In this work, we explore the ability of artificial neural networks to judge the grammatical acceptability of a sentence. Machine learning research of this kind is well placed to answer important open questions about the role of prior linguistic bias in language acquisition by providing a test for the Poverty of the Stimulus Argument. In service of this goal, we introduce the Corpus of Linguistic Acceptability (CoLA), a set of 10,657 English sentences labeled as grammatical or ungrammatical by expert linguists. We train several recurrent neural networks to do binary acceptability classification. These models set a baseline for the task. Error-analysis testing the models on specific grammatical phenomena reveals that they learn some systematic grammatical generalizations like subject-verb-object word order without any grammatical supervision. We find that neural sequence models show promise on the acceptability classification task. However, human-like performance across a wide range of grammatical constructions remains far off.
View on arXiv