Does Pre-training Induce Systematic Inference? How Masked Language Models Acquire Commonsense Knowledge

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4550 - 4557 July 10-15, 2022 ©2022 Association for Computational Linguistics Does Pre-training Induce Systematic Inference? How Masked Language Models Acquire Commonsense Knowledge Ian Porada∗ Mila, McGill University ian.porada@mail.mcgill.ca Alessandro Sordoni Microsoft Research Montréal alsordon@microsoft.com Jackie Chi Kit Cheung Mila, McGill University jcheung@cs.mcgill.ca Abstract Transformer models pre-trained with a masked- language-modeling objective (e.g., BERT) en- code commonsense knowledge as evidenced by behavioral probes; however, the extent to which this knowledge is acquired by system- atic inference over the semantics of the pre- training corpora is an open question. To an- swer this question, we selectively inject ver- balized knowledge into the pre-training mini- batches of BERT and evaluate how well the model generalizes to supported inferences af- ter pre-training on the injected knowledge. We find generalization does not improve over the course of pre-training BERT from scratch, sug- gesting that commonsense knowledge is ac- quired from surface-level, co-occurrence pat- terns rather than induced, systematic reasoning. 1 Introduction Pre-trained Transformers, such as BERT, encode knowledge about the world (Petroni et al., 2019; Zhou et al., 2020); e.g., BERT assigns relatively high probability to “fly” appearing in the context “robins can .” In this work, we investigate whether such knowledge is acquired during pre- training through systematic inference over the se- mantics of the pre-training corpora; e.g., can mod- els systematically infer “robins can fly” from the premises “birds can fly” and “robins are birds?” Resolving how models acquire commonsense knowledge has important implications. If models learn to make systematic inferences through pre- training, then scaling up pre-training is a promising direction for commonsense knowledge acquisition. If, instead, models only ever generalize based on superficial, surface-level patterns, then the majority of commonsense knowledge, which is only sup- ported implicitly, will never be acquired (Gordon and Van Durme, 2013; Forbes and Choi, 2017). ∗Work conducted while the author was an intern at Mi- crosoft Research Montréal. On the one hand, there is cursory evidence that pre-training might induce the ability to systemat- ically reason about the world. When fine-tuned on supervised training sets, pre-trained models can classify valid inferences better than strong base- lines (Clark et al., 2020; Talmor et al., 2020b); and, in zero-shot evaluations, pre-trained models perform relatively well on reasoning tasks that may require systematic reasoning, such as number comparison (Talmor et al., 2020a) and Winograd schemas (Sakaguchi et al., 2021). On the other hand, existing works have argued that pre-training does not generalize by systematic inference over semantics on the basis of theoreti- cal or synthetic results (Bender and Koller, 2020; Merrill et al., 2021; Traylor et al., 2021). Referring to physical commonsense knowledge acquired by BERT, Forbes et al. (2019) conclude that “neural language representations still only learn associa- tions that are explicitly written down.” Our main contribution is a direct evaluation of the training dynamics of BERT’s reasoning ability. We inject verbalized knowledge, such as “robins are …

Content truncated to 500 words