On the Robustness of Language Encoders against Grammatical Errors

Abstract

We conduct a thorough study to diagnose the behaviors of pre-trained language encoders (ELMo, BERT, and RoBERTa) when confronted with natural grammatical errors. Specifically, we collect real grammatical errors from non-native speakers and conduct adversarial attacks to simulate these errors on clean text data. We use this approach to facilitate debugging models on downstream applications. Results confirm that the performance of all tested models is affected but the degree of impact varies. To interpret model behaviors, we further design a linguistic acceptability task to reveal their abilities in identifying ungrammatical sentences and the position of errors. We find that fixed contextual encoders with a simple classifier trained on the prediction of sentence correctness are able to locate error positions. We also design a cloze test for BERT and discover that BERT captures the interaction between errors and specific tokens in context. Our results shed light on understanding the robustness and behaviors of language encoders against grammatical errors.

On the Adversarial Robustness of Vision Transformers

“…Transformer [35] architecture has achieved remarkable performance on many important Natural Language Processing (NLP) tasks, so the robustness of transformer has been studied on those NLP tasks. [17,19,29,22,12,43] conducted adversarial attacks on transformers including pretrained models, and in their experiments transformers usually show better robustness compared to models with structures such as LSTM or CNN, with a theoretical explanation provided in [17]. However, due to the discrete nature of NLP models, these studies are focusing on discrete perturbations (e.g., word or character substitutions) which are very different from small and continuous perturbations in computer vision tasks.…”