Enhancing Self-Consistency and Performance of Pre-Trained Language Models through Natural Language Inference

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 1754 - 1768 December 7-11, 2022 ©2022 Association for Computational Linguistics Enhancing Self-Consistency and Performance of Pre-Trained Language Models through Natural Language Inference Eric Mitchell, Joseph J. Noh, Siyan Li, William S. Armstrong, Ananth Agarwal, Patrick Liu, Chelsea Finn, Christopher D. Manning Stanford University eric.mitchell@cs.stanford.edu Abstract While large pre-trained language models are powerful, their predictions often lack logical consistency across test inputs. For example, a state-of-the-art Macaw question-answering (QA) model answers Yes to Is a sparrow a bird? and Does a bird have feet? but answers No to Does a sparrow have feet?. To address this failure mode, we propose a framework, Consistency Correction through Relation De- tection, or ConCoRD, for boosting the consis- tency and accuracy of pre-trained NLP models using pre-trained natural language inference (NLI) models without fine-tuning or re-training. Given a batch of test inputs, ConCoRD sam- ples several candidate outputs for each input and instantiates a factor graph that accounts for both the model’s belief about the likeli- hood of each answer choice in isolation and the NLI model’s beliefs about pair-wise answer choice compatibility. We show that a weighted MaxSAT solver can efficiently compute high- quality answer choices under this factor graph, improving over the raw model’s predictions. Our experiments demonstrate that ConCoRD consistently boosts accuracy and consistency of off-the-shelf closed-book QA and VQA mod- els using off-the-shelf NLI models, notably in- creasing accuracy of LXMERT on ConVQA by 5% absolute. See the project website1 for code and data. 1 Introduction Reliable and trustworthy AI systems should demon- strate internal self-consistency, in the sense that their predictions across inputs should imply logi- cally compatible beliefs about the world. However, even powerful large language models are known to lack self-consistency (Ray et al., 2019; Elazar et al., 2021; Kassner et al., 2021). For example, a question-answering (QA) model that answers the question Is a sparrow a bird? and Does a bird have 1https://ericmitchell.ai/emnlp-2022-concord/ Figure 1: ConCoRD first generates candidate outputs from the base pre-trained model, then estimates soft pairwise con- straints between output choices, and finally finds the most satisfactory choices of answers accounting for both the base model and NLI model’s beliefs. feet? with Yes is implicitly expressing the belief that A sparrow is a bird and A bird has feet. If the same model answers the question Does a sparrow have feet? with No, the model expresses the logi- cally incompatible belief A sparrow does not have feet. In such cases, ascertaining the model’s “true” belief is difficult, making interpreting and validat- ing its behavior correspondingly challenging. Prior work has improved model self-consistency by training with specialized loss functions (Elazar et al., 2021) or data augmentation (Ray et al., 2019), or alternatively re-ranking model predictions based on their mutual self-consistency using pre-written logical constraints, such as “all mammals have fur” (Kassner et al., 2021). However, the first class of methods requires expensive fine-tuning which might be impractical for many practitioners for …

Content truncated to 500 words