Assessing Phrasal Representation and Composition in Transformers
This addresses the need for interpretability in NLP for researchers and practitioners, but it is incremental as it builds on existing analysis methods.
The paper tackled the problem of understanding how transformer models represent phrases and whether they perform sophisticated composition like humans, finding that phrase representation relies heavily on word content with little evidence of nuanced composition.
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.