The DOI matches its Crossref record; 40 of the 41 references it cites were found in a registry or an index.
Transformers as Soft Reasoners over Language
Open accessAbstract
Beginning with McCarthy's Advice Taker (1959), AI has pursued the goal of providing a system with explicit, general knowledge and having the system reason over that knowledge. However, expressing the knowledge in a formal (logical or probabilistic) representation has been a major obstacle to this research. This paper investigates a modern approach to this problem where the facts and rules are provided as natural language sentences, thus bypassing a formal representation. We train transformers to reason (or emulate reasoning) over these sentences using synthetically generated data. Our models, that we call RuleTakers, provide the first empirical demonstration that this kind of soft reasoning over language is learnable, can achieve high (99%) accuracy, and generalizes to test data requiring substantially deeper chaining than seen during training (95%+ scores). We also demonstrate that the models transfer well to two hand-authored rulebases, and to rulebases paraphrased into more natural language. These findings are significant as it suggests a new role for transformers, namely as limited "soft theorem provers" operating over explicit theories in language. This in turn suggests new possibilities for explainability, correctability, and counterfactual reasoning in question-answering.
Overview
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“This process is algorithmic; a model that has learned a robust implementation of this algorithm should, in principle, generalize beyond the shallow depths seen in previous benchmarks (Clark et al., 2020;Tafjord et al., 2021;Saparov and He, 2023).”
From Probing for Long-Horizon Deductive Reasoning Capabilities in Language Models with Prolog (2026) · cites this paper as Clark et al., 2020
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“The PARARULE-Plus dataset [24] is a large-scale synthetic dataset for multi-step deductive reasoning over natural language, extending the original PARARULE [32] with deeper inference chains.”
From PrologMCP: A Standardized Prolog Tool Interface for LLM Agents (2026) · cites this paper as [32]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Unlike synthetically templated logic datasets (Han et al., 2022;Clark et al., 2020) in which all premises are stated as short clauses, NLR problems have explicit statements along with implicit information that must be inferred from common sense and context.”
From Reliable Reasoning Beyond Natural Language (2024) · cites this paper as Clark et al., 2020
The cited source says what the claim attributes to it.
In the citing paper
“Navigate, like many other benchmark datasets such as RuleTaker (Clark et al., 2020), is systematically generated.”
From Reliable Reasoning Beyond Natural Language (2024) · cites this paper as Clark et al., 2020
The cited source says what the claim attributes to it.
In the citing paper
“These benchmarks revealed that models at the time struggle as proof depth increases, but the depths studied remain shallow.”
From Probing for Long-Horizon Deductive Reasoning Capabilities in Language Models with Prolog (2026) · cites this paper as Clark et al., 2020
The cited source says what the claim attributes to it.
In the citing paper
“Early work on deductive reasoning benchmarks such as RuleTaker (Clark et al., 2020) and ProofWriter (Tafjord et al., 2021) established the basic setup of query answering over sets of rules and facts, with ProofWriter reaching a maximum proof depth of five steps.ProntoQA (Saparov and He, 2023) introduces proof-structure controls and formal error attribution, also evaluated up to five hops.”
From Probing for Long-Horizon Deductive Reasoning Capabilities in Language Models with Prolog (2026) · cites this paper as Clark et al., 2020
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
RoBERTa: A Robustly Optimized BERT Pretraining Approach
Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., & Stoyanov, V. (2019) · arXiv · doi:10.48550/arxiv.1907.11692 · Cited by 16,515
SQuAD: 100,000+ Questions for Machine Comprehension of Text
Rajpurkar, P., Zhang, J., Lopyrev, K., & Liang, P. (2016) · Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d16-1264 · Cited by 6,208
ASH, R. B. (1966) · Cited by 3,990
NIPS 2016 Tutorial: Generative Adversarial NetworksPreprint
Goodfellow, I. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1701.00160 · Cited by 2,048
A Decomposable Attention Model for Natural Language Inference
Parikh, A. P., Täckström, O., Das, D., & Uszkoreit, J. (2016) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/d16-1244 · Cited by 1,351
Enhanced LSTM for Natural Language Inference
Chen, Q., Zhu, X., Ling, Z.-H., Wei, S., Jiang, H., & Inkpen, D. Z. (2017) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/p17-1152 · Cited by 1,315
Towards a Theory of Declarative Knowledge
Apt, K. R., Blair, H. A., & Walker, A. (1988) · Foundations of Deductive Databases and Logic Programming · doi:10.1016/b978-0-934613-40-8.50006-3 · Cited by 1,151
Applications of circumscription to formalizing common-sense knowledge
McCarthy, J. (1986) · Artificial Intelligence · doi:10.1016/0004-3702(86)90032-9 · Cited by 1,114
RACE: Large-scale ReAding Comprehension Dataset From Examinations
Lai, G., Xie, Q., Liu, H., Yang, Y., & Hovy, E. (2017) · Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d17-1082 · Cited by 1,044
The logic theory machine-A complex information processing system
Newell, A., & Simon, H. (1956) · IEEE Transactions on Information Theory · doi:10.1109/tit.1956.1056797 · Cited by 846
McCarthy, J. (1960) · Cited by 799
SQuAD: 100,000+ Questions for Machine Comprehension of Text
Rajpurkar, P., Zhang, J., Lopyrev, K., & Liang, P. (2016) · arXiv · doi:10.48550/arxiv.1606.05250 · Cited by 678
Learning Deep Transformer Models for Machine Translation
Wang, Q., Li, B., Xiao, T., Zhu, J., Li, C., Wong, D. F., & Chao, L. S. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1176 · Cited by 673
Deep Learning for Symbolic MathematicsPreprint
Lample, G., & Charton, F. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1912.01412 · Cited by 405
Recognizing Textual Entailment
Dagan, I., Roth, D., Sammons, M., & Zanzotto, F. M. (2013) · Synthesis lectures on human language technologies · doi:10.2200/s00509ed1v01y201305hlt023 · Cited by 283
Towards AI-Complete Question Answering: A Set of Prerequisite Toy Tasks
Weston, J., Bordes, A., Chopra, S., Rush, A. M., van Merriënboer, B., Joulin, A., & Mikolov, T. (2016) · International Conference on Learning Representations · Cited by 261
Smodels - An Implementation of the Stable Model and Well-Founded Semantics for Normal LP
Niemelä, I., & Simons, P. (1997) · International Conference on Logic Programming · Cited by 261
Manning, C. D., & MacCartney, B. (2009) · Cited by 209
Improving Machine Reading Comprehension with General Reading Strategies
Sun, K., Yu, D., Yu, D., & Cardie, C. (2019) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/n19-1270 · Cited by 134
A Survey on Semantic ParsingPreprint
Kamath, A., & Das, R. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1812.00978 · Cited by 117
Reasoning Over Paragraph Effects in Situations
Lin, K., Tafjord, O., Clark, P. E., & Gardner, M. (2019) · Workshop on Machine Reading for Question Answering (MRQA) · doi:10.18653/v1/d19-5808 · Cited by 110
Self-citationCLUTRR: A Diagnostic Benchmark for Inductive Reasoning from Text
Sinha, K., Sodhani, S., Dong, J., Pineau, J., & Hamilton, W. L. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1458 · Cited by 110
Expert systems in production planning and scheduling: A state-of-the-art survey
Metaxiotis, K. S., Askounis, D., & Psarras, J. E. (2002) · Journal of Intelligent Manufacturing · doi:10.1023/a:1016064126976 · Cited by 98
NLProlog: Reasoning with Weak Unification for Question Answering in Natural Language
Weber, L., Minervini, P., Münchmeyer, J., Leser, U., & Rocktäschel, T. (2019) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/p19-1618 · Cited by 97
HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering
Yang, Z., Qi, P., Zhang, S., Bengio, Y., Cohen, W. W., Salakhutdinov, R., & Manning, C. D. (2018) · arXiv · doi:10.48550/arxiv.1809.09600 · Cited by 97
General logical databases and programs: Default logic semantics and stratification
Bidoít, N., & Froidevaux, C. (1991) · Information and Computation · doi:10.1016/0890-5401(91)90073-b · Cited by 89
QuaRTz: An Open-Domain Dataset of Qualitative Relationship Questions
Tafjord, O., Gardner, M., Lin, K., & Clark, P. E. (2019) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/d19-1608 · Cited by 72
Self-citationJansen, P. A., Wainwright, E., Marmorstein, S., & Morrison, C. T. (2018) · Language Resources and Evaluation Conference (LREC) · doi:10.63317/28283yn8kzpn · Cited by 56
He, H. W., & Choi, J. D. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1908.04943 · Cited by 37
Bunt, H., Bos, J., & Pulman, S. (1999) · Cited by 33
Probing Natural Language Inference Models through Semantic Fragments
Richardson, K., Hu, H., Moss, L. S., & Sabharwal, A. (2020) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v34i05.6397 · Cited by 24
Self-citationNatural Logic and Natural Language Inference
MacCartney, B., & Manning, C. D. (2013) · Text, speech and language technology · doi:10.1007/978-94-007-7284-7_8 · Cited by 24
Towards Neural Theorem Proving at ScalePreprint
Minervini, P., Bošnjak, M., Rocktäschel, T., & Riedel, S. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1807.08204 · Cited by 22
Improving Graph Neural Network Representations of Logical Formulae with Subgraph PoolingPreprint
Crouse, M., Abdelaziz, I., Cornelio, C., Thost, V., Wu, L., Forbus, K. D., & Fokoue, A. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1911.06904 · Cited by 22
Learning to Reason: Leveraging Neural Networks for Approximate DNF Counting
Abboud, R., Ceylan, İ. İ., & Lukasiewicz, T. (2020) · AAAI Publications (The Association for the Advancement of Artificial Intelligence (AAAI)) · doi:10.1609/aaai.v34i04.5705 · Cited by 22
Moss, L. S. (2009) · Lecture notes in computer science · doi:10.1007/978-3-642-14287-1_9 · Cited by 17
Musen, M. A., & van der Lei, J. (1988) · Machine intelligence and pattern recognition · doi:10.1016/b978-0-444-87137-4.50029-1 · Cited by 13
Differentiable Reasoning on Large Knowledge Bases and Natural Language
Minervini, P., Bošnjak, M., Rocktäschel, T., Riedel, S., & Grefenstette, E. (2020) · Studies on the semantic web · doi:10.3233/ssw200015 · Cited by 12
ADAPTIVE GENERATION OF PROGRAMMING PUZZLES
Kalyan, A., Polozov, O., & Kalai, A. T. (2019) · Cited by 1
Sentence Encoders on STILTs: Supplementary Training on Intermediate Labeled-data Tasks
Phang, J., Févry, T., & Bowman, S. R. (2018) · arXiv · doi:10.48550/arxiv.1811.01088
Research works printed but not linked: 5
- [Abdelaziz et al. , 2020] Ibrahim Abdelaziz, Veronika Thost, Maxwell Crouse, and Achille Fokoue. An experimental study of formula embeddings for automated theorem proving in first-order logic. arXiv , 2002.00423, 2020. · arXiv
- [Richardson and Sabharwal, 2019] Kyle Richardson and Ashish Sabharwal. What does my qa model know? devising controlled probes using expert knowledge. ArXiv , abs/1912.13337, 2019. · arXiv
- [Saxton et al. , 2019] David Saxton, Edward Grefenstette, Felix Hill, and Pushmeet Kohli. Analysing mathematical reasoning abilities of neural models. In ICLR , 2019.
- [Selsam et al. , 2019] Daniel Selsam, Matthew Lamm, Benedikt B¨ unz, Percy Liang, Leonardo de Moura, and David L. Dill. Learning a SAT solver from single-bit supervision. In ICLR , 2019.
- [Talmor et al. , 2019] A. Talmor, Y. Elazar, Y. Goldberg, and J. Berant. oLMpics - on what language model pre-training captures. ArXiv , abs/1912.13283, 2019. · arXiv
Cited by
Sorted by citations
Chain-Of-Thought Prompting Elicits Reasoning in Large Language Models
Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q. V., & Zhou, D. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-1800 · Cited by 4,505
“Another line of work aims to enable language models to perform logical or formal reasoning, often by verablizing the rules in natural language formal rules using language (Clark et al., 2020; Saeed et al., 2021; Liang et al., 2021).”
Text Data Augmentation for Deep Learning
Shorten, C., Khoshgoftaar, T. M., & Furht, B. (2021) · Journal Of Big Data · doi:10.1186/s40537-021-00492-0 · Cited by 1,852
Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models
Srivastava, A., Rastogi, A., Rao, A., Shoeb, A. A. M., Abid, A., Fisch, A., Brown, A. R., Santoro, A., Gupta, A., Garriga-Alonso, A., et al. (2022) · arXiv · doi:10.48550/arxiv.2206.04615 · Cited by 569
Self-citationA comprehensive survey on applications of transformers for deep learning tasks
Islam, S., Elmekki, H., Elsebai, A., Bentahar, J., Drawel, N., Rjoub, G., & Witold Pedrycz, C. (2023) · Expert Systems with Applications · doi:10.1016/j.eswa.2023.122666 · Cited by 590
Deep Learning applications for COVID-19
Shorten, C., Khoshgoftaar, T. M., & Furht, B. (2021) · Journal of Big Data · doi:10.1186/s40537-020-00392-9 · Cited by 350
Metadata
- Authors
- Clark, P., Tafjord, O., & Richardson, K.
- Affiliations
- Allen Institute for Artificial Intelligence (US)
- Venue
- Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence
- Published
- 2020-07
- Pages
- 3882-3890
- OpenAlex
- W3006188107
- Cited by
- 176
- References
- 41
- Citations by year
- 2026: 8 · 2025: 19 · 2024: 15 · 2023: 41 · 2022: 37 · 2021: 37 · 2020: 18
- FWCI
- 12.69
- Type
- conference paper
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- International Joint Conference on Artificial Intelligence (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Logic, Reasoning, and Knowledge
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Keywords
- natural language reasoning, transformer models, synthetic training data, rule-based reasoning, knowledge representation, theorem proving, question answering, explainability, counterfactual reasoning, knowledge distillation
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
