Linguistic Analysis of Pretrained Sentence Encoders with Acceptability Judgments
This work provides a detailed analysis of grammatical capabilities in sentence encoders for NLP researchers, but it is incremental as it builds on existing evaluation methods with a new dataset.
The authors tackled the problem of evaluating grammatical knowledge in pretrained sentence encoders by introducing a new analysis dataset with broad coverage of 13 syntactic phenomena, finding that models like BERT and GPT have strong command of complex structures like ditransitives and passives but struggle with long-distance dependencies like questions, though they outperform baselines.
Recent work on evaluating grammatical knowledge in pretrained sentence encoders gives a fine-grained view of a small number of phenomena. We introduce a new analysis dataset that also has broad coverage of linguistic phenomena. We annotate the development set of the Corpus of Linguistic Acceptability (CoLA; Warstadt et al., 2018) for the presence of 13 classes of syntactic phenomena including various forms of argument alternations, movement, and modification. We use this analysis set to investigate the grammatical knowledge of three pretrained encoders: BERT (Devlin et al., 2018), GPT (Radford et al., 2018), and the BiLSTM baseline from Warstadt et al. We find that these models have a strong command of complex or non-canonical argument structures like ditransitives (Sue gave Dan a book) and passives (The book was read). Sentences with long distance dependencies like questions (What do you think I ate?) challenge all models, but for these, BERT and GPT have a distinct advantage over the baseline. We conclude that recent sentence encoders, despite showing near-human performance on acceptability classification overall, still fail to make fine-grained grammaticality distinctions for many complex syntactic structures.