Elamur Research
☰
home / papers / 2022 / ROSCOE: A Suite of Metrics for Scoring Step-by-Step Reasoning

ROSCOE: A Suite of Metrics for Scoring Step-by-Step Reasoning

Written for the Research library (claude-opus-5-5) from the paper text, in our own words: the license "arXiv — non-exclusive license to distribute" does not allow redistributing the text itself. Original — arxiv.org/pdf/2212.07919

Summary

Language models that write out their reasoning before answering often do better on downstream tasks. However, there has been no dependable automatic way to check whether the intermediate steps are actually sound, separately from whether the final answer is right. Golovneva et al. (Meta AI Research) address this gap with ROSCOE, a collection of interpretable, largely unsupervised scores. Instead of judging a generated chain as one block of text, the scores examine it step by step.

The metrics fall into four groups:

  • Semantic alignment: how well the steps are grounded in the source context.
  • Logical inference: internal consistency and contradictions.
  • Semantic similarity: catching hallucinated or repeated content.
  • Language coherence: fluency and grammar.

To validate the suite, the authors define a taxonomy of reasoning errors and build two kinds of test data. One kind uses gold reasoning chains that are deliberately corrupted. The other uses model-generated chains rated by expert annotators.

ROSCOE consistently beats standard text-generation metrics on tasks that require reasoning over a context. This makes it a cheaper and more transparent alternative to manual review of rationales.

Key points

  • Metrics: ROSCOE contains 18 fine-grained metrics, each bounded in [0, 1]. Most are reference-free, comparing the hypothesis to the source, so no gold reasoning chains are needed.
  • Grounding scores: Faithfulness-Step averages, over all steps, the normalized cosine similarity between each step and its closest context sentence. The closest prior metrics are CTC and BARTScore.
  • Error taxonomy: The nine error types cover arithmetic, commonsense, factual and grammatical mistakes; hallucinated content; redundant content; repeated content; skipped (missing) steps; and lack of coherence.
  • Evaluation data: The suite is tested on 11 datasets, five human-annotated and six programmatically perturbed. Datasets include EntailmentBank, ProofWriter, EQASC, MATH and ASDIV, and outputs come from models such as GPT-3 (text-davinci-002).
  • Example results (Somers' D):
    • Info-Chain reaches 0.955 on negation perturbations in EntailmentBank.
    • Info-Step reaches 0.879 on shuffled operations in ASDIV.
    • The Grammar score is weak on MATH (0.392).
  • Availability: Code is released in ParlAI (projects/roscoe), and the annotated datasets are available for download.

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: Other

Cited by

120 papers of our library cite it (of the 140673 whose bibliographies we read); 10 of them have a page here, 110 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2212.07919 · arXiv 2212.07919

@misc{golovneva2022roscoe,
  title = {{ROSCOE: A Suite of Metrics for Scoring Step-by-Step Reasoning}},
  author = {Golovneva, Olga and Chen, Moya and Poff, Spencer and Corredor, Martin and Zettlemoyer, Luke and Fazel-Zarandi, Maryam and Celikyilmaz, Asli},
  year = {2022},
  doi = {10.48550/arxiv.2212.07919},
  eprint = {2212.07919},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2212.07919}
}

References

52 references, 21 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Shourya Aggarwal, Divyanshu Mandowara, Vishwajeet Agrawal, Dinesh Khandelwal, Parag Singla, and Dinesh Garg. Explanations for CommonsenseQA: New Dataset and Models. 2021. find
  2. Eneko Agirre, Daniel Cer, Mona Diab, Aitor Gonzalez-Agirre, and Weiwei Guo. *SEM 2013 shared task: Semantic textual similarity. In Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 1: Proceedings of the Main Conference and the Shared Task: Semantic Textual Similarity, pp.… find our card
  3. Alan Agresti. Analysis of ordinal categorical data, volume 656. John Wiley & Sons, 2010. find
  4. Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christopher D. Manning. A large annotated corpus for learning natural language inference. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pp. 632–642, Lisbon, Portugal, September 2015. Association for C… find
  5. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chri… find
  6. Oana-Maria Camburu, Tim Rocktäschel, Thomas Lukasiewicz, and Phil Blunsom. e-snli: Natural language inference with natural language explanations. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett (eds.), Advances in Neural Information Processing Systems 31, pp. 953… find our card
  7. Asli Celikyilmaz, Elizabeth Clark, and Jianfeng Gao. Evaluation of text generation: A survey. CoRR, abs/2006.14799, 2020. URL https://arxiv.org/abs/2006.14799. ↗
  8. Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, et al. Palm: Scaling language modeling with pathways. arXiv preprint arXiv:2204.02311, 2022. find
  9. Miruna-Adriana Clinciu, Arash Eshghi, and Helen Hastie. A study of automatic metrics for the evaluation of natural language explanations. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pp. 2376–2387, Online, April 2021. As… find
  10. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168, 2021. find our card
  11. Antonia Creswell, Murray Shanahan, and Irina Higgins. Selection-inference: Exploiting large language models for interpretable logical reasoning. arXiv, 2022. URL https://arxiv.org/abs/2205.09712. ↗
  12. Bhavana Dalvi, Peter Jansen, Oyvind Tafjord, Zhengnan Xie, Hannah Smith, Leighanna Pipatanangkura, and Peter Clark. Explaining answers with entailment trees. EMNLP, 2021. find our card
  13. Marina Danilevsky, Kun Qian, Ranit Aharonov, Yannis Katsis, Ban Kawas, and Prithviraj Sen. A survey of the state of explainable AI for natural language processing. In Proceedings of the 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 10th Internati… find
  14. Mingkai Deng, Bowen Tan, Zhengzhong Liu, Eric P. Xing, and Zhiting Hu. Compression, transduction, and creation: A unified framework for evaluating natural language generation. In EMNLP, 2021. URL https://aclanthology.org/2021.emnlp-main.599.pdf. ↗
  15. Daniel Deutsch, Rotem Dror, and Dan Roth. Re-examining system-level correlations of automatic summarization evaluation metrics. arXiv preprint arXiv:2204.10216, 2022. find
  16. Dheeru Dua, Yizhong Wang, Pradeep Dasigi, Gabriel Stanovsky, Sameer Singh, and Matt Gardner. DROP: A reading comprehension benchmark requiring discrete reasoning over paragraphs. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Hum… find our card
  17. Tianyu Gao, Xingcheng Yao, and Danqi Chen. Simcse: Simple contrastive learning of sentence embeddings. arXiv preprint arXiv:2104.08821, 2021. find our card
  18. Mor Geva, Daniel Khashabi, Elad Segal, Tushar Khot, Dan Roth, and Jonathan Berant. Did aristotle use a laptop? a question answering benchmark with implicit reasoning strategies. Transactions of the Association for Computational Linguistics, 9:346–361, 2021. find
  19. Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. Measuring mathematical problem solving with the math dataset. NeurIPS, 2021. find our card
  20. Lifu Huang, Ronan Le Bras, Chandra Bhagavatula, and Yejin Choi. Cosmos QA: Machine reading comprehension with contextual commonsense reasoning. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Pro… find our card
  21. Alon Jacovi and Yoav Goldberg. Aligning faithful interpretations with their social attribution. volume 9, pp. 294–310, Cambridge, MA, 2021. MIT Press. doi: 10.1162/tacl\_a\_00367. URL https: //aclanthology.org/2021.tacl-1.18. find
  22. Alon Jacovi, Swabha Swayamdipta, Shauli Ravfogel, Yanai Elazar, Yejin Choi, and Yoav Goldberg. Contrastive explanations for model interpretability. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp. 1597–1611, Online and Punta Cana, Dominican Republic, Nov… find
  23. Kalpesh Krishna, John Wieting, and Mohit Iyyer. Reformulating unsupervised style transfer as paraphrase generation. In Empirical Methods in Natural Language Processing, 2020. find our card
  24. Andrew K Lampinen, Nicholas Roy, Ishita Dasgupta, Stephanie Cy Chan, Allison Tam, James Mcclelland, Chen Yan, Adam Santoro, Neil C Rabinowitz, Jane Wang, and Felix Hill. Tell me why! Explanations support learning relational and causal structure. In Kamalika Chaudhuri, Stefanie Jegelka, Le Song, Csab… find
  25. Moritz Laurer, Wouter van Atteveldt, Andreu Casas, and Kasper Welbers. Less annotating, more classifying– addressing the data scarcity issue of supervised machine learning with deep transfer learning and bert-nli. 2022. find
  26. Christoph Leiter, Piyawat Lertvittayakumjorn, Marina Fomicheva, Wei Zhao, Yang Gao, and Steffen Eger. Towards explainable evaluation metrics for natural language generation. CoRR, abs/2203.11131, 2022. URL https://doi.org/10.48550/arXiv.2203.11131. — [doi:10.48550/arxiv.2203.11131](https://doi.org/10.48550/arxiv.2203.11131) ↗
  27. Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. In Proceedings of the 58th Annual Meeting of the Assoc… find our card
  28. Aitor Lewkowycz, Anders Andreassen, David Dohan, Ethan Dyer, Henryk Michalewski, Vinay Ramasesh, Ambrose Slone, Cem Anil, Imanol Schlag, Theo Gutman-Solo, et al. Solving quantitative reasoning problems with language models. arXiv preprint arXiv:2206.14858, 2022. find our card
  29. Yifei Li, Zeqi Lin, Shizhuo Zhang, Qiang Fu, Bei Chen, Jian-Guang Lou, and Weizhu Chen. On the advance of making language models better reasoners. arXiv, 2022. URL https://arxiv.org/abs/2206.02336. — [arXiv:2206.02336](https://arxiv.org/abs/2206.02336) ↗
  30. Chao-Chun Liang, Yu-Shiang Wong, Yi-Chung Lin, and Keh-Yih Su. A meaning-based statistical English math word problem solver. pp. 652–662, June 2018. doi: 10.18653/v1/N18-1060. URL https: //aclanthology.org/N18-1060. find
  31. Chin-Yew Lin. ROUGE: A package for automatic evaluation of summaries. In Text Summarization Branches Out, pp. 74–81, Barcelona, Spain, July 2004. Association for Computational Linguistics. URL https: //aclanthology.org/W04-1013. find
  32. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692, 2019. find
  33. Shen-yun Miao, Chao-Chun Liang, and Keh-Yih Su. A diverse corpus for evaluating and developing english math word problem solvers. pp. 975–984, 2020. find
  34. Maxwell Nye, Anders Johan Andreassen, Guy Gur-Ari, Henryk Michalewski, Jacob Austin, David Bieber, David Dohan, Aitor Lewkowycz, Maarten Bosma, David Luan, et al. Show your work: Scratchpads for intermediate computation with language models. arXiv preprint arXiv:2112.00114, 2021. find our card
  35. Juri Opitz and Anette Frank. Towards a decomposable metric for explainable evaluation of text generation from AMR. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pp. 1504–1518, Online, April 2021. Association for Computati… find our card
  36. Simon Ostermann, Michael Roth, Ashutosh Modi, Stefan Thater, and Manfred Pinkal. Semeval-2018 task 11: Machine comprehension using commonsense knowledge. In *SEMEVAL, 2018. URL https: //aclanthology.org/S18-1119.pdf. find
  37. Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9, 2019. find
  38. Nils Reimers and Iryna Gurevych. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 11 2019. URL https://arxiv.org/abs/1908.10084. ↗
  39. Ehud Reiter. Natural language generation challenges for explainable AI. In Proceedings of the 1st Workshop on Interactive Natural Language Technology for Explainable Artificial Intelligence (NL4XAI 2019), pp. 3–7. Association for Computational Linguistics, 2019. doi: 10.18653/v1/W19-8402. URL https:… find
  40. Christopher Michael Rytting and David Wingate. Leveraging the inductive bias of large language models for abstract textual reasoning. 2021. URL https://openreview.net/forum?id=urueR03mkng. ↗
  41. Thibault Sellam, Dipanjan Das, and Ankur Parikh. BLEURT: Learning robust metrics for text generation. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp. 7881–7892, Online, July 2020. Association for Computational Linguistics. doi: 10.18653/v1/2020.acl-mai… find
  42. Robert H Somers. A new asymmetric measure of association for ordinal variables. American sociological review, pp. 799–811, 1962. find
  43. Oyvind Tafjord, Bhavana Dalvi, and Peter Clark. ProofWriter: Generating implications, proofs, and abductive statements over natural language. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pp. 3621–3634, Online, August 2021. Association for Computational Linguistics.… find our card
  44. Brian Thompson and Matt Post. Automatic machine translation evaluation in many languages via zero-shot paraphrasing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 90–121, Online, November 2020. Association for Computational Linguistics. doi: 1… find our card
  45. Pauli Virtanen, Ralf Gommers, Travis E. Oliphant, Matt Haberland, Tyler Reddy, David Cournapeau, Evgeni Burovski, Pearu Peterson, Warren Weckesser, Jonathan Bright, Stéfan J. van der Walt, Matthew Brett, Joshua Wilson, K. Jarrod Millman, Nikolay Mayorov, Andrew R. J. Nelson, Eric Jones, Robert Kern,… find our card
  46. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, and Denny Zhou. Self-consistency improves chain of thought reasoning in language models. arXiv preprint arXiv:2203.11171, 2022. find our card
  47. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Ed Chi, Quoc Le, and Denny Zhou. Chain of thought prompting elicits reasoning in large language models. arXiv preprint arXiv:2201.11903, 2022. find our card
  48. Sean Welleck, Jiacheng Liu, Ximing Lu, Hannaneh Hajishirzi, and Yejin Choi. Naturalprover: Grounded mathematical proof generation with language models. 2022. find
  49. Peter West, Chandra Bhagavatula, Jack Hessel, Jena Hwang, Liwei Jiang, Ronan Le Bras, Ximing Lu, Sean Welleck, and Yejin Choi. Symbolic knowledge distillation: from general language models to commonsense models. In Proceedings of the 2022 Conference of the North American Chapter of the Association f… find our card
  50. Weizhe Yuan, Graham Neubig, and Pengfei Liu. Bartscore: Evaluating generated text as text generation. In M. Ranzato, A. Beygelzimer, Y. Dauphin, P.S. Liang, and J. Wortman Vaughan (eds.), Advances in Neural Information Processing Systems, volume 34, pp. 27263–27277. Curran Associates, Inc., 2021. UR… find our card
  51. Eric Zelikman, Yuhuai Wu, Jesse Mu, and Noah D. Goodman. Star: Bootstrapping reasoning with reasoning. arXiv, 2022. find our card
  52. Tianyi Zhang, Varsha Kishore, Felix Wu*, Kilian Q. Weinberger, and Yoav Artzi. Bertscore: Evaluating text generation with bert. In International Conference on Learning Representations, 2020. URL https: //openreview.net/forum?id=SkeHuCVFDr. find