FaiRR: Faithful and Robust Deductive Reasoning over Natural Language
Soumya
Sanyal
author
Harman
Singh
author
Xiang
Ren
author
2022-05
text
Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
Association for Computational Linguistics
Dublin, Ireland
conference publication
Transformers have been shown to be able to perform deductive reasoning on a logical rulebase containing rules and statements written in natural language. Recent works show that such models can also produce the reasoning steps (i.e., the proof graph) that emulate the model’s logical reasoning process. Currently, these black-box models generate both the proof graph and intermediate inferences within the same model and thus may be unfaithful. In this work, we frame the deductive logical reasoning task by defining three modular components: rule selection, fact selection, and knowledge composition. The rule and fact selection steps select the candidate rule and facts to be used and then the knowledge composition combines them to generate new inferences. This ensures model faithfulness by assured causal relation from the proof step to the inference reasoning. To test our framework, we propose FaiRR (Faithful and Robust Reasoner) where the above three components are independently modeled by transformers. We observe that FaiRR is robust to novel language perturbations, and is faster at inference than previous works on existing reasoning datasets. Additionally, in contrast to black-box generative models, the errors made by FaiRR are more interpretable due to the modular approach.
sanyal-etal-2022-fairr
10.18653/v1/2022.acl-long.77
https://aclanthology.org/2022.acl-long.77
2022-05
1075
1093