Syntactic Data Augmentation Increases Robustness to Inference Heuristics

Abstract

Pretrained neural models such as BERT, when fine-tuned to perform natural language inference (NLI), often show high accuracy on standard datasets, but display a surprising lack of sensitivity to word order on controlled challenge sets. We hypothesize that this issue is not primarily caused by the pretrained model's limitations, but rather by the paucity of crowdsourced NLI examples that might convey the importance of syntactic structure at the fine-tuning stage. We explore several methods to augment standard training sets with syntactically informative examples, generated by applying syntactic transformations to sentences from the MNLI corpus. The best-performing augmentation method, subject/object inversion, improved BERT's accuracy on controlled examples that diagnose sensitivity to word order from 0.28 to 0.73, without affecting performance on the MNLI test set. This improvement generalized beyond the particular construction used for data augmentation, suggesting that augmentation causes BERT to recruit abstract syntactic representations.

ACL 2020

References in corpus (7)

• Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
• The Effectiveness of Data Augmentation in Image Classification using Deep Learning
• Assessing BERT's Syntactic Abilities
• Learning and Evaluating General Linguistic Intelligence
• Adversarial NLI: A New Benchmark for Natural Language Understanding
• Improving Natural Language Inference with a Pretrained Parser
• HUBERT Untangles BERT to Improve Transfer across NLP Tasks

Cited by in corpus (1)

• An unsupervised learning approach to evaluate questionnaire data -- what one can learn from violations of measurement invariance