Pre-training by language modeling has become a popular and successful approach to NLP tasks, but we have yet to understand exactly what linguistic capacities these pretraining processes confer upon models. In this paper we introduce a suite of diagnostics drawn from human language experiments, which allow us to ask targeted questions about information used by language models for generating predictions in context. As a case study, we apply these diagnostics to the popular BERT model, finding that it can generally distinguish good from bad completions involving shared category or role reversal, albeit with less sensitivity than humans, and it robustly retrieves noun hypernyms, but it struggles with challenging inference and role-based event predictionand in particular, it shows clear insensitivity to the contextual impacts of negation.

We propose a diagnostic method for probing specific information captured in vector representations of sentence meaning, via simple classification tasks with strategically constructed sentence sets. We identify some key types of semantic information that we might expect to be captured in sentence composition, and illustrate example classification tasks for targeting this information.

Deep transformer models have pushed performance on NLP tasks to new limits, suggesting sophisticated treatment of complex linguistic inputs, such as phrases. However, we have limited understanding of how these models handle representation of phrases, and whether this reflects sophisticated composition of phrase meaning like that done by humans. In this paper, we present systematic analysis of phrasal representations in state-of-the-art pre-trained transformers. We use tests leveraging human judgments of phrase similarity and meaning shift, and compare results before and after control of word overlap, to tease apart lexical effects versus composition effects. We find that phrase representation in these models relies heavily on word content, with little evidence of nuanced composition. We also identify variations in phrase representation quality across models, layers, and representation types, and make corresponding recommendations for usage of representations from these models.

No abstract