Elamur Research⌕
☰
home / papers / 2023 / WorldSense: A Synthetic Benchmark for Grounded Reasoning in Large Language Models

WorldSense: A Synthetic Benchmark for Grounded Reasoning in Large Language Models

Paper · 2023 · Youssef Benchekroun, Megi Dervishi, Mark Ibrahim et al.

Summary

LLMs often seem to understand the physical world, but it is hard to tell whether they build coherent internal scene representations. Existing benchmarks suffer from bias and contamination, and they mix world knowledge with reasoning. WorldSense is a synthetic benchmark that probes such tacit world models through transitive inference over entities arranged in a linear order. It has three task families: grounded inference, consistency detection (is the described scene impossible?) and completeness detection (does the description leave the answer undetermined?). Each family has a trivial control. Problems are generated from abstract schemata and rendered with interchangeable lexical "skins." This keeps answers balanced and independent of wording. When GPT3.5, GPT4 and Llama2-chat were tested, all made errors even with three entities. They did especially poorly when the order had to be reconstructed from a description, and they rarely recognised contradictory or underspecified inputs. They also showed strong preferences for particular answers. Chain-of-thought prompting and in-context examples helped only marginally. Finetuning Llama2 improved results but did not extend beyond linear-order problems. The benchmark therefore offers a cleaner, contamination-resistant way to measure grounded reasoning.

Key points

  • Worlds contain up to five entities in a total order. Skins cover three domains (spatial, temporal, scalar) with six skins each, using two opposite binary relations and one ternary "in-between" relation.
  • Evaluated models were GPT3.5, GPT4 and Llama2-chat. GPT4 did fairly well when the order was given directly, but every model degraded sharply when it had to build the world state from text.
  • All models showed heavy response biases. Chain-of-thought prompting and in-context learning reduced errors only slightly and inconsistently.
  • Llama2 70B was finetuned on 100K and 1M examples. It generalised to new entities, relations, domains and larger problem sizes, with little sign of memorisation, and consistency problems were the hardest to learn.
  • Finetuning brought no gains on non-linear relations (e.g., parallel/perpendicular lines) or on external reasoning benchmarks. Data, the finetuning set and test splits are released at github.com/facebookresearch/worldsense.

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

🎓Université de MontréalMontreal · Canada68 works · 79,382 citations🔬Mila - Quebec Artificial Intelligence InstituteMontreal · Canada54 works · 13,565 citations🎓Université Paris Dauphine-PSLParis · France4 works · 16,983 citations

Also on the byline: Lamsade.

 Fields, topics and concepts

📚 Shelves 5The 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 works

🧭 Topics 3Research subjects and languages the work is about.Language models & pretraining · 3,320 worksReasoning · 1,426 worksEvaluation & benchmarks · 1,220 works

Cite

Cite as Youssef Benchekroun, Megi Dervishi, Mark Ibrahim, Jean-Baptiste Gaya, Xavier Martinet, Gregoire Mialon, Thomas Scialom, Emmanuel Dupoux, Dieuwke Hupkes and Pascal Vincent. 2023. WorldSense: A Synthetic Benchmark for Grounded Reasoning in Large Language Models. arXiv:2311.15930.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2311.15930 · arXiv 2311.15930

@misc{benchekroun2023worldsense,
  title = {{WorldSense: A Synthetic Benchmark for Grounded Reasoning in Large Language Models}},
  author = {Youssef Benchekroun, Megi Dervishi, Mark Ibrahim, Jean-Baptiste Gaya, Xavier Martinet, Gregoire Mialon, Thomas Scialom, Emmanuel Dupoux, Dieuwke Hupkes and Pascal Vincent},
  year = {2023},
  doi = {10.48550/arxiv.2311.15930},
  eprint = {2311.15930},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2311.15930}
}

References

37 references, 18 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jacob Austin, Augustus Odena, Maxwell Nye, Maarten Bosma, Henryk Michalewski, David Dohan, Ellen Jiang, Carrie Cai, Michael Terry, Quoc Le, et al. 2021. Program synthesis with large language models. arXiv preprint arXiv:2108.07732.
  2. Yonatan Bisk, Rowan Zellers, Ronan Le Bras, Jianfeng Gao, and Yejin Choi. 2020. PIQA: reasoning about physical commonsense in natural lan-Samuel R Bowman. 2023. Eight things to know about large language models. arXiv preprint arXiv:2304.00612. find
  3. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901.
  4. Peter E Bryant and Thomas Trabasso. 1971. Transitive inferences and memory in young children. Nature. find
  5. Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, et al. 2021. Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374.
  6. Amartya Chowdhury, Deepak K. T., Samudra Vijaya K, and S. R. Mahadeva Prasanna. 2022. Machine translation for a very low-resource language - layer freezing approach on transfer learning. In Proceedings of the Fifth Workshop on Technologies for Machine Translation of Low-Resource Languages (LoResMT 2… find
  7. Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. 2018. Think you have solved question answering? try arc, the ai2 reasoning challenge. arXiv preprint arXiv:1803.05457.
  8. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, et al. 2021. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168.
  9. Douglas J Gillan. 1981. Reasoning in the chimpanzee: Ii. transitive inference. Journal of Experimental Psychology: Animal Behavior Processes, 7(2):150. find
  10. Logan Grosenick, Tricia S Clement, and Russell D Fernald. 2007. Fish can infer social rank by observation alone. Nature, 445(7126):429–432. find
  11. Kristina Gulordava, Piotr Bojanowski, Edouard Grave, Tal Linzen, and Marco Baroni. 2018. Colorless green recurrent networks dream hierarchically. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume…
  12. Wes Gurnee and Max Tegmark. 2023. Language models represent space and time. arXiv preprint arXiv:2310.02207. find
  13. Suchin Gururangan, Swabha Swayamdipta, Omer Levy, Roy Schwartz, Samuel Bowman, and Noah A. Smith. 2018. Annotation artifacts in natural language inference data. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Techno…
  14. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2020. Measuring massive multitask language understanding. In International Conference on Learning Representations.
  15. Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. 2021. Measuring mathematical problem solving with the math dataset. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2).
  16. Mandar Joshi, Eunsol Choi, Daniel Weld, and Luke Zettlemoyer. 2017. TriviaQA: A large scale distantly supervised challenge dataset for reading comprehension. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1601–1611, Vancouver…
  17. Pride Kavumba, Ryo Takahashi, and Yusuke Oda. 2022. Are prompt-based models clueless? In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 2333–2352, Dublin, Ireland. Association for Computational Linguistics. find
  18. Olga F Lazareva and Edward A Wasserman. 2010. Nonverbal transitive inference: Effects of task and awareness on human performance. Behavioural Processes, 83(1):99–112. find
  19. Todor Mihaylov, Peter Clark, Tushar Khot, and Ashish Sabharwal. 2018. Can a suit of armor conduct electricity? a new dataset for open book question answering. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 2381–2391, Brussels, Belgium. Association fo… find
  20. Melanie Mitchell and David C. Krakauer. 2023. The debate over understanding in AI’s large language models. Proceedings of the National Academy of Sciences, 120(13):e2215907120. find
  21. Xenia Ohmer, Elia Bruni, and Dieuwke Hupkes. 2023. Evaluating task understanding through multilingual consistency: A chatgpt case study. arXiv preprint arXiv:2305.11662. — [arXiv:2305.11662](https://arxiv.org/abs/2305.11662) find
  22. Arkil Patel, Satwik Bhattamishra, and Navin Goyal. 2021. Are NLP models really able to solve simple math word problems? In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2080–2094, Online. Associat…
  23. Yasaman Razeghi, Robert L Logan IV, Matt Gardner, and Sameer Singh. 2022. Impact of pre-training term frequencies on few-shot numerical reasoning. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 840–854, Abu Dhabi, United Arab Emirates. Association for Computational L… find
  24. Jonathan Roberts, Timo Lüddecke, Sowmen Das, Kai Han, and Samuel Albanie. 2023. Gpt4geo: How a language model sees the world’s geography. arXiv preprint arXiv:2306.00020. find
  25. William A Roberts and Maria T Phelps. 1994. Transitive inference in rats: A test of the spatial coding hypothesis. Psychological Science, 5(6):368–374. find
  26. Oscar Sainz, Jon Ander Campos, Iker García-Ferrero, Julen Etxaniz, Oier Lopez de Lacalle, and Eneko Agirre. 2023. Nlp evaluation in trouble: On the need to measure llm data contamination for each benchmark. arXiv preprint arXiv:2310.18018.
  27. Maarten Sap, Hannah Rashkin, Derek Chen, Ronan Le Bras, and Yejin Choi. 2019. Social IQa: Commonsense reasoning about social interactions. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processi… find
  28. Freda Shi, Xinyun Chen, Kanishka Misra, Nathan Scales, David Dohan, Ed H Chi, Nathanael Schärli, and Denny Zhou. 2023. Large language models can be easily distracted by irrelevant context. In International Conference on Machine Learning, pages 31210–31227. PMLR.
  29. Alessandro Stolfo, Zhijing Jin, Kumar Shridhar, Bernhard Schoelkopf, and Mrinmaya Sachan. 2023. A causal framework to quantify the robustness of mathematical reasoning with language models. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pap…
  30. Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. 2019. CommonsenseQA: A question answering challenge targeting commonsense knowledge. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…
  31. 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.
  32. Lorenzo Von Fersen, Clive D Wynne, Juan D Delius, and John E Staddon. 1991. Transitive inference formation in pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 17(3):334. find
  33. Lucas Weber, Elia Bruni, and Dieuwke Hupkes. 2023. Mind the instructions: a holistic evaluation of consistency and interactions in prompt-based learning. In The SIGNLL Conference on Computational Natural Language Learning (CONLL).
  34. Jason Weston, Antoine Bordes, Sumit Chopra, and Tomás Mikolov. 2016. Towards ai-complete question answering: A set of prerequisite toy tasks. In 4th International Conference on Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings.
  35. Zhicheng Yang, Jinghui Qin, Jiaqi Chen, and Xiaodan Liang. 2022. Unbiased math word problems benchmark for mitigating solving bias. In Findings of the Association for Computational Linguistics: NAACL 2022, pages 1401–1408, Seattle, United States. Association for Computational Linguistics. find
  36. Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi, and Yejin Choi. 2019. HellaSwag: Can a machine really finish your sentence? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4791–4800, Florence, Italy. Association for Computational Linguistics. find
  37. Kun Zhou, Yutao Zhu, Zhipeng Chen, Wentong Chen, Wayne Xin Zhao, Xu Chen, Yankai Lin, Ji-Rong Wen, and Jiawei Han. 2023. Don’t make your llm an evaluation benchmark cheater. arXiv preprint arXiv:2311.01964. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/benchekroun-2023-worldsense-synthetic-benchmark-grounded-reasoning/