---
title: "Faithful Reasoning Using Large Language Models (2022)"
description: "Creswell and Shanahan (DeepMind) propose a forward-chaining system whose computation follows… 31 references, cited by 85 papers here."
authors: ["Antonia Creswell", "Murray Shanahan"]
year: 2022
venue: "arXiv:2208.14271"
source: "https://arxiv.org/abs/2208.14271"
doi: "10.48550/arxiv.2208.14271"
arxiv: "2208.14271"
licence: "CC BY 4.0 (arXiv)"
licence_url: "http://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/creswell-2022-faithful-reasoning-using-large-language/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/creswell-2022-faithful-reasoning-using-large-language/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2022](https://md.research.elamur.ai/free/en/papers/2022/page/4/) / Faithful Reasoning Using Large Language Models

# Faithful Reasoning Using Large Language Models

**Authors:** [Antonia Creswell](https://md.research.elamur.ai/free/en/authors/antonia-creswell/) and [Murray Shanahan](https://md.research.elamur.ai/free/en/authors/murray-shanahan/) · **Year:** 2022 · **Venue:** arXiv:2208.14271 · **Licence:** [CC BY 4.0 (arXiv)](http://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2208.14271) · [PDF](https://arxiv.org/pdf/2208.14271) · [DOI 10.48550/arxiv.2208.14271](https://doi.org/10.48550/arxiv.2208.14271) · [arXiv 2208.14271](https://arxiv.org/abs/2208.14271) · [OpenAlex](https://openalex.org/W4293998609)

**Citations:** 37 citations (OpenAlex, 2026-10-14) · [85](https://md.research.elamur.ai/free/en/papers/creswell-2022-faithful-reasoning-using-large-language/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
Large language models usually answer questions in a single opaque pass. Users cannot see which premises were used or check the logic, and multi-step problems suffer. Creswell and Shanahan (DeepMind) propose a forward-chaining system whose computation follows the formal definition of a valid reasoning trace.

Its core, Selection-Inference (SI), alternates two fine-tuned LMs:
- **Selection** picks supporting statements from the given context by their sentence labels, so it cannot invent facts.
- **Inference** derives a new statement from the selection alone, without seeing the question.

A **Halter** LM decides after each step whether the question can be answered and maps the latest inference onto the required answer format. A learned **value function** scores partial traces to steer a step-level beam search.

On Proof Writer and a QA version of EntailmentBank, the system beats baselines on final-answer accuracy. It also produces more accurate, checkable traces, hallucinates less and relies more on the context. This matters for trust and debugging, because users can audit each step instead of accepting an unexplained answer.

## Key points
- The architecture has four separately fine-tuned LMs: Selection, Inference, Halter and value function. None is trained end-to-end for final-answer accuracy.
- Connectedness of the trace is guaranteed because selections must come from context labels such as "sent 3". Validity holds only if each individual inference step is logically correct.
- If no answer is reached within a fixed step budget, the model returns "Unknown". Filtering out these abstentions noticeably raises precision.
- Beam search expands P candidate steps per trace, keeps the top B ≤ P by value score, and continues until all traces have halted. Search improved both SI and the Proof Only baseline.
- Limitations: the context is assumed to be supplied, and step correctness is not formally guaranteed. Future work aims to add retrieval to populate the context, framed as an instance of "algorithmic prompting".

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

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works

🧭 Topics (1): [🧮 Reasoning](https://md.research.elamur.ai/free/en/topics/reasoning/) · 1,685 works

## Cite

Cite as: Antonia Creswell and Murray Shanahan. 2022. Faithful Reasoning Using Large Language Models. arXiv:2208.14271.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@misc{creswell2022faithful,
  title = {{Faithful Reasoning Using Large Language Models}},
  author = {Antonia Creswell and Murray Shanahan},
  year = {2022},
  doi = {10.48550/arxiv.2208.14271},
  eprint = {2208.14271},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2208.14271}
}
```

- [← previous ATTEMPT: Parameter-Efficient Multi-task Tuning via Attentional Mixtures of Soft Prompts](https://md.research.elamur.ai/free/en/papers/asai-2022-attempt-parameter-efficient-multi-task/)
- [next → The Moral Integrity Corpus: A Benchmark for Ethical Dialogue Systems](https://md.research.elamur.ai/free/en/papers/ziems-2022-moral-integrity-corpus-benchmark-ethical/)
- [↑ back to papers of 2022](https://md.research.elamur.ai/free/en/papers/2022/page/4/)

## References

31 references, 17 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. J.-B. Alayrac, J. Donahue, P. Luc, A. Miech, I. Barr, Y. Hasson, K. Lenc, A. Mensch, K. Millican, M. Reynolds, et al. [Flamingo: a visual language model for few-shot learning](https://md.research.elamur.ai/free/en/papers/alayrac-2022-flamingo-visual-language-model-few/). arXiv preprint arXiv:2204.14198, 2022.
2. A. Banino, J. Balaguer, and C. Blundell. Pondernet: Learning to ponder. arXiv preprint arXiv:2107.05407, 2021. [find](https://scholar.google.com/scholar?q=A%2E+Banino%2C+J%2E+Balaguer%2C+and+C%2E+Blundell%2E+Pondernet%3A+Learning+to+ponder%2E+arXiv+preprint+arXiv%3A2107%2E05407%2C+2021%2E)
3. E. M. Bender, T. Gebru, A. McMillan-Major, and S. Shmitchell. On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, FAccT ’21, page 610–623, New York, NY, USA, 2021. Association for Computing Mac… — [doi:10.1145/3442188.3445922](https://doi.org/10.1145/3442188.3445922) [find](https://scholar.google.com/scholar?q=E%2E+M%2E+Bender%2C+T%2E+Gebru%2C+A%2E+McMillan%2DMajor%2C+and+S%2E+Shmitchell%2E+On+the+dangers+of+stochastic+parrots%3A+Can+language+models+be+too+big%3F+In+Proceedings+of+the+2021+ACM+Conference+on+Fairness%2C+Accountabilit)
4. G. Betz, C. Voigt, and K. Richardson. Critical thinking for language models. In Proceedings of the 14th International Conference on Computational Semantics (IWCS), pages 63– 75, Groningen, The Netherlands (online), June 2021. Association for Computational Linguistics. URL https://aclanthology.org/ 2… [↗](https://aclanthology.org/)
5. K. Bostrom, Z. Sprague, S. Chaudhuri, and G. Durrett. Natural language deduction through search over statement compositions. arXiv preprint arXiv:2201.06028, 2022. [find](https://scholar.google.com/scholar?q=K%2E+Bostrom%2C+Z%2E+Sprague%2C+S%2E+Chaudhuri%2C+and+G%2E+Durrett%2E+Natural+language+deduction+through+search+over+statement+compositions%2E+arXiv+preprint+arXiv%3A2201%2E06028%2C+2022%2E)
6. P. Clark, I. Cowhey, O. Etzioni, T. Khot, A. Sabharwal, C. Schoenick, and O. Tafjord. [Think you have solved question answering? try arc, the ai2 reasoning challenge](https://md.research.elamur.ai/free/en/papers/clark-2018-think-have-solved-question-answering/). arXiv preprint arXiv:1803.05457, 2018.
7. K. Cobbe, V. Kosaraju, M. Bavarian, J. Hilton, R. Nakano, C. Hesse, and J. Schulman. [Training verifiers to solve math word problems](https://md.research.elamur.ai/free/en/papers/cobbe-2021-training-verifiers-solve-math-word/). arXiv preprint arXiv:2110.14168, 2021.
8. A. Creswell, M. Shanahan, and I. Higgins. Selection-inference: Exploiting large language models for interpretable logical reasoning. arXiv preprint arXiv:2205.09712, 2022. [find](https://scholar.google.com/scholar?q=A%2E+Creswell%2C+M%2E+Shanahan%2C+and+I%2E+Higgins%2E+Selection%2Dinference%3A+Exploiting+large+language+models+for+interpretable+logical+reasoning%2E+arXiv+preprint+arXiv%3A2205%2E09712%2C+2022%2E)
9. B. Dalvi, P. Jansen, O. Tafjord, Z. Xie, H. Smith, L. Pipatanangkura, and P. Clark. [Explaining answers with entailment trees](https://md.research.elamur.ai/free/en/papers/dalvi-2021-explaining-answers-entailment-trees/). In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 7358–7370, Online and Punta Cana, Dominican Republic, Nov. 2021. Association…
10. B. Dalvi, O. Tafjord, and P. Clark. Towards teachable reasoning systems. arXiv preprint arXiv:2204.13074, 2022. [find](https://scholar.google.com/scholar?q=B%2E+Dalvi%2C+O%2E+Tafjord%2C+and+P%2E+Clark%2E+Towards+teachable+reasoning+systems%2E+arXiv+preprint+arXiv%3A2204%2E13074%2C+2022%2E)
11. I. Dasgupta, A. K. Lampinen, S. C. Chan, A. Creswell, D. Kumaran, J. L. McClelland, and F. Hill. Language models show human-like content effects on reasoning. arXiv preprint arXiv:2207.[07051, 2022](https://md.research.elamur.ai/free/en/papers/dasgupta-2022-language-models-show-human-like/).
12. A. Graves. Generating sequences with recurrent neural networks. arXiv preprint arXiv:1308.0850, 2013. [find](https://scholar.google.com/scholar?q=A%2E+Graves%2E+Generating+sequences+with+recurrent+neural+networks%2E+arXiv+preprint+arXiv%3A1308%2E0850%2C+2013%2E)
13. A. Graves. Adaptive computation time for recurrent neural networks. arXiv preprint arXiv:1603.08983, 2016. [find](https://scholar.google.com/scholar?q=A%2E+Graves%2E+Adaptive+computation+time+for+recurrent+neural+networks%2E+arXiv+preprint+arXiv%3A1603%2E08983%2C+2016%2E)
14. T. Kojima, S. S. Gu, M. Reid, Y. Matsuo, and Y. Iwasawa. [Large language models are zero-shot reasoners](https://md.research.elamur.ai/free/en/papers/kojima-2022-large-language-models-zero-shot/). arXiv preprint arXiv:2205.11916, 2022.
15. S. Kumar and P. Talukdar. Nile: Natural language inference with faithful natural language explanations. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 8730[–8742, 2020](https://md.research.elamur.ai/free/en/papers/kumar-2020-nile-natural-language-inference-faithful/).
16. R. Nakano, J. Hilton, S. Balaji, J. Wu, L. Ouyang, C. Kim, C. Hesse, S. Jain, V. Kosaraju, W. Saunders, et al. [Webgpt: Browser-assisted question-answering with human feedback](https://md.research.elamur.ai/free/en/papers/nakano-2021-webgpt-browser-assisted-question-answering/). arXiv preprint arXiv:2112.09332, 2021.
17. V. Gupta, S. Zhang, A. Vempala, Y. He, T. Choji, and V. Srikumar. Right for the right reason: Evidence extraction for trustworthy tabular reasoning. In Proceedings of the 60th Annual M. Nye, A. J. Andreassen, G. Gur-Ari, H. Michalewski, J. Austin, D. Bieber, D. Dohan, Meeting of the Association for… [find](https://scholar.google.com/scholar?q=V%2E+Gupta%2C+S%2E+Zhang%2C+A%2E+Vempala%2C+Y%2E+He%2C+T%2E+Choji%2C+and+V%2E+Srikumar%2E+Right+for+the+right+reason%3A+Evidence+extraction+for+trustworthy+tabular+reasoning%2E+In+Proceedings+of+the+60th+Annual+M%2E+Nye%2C+A%2E+J%2E+And)
18. Casas, L. A. Hendricks, J. Welbl, A. Clark, et al. [Training compute-optimal large language models](https://md.research.elamur.ai/free/en/papers/hoffmann-2022-training-compute-optimal-large-language/). arXiv preprint arXiv:2203.15556, 2022.
19. H. Jhamtani and P. Clark. Learning to explain: Datasets and models for identifying valid reasoning chains in multihop question-answering. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 137–150, Online, Nov. 2020. Association for Computational… [find](https://scholar.google.com/scholar?q=H%2E+Jhamtani+and+P%2E+Clark%2E+Learning+to+explain%3A+Datasets+and+models+for+identifying+valid+reasoning+chains+in+multihop+question%2Danswering%2E+In+Proceedings+of+the+2020+Conference+on+Empirical+Methods+in+)
20. S. Kadavath, T. Conerly, A. Askell, T. Henighan, D. Drain, E. Perez, N. Schiefer, Z. H. Dodds, N. DasSarma, E. Tran-Johnson, et al. Language models (mostly) know what they know. arXiv preprint arXiv:2207.05221, 2022. [find](https://scholar.google.com/scholar?q=S%2E+Kadavath%2C+T%2E+Conerly%2C+A%2E+Askell%2C+T%2E+Henighan%2C+D%2E+Drain%2C+E%2E+Perez%2C+N%2E+Schiefer%2C+Z%2E+H%2E+Dodds%2C+N%2E+DasSarma%2C+E%2E+Tran%2DJohnson%2C+et+al%2E+Language+models+%28mostly%29+know+what+they+know%2E+arXiv+preprint+arXiv%3A2)
21. D. Ribeiro, S. Wang, X. Ma, R. Dong, X. Wei, H. Zhu, X. Chen, Z. Huang, P. Xu, A. Arnold, et al. Entailment tree explanations via iterative retrieval-generation reasoner. arXiv preprint arXiv:2205.09224, 2022. [find](https://scholar.google.com/scholar?q=D%2E+Ribeiro%2C+S%2E+Wang%2C+X%2E+Ma%2C+R%2E+Dong%2C+X%2E+Wei%2C+H%2E+Zhu%2C+X%2E+Chen%2C+Z%2E+Huang%2C+P%2E+Xu%2C+A%2E+Arnold%2C+et+al%2E+Entailment+tree+explanations+via+iterative+retrieval%2Dgeneration+reasoner%2E+arXiv+preprint+arXiv%3A2205%2E092)
22. S. Saha, S. Ghosh, S. Srivastava, and M. Bansal. [Prover: Proof generation for interpretable reasoning over rules](https://md.research.elamur.ai/free/en/papers/saha-2020-prover-proof-generation-interpretable-reasoning/). In EMNLP (1), 2020.
23. T. Sellam, D. Das, and A. Parikh. [Bleurt: Learning robust metrics for text generation](https://md.research.elamur.ai/free/en/papers/sellam-2020-bleurt-learning-robust-metrics-text/). In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7881–7892, 2020.
24. O. Tafjord, B. Dalvi, and P. Clark. [Proofwriter: Generating implications, proofs, and abductive statements over natural language](https://md.research.elamur.ai/free/en/papers/tafjord-2021-proofwriter-generating-implications-proofs-abductive/). In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 3621–3634, 2021.
25. J. Wei, X. Wang, D. Schuurmans, M. Bosma, E. Chi, Q. Le, and D. Zhou. Chain of thought prompting elicits reasoning in large language models, 2022. — [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://md.research.elamur.ai/free/en/papers/wei-2022-chain-thought-prompting-elicits-reasoning/)
26. Uesato, Jordi Grau-Moya, Ramana Kumar and Irina Higgins for insightful discussions. L. Weidinger, J. Mellor, M. Rauh, C. Griffin, J. Uesato, P. Huang, M. Cheng, M. Glaese, B. Balle, A. Kasirzadeh, Z. Kenton, S. Brown, W. Hawkins, T. Stepleton, C. Biles, A. Birhane, J. Haas, L. Rimell, L. A. Hendrick… — [Ethical and social risks of harm from Language Models](https://md.research.elamur.ai/free/en/papers/weidinger-2021-ethical-social-risks-harm-language/)
27. Z. Xie, S. Thiem, J. Martin, E. Wainwright, S. Marmorstein, and P. Jansen. Worldtree v2: A corpus of science-domain structured explanations and inference patterns supporting multi-hop inference. In Proceedings of the 12th Language Resources and Evaluation Conference, pages 5456– 5473, 2020. [find](https://scholar.google.com/scholar?q=Z%2E+Xie%2C+S%2E+Thiem%2C+J%2E+Martin%2C+E%2E+Wainwright%2C+S%2E+Marmorstein%2C+and+P%2E+Jansen%2E+Worldtree+v2%3A+A+corpus+of+science%2Ddomain+structured+explanations+and+inference+patterns+supporting+multi%2Dhop+inference%2E+In+Pr)
28. E. Zelikman, Y. Wu, and N. D. Goodman. [Star: Bootstrapping reasoning with reasoning](https://md.research.elamur.ai/free/en/papers/zelikman-2022-star-bootstrapping-reasoning-reasoning/). arXiv preprint arXiv:2203.14465, 2022.
29. A. Zeng, A. Wong, S. Welker, K. Choromanski, F. Tombari, A. Purohit, M. Ryoo, V. Sindhwani, J. Lee, V. Vanhoucke, et al. [Socratic models: Composing zero-shot multimodal reasoning with language](https://md.research.elamur.ai/free/en/papers/zeng-2022-socratic-models-composing-zero-shot/). arXiv preprint arXiv:2204.00598, 2022.
30. H. Zhang, L. H. Li, T. Meng, K.-W. Chang, and G. V. d. Broeck. On the paradox of learning to reason from data. arXiv preprint arXiv:2205.11502, 2022. — [arXiv:2205.11502](https://arxiv.org/abs/2205.11502) [find](https://scholar.google.com/scholar?q=H%2E+Zhang%2C+L%2E+H%2E+Li%2C+T%2E+Meng%2C+K%2E%2DW%2E+Chang%2C+and+G%2E+V%2E+d%2E+Broeck%2E+On+the+paradox+of+learning+to+reason+from+data%2E+arXiv+preprint+arXiv%3A2205%2E11502%2C+2022%2E+%E2%80%94+%5BarXiv%3A2205%2E11502%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2205%2E11)
31. D. Zhou, N. Schärli, L. Hou, J. Wei, N. Scales, X. Wang, D. Schuurmans, O. Bousquet, Q. Le, and E. Chi. [Least-to-most prompting enables complex reasoning in large language models](https://md.research.elamur.ai/free/en/papers/al-2022-least-most-prompting-enables-complex/). arXiv preprint arXiv:2205.10625, 2022. \#\# Acknowledgements The authors would like to thank Angeliki Lazaridou, Charles B…

*Updated 2 November 2026*