. 1947. Individual comparisons of grouped data by ranking methods. find
  70. Alex Wilf, Sihyun Shawn Lee, Paul Pu Liang, and Louis-Philippe Morency. 2023. Think twice: Perspective-taking improves large language models’ theory-of-mind capabilities. arXiv preprint arXiv:2311.10227.
  71. Heinz Wimmer and Josef Perner. 1983. Beliefs about beliefs: Representation and constraining function of wrong beliefs in young children’s understanding of deception. Cognition, 13(1):103–128. find
  72. Yufan Wu, Yinghui He, Yilin Jia, Rada Mihalcea, Yulong Chen, and Naihao Deng. 2023. Hi-tom: A benchmark for evaluating higher-order theory of mind reasoning in large language models. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 10691–10706. find
  73. Chujie Zheng, Hao Zhou, Fandong Meng, Jie Zhou, and Minlie Huang. 2023. On large language models’ selection bias in multi-choice questions. arXiv preprint arXiv:2309.03882. find
  74. Pei Zhou, Aman Madaan, Srividya Pranavi Potharaju, Aditya Gupta, Kevin R McKee, Ari Holtzman, Jay Pujara, Xiang Ren, Swaroop Mishra, Aida Nematzadeh, et al. 2023. How far are large language models from agents with theory-of-mind? arXiv preprint arXiv:2310.03051.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/chen-2024-tombench-benchmarking-theory-mind-large/