What do you learn from context? Probing for sentence structure in contextualized word representations

TL;DR

This paper introduces a new probing method to analyze how contextualized word representations encode sentence structure, revealing strengths in syntax but limited improvements in semantics.

Contribution

It presents a novel edge probing task design and evaluates multiple models, providing insights into their encoding of syntactic and semantic information.

Findings

Models excel at syntactic phenomena

Limited semantic understanding improvements

Probing reveals strengths and weaknesses in representations

Abstract

Contextualized representation models such as ELMo (Peters et al., 2018a) and BERT (Devlin et al., 2018) have recently achieved state-of-the-art results on a diverse array of downstream NLP tasks. Building on recent token-level probing work, we introduce a novel edge probing task design and construct a broad suite of sub-sentence tasks derived from the traditional structured NLP pipeline. We probe word-level contextual representations from four recent models and investigate how they encode sentence structure across a range of syntactic, semantic, local, and long-range phenomena. We find that existing models trained on language modeling and translation produce strong representations for syntactic phenomena, but only offer comparably small improvements on semantic tasks over a non-contextual baseline.