StructBERT: Incorporating Language Structures into Pre-training for Deep Language Understanding

TL;DR

StructBERT enhances pre-trained language models by integrating language structures during pre-training, significantly improving performance across various natural language understanding tasks.

Contribution

It introduces a novel structural pre-training approach for BERT by adding auxiliary tasks that leverage word and sentence level language structures.

Findings

Achieves state-of-the-art on GLUE benchmark with 89.0 score

F1 score of 93.0 on SQuAD v1.1

91.7% accuracy on SNLI

Abstract

Recently, the pre-trained language model, BERT (and its robustly optimized version RoBERTa), has attracted a lot of attention in natural language understanding (NLU), and achieved state-of-the-art accuracy in various NLU tasks, such as sentiment classification, natural language inference, semantic textual similarity and question answering. Inspired by the linearization exploration work of Elman [8], we extend BERT to a new model, StructBERT, by incorporating language structures into pre-training. Specifically, we pre-train StructBERT with two auxiliary tasks to make the most of the sequential order of words and sentences, which leverage language structures at the word and sentence levels, respectively. As a result, the new model is adapted to different levels of language understanding required by downstream tasks. The StructBERT with structural pre-training gives surprisingly good empirical results on a variety of downstream tasks, including pushing the state-of-the-art on the GLUE benchmark to 89.0 (outperforming all published models), the F1 score on SQuAD v1.1 question answering to 93.0, the accuracy on SNLI to 91.7.